Lecture Notes On Classical Mechanics

Lecture Notes on Classical Mechanics
Version 0.1 – Not for Distribution!
Daniel Arovas
March 9, 2009
Contents
0.1 Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xvi
0.2 Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xvii
1 Introduction to Dynamics 1
1.1 Introduction and Review . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1.1 Newton’s laws of motion . . . . . . . . . . . . . . . . . . . . . . . . 2
1.1.2 Aside : inertial vs. gravitational mass . . . . . . . . . . . . . . . . . 3
1.2 Examples of Motion in One Dimension . . . . . . . . . . . . . . . . . . . . 4
1.2.1 Uniform force . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.2.2 Uniform force with linear frictional damping . . . . . . . . . . . . . 5
1.2.3 Uniform force with quadratic frictional damping . . . . . . . . . . . 6
1.2.4 Crossed electric and magnetic fields . . . . . . . . . . . . . . . . . . 7
1.3 Pause for Reflection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2 Systems of Particles 11
2.1 Work-Energy Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.2 Conservative and Nonconservative Forces . . . . . . . . . . . . . . . . . . . 12
2.2.1 Example : integrating F = −∇U . . . . . . . . . . . . . . . . . . . 14
2.3 Conservative Forces in Many Particle Systems . . . . . . . . . . . . . . . . 16
2.4 Linear and Angular Momentum . . . . . . . . . . . . . . . . . . . . . . . . 16
2.5 Scaling of Solutions for Homogeneous Potentials . . . . . . . . . . . . . . . 18
i
ii CONTENTS
2.5.1 Euler’s theorem for homogeneous functions . . . . . . . . . . . . . . 18
2.5.2 Scaled equations of motion . . . . . . . . . . . . . . . . . . . . . . . 18
2.6 Appendix I : Curvilinear Orthogonal Coordinates . . . . . . . . . . . . . . 20
2.6.1 Example : spherical coordinates . . . . . . . . . . . . . . . . . . . . 21
2.6.2 Vector calculus : grad, div, curl . . . . . . . . . . . . . . . . . . . . 21
2.6.3 Common curvilinear orthogonal systems . . . . . . . . . . . . . . . 23
3 One-Dimensional Conservative Systems 27
3.1 Description as a Dynamical System . . . . . . . . . . . . . . . . . . . . . . 27
3.1.1 Example : harmonic oscillator . . . . . . . . . . . . . . . . . . . . . 28
3.2 One-Dimensional Mechanics as a Dynamical System . . . . . . . . . . . . . 29
3.2.1 Sketching phase curves . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.3 Fixed Points and their Vicinity . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.3.1 Linearized dynamics in the vicinity of a fixed point . . . . . . . . . 31
3.4 Examples of Conservative One-Dimensional Systems . . . . . . . . . . . . . 33
3.4.1 Harmonic oscillator . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.4.2 Pendulum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.4.3 Other potentials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
4 Linear Oscillations 41
4.1 Damped Harmonic Oscillator . . . . . . . . . . . . . . . . . . . . . . . . . . 41
4.1.1 Classification of damped harmonic motion . . . . . . . . . . . . . . 42
4.1.2 Remarks on the case of critical damping . . . . . . . . . . . . . . . 44
4.2 Damped Harmonic Oscillator with Forcing . . . . . . . . . . . . . . . . . . 44
4.2.1 Resonant forcing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
4.2.2 R-L-C circuits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
4.2.3 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
CONTENTS iii
4.3 General solution by Green’s function method . . . . . . . . . . . . . . . . . 52
4.4 General Linear Autonomous Inhomogeneous ODEs . . . . . . . . . . . . . . 53
4.5 Kramers-Krönig Relations (advanced material) . . . . . . . . . . . . . . . . 57
5 Calculus of Variations 59
5.1 Snell’s Law . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
5.2 Functions and Functionals . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
5.3 Examples from the Calculus of Variations . . . . . . . . . . . . . . . . . . . 63
5.3.1 Example 1 : minimal surface of revolution . . . . . . . . . . . . . . 63
5.3.2 Example 2 : geodesic on a surface of revolution . . . . . . . . . . . 66
5.3.3 Example 3 : brachistochrone . . . . . . . . . . . . . . . . . . . . . . 67
5.3.4 Ocean waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
5.4 Appendix : More on Functionals . . . . . . . . . . . . . . . . . . . . . . . . 70
6 Lagrangian Mechanics 77
6.1 Generalized Coordinates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
6.2 Hamilton’s Principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
6.2.1 Momentum conservation . . . . . . . . . . . . . . . . . . . . . . . . 78
6.2.2 Invariance of the equations of motion . . . . . . . . . . . . . . . . . 78
6.2.3 Remarks on the order of the equations of motion . . . . . . . . . . 79
6.2.4 Lagrangian for a free particle . . . . . . . . . . . . . . . . . . . . . . 79
6.3 Remarks on the Choice of Generalized Coordinates . . . . . . . . . . . . . . 80
6.4 How to Solve Mechanics Problems . . . . . . . . . . . . . . . . . . . . . . . 81
6.5 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
6.5.1 One-dimensional motion . . . . . . . . . . . . . . . . . . . . . . . . 81
6.5.2 Central force in two dimensions . . . . . . . . . . . . . . . . . . . . 82
6.5.3 A sliding point mass on a sliding wedge . . . . . . . . . . . . . . . . 83

iv CONTENTS
6.5.4 A pendulum attached to a mass on a spring . . . . . . . . . . . . . 84
6.5.5 The double pendulum . . . . . . . . . . . . . . . . . . . . . . . . . . 86
6.5.6 The thingy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
6.6 Conserved Quantities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
6.6.1 Momentum conservation . . . . . . . . . . . . . . . . . . . . . . . . 90
6.6.2 Energy conservation . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
6.7 Appendix : Virial Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
7 Noether’s Theorem 95
7.1 Continuous Symmetry Implies Conserved Charges . . . . . . . . . . . . . . 95
7.1.1 Examples of one-parameter families of transformations . . . . . . . 96
7.2 Conservation of Linear and Angular Momentum . . . . . . . . . . . . . . . 97
7.3 Advanced Discussion : Invariance of L vs. Invariance of S . . . . . . . . . . 98
7.3.1 The Hamiltonian . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
7.3.2 Is H = T + U ? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
7.3.3 Example: A bead on a rotating hoop . . . . . . . . . . . . . . . . . 102
7.4 Charged Particle in a Magnetic Field . . . . . . . . . . . . . . . . . . . . . 104
7.5 Fast Perturbations : Rapidly Oscillating Fields . . . . . . . . . . . . . . . . 106
7.5.1 Example : pendulum with oscillating support . . . . . . . . . . . . 108
7.6 Field Theory: Systems with Several Independent Variables . . . . . . . . . 109
7.6.1 Gross-Pitaevskii model . . . . . . . . . . . . . . . . . . . . . . . . . 112
8 Constraints 115
8.1 Constraints and Variational Calculus . . . . . . . . . . . . . . . . . . . . . . 115
8.2 Constrained Extremization of Functions . . . . . . . . . . . . . . . . . . . . 117
8.3 Extremization of Functionals : Integral Constraints . . . . . . . . . . . . . 117
8.4 Extremization of Functionals : Holonomic Constraints . . . . . . . . . . . . 118

CONTENTS v
8.4.1 Examples of extremization with constraints . . . . . . . . . . . . . . 119
8.5 Application to Mechanics . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
8.5.1 Constraints and conservation laws . . . . . . . . . . . . . . . . . . . 122
8.6 Worked Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
8.6.1 One cylinder rolling off another . . . . . . . . . . . . . . . . . . . . 123
8.6.2 Frictionless motion along a curve . . . . . . . . . . . . . . . . . . . 125
8.6.3 Disk rolling down an inclined plane . . . . . . . . . . . . . . . . . . 128
8.6.4 Pendulum with nonrigid support . . . . . . . . . . . . . . . . . . . . 129
8.6.5 Falling ladder . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
8.6.6 Point mass inside rolling hoop . . . . . . . . . . . . . . . . . . . . . 135
9 Central Forces and Orbital Mechanics 141
9.1 Reduction to a one-body problem . . . . . . . . . . . . . . . . . . . . . . . 141
9.1.1 Center-of-mass (CM) and relative coordinates . . . . . . . . . . . . 141
9.1.2 Solution to the CM problem . . . . . . . . . . . . . . . . . . . . . . 142
9.1.3 Solution to the relative coordinate problem . . . . . . . . . . . . . . 142
9.2 Almost Circular Orbits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
9.3 Precession in a Soluble Model . . . . . . . . . . . . . . . . . . . . . . . . . . 146
9.4 The Kepler Problem: U (r) = −k r−1 . . . . . . . . . . . . . . . . . . . . . . 148
9.4.1 Geometric shape of orbits . . . . . . . . . . . . . . . . . . . . . . . 148
9.4.2 Laplace-Runge-Lenz vector . . . . . . . . . . . . . . . . . . . . . . . 148
9.4.3 Kepler orbits are conic sections . . . . . . . . . . . . . . . . . . . . 149
9.4.4 Period of bound Kepler orbits . . . . . . . . . . . . . . . . . . . . . 152
9.4.5 Escape velocity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
9.4.6 Satellites and spacecraft . . . . . . . . . . . . . . . . . . . . . . . . 153
9.4.7 Two examples of orbital mechanics . . . . . . . . . . . . . . . . . . 154
9.5 Appendix I : Mission to Neptune . . . . . . . . . . . . . . . . . . . . . . . . 156

vi CONTENTS
9.5.1 I. Earth to Jupiter . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
9.5.2 II. Encounter with Jupiter . . . . . . . . . . . . . . . . . . . . . . . 160
9.5.3 III. Jupiter to Neptune . . . . . . . . . . . . . . . . . . . . . . . . . 162
9.6 Appendix II : Restricted Three-Body Problem . . . . . . . . . . . . . . . . 163
10 Small Oscillations 171
10.1 Coupled Coordinates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171
10.2 Expansion about Static Equilibrium . . . . . . . . . . . . . . . . . . . . . . 172
10.3 Method of Small Oscillations . . . . . . . . . . . . . . . . . . . . . . . . . . 172
10.3.1 Can you really just choose an A so that both these wonderful things
happen in 10.13 and 10.14? . . . . . . . . . . . . . . . . . . . . . . . 173
10.3.2 Er...care to elaborate? . . . . . . . . . . . . . . . . . . . . . . . . . 173
10.3.3 Finding the modal matrix . . . . . . . . . . . . . . . . . . . . . . . 174
10.4 Example: Masses and Springs . . . . . . . . . . . . . . . . . . . . . . . . . . 176
10.5 Example: Double Pendulum . . . . . . . . . . . . . . . . . . . . . . . . . . 178
10.6 Zero Modes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
10.6.1 Example of zero mode oscillations . . . . . . . . . . . . . . . . . . . 179
10.7 Chain of Mass Points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182
10.7.1 Continuum limit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185
10.8 Appendix I : General Formulation . . . . . . . . . . . . . . . . . . . . . . . 186
10.9 Appendix II : Additional Examples . . . . . . . . . . . . . . . . . . . . . . 188
10.9.1 Right Triatomic Molecule . . . . . . . . . . . . . . . . . . . . . . . . 188
10.9.2 Triple Pendulum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 190
10.9.3 Equilateral Linear Triatomic Molecule . . . . . . . . . . . . . . . . 193
10.10 Aside : Christoffel Symbols . . . . . . . . . . . . . . . . . . . . . . . . . . . 197
11 Elastic Collisions 199
11.1 Center of Mass Frame . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199

CONTENTS vii
11.2 Central Force Scattering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203
11.2.1 Hard sphere scattering . . . . . . . . . . . . . . . . . . . . . . . . . 205
11.2.2 Rutherford scattering . . . . . . . . . . . . . . . . . . . . . . . . . . 206
11.2.3 Transformation to laboratory coordinates . . . . . . . . . . . . . . . 206
12 Noninertial Reference Frames 209
12.1 Accelerated Coordinate Systems . . . . . . . . . . . . . . . . . . . . . . . . 209
12.1.1 Translations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211
12.1.2 Motion on the surface of the earth . . . . . . . . . . . . . . . . . . . 211
12.2 Spherical Polar Coordinates . . . . . . . . . . . . . . . . . . . . . . . . . . . 213
12.3 Centrifugal Force . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214
12.3.1 Rotating tube of fluid . . . . . . . . . . . . . . . . . . . . . . . . . . 214
12.4 The Coriolis Force . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215
12.4.1 Foucault’s pendulum . . . . . . . . . . . . . . . . . . . . . . . . . . 218
13 Rigid Body Motion and Rotational Dynamics 221
13.1 Rigid Bodies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221
13.1.1 Examples of rigid bodies . . . . . . . . . . . . . . . . . . . . . . . . 221
13.2 The Inertia Tensor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 222
13.2.1 Coordinate transformations . . . . . . . . . . . . . . . . . . . . . . 224
13.2.2 The case of no fixed point . . . . . . . . . . . . . . . . . . . . . . . 224
13.3 Parallel Axis Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225
13.3.1 Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 226
13.3.2 General planar mass distribution . . . . . . . . . . . . . . . . . . . 227
13.4 Principal Axes of Inertia . . . . . . . . . . . . . . . . . . . . . . . . . . . . 228
13.5 Euler’s Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 229
13.5.1 Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 232

viii CONTENTS
13.6 Euler’s Angles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 233
13.6.1 Torque-free symmetric top . . . . . . . . . . . . . . . . . . . . . . . 235
13.6.2 Symmetric top with one point fixed . . . . . . . . . . . . . . . . . . 236
13.7 Rolling and Skidding Motion of Real Tops . . . . . . . . . . . . . . . . . . . 239
13.7.1 Rolling tops . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 239
13.7.2 Skidding tops . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 241
13.7.3 Tippie-top . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 242
14 Continuum Mechanics 245
14.1 Strings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 245
14.2 d’Alembert’s Solution to the Wave Equation . . . . . . . . . . . . . . . . . 247
14.2.1 Energy density and energy current . . . . . . . . . . . . . . . . . . 248
14.2.2 Reflection at an interface . . . . . . . . . . . . . . . . . . . . . . . . 249
14.2.3 Mass point on a string . . . . . . . . . . . . . . . . . . . . . . . . . 250
14.2.4 Interface between strings of different mass density . . . . . . . . . . 253
14.3 Finite Strings : Bernoulli’s Solution . . . . . . . . . . . . . . . . . . . . . . 255
14.4 Sturm-Liouville Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 257
14.4.1 Variational method . . . . . . . . . . . . . . . . . . . . . . . . . . . 259
14.5 Continua in Higher Dimensions . . . . . . . . . . . . . . . . . . . . . . . . . 261
14.5.1 Membranes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 262
14.5.2 Helmholtz equation . . . . . . . . . . . . . . . . . . . . . . . . . . . 263
14.5.3 Rectangles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 264
14.5.4 Circles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 265
14.5.5 Sound in fluids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 266
14.6 Dispersion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 268
14.7 Appendix I : Three Strings . . . . . . . . . . . . . . . . . . . . . . . . . . . 270
14.8 Appendix II : General Field Theoretic Formulation . . . . . . . . . . . . . . 273

CONTENTS ix
14.8.1 Euler-Lagrange equations for classical field theories . . . . . . . . . 273
14.8.2 Conserved currents in field theory . . . . . . . . . . . . . . . . . . . 274
14.8.3 Gross-Pitaevskii model . . . . . . . . . . . . . . . . . . . . . . . . . 275
14.9 Appendix III : Green’s Functions . . . . . . . . . . . . . . . . . . . . . . . . 277
14.9.1 Perturbation theory . . . . . . . . . . . . . . . . . . . . . . . . . . . 279
14.9.2 Perturbation theory for eigenvalues and eigenfunctions . . . . . . . 282
15 Special Relativity 285
15.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 285
15.1.1 Michelson-Morley experiment . . . . . . . . . . . . . . . . . . . . . 285
15.1.2 Einsteinian and Galilean relativity . . . . . . . . . . . . . . . . . . . 289
15.2 Intervals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 290
15.2.1 Proper time . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 292
15.2.2 Irreverent problem from Spring 2002 final exam . . . . . . . . . . . 293
15.3 Four-Vectors and Lorentz Transformations . . . . . . . . . . . . . . . . . . 294
15.3.1 Covariance and contravariance . . . . . . . . . . . . . . . . . . . . . 298
15.3.2 What to do if you hate raised and lowered indices . . . . . . . . . . 299
15.3.3 Comparing frames . . . . . . . . . . . . . . . . . . . . . . . . . . . . 300
15.3.4 Example I . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 300
15.3.5 Example II . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 301
15.3.6 Deformation of a rectangular plate . . . . . . . . . . . . . . . . . . 302
15.3.7 Transformation of velocities . . . . . . . . . . . . . . . . . . . . . . 303
15.3.8 Four-velocity and four-acceleration . . . . . . . . . . . . . . . . . . 304
15.4 Three Kinds of Relativistic Rockets . . . . . . . . . . . . . . . . . . . . . . 305
15.4.1 Constant acceleration model . . . . . . . . . . . . . . . . . . . . . . 305
15.4.2 Constant force with decreasing mass . . . . . . . . . . . . . . . . . 306
15.4.3 Constant ejecta velocity . . . . . . . . . . . . . . . . . . . . . . . . 307

x CONTENTS
15.5 Relativistic Mechanics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 308
15.5.1 Relativistic harmonic oscillator . . . . . . . . . . . . . . . . . . . . 310
15.5.2 Energy-momentum 4-vector . . . . . . . . . . . . . . . . . . . . . . 310
15.5.3 4-momentum for massless particles . . . . . . . . . . . . . . . . . . 312
15.6 Relativistic Doppler Effect . . . . . . . . . . . . . . . . . . . . . . . . . . . 312
15.6.1 Romantic example . . . . . . . . . . . . . . . . . . . . . . . . . . . . 313
15.7 Relativistic Kinematics of Particle Collisions . . . . . . . . . . . . . . . . . 315
15.7.1 Spontaneous particle decay into two products . . . . . . . . . . . . 316
15.7.2 Miscellaneous examples of particle decays . . . . . . . . . . . . . . . 318
15.7.3 Threshold particle production with a stationary target . . . . . . . 318
15.7.4 Transformation between frames . . . . . . . . . . . . . . . . . . . . 319
15.7.5 Compton scattering . . . . . . . . . . . . . . . . . . . . . . . . . . . 321
15.8 Covariant Electrodynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . 322
15.8.1 Lorentz force law . . . . . . . . . . . . . . . . . . . . . . . . . . . . 324
15.8.2 Gauge invariance . . . . . . . . . . . . . . . . . . . . . . . . . . . . 325
15.8.3 Transformations of fields . . . . . . . . . . . . . . . . . . . . . . . . 326
15.8.4 Invariance versus covariance . . . . . . . . . . . . . . . . . . . . . . 328
15.9 Appendix I : The Pole, the Barn, and Rashoman . . . . . . . . . . . . . . . 329
15.10 Appendix II : Photographing a Moving Pole . . . . . . . . . . . . . . . . . 331
16 Dynamical Systems 333
16.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 333
16.1.1 Phase space and phase curves . . . . . . . . . . . . . . . . . . . . . 333
16.1.2 Vector fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 333
16.1.3 Existence / uniqueness / extension theorems . . . . . . . . . . . . . 334
16.1.4 Linear differential equations . . . . . . . . . . . . . . . . . . . . . . 335
16.1.5 Lyapunov functions . . . . . . . . . . . . . . . . . . . . . . . . . . . 335

CONTENTS xi
16.2 N = 1 Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 336
16.2.1 Classification of fixed points (N = 1) . . . . . . . . . . . . . . . . . 337
16.2.2 Logistic equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 338
16.2.3 Singular f (u) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 339
16.2.4 Recommended exercises . . . . . . . . . . . . . . . . . . . . . . . . . 340
16.2.5 Non-autonomous ODEs . . . . . . . . . . . . . . . . . . . . . . . . . 340
16.3 Flows on the Circle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 341
16.3.1 Nonuniform oscillator . . . . . . . . . . . . . . . . . . . . . . . . . . 341
16.4 Appendix I : Evolution of Phase Space Volumes . . . . . . . . . . . . . . . 342
16.5 Appendix II : Lyapunov Characteristic Exponents . . . . . . . . . . . . . . 343
17 Bifurcations 347
17.1 Types of Bifurcations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 347
17.1.1 Saddle-node bifurcation . . . . . . . . . . . . . . . . . . . . . . . . . 347
17.1.2 Transcritical bifurcation . . . . . . . . . . . . . . . . . . . . . . . . 348
17.1.3 Pitchfork bifurcation . . . . . . . . . . . . . . . . . . . . . . . . . . 350
17.1.4 Imperfect bifurcation . . . . . . . . . . . . . . . . . . . . . . . . . . 351
17.2 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 354
17.2.1 Magnetization dynamics . . . . . . . . . . . . . . . . . . . . . . . . 354
17.2.2 Population dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . 357
17.3 Appendix I : The Bletch . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 361
17.4 Appendix II : Landau Theory of Phase Transitions . . . . . . . . . . . . . . 363
17.4.1 Cubic terms in Landau theory : first order transitions . . . . . . . . 364
17.4.2 Sixth order Landau theory : tricritical point . . . . . . . . . . . . . 366
17.4.3 Hysteresis for the sextic potential . . . . . . . . . . . . . . . . . . . 367
18 Two-Dimensional Phase Flows 371

xii CONTENTS
18.1 Harmonic Oscillator and Pendulum . . . . . . . . . . . . . . . . . . . . . . 371
18.1.1 Simple harmonic oscillator . . . . . . . . . . . . . . . . . . . . . . . 371
18.1.2 Pendulum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 373
18.2 General N = 2 Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 374
18.2.1 The damped driven pendulum . . . . . . . . . . . . . . . . . . . . . 374
18.2.2 Classification of N = 2 fixed points . . . . . . . . . . . . . . . . . . 377
18.2.3 The fixed point zoo . . . . . . . . . . . . . . . . . . . . . . . . . . . 379
18.2.4 Fixed points for N = 3 systems . . . . . . . . . . . . . . . . . . . . 381
18.3 Andronov-Hopf Bifurcation . . . . . . . . . . . . . . . . . . . . . . . . . . . 382
18.4 Population Biology : Lotka-Volterra Models . . . . . . . . . . . . . . . . . . 384
18.4.1 Rabbits and foxes . . . . . . . . . . . . . . . . . . . . . . . . . . . . 384
18.4.2 Rabbits and sheep . . . . . . . . . . . . . . . . . . . . . . . . . . . . 385
18.5 Index Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 387
18.5.1 Gauss-Bonnet Theorem . . . . . . . . . . . . . . . . . . . . . . . . . 391
18.6 Poincaré-Bendixson Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . 393
18.7 Appendix : Example Problem . . . . . . . . . . . . . . . . . . . . . . . . . 394
19 Nonlinear Oscillators 397
19.1 Weakly Perturbed Linear Oscillators . . . . . . . . . . . . . . . . . . . . . . 397
19.1.1 Naı̈ve Perturbation theory and its failure . . . . . . . . . . . . . . . 398
19.1.2 Poincaré-Lindstedt method . . . . . . . . . . . . . . . . . . . . . . . 399
19.2 Multiple Time Scale Method . . . . . . . . . . . . . . . . . . . . . . . . . . 401
19.2.1 Duffing oscillator . . . . . . . . . . . . . . . . . . . . . . . . . . . . 403
19.2.2 Forced Duffing oscillator . . . . . . . . . . . . . . . . . . . . . . . . 404
19.2.3 Van der Pol oscillator . . . . . . . . . . . . . . . . . . . . . . . . . . 407
19.2.4 Forced van der Pol oscillator . . . . . . . . . . . . . . . . . . . . . . 409
19.3 Relaxation Oscillations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 413

CONTENTS xiii
19.3.1 Example problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . 415
19.3.2 Multiple limit cycles . . . . . . . . . . . . . . . . . . . . . . . . . . 417
19.3.3 Example problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . 418
19.4 Parametric Oscillator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 419
19.5 Appendix I : Multiple Time Scale Analysis to O(2 ) . . . . . . . . . . . . . 422
19.6 Appendix II : MSA and Poincaré-Lindstedt Methods . . . . . . . . . . . . . 424
19.6.1 Problem using multiple time scale analysis . . . . . . . . . . . . . . 424
19.6.2 Solution using Poincaré-Lindstedt method . . . . . . . . . . . . . . 428
19.7 Appendix III : Modified van der Pol Oscillator . . . . . . . . . . . . . . . . 430
20 Hamiltonian Mechanics 437
20.1 The Hamiltonian . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 437
20.2 Modified Hamilton’s Principle . . . . . . . . . . . . . . . . . . . . . . . . . 439
20.3 Phase Flow is Incompressible . . . . . . . . . . . . . . . . . . . . . . . . . . 439
20.4 Poincaré Recurrence Theorem . . . . . . . . . . . . . . . . . . . . . . . . . 440
20.5 Kac Ring Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 441
20.6 Poisson Brackets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 444
20.7 Canonical Transformations . . . . . . . . . . . . . . . . . . . . . . . . . . . 445
20.7.1 Point transformations in Lagrangian mechanics . . . . . . . . . . . 445
20.7.2 Canonical transformations in Hamiltonian mechanics . . . . . . . . 447
20.7.3 Hamiltonian evolution . . . . . . . . . . . . . . . . . . . . . . . . . 447
20.7.4 Symplectic structure . . . . . . . . . . . . . . . . . . . . . . . . . . 448
20.7.5 Generating functions for canonical transformations . . . . . . . . . 449
20.8 Hamilton-Jacobi Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 452
20.8.1 The action as a function of coordinates and time . . . . . . . . . . . 452
20.8.2 The Hamilton-Jacobi equation . . . . . . . . . . . . . . . . . . . . . 454
20.8.3 Time-independent Hamiltonians . . . . . . . . . . . . . . . . . . . . 455

xiv CONTENTS
20.8.4 Example: one-dimensional motion . . . . . . . . . . . . . . . . . . . 456
20.8.5 Separation of variables . . . . . . . . . . . . . . . . . . . . . . . . . 456
20.8.6 Example #2 : point charge plus electric field . . . . . . . . . . . . . 458
20.8.7 Example #3 : Charged Particle in a Magnetic Field . . . . . . . . . 460
20.9 Action-Angle Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 462
20.9.1 Circular Phase Orbits: Librations and Rotations . . . . . . . . . . . 462
20.9.2 Action-Angle Variables . . . . . . . . . . . . . . . . . . . . . . . . . 464
20.9.3 Canonical Transformation to Action-Angle Variables . . . . . . . . 464
20.9.4 Example : Harmonic Oscillator . . . . . . . . . . . . . . . . . . . . 465
20.9.5 Example : Particle in a Box . . . . . . . . . . . . . . . . . . . . . . 466
20.9.6 Kepler Problem in Action-Angle Variables . . . . . . . . . . . . . . 469
20.9.7 Charged Particle in a Magnetic Field . . . . . . . . . . . . . . . . . 471
20.9.8 Motion on Invariant Tori . . . . . . . . . . . . . . . . . . . . . . . . 472
20.10 Canonical Perturbation Theory . . . . . . . . . . . . . . . . . . . . . . . . . 473
20.10.1 Canonical Transformations and Perturbation Theory . . . . . . . . 473
20.10.2 Canonical Perturbation Theory for n = 1 Systems . . . . . . . . . . 475
20.10.3 Example : Nonlinear Oscillator . . . . . . . . . . . . . . . . . . . . 477
20.10.4 n > 1 Systems : Degeneracies and Resonances . . . . . . . . . . . . 479
20.10.5 Particle-Wave Interaction . . . . . . . . . . . . . . . . . . . . . . . . 481
20.11 Adiabatic Invariants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 484
20.11.1 Example: mechanical mirror . . . . . . . . . . . . . . . . . . . . . . 485
20.11.2 Example: magnetic mirror . . . . . . . . . . . . . . . . . . . . . . . 486
20.11.3 Resonances . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 488
20.12 Appendix : Example Problem in Canonical Perturbation Theory . . . . . . 488
21 Strange Attractors and Chaos : A Preview 491
21.1 The Lorenz Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 491

CONTENTS xv
21.1.1 Fixed point analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . 493
22 Physics 110A-B Midterm and Final Examinations 497
22.1 F05 Physics 110A Midterm #1 . . . . . . . . . . . . . . . . . . . . . . . . . 498
22.3 F05 Physics 110A Final Exam . . . . . . . . . . . . . . . . . . . . . . . . . 509
22.6 F07 Physics 110A Final Exam . . . . . . . . . . . . . . . . . . . . . . . . . 527
22.7 W08 Physics 110B Midterm Exam . . . . . . . . . . . . . . . . . . . . . . . 538
22.8 W08 Physics 110B Final Exam . . . . . . . . . . . . . . . . . . . . . . . . . 543

xvi CONTENTS
0.1 Preface
These lecture notes are based on material presented in both graduate and undergraduate
mechanics classes which I have taught on several occasions during the past 20 years at
UCSD (Physics 110A-B and Physics 200A-B).
I have been consciously and very likely unconsciously influenced by the presentations in
many fine books on the subject. My favorites are:
• A. Fetter and J. D. Walecka, Theoretical Mechanics of Particles and Continua : An

excellent graduate level text, much of which is suitable for advanced undergraduates.
• V. I. Arnold, Ordinary Differential Equations : This book is a gem. If I had to give
one example of a technical book that “I couldn’t put down,” this would be it.
• V. D. Barger and M. G. Olsson, Classical Mechanics : A Modern Perspective : Some-
what quirky in its presentation, but replete with physical insights and interesting
worked examples. A good companion text for a junior level mechanics course.
• S. H. Strogatz, Nonlinear Dynamics and Chaos : A superb introduction to the subject,
and truly a pleasure to read. I appreciate authors who let their sense of humor come
through in scholarly works.
I learned the subject from Hartmut Sadrozinski’s Physics 205 class at Princeton during the
Fall of 1978. We used the first edition of the book by H. Goldstein, which for decades was
the standard graduate level text. (To terrorize its students, the department at Princeton
would inflict Goldstein on the sophomore undergraduates.) I still remember Goldstein’s
dedication, “to Louis J. Klein, teacher and friend,” which made my eyes moist. Hartmut’s
kindness and dedication to his students have served as a model for me. And he has remained
a friend for the past 30 years. But if he thinks I’m going to compose some sappy dedication
like Goldstein’s, he’s got another thing coming.
The level of these notes is appropriate for an advanced undergraduate or first year graduate
course in classical mechanics. In some instances, I’ve tried to collect the discussion of more
advanced material into separate sections, but in many cases this proves inconvenient, and
so the level of the presentation fluctuates somewhat. In addition, there are no collections of
problems at the end of each chapter. On the other hand, there are a great many problems
worked out in detail throughout. In my view, problem solving is essential toward learning
basic physics. In some (exceedingly rare) cases, brilliant individuals might apprehend the
fundamentals of mechanics through deep contemplation of pellucid lecture presentations.
However, the vast majority of us acquire physical intuition more slowly, and it is through
problem solving that one accumulates arrows in one’s quiver. Eventually, through our
brains’ uncanny ability to sense patterns and structure, something clicks, and we develop a
broader and deeper physical understanding.
More books: I should probably mention Landau and Lifshitz volume 1, lest Vladimir Putin
send someone over here to put plutonium in my tuna salad. Surely it is a great book,
0.2. ACKNOWLEDGMENTS xvii
although more at a graduate level, and it is most useful after one has already gained
some facility with the subject. The book by Thornton and Marion is widely used by
undergraduate Physics majors. Recently at UCSD we switched to a newer book by Taylor.
Both these texts are excellent.
So why, might you ask, does the world need yet another undergraduate mechanics text?
Well, you see, this one is free. My intention is to make these notes available to any instruc-
tors and students who might find them useful, perhaps as a companion resource to a more
proper text like Taylor or Thornton and Marion, each of which includes a good collection
of problems.
In addition to the text, virtually all of the figures I have composed myself, usually with
Keynote on my Mac. On rare occasions I’ve grabbed images from the web, such as the dreidl
in fig. 13.7. I hope that the original source for the dreidl won’t hunt me down and sue me.
You’d think whoever it is would cut me some slack since I am in fact Jewish. Of course,
with us Jews you never know. I’ve also reiterated and extended some discussions from
other texts, such as Barger and Olsson’s nice discussion and application of the gravitational
swing-by effect, and their discussion of rolling and skidding tops.
My only request, to those who would use these notes: please contact me if you find er-
rors or typos, or if you have suggestions for additional material. My email address is
darovas@ucsd.edu. I plan to update and extend these notes as my time and inclination
permit.
0.2 Acknowledgments
Over the years I have greatly benefited from the expertise and hard work of some excellent
teaching assistants, such as Jesse Goldberg, Ben Schmidel, and Andrew Meyertholen. I’m
especially grateful to Ben for writing a nifty javascript for plotting phase curves for one-
dimensional systems, which students have found helpful and illustrative.
I’ve been fortunate to have taught dozens of outstanding students. A handful deserve special
mention: Roland Roeder, Pavel Kolinko, Carrie Jiang, Andre Gomez, and very recently Aris
Alexandradinata and Alex Frenzel.
I’d like to take this opportunity to give thanks and praise to the most outstanding teachers
from whom I have learned, from junior high to graduate school: Norbert Lux, Dick Johnson,
Bob Gunning, Jim Peebles, John Cardy, and Doug Moore.
Many of my colleagues have developed and maintained a deep commitment to physics

teaching, simultaneous with their careers as leading researchers. Over the years I have
drawn on the expertise of many, but Kim Griest, Ben Grinstein, Tom O’Neil, and Vivek
Sharma have consistently been willing and even eager to discuss pedagogical matters, and
have always been helpful and generous with their time.
xviii CONTENTS
The staff support that lecturers receive in their teaching duties is often overlooked by
students and faculty alike. For me, there have been many occasions where the assistance
of our student affairs staff has made a significant positive difference in the quality of my
classes. Over the years, Patti Hey and Barbara Lowe have been extremely dependable
sources of support. I’d also like to thank Patti for all the chocolates, and to heap special
praise upon Barbara for her supreme competence and professionalism.
I thank my Psychology department colleague Hal Pashler for the many diverting political
arguments we’ve had over sushi during the past 20 years.
I am grateful to my wife Joyce and to my children Ezra and Lily for putting up with all the
outrageous lies I’ve told them about getting off the computer in ‘just a few minutes’ while
working on these notes.
This book is dedicated to the only two creatures I know who are never angry with me: my
father (Louis) and my dog (Henry).
Figure 1: My father and my dog.

Chapter 1
Introduction to Dynamics
1.1 Introduction and Review
Dynamics is the science of how things move. A complete solution to the motion of a system
means that we know the coordinates of all its constituent particles as functions of time.
For a single point particle moving in three-dimensional space, this means we want to know
its position vector r(t) as a function of time. If there are many particles, the motion is
described by a set of functions ri (t), where i labels which particle we are talking about. So
generally speaking, solving for the motion means being able to predict where a particle will
be at any given instant of time. Of course, knowing the function ri (t) means we can take
its derivative and obtain the velocity vi (t) = dri /dt at any time as well.
The complete motion for a system is not given to us outright, but rather is encoded in a
set of differential equations, called the equations of motion. An example of an equation of
motion is
d2x
m = −mg (1.1)
dt2
with the solution
x(t) = x0 + v0 t − 12 gt2 (1.2)
where x0 and v0 are constants corresponding to the initial boundary conditions on the
position and velocity: x(0) = x0 , v(0) = v0 . This particular solution describes the vertical
motion of a particle of mass m moving near the earth’s surface.
In this class, we shall discuss a general framework by which the equations of motion may
be obtained, and methods for solving them. That “general framework” is Lagrangian Dy-
namics, which itself is really nothing more than an elegant restatement of Isaac Newton’s
Laws of Motion.
1
2 CHAPTER 1. INTRODUCTION TO DYNAMICS
1.1.1 Newton’s laws of motion
Aristotle held that objects move because they are somehow impelled to seek out their
natural state. Thus, a rock falls because rocks belong on the earth, and flames rise because
fire belongs in the heavens. To paraphrase Wolfgang Pauli, such notions are so vague as to
be “not even wrong.” It was only with the publication of Newton’s Principia in 1687 that
a theory of motion which had detailed predictive power was developed.
Newton’s three Laws of Motion may be stated as follows:

I. A body remains in uniform motion unless acted on by a force.
II. Force equals rate of change of momentum: F = dp/dt.
III. Any two bodies exert equal and opposite forces on each other.
Newton’s First Law states that a particle will move in a straight line at constant (possibly
zero) velocity if it is subjected to no forces. Now this cannot be true in general, for suppose
we encounter such a “free” particle and that indeed it is in uniform motion, so that r(t) =
r0 + v0 t. Now r(t) is measured in some coordinate system, and if instead we choose
to measure r(t) in a different coordinate system whose origin R moves according to the
function R(t), then in this new “frame of reference” the position of our particle will be
r 0 (t) = r(t) − R(t)
= r0 + v0 t − R(t) . (1.3)
If the acceleration d2R/dt2 is nonzero, then merely by shifting our frame of reference we have
apparently falsified Newton’s First Law – a free particle does not move in uniform rectilinear
motion when viewed from an accelerating frame of reference. Thus, together with Newton’s
Laws comes an assumption about the existence of frames of reference – called inertial frames
– in which Newton’s Laws hold. A transformation from one frame K to another frame K0
which moves at constant velocity V relative to K is called a Galilean transformation. The
equations of motion of classical mechanics are invariant (do not change) under Galilean
transformations.
At first, the issue of inertial and noninertial frames is confusing. Rather than grapple with
this, we will try to build some intuition by solving mechanics problems assuming we are
in an inertial frame. The earth’s surface, where most physics experiments are done, is not
an inertial frame, due to the centripetal accelerations associated with the earth’s rotation
about its own axis and its orbit around the sun. In this case, not only is our coordinate
system’s origin – somewhere in a laboratory on the surface of the earth – accelerating, but
the coordinate axes themselves are rotating with respect to an inertial frame. The rotation
of the earth leads to fictitious “forces” such as the Coriolis force, which have large-scale
consequences. For example, hurricanes, when viewed from above, rotate counterclockwise
in the northern hemisphere and clockwise in the southern hemisphere. Later on in the course
we will devote ourselves to a detailed study of motion in accelerated coordinate systems.
Newton’s “quantity of motion” is the momentum p, defined as the product p = mv of a

particle’s mass m (how much stuff there is) and its velocity (how fast it is moving). In
1.1. INTRODUCTION AND REVIEW 3
order to convert the Second Law into a meaningful equation, we must know how the force
F depends on the coordinates (or possibly velocities) themselves. This is known as a force
law. Examples of force laws include:
Constant force: F = −mg
Hooke’s Law: F = −kx
Gravitation: F = −GM m r̂/r2
v
Lorentz force: F = qE +q ×B
c
Fluid friction (v small): F = −b v .
Note that for an object whose mass does not change we can write the Second Law in the
familiar form F = ma, where a = dv/dt = d2r/dt2 is the acceleration. Most of our initial
efforts will lie in using Newton’s Second Law to solve for the motion of a variety of systems.
The Third Law is valid for the extremely important case of central forces which we will
discuss in great detail later on. Newtonian gravity – the force which makes the planets orbit
the sun – is a central force. One consequence of the Third Law is that in free space two
isolated particles will accelerate in such a way that F1 = −F2 and hence the accelerations
are parallel to each other, with
a1 m2
=− , (1.4)
a2 m1
where the minus sign is used here to emphasize that the accelerations are in opposite
directions. We can also conclude that the total momentum P = p1 + p2 is a constant, a
result known as the conservation of momentum.
1.1.2 Aside : inertial vs. gravitational mass
In addition to postulating the Laws of Motion, Newton also deduced the gravitational force
law, which says that the force Fij exerted by a particle i by another particle j is
ri − rj
Fij = −Gmi mj , (1.5)
|ri − rj |3
where G, the Cavendish constant (first measured by Henry Cavendish in 1798), takes the
value
G = (6.6726 ± 0.0008) × 10−11 N · m2 /kg2 . (1.6)
Notice Newton’s Third Law in action: Fij + Fji = 0. Now a very important and special
feature of this “inverse square law” force is that a spherically symmetric mass distribution
has the same force on an external body as it would if all its mass were concentrated at its
center. Thus, for a particle of mass m near the surface of the earth, we can take mi = m
and mj = Me , with ri − rj ' Re r̂ and obtain
F = −mg r̂ ≡ −mg (1.7)
where r̂ is a radial unit vector pointing from the earth’s center and g = GMe /Re2 ' 9.8 m/s2
is the acceleration due to gravity at the earth’s surface. Newton’s Second Law now says
that a = −g, i.e. objects accelerate as they fall to earth. However, it is not a priori clear
why the inertial mass which enters into the definition of momentum should be the same
as the gravitational mass which enters into the force law. Suppose, for instance, that the
gravitational mass took a different value, m0 . In this case, Newton’s Second Law would
predict
m0
a=− g (1.8)
m
and unless the ratio m0 /m were the same number for all objects, then bodies would fall
with different accelerations. The experimental fact that bodies in a vacuum fall to earth at
the same rate demonstrates the equivalence of inertial and gravitational mass, i.e. m0 = m.
1.2 Examples of Motion in One Dimension
To gain some experience with solving equations of motion in a physical setting, we consider
some physically relevant examples of one-dimensional motion.
1.2.1 Uniform force
With F = −mg, appropriate for a particle falling under the influence of a uniform gravita-
tional field, we have m d2x/dt2 = −mg, or ẍ = −g. Notation:
dx d2x 7
¨¨˙ = d x ,
ẋ ≡ , ẍ ≡ , ẍ etc. (1.9)
dt dt2 dt7
With v = ẋ, we solve dv/dt = −g:
Zv(t) Zt
dv = ds (−g) (1.10)
v(0) 0
v(t) − v(0) = −gt . (1.11)
Note that there is a constant of integration, v(0), which enters our solution.
1.2. EXAMPLES OF MOTION IN ONE DIMENSION 5
We are now in position to solve dx/dt = v:
Zx(t) Zt
dx = ds v(s) (1.12)
x(0) 0
Zt

x(t) = x(0) + ds v(0) − gs (1.13)
0
= x(0) + v(0)t − 12 gt2 . (1.14)
Note that a second constant of integration, x(0), has appeared.
1.2.2 Uniform force with linear frictional damping
In this case,
dv
m = −mg − γv (1.15)
dt
which may be rewritten
dv γ
= − dt (1.16)
v + mg/γ m
d ln(v + mg/γ) = −(γ/m)dt . (1.17)
Integrating then gives

v(t) + mg/γ
ln = −γt/m (1.18)
v(0) + mg/γ

mg mg
v(t) = − + v(0) + e−γt/m . (1.19)
γ γ
Note that the solution to the first order ODE mv̇ = −mg − γv entails one constant of
integration, v(0).
One can further integrate to obtain the motion

m mg mg
x(t) = x(0) + v(0) + (1 − e−γt/m ) − t. (1.20)
γ γ γ
The solution to the second order ODE mẍ = −mg − γ ẋ thus entails two constants of
integration: v(0) and x(0). Notice that as t goes to infinity the velocity tends towards
the asymptotic value v = −v∞ , where v∞ = mg/γ. This is known as the terminal veloc-
ity. Indeed, solving the equation v̇ = 0 gives v = −v∞ . The initial velocity is effectively
“forgotten” on a time scale τ ≡ m/γ.
Electrons moving in solids under the influence of an electric field also achieve a terminal
velocity. In this case the force is not F = −mg but rather F = −eE, where −e is the
electron charge (e > 0) and E is the electric field. The terminal velocity is then obtained
from
v∞ = eE/γ = eτ E/m . (1.21)
The current density is a product:
current density = (number density) × (charge) × (velocity)
j = n · (−e) · (−v∞ )
ne2 τ
= E. (1.22)
m
The ratio j/E is called the conductivity of the metal, σ. According to our theory, σ =
ne2 τ /m. This is one of the most famous equations of solid state physics! The dissipation
is caused by electrons scattering off impurities and lattice vibrations (“phonons”). In high
purity copper at low temperatures (T < ∼ 4 K), the scattering time τ is about a nanosecond
(τ ≈ 10−9 s).
1.2.3 Uniform force with quadratic frictional damping
At higher velocities, the frictional damping is proportional to the square of the velocity.
The frictional force is then Ff = −cv 2 sgn (v), where sgn (v) is the sign of v: sgn (v) = +1
if v > 0 and sgn (v) = −1 if v < 0. (Note one can also write sgn (v) = v/|v| where |v| is
the absolute value.) Why all this trouble with sgn (v)? Because it is important that the
frictional force dissipate energy, and therefore that Ff be oppositely directed with respect to
the velocity v. We will assume that v < 0 always, hence Ff = +cv 2 .
2
p is a terminal velocity, since setting v̇ = −g + (c/m)v = 0 gives v = ±v∞ ,
Notice that there
where v∞ = mg/c. One can write the equation of motion as
dv g
= 2 (v 2 − v∞
2
) (1.23)
dt v∞
and using
1 1 1 1
= − (1.24)
v 2 − v∞
2 2v∞ v − v∞ v + v∞
we obtain
dv 1 dv 1 dv
= −
v2 2
− v∞ 2v∞ v − v∞ 2v∞ v + v∞

1 v∞ − v
= d ln
2v∞ v∞ + v
g
= 2 dt . (1.25)
v∞
1.2. EXAMPLES OF MOTION IN ONE DIMENSION 7
Assuming v(0) = 0, we integrate to obtain

1 v∞ − v(t) gt
ln = 2 (1.26)
2v∞ v∞ + v(t) v∞
which may be massaged to give the final result
v(t) = −v∞ tanh(gt/v∞ ) . (1.27)
Recall that the hyperbolic tangent function tanh(x) is given by

sinh(x) ex − e−x
tanh(x) = = x . (1.28)
cosh(x) e + e−x
Again, as t → ∞ one has v(t) → −v∞ , i.e. v(∞) = −v∞ .
Advanced Digression: To gain an understanding of the constant c, consider a flat surface

of area S moving through a fluid at velocity v (v > 0). During a time ∆t, all the fluid
molecules inside the volume ∆V = S · v∆t will have executed an elastic collision with the
moving surface. Since the surface is assumed to be much more massive than each fluid
molecule, the center of mass frame for the surface-molecule collision is essentially the frame
of the surface itself. If a molecule moves with velocity u is the laboratory frame, it moves
with velocity u − v in the center of mass (CM) frame, and since the collision is elastic, its
final CM frame velocity is reversed, to v − u. Thus, in the laboratory frame the molecule’s
velocity has become 2v − u and it has suffered a change in velocity of ∆u = 2(v − u). The
total momentum change is obtained by multiplying ∆u by the total mass M = %∆V , where
% is the mass density of the fluid. But then the total momentum imparted to the fluid is
∆P = 2(v − u) · % S v∆t (1.29)
and the force on the fluid is

∆P
F = = 2S % v(v − u) . (1.30)
∆t
Now it is appropriate to average this expression over the microscopic distribution of molec-
ular velocities u, and since on average hui = 0, we obtain the result hF i = 2S%v 2 , where
h· · · i denotes a microscopic average over the molecular velocities in the fluid. (There is a
subtlety here concerning the effect of fluid molecules striking the surface from either side –
you should satisfy yourself that this derivation is sensible!) Newton’s Third Law then states
that the frictional force imparted to the moving surface by the fluid is Ff = −hF i = −cv 2 ,
where c = 2S%. In fact, our derivation is too crude to properly obtain the numerical prefac-
tors, and it is better to write c = µ%S, where µ is a dimensionless constant which depends
on the shape of the moving object.
1.2.4 Crossed electric and magnetic fields
Consider now a three-dimensional example of a particle of charge q moving in mutually

perpendicular E and B fields. We’ll throw in gravity for good measure. We take E = E x̂,
B = B ẑ, and g = −g ẑ. The equation of motion is Newton’s 2nd Law again:
m r̈ = mg + qE + qc ṙ × B . (1.31)
The RHS (right hand side) of this equation is a vector sum of the forces due to gravity plus
the Lorentz force of a moving particle in an electromagnetic field. In component notation,
we have
qB
mẍ = qE + ẏ (1.32)
c
qB
mÿ = − ẋ (1.33)
c
mz̈ = −mg . (1.34)
The equations for coordinates x and y are coupled, while that for z is independent and may
be immediately solved to yield
z(t) = z(0) + ż(0) t − 12 gt2 . (1.35)
The remaining equations may be written in terms of the velocities vx = ẋ and vy = ẏ:
v̇x = ωc (vy + uD ) (1.36)

v̇y = −ωc vx , (1.37)
where ωc = qB/mc is the cyclotron frequency and uD = cE/B is the drift speed for the
particle. As we shall see, these are the equations for a harmonic oscillator. The solution is

vx (t) = vx (0) cos(ωc t) + vy (0) + uD sin(ωc t) (1.38)

vy (t) = −uD + vy (0) + uD cos(ωc t) − vx (0) sin(ωc t) . (1.39)
Integrating again, the full motion is given by:
x(t) = x(0) + A sin δ + A sin(ωc t − δ) (1.40)

y(r) = y(0) − uD t − A cos δ + A cos(ωc t − δ) , (1.41)
where

1 ẏ(0) + uD
q 2 −1
A= ẋ2 (0) + ẏ(0) + uD , δ = tan . (1.42)
ωc ẋ(0)
Thus, in the full solution of the motion there are six constants of integration:
x(0) , y(0) , z(0) , A , δ , ż(0) . (1.43)
Of course instead of A and δ one may choose as constants of integration ẋ(0) and ẏ(0).
1.3. PAUSE FOR REFLECTION 9
1.3 Pause for Reflection
In mechanical systems, for each coordinate, or “degree of freedom,” there exists a cor-
responding second order ODE. The full solution of the motion of the system entails two
constants of integration for each degree of freedom.
Chapter 2
Systems of Particles
2.1 Work-Energy Theorem
Consider a system of many particles, with positions ri and velocities ṙi . The kinetic energy
of this system is X X
1 2
T = Ti = 2 mi ṙi . (2.1)
i i
Now let’s consider how the kinetic energy of the system changes in time. Assuming each
mi is time-independent, we have
dTi
= mi ṙi · r̈i . (2.2)
dt
Here, we’ve used the relation
d dA
A2 = 2 A ·

. (2.3)
dt dt
We now invoke Newton’s 2nd Law, mi r̈i = Fi , to write eqn. 2.2 as Ṫi = Fi · ṙi . We integrate
this equation from time tA to tB :
ZtB
(B) (A) dTi
Ti − Ti = dt
dt
tA
ZtB X (A→B)
= dt Fi · ṙi ≡ Wi , (2.4)
tA i
where Wi(A→B) is the total work done on particle

P i during its motion from state A to state
B, Clearly P
the total kinetic energy is T = i Ti and the total work done on all particles is
W (A→B) = i Wi(A→B) . Eqn. 2.4 is known as the work-energy theorem. It says that
In the evolution of a mechanical system, the change in total kinetic energy is equal to the
total work done: T (B) − T (A) = W (A→B) .
11
12 CHAPTER 2. SYSTEMS OF PARTICLES
Figure 2.1: Two paths joining points A and B.
2.2 Conservative and Nonconservative Forces
For the sake of simplicity, consider a single particle with kinetic energy T = 12 mṙ 2 . The
work done on the particle during its mechanical evolution is
ZtB
(A→B)
W = dt F · v , (2.5)
tA
where v = ṙ. This is the most general expression for the work done. If the force F depends
only on the particle’s position r, we may write dr = v dt, and then
ZrB
W (A→B) = dr · F (r) . (2.6)
rA
Consider now the force

F (r) = K1 y x̂ + K2 x ŷ , (2.7)
where K1,2 are constants. Let’s evaluate the work done along each of the two paths in fig.
2.1:
ZxB ZyB
W (I) = K1 dx yA + K2 dy xB = K1 yA (xB − xA ) + K2 xB (yB − yA ) (2.8)
xA yA
ZxB ZyB
(II)
W = K1 dx yB + K2 dy xA = K1 yB (xB − xA ) + K2 xA (yB − yA ) . (2.9)
xA yA
2.2. CONSERVATIVE AND NONCONSERVATIVE FORCES 13
Note that in general W (I) 6= W (II) . Thus, if we start at point A, the kinetic energy at point
B will depend on the path taken, since the work done is path-dependent.
The difference between the work done along the two paths is
W (I) − W (II) = (K2 − K1 ) (xB − xA ) (yB − yA ) . (2.10)
Thus, we see that if K1 = K2 , the work is the same for the two paths. In fact, if K1 = K2 ,
the work would be path-independent, and would depend only on the endpoints. This is
true for any path, and not just piecewise linear paths of the type depicted in fig. 2.1. The
reason for this is Stokes’ theorem:
I Z
d` · F = dS n̂ · ∇ × F . (2.11)
∂C C
Here, C is a connected region in three-dimensional space, ∂C is mathematical notation for

the boundary of C, which is a closed path1 , dS is the scalar differential area element, n̂ is
the unit normal to that differential area element, and ∇ × F is the curl of F :
x̂ ŷ ẑ
 
∂ ∂ ∂ 
∇ × F = det  ∂x ∂y ∂z
Fx Fy Fz

∂Fy ∂Fz ∂Fz ∂Fx ∂Fx ∂Fy
= − x̂ + − ŷ + − ẑ . (2.12)
∂z ∂y ∂x ∂z ∂y ∂x
For the force under consideration, F (r) = K1 y x̂ + K2 x ŷ, the curl is
∇ × F = (K2 − K1 ) ẑ , (2.13)
which is a constant. The RHS of eqn. 2.11 is then simply proportional to the area enclosed
H
by C. When we compute the work difference in eqn. 2.10, we evaluate the integral d` · F
C
−1
along the path γII ◦ γI , which is to say path I followed by the inverse of path II. In this
case, n̂ = ẑ and the integral of n̂ · ∇ × F over the rectangle C is given by the RHS of eqn.
2.10.
When ∇ × F = 0 everywhere in space, we can always write F = −∇U , where U (r) is the
potential energy. Such forces are called conservative forces because the total energy of the
system, E = T + U , is then conserved during its motion. We can see this by evaluating the
1
If C is multiply connected, then ∂C is a set of closed paths. For example, if C is an annulus, ∂C is two
circles, corresponding to the inner and outer boundaries of the annulus.
work done,
ZrB
W (A→B) = dr · F (r)
rA
ZrB
= − dr · ∇U
rA
= U (rA ) − U (rB ) . (2.14)
The work-energy theorem then gives
T (B) − T (A) = U (rA ) − U (rB ) , (2.15)
which says
E (B) = T (B) + U (rB ) = T (A) + U (rA ) = E (A) . (2.16)
Thus, the total energy E = T + U is conserved.
2.2.1 Example : integrating F = −∇U
If ∇ × F = 0, we can compute U (r) by integrating, viz.

Zr
U (r) = U (0) − dr 0 · F (r 0 ) . (2.17)
0
The integral does not depend on the path chosen connecting 0 and r. For example, we can
take
(x,0,0)
Z (x,y,0)
Z (x,y,z)
Z
U (x, y, z) = U (0, 0, 0) − dx0 Fx (x0 , 0, 0) − dy 0 Fy (x, y 0 , 0) − dz 0 Fz (x, y, z 0 ) . (2.18)
(0,0,0) (x,0,0) (z,y,0)
The constant U (0, 0, 0) is arbitrary and impossible to determine from F alone.
As an example, consider the force
F (r) = −ky x̂ − kx ŷ − 4bz 3 ẑ , (2.19)
where k and b are constants. We have

∂Fz
∂Fy
∇×F x = − =0 (2.20)
∂y ∂z

∂Fx ∂Fz
∇×F y = − =0 (2.21)
∂z ∂x

∂Fy ∂Fx
∇×F z = − =0, (2.22)
∂x ∂y
2.2. CONSERVATIVE AND NONCONSERVATIVE FORCES 15
so ∇ × F = 0 and F must be expressible as F = −∇U . Integrating using eqn. 2.18, we

have
(x,0,0)
Z (x,y,0)
Z (x,y,z)
Z
0 0 0 3
U (x, y, z) = U (0, 0, 0) + dx k · 0 + dy kxy + dz 0 4bz 0 (2.23)
(0,0,0) (x,0,0) (z,y,0)
= U (0, 0, 0) + kxy + bz 4 . (2.24)
Another approach is to integrate the partial differential equation ∇U = −F . This is in fact

three equations, and we shall need all of them to obtain the correct answer. We start with
the x̂-component,
∂U
= ky . (2.25)
∂x
Integrating, we obtain
U (x, y, z) = kxy + f (y, z) , (2.26)
where f (y, z) is at this point an arbitrary function of y and z. The important thing is that
it has no x-dependence, so ∂f /∂x = 0. Next, we have
∂U
= kx =⇒ U (x, y, z) = kxy + g(x, z) . (2.27)
∂y
Finally, the z-component integrates to yield
∂U
= 4bz 3 =⇒ U (x, y, z) = bz 4 + h(x, y) . (2.28)
∂z
We now equate the first two expressions:
kxy + f (y, z) = kxy + g(x, z) . (2.29)
Subtracting kxy from each side, we obtain the equation f (y, z) = g(x, z). Since the LHS is
independent of x and the RHS is independent of y, we must have
f (y, z) = g(x, z) = q(z) , (2.30)
where q(z) is some unknown function of z. But now we invoke the final equation, to obtain
bz 4 + h(x, y) = kxy + q(z) . (2.31)
The only possible solution is h(x, y) = C + kxy and q(z) = C + bz 4 , where C is a constant.
Therefore,
U (x, y, z) = C + kxy + bz 4 . (2.32)
Note that it would be very wrong to integrate ∂U/∂x = ky and obtain U (x, y, z) = kxy +
C 0 , where C 0 is a constant. As we’ve seen, the ‘constant of integration’ we obtain upon
integrating this first order PDE is in fact a function of y and z. The fact that f (y, z) carries
no explicit x dependence means that ∂f /∂x = 0, so by construction U = kxy + f (y, z) is a
solution to the PDE ∂U/∂x = ky, for any arbitrary function f (y, z).
2.3 Conservative Forces in Many Particle Systems
X
1 2
T = 2 mi ṙi (2.33)
i
X X
U= V (ri ) + v |ri − rj | . (2.34)
i i<j
Here, V (r) is the external (or one-body) potential, and v(r−r 0 ) is the interparticle potential,
which we assume to be central, depending only on the distance between any pair of particles.
The equations of motion are
mi r̈i = Fi(ext) + Fi(int) , (2.35)
with
∂V (ri )
Fi(ext) = − (2.36)
∂ri

X ∂v |ri − rj | X (int)
Fi(int) =− ≡ Fij . (2.37)
ri
j j
Here, Fij(int) is the force exerted on particle i by particle j:

(int) ∂v |ri − rj | ri − rj 0
Fij = =− v |ri − rj | . (2.38)
∂ri |ri − rj |
Note that Fij(int) = −Fji(int) , otherwise known as Newton’s Third Law. It is convenient to
abbreviate rij ≡ ri − rj , in which case we may write the interparticle force as Fij(int) =
−r̂ij v 0 rij .

2.4 Linear and Angular Momentum

P
Consider now the total momentum of the system, P = i pi . Its rate of change is
(int) (int)
Fij +Fji =0
dP X X (ext) zX}| {
= ṗi = Fi + Fij(int) = Ftot
(ext)
, (2.39)
dt
i i i6=j
since the sum over all internal forces cancels as a result of Newton’s Third Law. We write
X
P = mi ṙi = M Ṙ (2.40)
i
X
M= mi (total mass) (2.41)
i
P
i mi ri
R= P (center-of-mass) . (2.42)
i mi
2.4. LINEAR AND ANGULAR MOMENTUM 17
Next, consider the total angular momentum,

X X
L= ri × pi = mi ri × ṙi . (2.43)
i i
The rate of change of L is then
dL X
= mi ṙi × ṙi + mi ri × r̈i
dt
i
X X
= ri × Fi(ext) + ri × Fij(int)
i i6=j
(int)
rij ×Fij =0
X zX }| {
1
= ri × Fi(ext) + 2
(int)
(ri − rj ) × Fij
i i6=j
(ext)
= Ntot . (2.44)
Finally, it is useful to establish the result

X X 2
T = 1
2 mi ṙi2 = 12 M Ṙ2 + 1
2 mi ṙi − Ṙ , (2.45)
i i
which says that the kinetic energy may be written as a sum of two terms, those being the
kinetic energy of the center-of-mass motion, and the kinetic energy of the particles relative
to the center-of-mass.
Recall the “work-energy theorem” for conservative systems,
final
Z final
Z final
Z
0= dE = dT + dU
initial initial
initial
XZ
= T (B) − T (A) − dri · Fi (2.46)
i
XZ
(B) (A)
∆T = T −T = dri · Fi . (2.47)
i
Note that for continuous systems, we replace

X Z
d3r ρ(r) φ(r) ,

mi φ ri −→ (2.48)
i
where ρ(r) is the mass density, and φ(r) is any function.

2.5 Scaling of Solutions for Homogeneous Potentials
2.5.1 Euler’s theorem for homogeneous functions
In certain cases of interest, the potential is a homogeneous function of the coordinates. This
means
U λ r1 , . . . , λ rN = λk U r1 , . . . , rN .

(2.49)
Here, k is the degree of homogeneity of U . Familiar examples include gravity,
X mi mj
U r1 , . . . , rN = −G ; k = −1 , (2.50)
|ri − rj |
i<j
and the harmonic oscillator,

X
1

U q1 , . . . , q n = 2 Vσσ0 qσ qσ0 ; k = +2 . (2.51)
σ,σ 0
The sum of two homogeneous functions is itself homogeneous only if the component func-
tions themselves are of the same degree of homogeneity. Homogeneous functions obey a
special result known as Euler’s Theorem, which we now prove. Suppose a multivariable
function H(x1 , . . . , xn ) is homogeneous:
H(λ x1 , . . . , λ xn ) = λk H(x1 , . . . , xn ) . (2.52)
Then
n
d X ∂H
H λ x1 , . . . , λ xn = xi = kH (2.53)
dλ ∂xi
λ=1 i=1
2.5.2 Scaled equations of motion
Now suppose the we rescale distances and times, defining
ri = α r̃i , t = β t̃ . (2.54)
Then
dri α dr̃i d2 ri α d2 r̃i
= , = . (2.55)
dt β dt̃ dt2 β 2 dt̃2
The force Fi is given by
∂
Fi = − U r1 , . . . , rN
∂ri
∂
αk U r̃1 , . . . , r̃N

=−
∂(αr̃i )
= αk−1 F̃i . (2.56)
2.5. SCALING OF SOLUTIONS FOR HOMOGENEOUS POTENTIALS 19
Thus, Newton’s 2nd Law says
α d2 r̃i
m = αk−1 F̃i . (2.57)
β 2 i dt̃2
If we choose β such that
We now demand
α 1
= αk−1 ⇒ β = α1− 2 k , (2.58)
β2
then the equation of motion is invariant under the rescaling transformation! This means
1
k−1
that if r(t) is a solution to the equations of motion, then so is α r α 2 t . This gives us
an entire one-parameter family of solutions, for all real positive α.
1
If r(t) is periodic with period T , the ri (t; α) is periodic with period T 0 = α1− 2 k T . Thus,
1− 1 k
T0 L0

2
= . (2.59)
T L
Here, α = L0 /L is the ratio of length scales. Velocities, energies and angular momenta scale
accordingly:
v0 L0 T 0

L 1
v = ⇒ = = α2k (2.60)
T v L T
0 2 0 2
M L2 E0 L T
E = 2
⇒ = = αk (2.61)
T E L T
0 2 0
M L2 |L0 | L T 1
L = ⇒ = = α(1+ 2 k) . (2.62)
T |L| L T
As examples, consider:
(i) Harmonic Oscillator : Here k = 2 and therefore
qσ (t) −→ qσ (t; α) = α qσ (t) . (2.63)
Thus, rescaling lengths alone gives another solution.
(ii) Kepler Problem : This is gravity, for which k = −1. Thus,
r(t) −→ r(t; α) = α r α−3/2 t .

(2.64)
Thus, r3 ∝ t2 , i.e.
3 2
L0 T0

= , (2.65)
L T
also known as Kepler’s Third Law.
2.6 Appendix I : Curvilinear Orthogonal Coordinates
The standard cartesian coordinates are {x1 , . . . , xd }, where d is the dimension of space.
Consider a different set of coordinates, {q1 , . . . , qd }, which are related to the original coor-
dinates xµ via the d equations

qµ = qµ x1 , . . . , xd . (2.66)
In general these are nonlinear equations.
Let ê0i = x̂i be the Cartesian set of orthonormal unit vectors, and define êµ to be the unit
vector perpendicular to the surface dqµ = 0. A differential change in position can now be
described in both coordinate systems:
d
X d
X
ds = ê0i dxi = êµ hµ (q) dqµ , (2.67)
i=1 µ=1
where each hµ (q) is an as yet unknown function of all the components qν . Finding the
coefficient of dqµ then gives
d d
X ∂xi 0 X
hµ (q) êµ = ê ⇒ êµ = Mµ i ê0i , (2.68)
∂qµ i
i=1 i=1
where
1 ∂xi
Mµi (q) = . (2.69)
hµ (q) ∂qµ
The dot product of unit vectors in the new coordinate system is then
d
1 X ∂xi ∂xi
êµ · êν = M M t

µν
= . (2.70)
hµ (q) hν (q) ∂qµ ∂qν
i=1
The condition that the new basis be orthonormal is then

d
X ∂xi ∂xi
= h2µ (q) δµν . (2.71)
∂qµ ∂qν
i=1
This gives us the relation v

u d
uX ∂xi 2
hµ (q) = t . (2.72)
∂qµ
i=1
Note that
d
X
(ds)2 = h2µ (q) (dqµ )2 . (2.73)
µ=1
2.6. APPENDIX I : CURVILINEAR ORTHOGONAL COORDINATES 21
For general coordinate systems, which are not necessarily orthogonal, we have
d
X
2
(ds) = gµν (q) dqµ dqν , (2.74)
µ,ν=1
where gµν (q) is a real, symmetric, positive definite matrix called the metric tensor .
2.6.1 Example : spherical coordinates
Consider spherical coordinates (ρ, θ, φ):

x = ρ sin θ cos φ , y = ρ sin θ sin φ , z = ρ cos θ . (2.75)
It is now a simple matter to derive the results
h2ρ = 1 , h2θ = ρ2 , h2φ = ρ2 sin2 θ . (2.76)
Thus,
ds = ρ̂ dρ + ρ θ̂ dθ + ρ sin θ φ̂ dφ . (2.77)
2.6.2 Vector calculus : grad, div, curl
Here we restrict our attention to d = 3. The gradient ∇U of a function U (q) is defined by

∂U ∂U ∂U
dU = dq1 + dq2 + dq
∂q1 ∂q2 ∂q3 3
≡ ∇U · ds . (2.78)
Thus,
ê1 ∂ ê2 ∂ ê3 ∂
∇= + + . (2.79)
h1 (q) ∂q1 h2 (q) ∂q2 h3 (q) ∂q3
For the divergence, we use the divergence theorem, and we appeal to fig. 19.15:
Z Z
dV ∇ · A = dS n̂ · A , (2.80)
Ω ∂Ω
where Ω is a region of three-dimensional space and ∂Ω is its closed two-dimensional bound-

ary. The LHS of this equation is
LHS = ∇ · A · (h1 dq1 ) (h2 dq2 ) (h3 dq3 ) . (2.81)
The RHS is
q +dq q +dq q +dq
1 1 2 2 1 3
RHS = A1 h2 h3 dq2 dq3 + A2 h1 h3 dq1 dq3 + A3 h1 h2 dq1 dq2
q1 q2 q3

∂ ∂ ∂
= A h h + A h h + A h h dq1 dq2 dq3 . (2.82)
∂q1 1 2 3 ∂q2 2 1 3 ∂q3 3 1 2
Figure 2.2: Volume element Ω for computing divergences.
We therefore conclude

1 ∂ ∂ ∂
∇·A= A1 h 2 h 3 + A2 h 1 h 3 + A3 h 1 h 2 . (2.83)
h1 h2 h3 ∂q1 ∂q2 ∂q3
To obtain the curl ∇ × A, we use Stokes’ theorem again,

Z I
dS n̂ · ∇ × A = d` · A , (2.84)
Σ ∂Σ
where Σ is a two-dimensional region of space and ∂Σ is its one-dimensional boundary. Now

consider a differential surface element satisfying dq1 = 0, i.e. a rectangle of side lengths
h2 dq2 and h3 dq3 . The LHS of the above equation is
LHS = ê1 · ∇ × A (h2 dq2 ) (h3 dq3 ) . (2.85)
The RHS is
q +dq q +dq
2 2 3 3
RHS = A3 h3 dq3 − A2 h2 dq2
q2 q3

∂ ∂
= A3 h 3 − A2 h2 dq2 dq3 . (2.86)
∂q2 ∂q3
Therefore
1 ∂(h3 A3 ) ∂(h2 A2 )
(∇ × A)1 = − . (2.87)
h2 h3 ∂q2 ∂q3
This is one component of the full result
 
h1 ê1 h2 ê2 h3 ê3
1
∇×A= det  ∂q∂1 ∂ ∂ 
∂q3  . (2.88)

∂q2
h1 h2 h2
h1 A1 h2 A2 h3 A3
The Laplacian of a scalar function U is given by
∇2 U = ∇ · ∇U
( )
1 ∂ h2 h3 ∂U ∂ h1 h3 ∂U ∂ h1 h2 ∂U
= + + . (2.89)
h1 h2 h 3 ∂q1 h1 ∂q1 ∂q2 h2 ∂q2 ∂q3 h3 ∂q3
2.6.3 Common curvilinear orthogonal systems
Rectangular coordinates
In rectangular coordinates (x, y, z), we have
h x = hy = hz = 1 . (2.90)
Then the gradient is

∂U ∂U ∂U
∇U = x̂ + ŷ + ẑ . (2.91)
∂x ∂y ∂z
The divergence is
∂Ax ∂Ay ∂Az
∇·A= + + . (2.92)
∂x ∂y ∂z
The curl is

∂Az ∂Ay ∂Ax ∂Az ∂Ay ∂Ax
∇×A= − x̂ + − ŷ + − ẑ . (2.93)
∂y ∂z ∂z ∂x ∂x ∂y
The Laplacian is
∂2U ∂2U ∂2U
∇2 U = + + . (2.94)
∂x2 ∂y 2 ∂z 2
Cylindrical coordinates
In cylindrical coordinates (ρ, φ, z), we have
ρ̂ = x̂ cos φ + ŷ sin φ x̂ = ρ̂ cos φ − φ̂ sin φ dρ̂ = φ̂ dφ (2.95)

φ̂ = −x̂ sin φ + ŷ cos φ ŷ = ρ̂ sin φ + φ̂ cos φ dφ̂ = −ρ̂ dφ . (2.96)
The metric is given in terms of
hρ = 1 , hφ = ρ , hz = 1 . (2.97)

∂U φ̂ ∂U ∂U
∇U = ρ̂ + + ẑ . (2.98)
∂ρ ρ ∂φ ∂z
The divergence is
1 ∂(ρ Aρ ) 1 ∂Aφ ∂Az
∇·A= + + . (2.99)
ρ ∂ρ ρ ∂φ ∂z
The curl is

1 ∂Az ∂Aφ ∂Aρ ∂Az 1 ∂(ρAφ ) 1 ∂Aρ
∇×A= − ρ̂ + − φ̂ + − ẑ . (2.100)
ρ ∂φ ∂z ∂z ∂ρ ρ ∂ρ ρ ∂φ
The Laplacian is
1 ∂2U ∂2U

2 1 ∂ ∂U
∇ U= ρ + 2 + . (2.101)
ρ ∂ρ ∂ρ ρ ∂φ2 ∂z 2
Spherical coordinates
In spherical coordinates (r, θ, φ), we have
r̂ = x̂ sin θ cos φ + ŷ sin θ sin φ + ẑ sin θ (2.102)

θ̂ = x̂ cos θ cos φ + ŷ cos θ sin φ − ẑ cos θ (2.103)
φ̂ = −x̂ sin φ + ŷ cos φ , (2.104)
for which
r̂ × θ̂ = φ̂ , θ̂ × φ̂ = r̂ , φ̂ × r̂ = θ̂ . (2.105)
The inverse is
x̂ = r̂ sin θ cos φ + θ̂ cos θ cos φ − φ̂ sin φ (2.106)

ŷ = r̂ sin θ sin φ + θ̂ cos θ sin φ + φ̂ cos φ (2.107)
ẑ = r̂ cos θ − θ̂ sin θ . (2.108)
The differential relations are
dr̂ = θ̂ dθ + sin θ φ̂ dφ (2.109)

dθ̂ = −r̂ dθ + cos θ φ̂ dφ (2.110)

dφ̂ = − sin θ r̂ + cos θ θ̂ dφ (2.111)
The metric is given in terms of
hr = 1 , hθ = r , hφ = r sin θ . (2.112)

∂U θ̂ ∂U φ̂ ∂U
∇U = r̂ + + . (2.113)
∂ρ r ∂θ r sin θ ∂φ
The divergence is
1 ∂(r2 Ar ) 1 ∂(sin θ Aθ ) 1 ∂Aφ
∇·A= + + . (2.114)
r2 r r sin θ ∂θ r sin θ ∂φ
The curl is

1 ∂(sin θ Aφ ) ∂Aθ 1 1 ∂Ar ∂(rAφ )
∇×A= − r̂ + − θ̂
r sin θ ∂r ∂φ r sin θ ∂φ ∂r

1 ∂(rAθ ) ∂Ar
+ − φ̂ . (2.115)
r ∂r ∂θ
The Laplacian is
∂2U

2 1 ∂ 2 ∂U 1 ∂ ∂U 1
∇ U= 2 r + 2 sin θ + 2 2 . (2.116)
r ∂r ∂r r sin θ ∂θ ∂θ r sin θ ∂φ2
Chapter 3
One-Dimensional Conservative
Systems
3.1 Description as a Dynamical System
For one-dimensional mechanical systems, Newton’s second law reads
mẍ = F (x) . (3.1)
A system is conservative if the force is derivable from a potential: F = −dU/dx. The total
energy,
E = T + U = 12 mẋ2 + U (x) , (3.2)
is then conserved. This may be verified explicitly:
dE d h1 2
i
= mẋ + U (x)
dt dt 2
h i
= mẍ + U 0 (x) ẋ = 0 . (3.3)
Conservation of energy allows us to reduce the equation of motion from second order to
first order: v
u !
dx u2
= ±t E − U (x) . (3.4)
dt m
Note that the constant E is a constant of integration. The ± sign above depends on the
direction of motion. Points x(E) which satisfy
E = U (x) ⇒ x(E) = U −1 (E) , (3.5)
where U −1 is the inverse function, are called turning points. When the total energy is E,
the motion of the system is bounded by the turning points, and confined to the region(s)
27
28 CHAPTER 3. ONE-DIMENSIONAL CONSERVATIVE SYSTEMS
U (x) ≤ E. We can integrate eqn. 3.4 to obtain
r Zx
m dx0
t(x) − t(x0 ) = ± p . (3.6)
2 E − U (x0 )
x0
This is to be inverted to obtain the function x(t). Note that there are now two constants
of integration, E and x0 . Since
E = E0 = 21 mv02 + U (x0 ) , (3.7)
we could also consider x0 and v0 as our constants of integration, writing E in terms of x0

and v0 . Thus, there are two independent constants of integration.
For motion confined between two turning points x± (E), the period of the motion is given
by
xZ
+ (E)
√ dx0
T (E) = 2m p . (3.8)
E − U (x0 )
x− (E)
3.1.1 Example : harmonic oscillator
In the case of the harmonic oscillator, we have U (x) = 21 kx2 , hence

r
dt m
=± . (3.9)
dx 2E − kx2
p
The turning points are x ± (E) = ± 2E/k, for E ≥ 0. To solve for the motion, let us
substitute r
2E
x= sin θ . (3.10)
k
We then find r
m
dt = dθ , (3.11)
k
with solution
θ(t) = θ0 + ωt , (3.12)
p
where ω = k/m is the harmonic oscillator frequency. Thus, the complete motion of the
system is given by
r
2E
x(t) = sin(ωt + θ0 ) . (3.13)
k
Note the two constants of integration, E and θ0 .
3.2. ONE-DIMENSIONAL MECHANICS AS A DYNAMICAL SYSTEM 29
3.2 One-Dimensional Mechanics as a Dynamical System
Rather than writing the equation of motion as a single second order ODE, we can instead
write it as two coupled first order ODEs, viz.
dx
=v (3.14)
dt
dv 1
= F (x) . (3.15)
dt m
This may be written in matrix-vector form, as

d x v
= 1 . (3.16)
dt v m F (x)
This is an example of a dynamical system, described by the general form

dϕ
= V (ϕ) , (3.17)
dt
where ϕ = (ϕ1 , . . . , ϕN ) is an N -dimensional vector in phase space. For the model of eqn.
3.16, we evidently have N = 2. The object V (ϕ) is called a vector field . It is itself a vector,
existing at every point in phase space, RRN . Each of the components of V (ϕ) is a function
(in general) of all the components of ϕ:
Vj = Vj (ϕ1 , . . . , ϕN ) (j = 1, . . . , N ) . (3.18)
Solutions to the equation ϕ̇ = V (ϕ) are called integral curves. Each such integral curve
ϕ(t) is uniquely determined by N constants of integration, which may be taken to be the
initial value ϕ(0). The collection of all integral curves is known as the phase portrait of the
dynamical system.
In plotting the phase portrait of a dynamical system, we need to first solve for its motion,
starting from arbitrary initial conditions. In general this is a difficult problem, which can
only be treated numerically. But for conservative mechanical systems in d = 1, it is a trivial
matter! The reason is that energy conservation completely determines theq phase portraits.
2

The velocity becomes a unique double-valued function of position, v(x) = ± m E − U (x) .
The phase curves are thus curves of constant energy.
3.2.1 Sketching phase curves
To plot the phase curves,
(i) Sketch the potential U (x).

Figure 3.1: A potential U (x) and the corresponding phase portraits. Separatrices are shown
in red.
q
2

(ii) Below this plot, sketch v(x; E) = ± m E − U (x) .
(iii) When E lies at a local extremum of U (x), the system is at a fixed point.
(a) For E slightly above Emin , the phase curves are ellipses.
(b) For E slightly below Emax , the phase curves are (locally) hyperbolae.
(c) For E = Emax the phase curve is called a separatrix .
(iv) When E > U (∞) or E > U (−∞), the motion is unbounded .
(v) Draw arrows along the phase curves: to the right for v > 0 and left for v < 0.
The period of the orbit T (E) has a simple geometric interpretation. The area A in phase
space enclosed by a bounded phase curve is
I xZ
+ (E)
q
dx0
p
8
A(E) = v dx = m E − U (x0 ) . (3.19)
E x− (E)
Thus, the period is proportional to the rate of change of A(E) with E:

∂A
T =m . (3.20)
∂E
3.3 Fixed Points and their Vicinity
A fixed point (x∗ , v ∗ ) of the dynamics satisfies U 0 (x∗ ) = 0 and v ∗ = 0. Taylor’s theorem
then allows us to expand U (x) in the vicinity of x∗ :
U (x) = U (x∗ ) + U 0 (x∗ ) (x − x∗ ) + 21 U 00 (x∗ ) (x − x∗ )2 + 16 U 000 (x∗ ) (x − x∗ )3 + . . . . (3.21)
3.3. FIXED POINTS AND THEIR VICINITY 31
Since U 0 (x∗ ) = 0 the linear term in δx = x − x∗ vanishes. If δx is sufficiently small, we can

ignore the cubic, quartic, and higher order terms, leaving us with
U (δx) ≈ U0 + 21 k(δx)2 , (3.22)
where U0 = U (x∗ ) and k = U 00 (x∗ ) > 0. The solutions to the motion in this potential are:
δv0
U 00 (x∗ ) > 0 : δx(t) = δx0 cos(ωt) + sin(ωt) (3.23)
ω
δv0
U 00 (x∗ ) < 0 : δx(t) = δx0 cosh(γt) + sinh(γt) , (3.24)
γ
p p
where ω = k/m for k > 0 and γ = −k/m for k < 0. The energy is
E = U0 + 12 m (δv0 )2 + 12 k (δx0 )2 . (3.25)
For a separatrix, we have E = U0 and U 00 (x∗ ) < 0. From the equation for the energy, we
obtain δv0 = ±γ δx0 . Let’s take δv0 = −γ δx0 , so that the initial velocity is directed toward
the unstable fixed point (UFP). I.e. the initial velocity is negative if we are to the right of
the UFP (δx0 > 0) and positive if we are to the left of the UFP (δx0 < 0). The motion of
the system is then
δx(t) = δx0 exp(−γt) . (3.26)
The particle gets closer and closer to the unstable fixed point at δx = 0, but it takes an
infinite amount of time to actually get there. Put another way, the time it takes to get from
δx0 to a closer point δx < δx0 is

−1 δx0
t=γ ln . (3.27)
δx
This diverges logarithmically as δx → 0. Generically, then, the period of motion along a

separatrix is infinite.
3.3.1 Linearized dynamics in the vicinity of a fixed point
Linearizing in the vicinity of such a fixed point, we write δx = x − x∗ and δv = v − v ∗ ,

obtaining
d δx 0 1 δx
= + ..., (3.28)
dt δv −m 1
U 00 (x∗ ) 0 δv
This is a linear equation, which we can solve completely.
Consider the general linear equation ϕ̇ = A ϕ, where A is a fixed real matrix. Now whenever
we have a problem involving matrices, we should start thinking about eigenvalues and
Figure 3.2: Phase curves in the vicinity of centers and saddles.
eigenvectors. Invariably, the eigenvalues and eigenvectors will prove to be useful, if not
essential, in solving the problem. The eigenvalue equation is
A ψα = λα ψα . (3.29)
Here ψα is the αth right eigenvector 1 of A. The eigenvalues are roots of the characteristic
equation P (λ) = 0, where P (λ) = det(λ · 11 − A). Let’s expand ϕ(t) in terms of the right
eigenvectors of A: X
ϕ(t) = Cα (t) ψα . (3.30)
α
Assuming, for the purposes of this discussion, that A is nondegenerate, and its eigenvectors
span RRN , the dynamical system can be written as a set of decoupled first order ODEs for
the coefficients Cα (t):
Ċα = λα Cα , (3.31)
with solutions
Cα (t) = Cα (0) exp(λα t) . (3.32)
If Re(λα ) > 0, Cα (t) flows off to infinity, while if Re(λα ) > 0, Cα (t) flows to zero. If
|λα | = 1, then Cα (t) oscillates with frequency Im(λα ).
For a two-dimensional matrix, it is easy to show – an exercise for the reader – that
P (λ) = λ2 − T λ + D , (3.33)
where T = Tr(A) and D = det(A). The eigenvalues are then

p
λ± = 21 T ± 12 T 2 − 4D . (3.34)
1
If A is symmetric, the right and left eigenvectors are the same. If A is not symmetric, the right and left
eigenvectors differ, although the set of corresponding eigenvalues is the same.
3.4. EXAMPLES OF CONSERVATIVE ONE-DIMENSIONAL SYSTEMS 33
We’ll study the general case in Physics 110B. For now, we focus on our conservative me-
chanical system of eqn. 3.28. The trace and determinant of the above matrix are T = 0 and
D=m 1
U 00 (x∗ ). Thus, there are only two (generic) possibilities: centers, when U 00 (x∗ ) > 0,
and saddles, when U 00 (x∗ ) < 0. Examples of each are shown in Fig. 3.1.
3.4 Examples of Conservative One-Dimensional Systems
3.4.1 Harmonic oscillator
Recall again the harmonic oscillator, discussed in lecture 3. The potential energy is U (x) =
1 2
2 kx . The equation of motion is
d2 x dU
m =− = −kx , (3.35)
dt2 dx
where m is the mass and k the force constant (of a spring). With v = ẋ, this may be written
as the N = 2 system,

d x 0 1 x v
= = , (3.36)
dt v −ω 2 0 v −ω 2 x
p
where ω = k/m has the dimensions of frequency (inverse time). The solution is well
known:
v0
x(t) = x0 cos(ωt) + sin(ωt) (3.37)
ω
v(t) = v0 cos(ωt) − ω x0 sin(ωt) . (3.38)
The phase curves are ellipses:
ω0 x2 (t) + ω0−1 v 2 (t) = C , (3.39)
where C is a constant, independent of time. A sketch of the phase curves and of the phase
flow is shown in Fig. 18.1. Note that the x and v axes have different dimensions.
Energy is conserved:
E = 21 mv 2 + 12 kx2 . (3.40)
Therefore we may find the length of the semimajor and semiminor axes by setting v = 0 or
x = 0, which gives r r
2E 2E
xmax = , vmax = . (3.41)
k m
The area of the elliptical phase curves is thus
2πE
A(E) = π xmax vmax = √ . (3.42)
mk
Figure 3.3: Phase curves for the harmonic oscillator.
The period of motion is therefore

r
∂A m
T (E) = m = 2π , (3.43)
∂E k
which is independent of E.
3.4.2 Pendulum
Next, consider the simple pendulum, composed of a mass point m affixed to a massless rigid
rod of length `. The potential is U (θ) = −mg` cos θ, hence
dU
m`2 θ̈ = − = −mg` sin θ . (3.44)
dθ
This is equivalent to
θ d ω
= , (3.45)
ω dt −ω02 sin θ
p
where ω = θ̇ is the angular velocity, and where ω0 = g/` is the natural frequency of small
oscillations.
The conserved energy is

E= 1
2 m`2 θ̇2 + U (θ) . (3.46)
Assuming the pendulum is released from rest at θ = θ0 ,
2E
= θ̇2 − 2ω02 cos θ = −2ω02 cos θ0 . (3.47)
m`2
The period for motion of amplitude θ0 is then
√ Zθ0
8 dθ 4
K sin2 12 θ0 ,

T θ0 = √ = (3.48)
ω0 cos θ − cos θ0 ω0
0
Figure 3.4: Phase curves for the simple pendulum. The separatrix divides phase space into
regions of vibration and libration.
where K(z) is the complete elliptic integral of the first kind. Expanding K(z), we have

2π 1 2 1
9 4 1

T θ0 = 1+ 4 sin 2 θ0 + 64 sin 2 θ0 + ... . (3.49)
ω0
For θ0 → 0, the period approaches the usual result 2π/ω0 , valid for the linearized equation
θ̈ = −ω02 θ. As θ0 → π2 , the period diverges logarithmically.
The phase curves for the pendulum are shown in Fig. 18.2. The small oscillations of the
pendulum are essentially the same as those of a harmonic oscillator. Indeed, within the
small angle approximation, sin θ ≈ θ, and the pendulum equations of motion are exactly
those of the harmonic oscillator. These oscillations are called librations. They involve
a back-and-forth motion in real space, and the phase space motion is contractable to a
point, in the topological sense. However, if the initial angular velocity is large enough, a
qualitatively different kind of motion is observed, whose phase curves are rotations. In this
case, the pendulum bob keeps swinging around in the same direction, because, as we’ll see
in a later lecture, the total energy is sufficiently large. The phase curve which separates
these two topologically distinct motions is called a separatrix .
3.4.3 Other potentials
Using the phase plotter application written by Ben Schmidel, available on the Physics 110A
course web page, it is possible to explore the phase curves for a wide variety of potentials.
Three examples are shown in the following pages. The first is the effective potential for the
Kepler problem,
k `2
Ueff (r) = − + , (3.50)
r 2µr2
about which we shall have much more to say when we study central forces. Here r is the
separation between two gravitating bodies of masses m1,2 , µ = m1 m2 /(m1 + m2 ) is the
‘reduced mass’, and k = Gm1 m2 , where G is the Cavendish constant. We can then write

1 1
Ueff (r) = U0 − + 2 , (3.51)
x 2x
where r0 = `2 /µk has the dimensions of length, and x ≡ r/r0 , and where U0 = k/r0 =
µk 2 /`2 . Thus, if distances are measured in units of r0 and the potential in units of U0 , the
potential may be written in dimensionless form as U(x) = − x1 + 2x12 .
The second is the hyperbolic secant potential,
U (x) = −U0 sech2 (x/a) , (3.52)
which, in dimensionless form, is U(x) = −sech2 (x), after measuring distances in units of a
and potential in units of U0 .
The final example is

x x
U (x) = U0 cos + . (3.53)
a 2a
Again measuring x in units of a and U in units of U0 , we arrive at U(x) = cos(x) + 21 x.
Figure 3.5: Phase curves for the Kepler effective potential U (x) = −x−1 + 21 x−2 .
Figure 3.6: Phase curves for the potential U (x) = −sech2 (x).
Figure 3.7: Phase curves for the potential U (x) = cos(x) + 21 x.

Chapter 4
Linear Oscillations
Harmonic motion is ubiquitous in Physics. The reason is that any potential energy function,
when expanded in a Taylor series in the vicinity of a local minimum, is a harmonic function:
q ∗ )=0
∇U (~
N
z }| { N
∂ 2 U

∗
X ∂U ∗
X
U (~q ) = U (~q ) + 1
(q − qj ) + 2 (q − qj∗ ) (qk − qk∗ ) + . . . , (4.1)
∂qj q~=~q ∗ j ∂qj ∂qk q~=~q ∗ j
j=1 j,k=1
where the {qj } are generalized coordinates – more on this when we discuss Lagrangians. In
one dimension, we have simply
U (x) = U (x∗ ) + 21 U 00 (x∗ ) (x − x∗ )2 + . . . . (4.2)
Provided the deviation η = x − x∗ is small enough in magnitude, the remaining terms in

the Taylor expansion may be ignored. Newton’s Second Law then gives
m η̈ = −U 00 (x∗ ) η + O(η 2 ) . (4.3)
This, to lowest order, is the equation of motion for a harmonic oscillator. If U 00 (x∗ ) > 0,
the equilibrium point x = x∗ is stable, since for small deviations from equilibrium the
restoring force pushes the system back toward the equilibrium point. When U 00 (x∗ ) < 0,
the equilibrium is unstable, and the forces push one further away from equilibrium.
4.1 Damped Harmonic Oscillator
In the real world, there are frictional forces, which we here will approximate by F = −γv.
We begin with the homogeneous equation for a damped harmonic oscillator,
d2x dx
2
+ 2β + ω02 x = 0 , (4.4)
dt dt
41
42 CHAPTER 4. LINEAR OSCILLATIONS
where γ = 2βm. To solve, write x(t) = i Ci e−iωi t . This renders the differential equation
P
4.4 an algebraic equation for the two eigenfrequencies ωi , each of which must satisfy
ω 2 + 2iβω − ω02 = 0 , (4.5)
hence
ω± = −iβ ± (ω02 − β 2 )1/2 . (4.6)
The most general solution to eqn. 4.4 is then
x(t) = C+ e−iω+ t + C− e−iω− t (4.7)
where C± are arbitrary constants. Notice that the eigenfrequencies are in general complex,
with a negative imaginary part (so long as the damping coefficient β is positive). Thus
e−iω± t decays to zero as t → ∞.
4.1.1 Classification of damped harmonic motion
We identify three classes of motion:

(i) Underdamped (ω02 > β 2 ) : x(t) = C e−βt cos(νt) + D e−βt sin(νt)
(ii) Overdamped (ω02 < β 2 ) : x(t) = C e−βt cosh(ν̃t) + D e−βt sinh(ν̃t)
(iii) Critically Damped (ω02 = β 2 ) : x(t) = C e−βt + D t e−βt
where ν ≡ (ω02 − β 2 )1/2 and ν̃ ≡ iν = (β 2 − ω02 )1/2 . Note that for case (i) ν is real and ν̃ is
imaginary, while for case (ii) ν̃ is real and ν is imaginary. Note also that the form for x(t)
in case (i) can be applied to case (ii), and vice versa, since
cos(νt) = cos(−iν̃t) = cosh(ν̃t)
sin(νt) = sin(−iν̃t) = sinh(ν̃t) .
The three types of behavior are depicted in fig. 4.1. To concretize these cases in one’s
mind, it is helpful to think of the case of a screen door or a shock absorber. If the hinges on
the door are underdamped, the door will swing back and forth (assuming it doesn’t have a
rim which smacks into the door frame) several times before coming to a stop. If the hinges
are overdamped, the door may take a very long time to close. To see this, note that the
overdamped solution can also be written as
√ √
− β− β 2 −ω02 t − β+ β 2 −ω02 t
x(t) = A e +Be , (4.8)
p
with A = 21 (C + D) and B = 12 (C − D). We now expand the expression β 2 − ω02 in powers
of ω02 /β 2 :
ω02 −1/2
q
2 2
β − ω0 = β 1 − 2
β
ω02 ω04

= β 1 − 2 − 4 + ... , (4.9)
2β 8β
4.1. DAMPED HARMONIC OSCILLATOR 43
Figure 4.1: Three classifications of damped harmonic motion. The initial conditions are
x(0) = 1, ẋ(0) = 0.
which leads to
ω02 ω4
q
β− β 2 − ω02 =+ 03 + . . .
2β 8β
ω2
q
β + β 2 − ω02 = 2β − 0 − + . . . . (4.10)
2β
Thus, we can write

x(t) = A e−t/τ1 + B e−t/τ2 , (4.11)
with
1 2β
τ1 = ≈ (4.12)
ω02
p
β− β 2 − ω02
1 1
τ2 = p ≈ . (4.13)
β+ β2 − ω02 2β
Thus x(t) is a sum of exponentials, with decay times τ1,2 . For β ω0 , we have that τ1 is
much larger than τ2 – the ratio is τ1 /τ2 ≈ 4β 2 /ω02 1. Thus, on time scales on the order of
τ1 , the second term has completely damped away. The decay time τ1 , though, is very long,
since β is so large. So a highly overdamped oscillator will take a very long time to come to
equilbrium.
4.1.2 Remarks on the case of critical damping
Define the first order differential operator

d
Dt = +β . (4.14)
dt
The solution to Dt x(t) = 0 is x̃(t) = A e−βt , where A is a constant. Note that the commu-
tator of Dt and t is unity:

Dt , t = 1 , (4.15)
where [A, B] ≡ AB − BA. The simplest way to verify eqn. 4.15 is to compute its action
upon an arbitrary function f (t):

d d
Dt , t f (t) = + β t f (t) − t + β f (t)
dt dt
d d
= t f (t) − t f (t) = f (t) . (4.16)
dt dt
We know that x(t) = x̃(t) = A e−βt satisfies Dt x(t) = 0. Therefore

0 = Dt Dt , t x̃(t)
0
z }| {
= Dt2 t x̃(t) − Dt t Dt x̃(t)

= Dt2 t x̃(t) . (4.17)
We already know that Dt2 x̃(t) = Dt Dt x̃(t) = 0. The above equation establishes that the
second independent solution to the second order ODE Dt2 x(t) = 0 is x(t) = t x̃(t). Indeed,
we can keep going, and show that

Dtn tn−1 x̃(t) = 0 . (4.18)
Thus, the n independent solutions to the nth order ODE

n
d
+ β x(t) = 0 (4.19)
dt
are
xk (t) = A tk e−βt , k = 0, 1, . . . , n − 1 . (4.20)
4.2 Damped Harmonic Oscillator with Forcing
When forced, the equation for the damped oscillator becomes

d2x dx
2
+ 2β + ω02 x = f (t) , (4.21)
dt dt
4.2. DAMPED HARMONIC OSCILLATOR WITH FORCING 45
where f (t) = F (t)/m. Since this equation is linear in x(t), we can, without loss of generality,
restrict out attention to harmonic forcing terms of the form
h i
f (t) = f0 cos(Ωt + ϕ0 ) = Re f0 e−iϕ0 e−iΩt (4.22)
where Re stands for “real part”. Here, Ω is the forcing frequency.
Consider first the complex equation

d2z dz
2
+ 2β + ω02 z = f0 e−iϕ0 e−iΩt . (4.23)
dt dt
We try a solution z(t) = z0 e−iΩt . Plugging in, we obtain the algebraic equation
f0 e−iϕ0
z0 = ≡ A(Ω) eiδ(Ω) f0 e−iϕ0 . (4.24)
ω02 − 2iβΩ − Ω 2
The amplitude A(Ω) and phase shift δ(Ω) are given by the equation
1
A(Ω) eiδ(Ω) = . (4.25)
ω02 − 2iβΩ − Ω 2
A basic fact of complex numbers:
−1
1 a + ib ei tan (b/a)
= 2 = √ . (4.26)
a − ib a + b2 a2 + b2
Thus,
−1/2
A(Ω) = (ω02 − Ω 2 )2 + 4β 2 Ω 2 (4.27)

−1 2βΩ
δ(Ω) = tan 2 . (4.28)
ω0 − Ω 2
Now since the coefficients β and ω02 are real, we can take the complex conjugate of eqn.
4.23, and write
z̈ + 2β ż + ω02 z = f0 e−iϕ0 e−iΩt (4.29)

z̄¨ + 2β z̄˙ + ω02 z̄ = f0 e+iϕ0 e+iΩt , (4.30)
where z̄ is the complex conjugate of z. We now add these two equations and divide by two
to arrive at
ẍ + 2β ẋ + ω02 x = f0 cos(Ωt + ϕ0 ) . (4.31)
Therefore, the real, physical solution we seek is
h i
xinh (t) = Re A(Ω) eiδ(Ω) · f0 e−iϕ0 e−iΩt

= A(Ω) f0 cos Ωt + ϕ0 − δ(Ω) . (4.32)
Figure 4.2: Amplitude and phase shift versus oscillator frequency (units of ω0 ) for β/ω0
values of 0.1 (red), 0.25 (magenta), 1.0 (green), and 2.0 (blue).
The quantity A(Ω) is the amplitude of the response (in units of f0 ), while δ(Ω) is the
(dimensionless) phase lag.
The maximum of the amplitude A(Ω) occurs when A0 (Ω) = 0. From
dA 2Ω 2 2 2

= − 3 Ω − ω0 + 2β , (4.33)
dΩ A(Ω)
we conclude that A0 (Ω) = 0 for Ω = 0 and for Ω = ΩR , where

q
ΩR = ω02 − 2β 2 . (4.34)
The solution at Ω = ΩR pertains only if ω02 > 2β 2 , of course, in which case Ω = 0 is a local
minimum and Ω = ΩR a local maximum. If ω02 < 2β 2 there is only a local maximum, at
Ω = 0. See Fig. 4.2.
Since equation 4.21 is linear, we can add a solution to the homogeneous equation to xinh (t)
and we will still have a solution. Thus, the most general solution to eqn. 4.21 is
x(t) = xinh (t) + xhom (t)

h i
= Re A(Ω) eiδ(Ω) · f0 e−iϕ0 e−iΩt + C+ e−iω+ t + C− e−iω− t
xinh (t) hom x (t)
z }| { z −βt }| {
=A(Ω) f0 cos Ωt + ϕ0 − δ(Ω) + C e cos(νt) + D e−βt sin(νt) . (4.35)
The last two terms in eqn. 4.35 are the solution to the homogeneous equation, i.e. with
f (t) = 0. They are necessary to include because they carry with them the two constants
of integration which always arise in the solution of a second order ODE. That is, C and D
are adjusted so as to satisfy x(0) = x0 and ẋ0 = v0 . However, due to their e−βt prefactor,
these terms decay to zero once t reaches a relatively low multiple of β −1 . They are called
transients, and may be set to zero if we are only interested in the long time behavior of the
system. This means, incidentally, that the initial conditions are effectively forgotten over a
time scale on the order of β −1 .
For ΩR > 0, one defines the quality factor , Q, of the oscillator by Q = ΩR /2β. Q is a rough
measure of how many periods the unforced oscillator executes before its initial amplitude is
damped down to a small value. For a forced oscillator driven near resonance, and for weak
damping, Q is also related to the ratio of average energy in the oscillator to the energy lost
per cycle by the external source. To see this, let us compute the energy lost per cycle,
2π/Ω
Z
∆E = m dt ẋ f (t)
0
2π/Ω
Z
= −m dt Ω A f02 sin(Ωt + ϕ0 − δ) cos(Ωt + ϕ0 )
0
= πA f02 m sin δ
= 2πβ m Ω A2 (Ω) f02 , (4.36)

since sin δ(Ω) = 2βΩ A(Ω). The oscillator energy, averaged over the cycle, is
2π/Ω
Z
Ω
dt 12 m ẋ2 + ω02 x2

E =
2π
0
2
= 1
4 m (Ω + ω02 ) A2 (Ω) f02 . (4.37)
Thus, we have
2πhEi Ω 2 + ω02
= . (4.38)
∆E 4βΩ
Thus, for Ω ≈ ΩR and β 2 ω02 , we have
2πhEi ω0
Q≈ ≈ . (4.39)
∆E 2β
4.2.1 Resonant forcing
When the damping β vanishes, the response diverges at resonance. The solution to the
resonantly forced oscillator
ẍ + ω02 x = f0 cos(ω0 t + ϕ0 ) (4.40)
is given by
xhom (t)
f0
z }| {
x(t) = t sin(ω0 t + ϕ0 )+ A cos(ω0 t) + B sin(ω0 t) . (4.41)
2ω0
The amplitude of this solution grows linearly due to the energy pumped into the oscillator by
the resonant external forcing. In the real world, nonlinearities can mitigate this unphysical,
unbounded response.
4.2.2 R-L-C circuits
Consider the R-L-C circuit of Fig. 4.3. When the switch is to the left, the capacitor is
charged, eventually to a steady state value Q = CV . At t = 0 the switch is thrown to the
right, completing the R-L-C circuit. Recall that the sum of the voltage drops across the
three elements must be zero:
dI Q
L + IR + =0. (4.42)
dt C
We also have Q̇ = I, hence
d2 Q R dQ 1
2
+ + Q=0, (4.43)
dt L dt LC
which is the equation for a damped harmonic oscillator, with ω0 = (LC)−1/2 and β = R/2L.
The boundary conditions at t = 0 are Q(0) = CV and Q̇(0) = 0. Under these conditions,
the full solution at all times is
β
Q(t) = CV e−βt cos νt + sin νt (4.44)
ν
ω 2
I(t) = −CV 0 e−βt sin νt . (4.45)
ν
Figure 4.3: An R-L-C circuit which behaves as a damped harmonic oscillator.

4.2.3 Examples
Third order linear ODE with forcing
The problem is to solve the equation

...
Lt x ≡ x + (a + b + c) ẍ + (ab + ac + bc) ẋ + abc x = f0 cos(Ωt) . (4.46)
The key to solving this is to note that the differential operator Lt factorizes:
d3 d2 d
Lt = 3
+ (a + b + c) 2
+ (ab + ac + bc) + abc
dt dt dt
d d d
= +a +b +c , (4.47)
dt dt dt
which says that the third order differential operator appearing in the ODE is in fact a
product of first order differential operators. Since
dx
+ αx = 0 =⇒ x(t) = A e−αx , (4.48)
dt
we see that the homogeneous solution takes the form
xh (t) = A e−at + B e−bt + C e−ct , (4.49)
where A, B, and C are constants.
To find the inhomogeneous solution, we solve Lt x = f0 e−iΩt and take the real part. Writing
x(t) = x0 e−iΩt , we have
Lt x0 e−iΩt = (a − iΩ) (b − iΩ) (c − iΩ) x0 e−iΩt (4.50)
and thus
f0 e−iΩt
x0 = ≡ A(Ω) eiδ f0 e−iΩt ,
(a − iΩ)(b − iΩ)(c − iΩ)
where
h i−1/2
A(Ω) = (a2 + Ω 2 ) (b2 + Ω 2 ) (c2 + Ω 2 ) (4.51)
Ω Ω Ω
δ(Ω) = tan−1 + tan−1 + tan−1 . (4.52)
a b c
Thus, the most general solution to Lt x(t) = f0 cos(Ωt) is
x(t) = A(Ω) f0 cos Ωt − δ(Ω) + A e−at + B e−bt + C e−ct .

(4.53)
Note that the phase shift increases monotonically from δ(0) = 0 to δ(∞) = 23 π.
Figure 4.4: A driven L-C-R circuit, with V (t) = V0 cos(ωt).
Mechanical analog of RLC circuit
Consider the electrical circuit in fig. 4.4. Our task is to construct its mechanical analog.
To do so, we invoke Kirchoff’s laws around the left and right loops:
Q1
L1 I˙1 + + R1 (I1 − I2 ) = 0 (4.54)
C1
L2 I˙2 + R2 I2 + R1 (I2 − I1 ) = V (t) . (4.55)
Let Q1 (t) be the charge on the left plate of capacitor C1 , and define
Zt
Q2 (t) = dt0 I2 (t0 ) . (4.56)
0
Then Kirchoff’s laws may be written

R1 1
Q̈1 + (Q̇1 − Q̇2 ) + Q1 = 0 (4.57)
L1 L1 C1
R2 R1 V (t)
Q̈2 + Q̇2 + (Q̇2 − Q̇1 ) = . (4.58)
L2 L2 L2
Now consider the mechanical system in Fig. 4.5. The blocks have masses M1 and M2 .
The friction coefficient between blocks 1 and 2 is b1 , and the friction coefficient between
block 2 and the floor is b2 . Here we assume a velocity-dependent frictional force Ff = −bẋ,
rather than the more conventional constant Ff = −µ W , where W is the weight of an
object. Velocity-dependent friction is applicable when the relative velocity of an object and
a surface is sufficiently large. There is a spring of spring constant k1 which connects block 1
to the wall. Finally, block 2 is driven by a periodic acceleration f0 cos(ωt). We now identify
R1 R2 1
X1 ↔ Q1 , X2 ↔ Q2 , b1 ↔ , b2 ↔ , k1 ↔ , (4.59)
L1 L2 L1 C1
Figure 4.5: The equivalent mechanical circuit for fig. 4.4.
as well as f (t) ↔ V (t)/L2 .
The solution again proceeds by Fourier transform. We write
Z∞
dω
V (t) = V̂ (ω) e−iωt (4.60)
2π
−∞
and
Z∞
Q1 (t) dω Q̂1 (ω)
= e−iωt (4.61)
Iˆ2 (t) 2π Iˆ2 (ω)
−∞
The frequency space version of Kirchoff’s laws for this problem is
Ĝ(ω)
z }| { 
−ω 2 − iω R1 /L1 + 1/L1 C1
   
R1 /L1 Q̂1 (ω) 0
  =  (4.62)
iω R1 /L2 −iω + (R1 + R2 )/L2 ˆ
I2 (ω) V̂ (ω)/L2
The homogeneous equation has eigenfrequencies given by the solution to det Ĝ(ω) = 0,
which is a cubic equation. Correspondingly, there are three initial conditions to account
for: Q1 (0), I1 (0), and I2 (0). As in the case of the single damped harmonic oscillator, these
transients are damped, and for large times may be ignored. The solution then is
   2 −1  
Q̂1 (ω) −ω − iω R1 /L1 + 1/L1 C1 R1 /L1 0
 =    .
Iˆ (ω)
2
iω R1 /L2 −iω + (R1 + R2 )/L2 V̂ (ω)/L2
(4.63)
To obtain the time-dependent Q1 (t) and I2 (t), we must compute the Fourier transform back
to the time domain.
4.3 General solution by Green’s function method
For a general forcing function f (t), we solve by Fourier transform. Recall that a function
F (t) in the time domain has a Fourier transform F̂ (ω) in the frequency domain. The relation
between the two is:
Z∞ Z∞
dω −iωt
F (t) = e F̂ (ω) ⇐⇒ F̂ (ω) = dt e+iωt F (t) . (4.64)
2π
−∞ −∞
We can convert the differential equation 4.2 to an algebraic equation in the frequency
domain, x̂(ω) = Ĝ(ω) fˆ(ω), where
1
Ĝ(ω) = (4.65)
ω02 − 2iβω − ω 2
is the Green’s function in the frequency domain. The general solution is written
Z∞
dω −iωt
x(t) = e Ĝ(ω) fˆ(ω) + xh (t) , (4.66)
2π
−∞
where xh (t) = i Ci e−iωi t is a solution to the homogeneous equation. We may also write
P
the above integral over the time domain:
Z∞
x(t) = dt0 G(t − t0 ) f (t0 ) + xh (t) (4.67)
−∞
Z∞
dω −iωs
G(s) = e Ĝ(ω)
2π
−∞
−1
= ν exp(−βs) sin(νs) Θ(s) (4.68)
where Θ(s) is the step function,

1 if s ≥ 0
Θ(s) = (4.69)
0 if s < 0
Example: force pulse
Consider a pulse force

f0 if 0 ≤ t ≤ T
f (t) = f0 Θ(t) Θ(T − t) = (4.70)
0 otherwise.
In the underdamped regime, for example, we find the solution

f0 −βt β −βt
x(t) = 2 1 − e cos νt − e sin νt (4.71)
ω0 ν
4.4. GENERAL LINEAR AUTONOMOUS INHOMOGENEOUS ODES 53
Figure 4.6: Response of an underdamped oscillator to a pulse force.
if 0 ≤ t ≤ T and
(
f0
x(t) = e−β(t−T ) cos ν(t − T ) − e−βt cos νt
ω02
)
β −β(t−T )
+ e sin ν(t − T ) − e−βt sin νt (4.72)
ν
if t > T .
4.4 General Linear Autonomous Inhomogeneous ODEs
This method immediately generalizes to the case of general autonomous linear inhomoge-
neous ODEs of the form
dnx dn−1 x dx
n
+ an−1 n−1
+ . . . + a1 + a0 x = f (t) . (4.73)
dt dt dt
We can write this as
Lt x(t) = f (t) , (4.74)
where Lt is the nth order differential operator
dn dn−1 d
Lt = n
+ an−1 n−1
+ . . . + a1 + a0 . (4.75)
dt dt dt
The general solution to the inhomogeneous equation is given by

Z∞
x(t) = xh (t) + dt0 G(t, t0 ) f (t0 ) , (4.76)
−∞
where G(t, t0 ) is the Green’s function. Note that Lt xh (t) = 0. Thus, in order for eqns. 4.74
and 4.92 to be true, we must have
this vanishes
z }| { Z∞
Lt x(t) = Lt xh (t) + dt0 Lt G(t, t0 ) f (t0 ) = f (t) , (4.77)
−∞
which means that

Lt G(t, t0 ) = δ(t − t0 ) , (4.78)
where δ(t − t0 ) is the Dirac δ-function. Some properties of δ(x):

Zb f (y) if a < y < b

dx f (x) δ(x − y) = (4.79)

0 if y < a or y > b .

a
X δ(x − x )
i ,

δ g(x) = (4.80)
g 0 (x )
i
x with
i
g(x )=0
i
valid for any functions f (x) and g(x). The sum in the second equation is over the zeros xi
of g(x).
Incidentally, the Dirac δ-function enters into the relation between a function and its Fourier
transform, in the following sense. We have
Z∞
dω −iωt ˆ
f (t) = e f (ω) (4.81)
2π
−∞
Z∞
fˆ(ω) = dt e+iωt f (t) . (4.82)
−∞
Substituting the second equation into the first, we have

Z∞ Z∞
dω −iωt 0
f (t) = e dt0 eiωt f (t0 )
2π
−∞ −∞
Z∞ ( Z∞ )
dω iω(t0 −t)
= dt0 e f (t0 ) , (4.83)
2π
−∞ −∞
4.4. GENERAL LINEAR AUTONOMOUS INHOMOGENEOUS ODES 55
which is indeed correct because the term in brackets is a representation of δ(t − t0 ):

Z∞
dω iωs
e = δ(s) . (4.84)
2π
−∞
If the differential equation Lt x(t) = f (t) is defined over some finite t interval with prescribed
boundary conditions on x(t) at the endpoints, then G(t, t0 ) will depend on t and t0 separately.
For the case we are considering, the interval is the entire real line t ∈ (−∞, ∞), and
G(t, t0 ) = G(t − t0 ) is a function of the single variable t − t0 .
d d

Note that Lt = L dt may be considered a function of the differential operator dt . If we
now Fourier transform the equation Lt x(t) = f (t), we obtain
Z∞ Z∞ n
dn−1

iωt iωt d d
dt e f (t) = dt e + an−1 n−1 + . . . + a1 + a0 x(t) (4.85)
dtn dt dt
−∞ −∞
Z∞ (
iωt n n−1
= dt e (−iω) + an−1 (−iω) + . . . + a1 (−iω) + a0 x(t) ,
−∞
where we integrate by parts on t, assuming the boundary terms at t = ±∞ vanish, i.e.

x(±∞) = 0, so that, inside the t integral,
k " k #
d d
eiωt x(t) → − eiωt x(t) = (−iω)k eiωt x(t) . (4.86)
dt dt
Thus, if we define
n
X
L̂(ω) = ak (−iω)k , (4.87)
k=0
then we have
L̂(ω) x̂(ω) = fˆ(ω) , (4.88)
where an ≡ 1. According to the Fundamental Theorem of Algebra, the nth degree poly-
nomial L̂(ω) may be uniquely factored over the complex ω plane into a product over n
roots:
L̂(ω) = (−i)n (ω − ω1 )(ω − ω2 ) · · · (ω − ωn ) . (4.89)
∗
If the {ak } are all real, then L̂(ω) = L̂(−ω ∗ ), hence if Ω is a root then so is −Ω ∗ . Thus,

the roots appear in pairs which are symmetric about the imaginary axis. I.e. if Ω = a + ib
is a root, then so is −Ω ∗ = −a + ib.
The general solution to the homogeneous equation is

n
X
xh (t) = Ai e−iωi t , (4.90)
i=1
which involves n arbitrary complex constants Ai . The susceptibility, or Green’s function in

Fourier space, Ĝ(ω) is then
1 in
Ĝ(ω) = = , (4.91)
L̂(ω) (ω − ω1 )(ω − ω2 ) · · · (ω − ωn )
and the general solution to the inhomogeneous equation is again given by

Z∞
x(t) = xh (t) + dt0 G(t − t0 ) f (t0 ) , (4.92)
−∞
where xh (t) is the solution to the homogeneous equation, i.e. with zero forcing, and where
Z∞
dω −iωs
G(s) = e Ĝ(ω)
2π
−∞
Z∞
dω e−iωs
= in
2π (ω − ω1 )(ω − ω2 ) · · · (ω − ωn )
−∞
n
X e−iωj s
= Θ(s) , (4.93)
j=1 i L0 (ωj )
where we assume that Im ωj < 0 for all j. The integral above was done using Cauchy’s the-
orem and the calculus of residues – a beautiful result from the theory of complex functions.
As an example, consider the familiar case
L̂(ω) = ω02 − 2iβω − ω 2

= −(ω − ω+ ) (ω − ω− ) , (4.94)
with ω± = −iβ ± ν, and ν = (ω02 − β 2 )1/2 . This yields
L0 (ω± ) = ∓(ω+ − ω− ) = ∓2ν . (4.95)
Then according to equation 4.93,

( )
e−iω+ s e−iω− s
G(s) = + Θ(s)
iL0 (ω+ ) iL0 (ω− )
e−βs e−iνs e−βs eiνs

= + Θ(s)
−2iν 2iν
= ν −1 e−βs sin(νs) Θ(s) , (4.96)
exactly as before.
4.5. KRAMERS-KRÖNIG RELATIONS (ADVANCED MATERIAL) 57
4.5 Kramers-Krönig Relations (advanced material)
Suppose χ̂(ω) ≡ Ĝ(ω) is analytic in the UHP1 . Then for all ν, we must have
Z∞
dν χ̂(ν)
=0, (4.97)
2π ν − ω + i
−∞
where is a positive infinitesimal. The reason is simple: just close the contour in the UHP,
assuming χ̂(ω) vanishes sufficiently rapidly that Jordan’s lemma can be applied. Clearly
this is an extremely weak restriction on χ̂(ω), given the fact that the denominator already
causes the integrand to vanish as |ω|−1 .
Let us examine the function

1 ν−ω i
= 2 2
− . (4.98)
ν − ω + i (ν − ω) + (ν − ω)2 + 2
which we have separated into real and imaginary parts. Under an integral sign, the first
term, in the limit → 0, is equivalent to taking a principal part of the integral. That is, for
any function F (ν) which is regular at ν = ω,
Z∞ Z∞
dν ν−ω dν F (ν)
lim F (ν) ≡ P . (4.99)
→0 2π (ν − ω)2 + 2 2π ν − ω
−∞ −∞
The principal part symbol P means that the singularity at ν = ω is elided, either by
smoothing out the function 1/(ν−) as above, or by simply cutting out a region of integration
of width on either side of ν = ω.
The imaginary part is more interesting. Let us write

h(u) ≡ . (4.100)
u2 + 2
For |u| , h(u) ' /u2 , which vanishes as → 0. For u = 0, h(0) = 1/ which diverges as
→ 0. Thus, h(u) has a huge peak at u = 0 and rapidly decays to 0 as one moves off the
peak in either direction a distance greater that . Finally, note that
Z∞
du h(u) = π , (4.101)
−∞
a result which itself is easy to show using contour integration. Putting it all together, this
tells us that

lim = πδ(u) . (4.102)
→0 u2 + 2
1
In this section, we use the notation χ̂(ω) for the susceptibility, rather than Ĝ(ω)
Thus, for positive infinitesimal ,

1 1
= P ∓ iπδ(u) , (4.103)
u ± i u
a most useful result.
We now return to our initial result 4.97, and we separate χ̂(ω) into real and imaginary
parts:
χ̂(ω) = χ̂0 (ω) + iχ̂00 (ω) . (4.104)
(In this equation, the primes do not indicate differentiation with respect to argument.) We
therefore have, for every real value of ω,
Z∞
dν h 0 ih 1 i
0= χ (ν) + iχ00 (ν) P − iπδ(ν − ω) . (4.105)
2π ν−ω
−∞
Taking the real and imaginary parts of this equation, we derive the Kramers-Krönig rela-
tions:
Z∞ 00
dν χ̂ (ν)
χ0 (ω) = +P (4.106)
π ν−ω
−∞
Z∞ 0
dν χ̂ (ν)
χ00 (ω) = −P . (4.107)
π ν−ω
−∞
Chapter 5
Calculus of Variations
5.1 Snell’s Law
Warm-up problem: You are standing at point (x1 , y1 ) on the beach and you want to get to
a point (x2 , y2 ) in the water, a few meters offshore. The interface between the beach and
the water lies at x = 0. What path results in the shortest travel time? It is not a straight
line! This is because your speed v1 on the sand is greater than your speed v2 in the water.
The optimal path actually consists of two line segments, as shown in Fig. 5.1. Let the path
pass through the point (0, y) on the interface. Then the time T is a function of y:
1 1
q q
T (y) = x21 + (y − y1 )2 + x22 + (y2 − y)2 . (5.1)
v1 v2
To find the minimum time, we set
dT 1 y − y1 1 y2 − y
=0= q + q
dy v1 x2 + (y − y )2 v2 x2 + (y − y)2
1 1 2 2
sin θ1 sin θ2
= − . (5.2)
v1 v2
Thus, the optimal path satisfies

sin θ1 v1
= , (5.3)
sin θ2 v2
which is known as Snell’s Law.
Snell’s Law is familiar from optics, where the speed of light in a polarizable medium is
written v = c/n, where n is the index of refraction. In terms of n,
n1 sin θ1 = n2 sin θ2 . (5.4)
59
60 CHAPTER 5. CALCULUS OF VARIATIONS
Figure 5.1: The shortest path between (x1 , y1 ) and (x2 , y2 ) is not a straight line, but rather
two successive line segments of different slope.
If there are several interfaces, Snell’s law holds at each one, so that
ni sin θi = ni+1 sin θi+1 , (5.5)
at the interface between media i and i + 1.
Now let us imagine that there are many such interfaces between regions of very small
thicknesses. We can then regard n and θ as continuous functions of the coordinate x. The
differential form of Snell’s law is

n(x) sin θ(x) = n(x + dx) sin θ(x + dx)
= (n + n0 dx) sin θ + cos θ θ0 dx

= n sin θ + n0 sin θ + n cos θ θ0 dx .

(5.6)
Thus,
dθ 1 dn
ctn θ=− . (5.7)
dx n dx
If we write the path as y = y(x), then tan θ = y 0 , and
d y 00
θ0 = tan−1 y 0 = , (5.8)
dx 1 + y02
which yields
1 y 00 n0
− · = . (5.9)
y0 1 + y02 n
This is a differential equation that y(x) must satisfy if the functional
Z Zx2
ds 1
q
dx n(x) 1 + y 0 2

T y(x) = = (5.10)
v c
x1
5.2. FUNCTIONS AND FUNCTIONALS 61
Figure 5.2: The path of shortest length is composed of three line segments. The relation
between the angles at each interface is governed by Snell’s Law.
is to be minimized.
5.2 Functions and Functionals
A function is a mathematical object which takes a real (or complex) variable, or several
such variables, and returns a real (or complex) number. A functional is a mathematical
object which takes an entire function and returns a number. In the case at hand, we have
Zx2
dx L(y, y 0 , x) ,

T y(x) = (5.11)
x1
where the function L(y, y 0 , x) is given by

q
L(y, y 0 , x) = c−1 n(x) 1 + y02 . (5.12)
Here n(x) is a given function characterizing the medium, and y(x) is the path whose time
is to be evaluated.
In ordinary calculus, we extremize a function f (x) by demanding that f not change to

lowest order when we change x → x + dx:
f (x + dx) = f (x) + f 0 (x) dx + 21 f 00 (x) (dx)2 + . . . . (5.13)
We say that x = x∗ is an extremum when f 0 (x∗ ) = 0.
For a functional, the first functional variation is obtained by sending y(x) → y(x) + δy(x),
Figure 5.3: A path y(x) and its variation y(x) + δy(x).
and extracting the variation in the functional to order δy. Thus, we compute
Zx2
dx L(y + δy, y 0 + δy 0 , x)

T y(x) + δy(x) =
x1
Zx2
∂L ∂L
δy + 0 δy 0 + O (δy)2

= dx L +
∂y ∂y
x1
Zx2
∂L ∂L d
= T y(x) + dx δy + 0 δy
∂y ∂y dx
x1
Zx2 " # x2
∂L d ∂L ∂L
= T y(x) + dx − δy + δy . (5.14)

∂y dx ∂y 0 ∂y 0
x1 x1
Now one very important thing about the variation δy(x) is that it must vanish at the
endpoints: δy(x1 ) = δy(x2 ) = 0. This is because the space of functions under consideration
satisfy fixed boundary conditions y(x1 ) = y1 and y(x2 ) = y2 . Thus, the last term in the
above equation vanishes, and we have
Zx2 " #
∂L d ∂L
δT = dx − δy . (5.15)
∂y dx ∂y 0
x1
We say that the first functional derivative of T with respect to y(x) is

" #
δT ∂L d ∂L
= − , (5.16)
δy(x) ∂y dx ∂y 0
x
where the subscript indicates

that
the expression inside the square brackets is to be evaluated
at x. The functional T y(x) is extremized when its first functional derivative vanishes,
5.3. EXAMPLES FROM THE CALCULUS OF VARIATIONS 63
which results in a differential equation for y(x),

∂L d ∂L
− =0, (5.17)
∂y dx ∂y 0
known as the Euler-Lagrange equation. Since L is independent of y, we have
" #
n y0

d ∂L 1 d
0= =
dx ∂y 0
p
c dx 1 + y02
n0 y0 n y 00
= p + . (5.18)
c 1 + y02 c 1 + y 0 2 3/2
We thus recover the second order equation in 5.9. However, note that the above equation
directly gives
n(x) sin θ(x) = const. , (5.19)
which follows from the relation y 0 = tan θ. For y(x) we obtain
n2 y 0 2 2 dy α
0 2 ≡ α = const. ⇒
dx
=p . (5.20)
1+y n (x) − α2
2
In general, we may expand a functional F [y + δy] in a functional Taylor series,

Z Z Z
F [y + δy] = F [y] + dx1 K1 (x1 ) δy(x1 ) + 21! dx1 dx2 K2 (x1 , x2 ) δy(x1 ) δy(x2 )
Z Z Z
+ 31! dx1 dx2 dx3 K3 (x1 , x2 , x3 ) δy(x1 ) δy(x2 ) δy(x3 ) + . . . (5.21)
and we write
δ nF
Kn (x1 , . . . , xn ) ≡ (5.22)
δy(x1 ) · · · δy(xn )
for the nth functional derivative.
5.3 Examples from the Calculus of Variations
Here we present three useful examples of variational calculus as applied to problems in

mathematics and physics.
5.3.1 Example 1 : minimal surface of revolution
Consider a surface formed by rotating the function y(x) about the x-axis. The area is then
Zx2
s 2
dy
A y(x) = dx 2πy 1 + , (5.23)
dx
x1
p
and is a functional of the curve y(x). Thus we can define L(y, y 0 ) = 2πy 1 + y 0 2 and make
the identification y(x) ↔ q(t). We can then apply what we have derived for the mechanical
action, with L = L(q, q̇, t), mutatis mutandis. Thus, the equation of motion is

d ∂L ∂L
0
= , (5.24)
dx ∂y ∂y
which is a second order ODE for y(x). Rather than treat the second order equation, though,
we can integrate once to obtain a first order equation, by noticing that

d 0 ∂L 00 ∂L 0 d ∂L ∂L ∂L 0 ∂L
y 0
−L =y 0
+y 0
− 0 y 00 − y −
dx ∂y ∂y dx ∂y ∂y ∂y ∂x

d ∂L ∂L ∂L
= y0 − − . (5.25)
dx ∂y 0 ∂y ∂x
In the second line above, the term in square brackets vanishes, thus
∂L dJ ∂L
J = y0 −L ⇒ =− , (5.26)
∂y 0 dx ∂x
and when L has no explicit x-dependence, J is conserved. One finds
y02 2πy
q
J = 2πy · p − 2πy 1 + y02 = − p . (5.27)
1+y 02 1 + y02
Solving for y 0 , s
2πy 2

dy
=± −1 , (5.28)
dx J
J
which may be integrated with the substitution y = 2π cosh χ, yielding

x−a
y(x) = b cosh , (5.29)
b
J
where a and b = 2π are constants of integration. Note there are two such constants, as
the original equation was second order. This shape is called a catenary. As we shall later
find, it is also the shape of a uniformly dense rope hanging between two supports, under
the influence of gravity. To fix the constants a and b, we invoke the boundary conditions
y(x1 ) = y1 and y(x2 ) = y2 .
Consider the case where −x1 = x2 ≡ x0 and y1 = y2 ≡ y0 . Then clearly a = 0, and we have
x
0
y0 = b cosh ⇒ γ = κ−1 cosh κ , (5.30)
b
with γ ≡ y0 /x0 and κ ≡ x0 /b. One finds that for any γ > 1.5089 there are two solutions,
one of which is a local minimum and one of which is a saddle point of A[y(x)]. The solution
Figure 5.4: Minimal surface solution, with y(x) = b cosh(x/b) and y(x0 ) = y0 . Top panel:
A/2πy02 vs. y0 /x0 . Bottom panel: sech(x0 /b) vs. y0 /x0 . The blue curve corresponds to a
local minimum of A[y(x)], and the red curve to a saddle point.
with the smaller value of κ (i.e. the larger value of sech κ) yields the smaller value of A, as
shown in Fig. 5.4. Note that
y cosh(x/b)
= , (5.31)
y0 cosh(x0 /b)
so y(x = 0) = y0 sech(x0 /b).
When extremizing functions that are defined over a finite or semi-infinite interval, one
must take care to evaluate the function at the boundary, for it may be that the boundary
yields a global extremum even though the derivative may not vanish there. Similarly, when
extremizing functionals, one must investigate the functions at the boundary of function
space. In this case, such a function would be the discontinuous solution, with


 y1 if x = x1





y(x) = 0 if x1 < x < x2 (5.32)






y2 if x = x2 .

This solution corresponds to a surface consisting of two discs of radii y1 and y2 , joined
by an infinitesimally thin thread. The area functional evaluated for this particular y(x)
is clearly A = π(y12 + y22 ). In Fig. 5.4, we plot A/2πy02 versus the parameter γ = y0 /x0 .
For γ > γc ≈ 1.564, one of the catenary solutions is the global minimum. For γ < γc , the
minimum area is achieved by the discontinuous solution.
Note that the functional derivative,

( )
2π 1 + y 0 2 − yy 00

δA ∂L d ∂L
K1 (x) = = − = , (5.33)
δy(x) ∂y dx ∂y 0 (1 + y 0 2 )3/2
indeed vanishes for the catenary solutions, but does not vanish for the discontinuous solu-
tion, where K1 (x) = 2π throughout the interval (−x0 , x0 ). Since y = 0 on this interval, y
cannot be decreased. The fact that K1 (x) > 0 means that increasing y will result in an
increase in A, so the boundary value for A, which is 2πy02 , is indeed a local minimum.
We furthermore see in Fig. 5.4 that for γ < γ∗ ≈ 1.5089 the local minimum and saddle
are no longer present. This is the familiar saddle-node bifurcation, here in function space.
Thus, for γ ∈ [0, γ∗ ) there are no extrema of A[y(x)], and the minimum area occurs for the
discontinuous y(x) lying at the boundary of function space. For γ ∈ (γ∗ , γc ), two extrema
exist, one of which is a local minimum and the other a saddle point. Still, the area is
minimized for the discontinuous solution. For γ ∈ (γc , ∞), the local minimum is the global
minimum, and has smaller area than for the discontinuous solution.
5.3.2 Example 2 : geodesic on a surface of revolution
We use cylindrical coordinates (ρ, φ, z) on the surface z = z(ρ). Thus,
ds2 = dρ2 + ρ2 dφ2 + dx2

n 2 o
= 1 + z 0 (ρ) dρ + ρ2 dφ2 ,

(5.34)

and the distance functional D φ(ρ) is
Zρ2
dρ L(φ, φ0 , ρ) ,

D φ(ρ) = (5.35)
ρ1
where q
L(φ, φ0 , ρ) = 1 + z 0 2 (ρ) + ρ2 φ0 2 (ρ) . (5.36)
The Euler-Lagrange equation is

∂L d ∂L ∂L
− =0 ⇒ = const. (5.37)
∂φ dρ ∂φ0 ∂φ0
Thus,
∂L ρ2 φ0
= =a, (5.38)
∂φ0
p
1 + z 0 2 + ρ2 φ0 2
where a is a constant. Solving for φ0 , we obtain
q 2
a 1 + z 0 (ρ)
dφ = p dρ , (5.39)
ρ ρ2 − a2
which we must integrate to find φ(ρ), subject to boundary conditions φ(ρi ) = φi , with
i = 1, 2.
On a cone, z(ρ) = λρ, and we have

r
p dρ p ρ2
dφ = a 1 + λ2 = 1 + λ2 d tan−1 −1 , (5.40)
a2
p
ρ ρ2 − a2
which yields r
p
−1 ρ2
φ(ρ) = β + 1+ λ2 tan −1 , (5.41)
a2
which is equivalent to
φ−β
ρ cos √ =a. (5.42)
1 + λ2
The constants β and a are determined from φ(ρi ) = φi .
5.3.3 Example 3 : brachistochrone
Problem: find the path between (x1 , y1 ) and (x2 , y2 ) which a particle sliding frictionlessly
and under constant gravitational acceleration will traverse in the shortest time. To solve
this we first must invoke some elementary mechanics. Assuming the particle is released
from (x1 , y1 ) at rest, energy conservation says
1 2
2 mv − mgy = mgy1 . (5.43)
Then the time, which is a functional of the curve y(x), is

Zx2 Zx2
s
ds 1 1 + y02
T y(x) = =√ dx (5.44)
v 2g y1 − y
x1 x1
Zx2
≡ dx L(y, y 0 , x) ,
x1
with s
1 + y02
L(y, y 0 , x) = . (5.45)
2g(y1 − y)
Since L is independent of x, eqn. 5.25, we have that
∂L h
02
i−1/2
J = y0 − L = − 2g (y 1 − y) 1 + y (5.46)
∂y 0
is conserved. This yields r
y1 − y
dx = − dy , (5.47)
2a − y1 + y
with a = (4gJ 2 )−1 . This may be integrated parametrically, writing
y1 − y = 2a sin2 ( 21 θ) ⇒ dx = 2a sin2 ( 12 θ) dθ , (5.48)
which results in the parametric equations

x − x1 = a θ − sin θ (5.49)
y − y1 = −a (1 − cos θ) . (5.50)
This curve is known as a cycloid.
5.3.4 Ocean waves
Surface waves in fluids propagate with a definite relation between their angular frequency
ω and their wavevector k = 2π/λ, where λ is the wavelength. The dispersion relation is a
function ω = ω(k). The group velocity of the waves is then v(k) = dω/dk.
In a fluid with a flat bottom at depth h, the dispersion relation turns out to be
√
p  gh k shallow (kh 1)

ω(k) = gk tanh kh ≈ (5.51)
√

gk deep (kh 1) .
Suppose we are in the shallow case, where the wavelength λ is significantly greater than
the depth h of the fluid. This is the case for ocean waves which break at√the shore. The
phase velocity and group velocity are then identical, and equal to v(h) = gh. The waves
propagate more slowly as they approach the shore.
Let us choose the following coordinate system: x represents the distance parallel to the
shoreline, y the distance perpendicular to the shore (which lies at y = 0), and h(y) is the
depth profile of the bottom. We assume h(y) to be a slowly varying function of y which
satisfies h(0) = 0. Suppose a disturbance in the ocean at position (x2 , y2 ) propagates until
it reaches the shore at (x1 , y1 = 0). The time of propagation is
Zx2
s
1 + y02
Z
ds
T y(x) = = dx . (5.52)
v g h(y)
x1
We thus identify the integrand

s
1 + y02
L(y, y 0 , x) = . (5.53)
g h(y)
As with the brachistochrone problem, to which this bears an obvious resemblance, L is
cyclic in the independent variable x, hence
∂L 2 −1/2
h i
J = y 0 0 − L = − g h(y) 1 + y 0 (5.54)
∂y
√
Figure 5.5: For shallow water waves, v = gh. To minimize the propagation time from a
source to the shore, the waves break parallel to the shoreline.
is constant. Solving for y 0 (x), we have

r
dy a
tan θ = = −1 , (5.55)
dx h(y)
where a = (gJ )−1 is a constant, and where θ is the local slope of the function y(x). Thus,
we conclude that near y = 0, where h(y) → 0, the waves come in parallel to the shoreline.
If h(y) = αy has a linear profile, the solution is again a cycloid, with
x(θ) = b (θ − sin θ) (5.56)
y(θ) = b (1 − cos θ) , (5.57)
where b = 2a/α and where the shore lies at θ = 0. Expanding in a Taylor series in θ for
small θ, we may eliminate θ and obtain y(x) as
1/3 1/3 2/3
y(x) = 29 b x + ... . (5.58)
A tsunami is a shallow water wave that manages propagates in deep water. This requires
λ > h, as we’ve seen, which means the disturbance must have a very long spatial extent out
in the open ocean, where h ∼ 10 km. An undersea earthquake is the only possible source;
the characteristic length of √
earthquake fault lines can be hundreds of kilometers. If we take
h = 10 km, we obtain v = gh ≈ 310 m/s or 1100 km/hr. At these speeds, a tsunami can
cross the Pacific Ocean in less than a day.
√
As the wave approaches the shore, it must slow down, since v = gh is diminishing. But
energy is conserved, which means that the amplitude must concomitantly rise. In extreme
cases, the water level rise at shore may be 20 meters or more.
5.4 Appendix : More on Functionals
We remarked in section 5.2 that a function f is an animal which gets fed a real number x
and excretes a real number f (x). We say f maps the reals to the reals, or
f: R→R (5.59)
Of course we also have functions g : C → C which eat and excrete complex numbers,
multivariable functions h : RN → R which eat N -tuples of numbers and excrete a single
number, etc.
A functional F [f (x)] eats entire functions (!) and excretes numbers. That is,
n o
F : f (x) x ∈ R → R (5.60)
This says that F operates on the set of real-valued functions of a single real variable, yielding
a real number. Some examples:
Z∞
2
F [f (x)] = 12 dx f (x)

(5.61)
−∞
Z∞ Z∞
F [f (x)] = 1
2 dx dx0 K(x, x0 ) f (x) f (x0 ) (5.62)
−∞ −∞
Z∞ 2
df
F [f (x)] = dx 12 A f 2 (x) + 12 B . (5.63)
dx
−∞
In classical mechanics, the action S is a functional of the path q(t):
Ztb n o
S[q(t)] = dt 21 mq̇ 2 − U (q) . (5.64)
ta
We can also have functionals which feed on functions of more than one independent variable,
such as
Ztb Zxb ( 2 2 )
∂y ∂y
S[y(x, t)] = dt dx 12 µ − 12 τ , (5.65)
∂t ∂x
ta xa
which happens to be the functional for a string of mass density µ under uniform tension τ .
Another example comes from electrodynamics:
Z Z
1 1
S[Aµ (x, t)] = − d3x dt Fµν F µν + jµ Aµ , (5.66)
16π c
which is a functional of the four fields {A0 , A1 , A2 , A3 }, where A0 = cφ. These are the
components of the 4-potential, each of which is itself a function of four independent variables
5.4. APPENDIX : MORE ON FUNCTIONALS 71
Figure 5.6: A functional S[q(t)] is the continuum limit of a function of a large number of
variables, S(q1 , . . . , qM ).
(x0 , x1 , x2 , x3 ), with x0 = ct. The field strength tensor is written in terms of derivatives of
the Aµ : Fµν = ∂µ Aν − ∂ν Aµ , where we use a metric gµν = diag(+, −, −, −) to raise and
lower indices. The 4-potential couples linearly to the source term Jµ , which is the electric
4-current (cρ, J ).
We extremize functions by sending the independent variable x to x + dx and demanding

that the variation df = 0 to first order in dx. That is,
f (x + dx) = f (x) + f 0 (x) dx + 12 f 00 (x)(dx)2 + . . . , (5.67)
whence df = f 0 (x) dx + O (dx)2 and thus

f 0 (x∗ ) = 0 ⇐⇒ x∗ an extremum. (5.68)
We extremize functionals by sending
f (x) → f (x) + δf (x) (5.69)
and demanding that the variation δF in the functional F [f (x)] vanish to first order in δf (x).
The variation δf (x) must sometimes satisfy certain boundary conditions. For example, if
F [f (x)] only operates on functions which vanish at a pair of endpoints, i.e. f (xa ) = f (xb ) =
0, then when we extremize the functional F we must do so within the space of allowed
functions. Thus, we would in this case require δf (xa ) = δf (xb ) = 0. We may expand the
functional F [f + δf ] in a functional Taylor series,
Z Z Z
1
F [f + δf ] = F [f ] + dx1 K1 (x1 ) δf (x1 ) + 2 ! dx1 dx2 K2 (x1 , x2 ) δf (x1 ) δf (x2 )
Z Z Z
1
+ 3 ! dx1 dx2 dx3 K3 (x1 , x2 , x3 ) δf (x1 ) δf (x2 ) δf (x3 ) + . . . (5.70)
and we write
δ nF
Kn (x1 , . . . , xn ) ≡ . (5.71)
δf (x1 ) · · · δf (xn )
In a more general case, F = F [{fi (x)} is a functional of several functions, each of which is
a function of several independent variables.1 We then write
Z
F [{fi + δfi }] = F [{fi }] + dx1 K1i (x1 ) δfi (x1 )
Z Z
+ 2 ! dx1 dx2 K2ij (x1 , x2 ) δfi (x1 ) δfj (x2 )
1
Z Z Z
+ 3 ! dx1 dx2 dx3 K3ijk (x1 , x2 , x3 ) δfi (x1 ) δfj (x2 ) δfk (x3 ) + . . . , (5.72)
1
with
i i2 ···in δ nF
Kn1 (x1 , x2 , . . . , xn ) = . (5.73)
δfi (x1 ) δfi (x2 ) δfi (xn )
1 2 n
Another way to compute functional derivatives is to send
f (x) → f (x) + 1 δ(x − x1 ) + . . . + n δ(x − xn ) (5.74)
and then differentiate n times with respect to 1 through n . That is,

δ nF ∂n
= F f (x) + 1 δ(x − x1 ) + . . . + n δ(x − xn ) . (5.75)

δf (x1 ) · · · δf (xn ) ∂1 · · · ∂n = =··· =0

1 2 n
Let’s see how this works. As an example, we’ll take the action functional from classical
mechanics,
Ztb n o
S[q(t)] = dt 12 mq̇ 2 − U (q) . (5.76)
ta
To compute the first functional derivative, we replace the function q(t) with q(t)+ δ(t−t1 ),
and expand in powers of :
Ztb n o
S q(t) + δ(t − t1 ) = S[q(t)] + dt m q̇ δ 0 (t − t1 ) − U 0 (q) δ(t − t1 )

ta
n o
= − m q̈(t1 ) + U 0 q(t1 ) , (5.77)
hence
δS n o
= − m q̈(t) + U 0 q(t) (5.78)
δq(t)
1
It may be also be that different functions depend on a different number of independent variables. E.g.
F = F [f (x), g(x, y), h(x, y, z)].
and setting the first functional derivative to zero yields Newton’s Second Law, mq̈ = −U 0 (q),
for all t ∈ [ta , tb ]. Note that we have used the result
Z∞
dt δ 0 (t − t1 ) h(t) = −h0 (t1 ) , (5.79)
−∞
which is easily established upon integration by parts.
To compute the second functional derivative, we replace
q(t) → q(t) + 1 δ(t − t1 ) + 2 δ(t − t2 ) (5.80)
and extract the term of order 1 2 in the double Taylor expansion. One finds this term to
be
Ztb n o
1 2 dt m δ 0 (t − t1 ) δ 0 (t − t2 ) − U 00 (q) δ(t − t1 ) δ(t − t2 ) . (5.81)
ta
Note that we needn’t bother with terms proportional to 21 or 22 since the recipe is to
differentiate once with respect to each of 1 and 2 and then to set 1 = 2 = 0. This
procedure uniquely selects the term proportional to 1 2 , and yields
δ2S n o
= − m δ 00 (t1 − t2 ) + U 00 q(t1 ) δ(t1 − t2 ) .

(5.82)
δq(t1 ) δq(t2 )
In multivariable calculus, the stability of an extremum is assessed by computing the matrix

of second derivatives at the extremal point, known as the Hessian matrix. One has
∂ 2f

∂f
=0 ∀i ; Hij = . (5.83)
∂xi x∗ ∂xi ∂xj x∗
The eigenvalues of the Hessian Hij determine the stability of the extremum. Since Hij is
a symmetric matrix, its eigenvectors η α may be chosen to be orthogonal. The associated
eigenvalues λα , defined by the equation
Hij ηjα = λα ηiα , (5.84)
are the respective curvatures in the directions η α , where α ∈ {1, . . . , n} where n is the
number of variables. The extremum is a local minimum if all the eigenvalues λα are positive,
a maximum if all are negative, and otherwise is a saddle point. Near a saddle point, there
are some directions in which the function increases and some in which it decreases.
In the case of functionals, the second functional derivative K2 (x1 , x2 ) defines an eigenvalue
problem for δf (x):
Zxb
dx2 K2 (x1 , x2 ) δf (x2 ) = λ δf (x1 ) . (5.85)
xa
In general there are an infinite number of solutions to this equation which form a basis in
function space, subject to appropriate boundary conditions at xa and xb . For example, in
the case of the action functional from classical mechanics, the above eigenvalue equation
becomes a differential equation,
d2

00 ∗

− m 2 + U q (t) δq(t) = λ δq(t) , (5.86)
dt
where q ∗ (t) is the solution to the Euler-Lagrange equations. As with the case of ordinary
multivariable functions, the functional extremum is a local minimum (in function space)
if every eigenvalue λα is positive, a local maximum if every eigenvalue is negative, and a
saddle point otherwise.
Consider the simple harmonic oscillator, for which U (q) = 12 mω02 q 2 . Then U 00 q ∗ (t) =

m ω02 ; note that we don’t even need to know the solution q ∗ (t) to obtain the second functional
derivative in this special case. The eigenvectors obey m(δ q̈ + ω02 δq) = −λ δq, hence
q
δq(t) = A cos ω02 + (λ/m) t + ϕ , (5.87)
where A and ϕ are constants. Demanding δq(ta ) = δq(tb ) = 0 requires
q
ω02 + (λ/m) tb − ta ) = nπ , (5.88)
where n is an integer. Thus, the eigenfunctions are

t − ta
δqn (t) = A sin nπ · , (5.89)
t b − ta
and the eigenvalues are

nπ 2
λn = m − mω02 , (5.90)
T
where T = tb − ta . Thus, so long as T > π/ω0 , there is at least one negative eigenvalue.
(n+1)π
Indeed, for nπ
ω0 < T < ω0 there will be n negative eigenvalues. This means the action
is generally not a minimum, but rather lies at a saddle point in the (infinite-dimensional)
function space.
To test this explicitly, consider a harmonic oscillator with the boundary conditions q(0) = 0
and q(T ) = Q. The equations of motion, q̈ + ω02 q = 0, along with the boundary conditions,
determine the motion,
Q sin(ω0 t)
q ∗ (t) = . (5.91)
sin(ω0 T )
The action for this path is then
ZT n o
∗
S[q (t)] = dt 21 m q̇ ∗2 − 12 mω02 q ∗2
0
ZT n
m ω02 Q2 o
= dt cos2 ω0 t − sin2 ω0 t
2 sin2 ω0 T
0
1 2
= 2 mω0 Q ctn (ω0 T ) . (5.92)
Next consider the path q(t) = Q t/T which satisfies the boundary conditions but does not
satisfy the equations of motion (it proceeds with constant velocity). One finds the action
for this path is
!
1 2 1 1
S[q(t)] = 2 mω0 Q − 3 ω0 T . (5.93)
ω0 T
Thus, provided ω0 T 6= nπ, in the limit T → ∞ we find that the constant velocity path has
lower action.
Finally, consider the general mechanical action,
Ztb

S q(t) = dt L(q, q̇, t) . (5.94)
ta
We now evaluate the first few terms in the functional Taylor series:
Ztb (
∂L ∂L
S q ∗ (t) + δq(t) = ∗ ∗

dt L(q , q̇ , t) + δqi + δ q̇i (5.95)
∂qi ∗ ∂ q̇i ∗
ta q q
)
1 ∂ 2L ∂ 2L 1 ∂ 2L
+ δqi δqj + δqi δ q̇j + δ q̇i δ q̇j + . . . .
2 ∂qi ∂qj ∗ ∂qi ∂ q̇j ∗ 2 ∂ q̇i ∂ q̇j ∗
q q q
To identify the functional derivatives, we integrate by parts. Let Φ... (t) be an arbitrary
function of time. Then

Ztb Ztb
dt Φi (t) δ q̇i (t) = − dt Φ̇i (t) δqi (t) (5.96)
ta ta
Ztb Ztb Ztb
d
dt Φij (t) δqi (t) δ q̇j (t) = dt dt0 Φij (t) δ(t − t0 ) 0 δqi (t) δqj (t0 )
dt
ta ta ta
Ztb Ztb
= dt dt0 Φij (t)) δ 0 (t − t0 ) δqi (t) δqj (t0 ) (5.97)
ta ta
Ztb Ztb Ztb
d d
dt Φij (t) dq̇i (t) δ q̇j (t) = dt dt0 Φij (t) δ(t − t0 ) δqi (t) δqj (t0 )
dt dt0
ta ta ta
Ztb Ztb
= − dt dt Φ̇ij (t) δ (t − t ) + Φij (t) δ (t − t ) δqi (t) δqj (t0 ) .
0 0 0 00 0
ta ta
(5.98)
Thus,
" #
δS ∂L d ∂L
= − (5.99)
δqi (t) ∂qi dt ∂ q̇i
q ∗ (t)
(
δ 2S ∂ 2L ∂ 2L
= δ(t − t0 ) − δ 00 (t − t0 )

0
δqi (t) δqj (t ) ∂qi ∂qj ∗ ∂ q̇i ∂ q̇j ∗
q (t) q (t)
" # )
∂ 2L ∂ 2L

d
+ 2 − δ 0 (t − t0 ) . (5.100)
∂qi ∂ q̇j dt ∂ q̇i ∂ q̇j ∗q (t)
Chapter 6
Lagrangian Mechanics
6.1 Generalized Coordinates
A set of generalized coordinates q1 , . . . , qn completely describes the positions of all particles

in a mechanical system. In a system with df degrees of freedom and k constraints, n = df −k
independent generalized coordinates are needed to completely specify all the positions. A
constraint is a relation among coordinates, such as x2 + y 2 + z 2 = a2 for a particle moving
on a sphere of radius a. In this case, df = p3 and k = 1. In this case, we could eliminate z
in favor of x and y, i.e. by writing z = ± a2 − x2 − y 2 , or we could choose as coordinates
the polar and azimuthal angles θ and φ.
For the moment we will assume that n = df − k, and that the generalized coordinates are
independent, satisfying no additional constraints among them. Later on we will learn how
to deal with any remaining constraints among the {q1 , . . . , qn }.
The generalized coordinates may have units of length, or angle, or perhaps something totally
different. In the theory of small oscillations, the normal coordinates are conventionally
chosen to have units of (mass)1/2 ×(length). However, once a choice of generalized coordinate
is made, with a concomitant set of units, the units of the conjugate momentum and force
are determined:
M L2 1 M L2 1
pσ = · , Fσ = · , (6.1)
T qσ T2 qσ

where A means ‘the units of A’, and where M , L, and T stand for mass, length, and time,
respectively. Thus, if qσ has dimensions of length, then pσ has dimensions of momentum
and Fσ has dimensions of force. If qσ is dimensionless, as is the case for an angle, pσ has
dimensions of angular momentum (M L2 /T ) and Fσ has dimensions of torque (M L2 /T 2 ).
77
78 CHAPTER 6. LAGRANGIAN MECHANICS
6.2 Hamilton’s Principle
The equations of motion of classical mechanics are embodied in a variational principle,

called Hamilton’s principle. Hamilton’s principle states that the motion of a system is such
that the action functional
Zt2

S q(t) = dt L(q, q̇, t) (6.2)
t1
is an extremum, i.e. δS = 0. Here, q = {q1 , . . . , qn } is a complete set of generalized

coordinates for our mechanical system, and
L=T −U (6.3)
is the Lagrangian, where T is the kinetic energy and U is the potential energy. Setting the
first variation of the action to zero gives the Euler-Lagrange equations,
momentum pσ force F
σ
z }| { z}|{
d ∂L ∂L
= . (6.4)
dt ∂ q̇σ ∂qσ
Thus, we have the familiar ṗσ = Fσ , also known as Newton’s second law. Note, however,
that the {qσ } are generalized coordinates, so pσ may not have dimensions of momentum,
nor Fσ of force. For example, if the generalized coordinate in question is an angle φ, then
the corresponding generalized momentum is the angular momentum about the axis of φ’s
rotation, and the generalized force is the torque.
6.2.1 Momentum conservation
∂L
Whenever L is independent of a generalized coordinate qσ , the conjugate force Fσ = ∂qσ
vanishes and therefore the conjugate momentum pσ = ∂∂L
q̇σ is conserved. This is an example
of a deep result known as Noether’s theorem which we will explore more fully next week.
Noether’s theorem guarantees that to every continuous symmetry of L there corresponds
an associated conserved quantity.
6.2.2 Invariance of the equations of motion
Suppose
d
L̃(q, q̇, t) = L(q, q̇, t) + G(q, t) . (6.5)
dt
Then
S̃[q(t)] = S[q(t)] + G(qb , tb ) − G(qa , ta ) . (6.6)
Since the difference S̃ − S is a function only of the endpoint values {qa , qb }, their variations
are identical: δ S̃ = δS. This means that L and L̃ result in the same equations of motion.
6.2. HAMILTON’S PRINCIPLE 79
Thus, the equations of motion are invariant under a shift of L by a total time derivative of
a function of coordinates and time.
6.2.3 Remarks on the order of the equations of motion
The equations of motion are second order in time. This follows from the fact that L =
L(q, q̇, t). Using the chain rule,
∂ 2L ∂ 2L ∂ 2L

d ∂L
= q̈σ0 + q̇σ0 + . (6.7)
dt ∂ q̇σ ∂ q̇σ ∂ q̇σ0 ∂ q̇σ ∂qσ0 ∂ q̇σ ∂t
That the equations are second order in time can be regarded as an empirical fact. It follows,
as we have just seen, from the fact that L depends on q and on q̇, but on no higher time
derivative terms. Suppose the Lagrangian did depend on the generalized accelerations q̈ as
well. What would the equations of motion look like?
Taking the variation of S,

Ztb tb
∂L ∂L d ∂L
δ dt L(q, q̇, q̈, t) = δqσ + δ q̇σ − δqσ
∂ q̇σ ∂ q̈σ dt ∂ q̈σ ta
ta
Ztb ( )
d2 ∂L

∂L d ∂L
+ dt − + 2 δqσ . (6.8)
∂qσ dt ∂ q̇σ dt ∂ q̈σ
ta
The boundary term vanishes if we require δqσ (ta ) = δqσ (tb ) = δ q̇σ (ta ) = δ q̇σ (tb ) = 0 ∀ σ.
The equations of motion would then be fourth order in time.
6.2.4 Lagrangian for a free particle
For a free particle, we can use Cartesian coordinates for each particle as our system of
generalized coordinates. For a single particle, the Lagrangian L(x, v, t) must be a function
solely of v 2 . This is because homogeneity with respect to space and time preclude any
dependence of L on x or on t, and isotropy of space means L must depend on v 2 . We
next invoke Galilean relativity, which says that the equations of motion are invariant under
transformation to a reference frame moving with constant velocity. Let V be the velocity
of the new reference frame K0 relative to our initial reference frame K. Then x0 = x − V t,
and v 0 = v − V . In order that the equations of motion be invariant under the change in
reference frame, we demand
d
L0 (v 0 ) = L(v) + G(x, t) . (6.9)
dt
The only possibility is L = 12 mv 2 , where the constant m is the mass of the particle. Note:
d 1 dG
L0 = 12 m(v − V )2 = 12 mv 2 + 2 mV 2
t − mV · x =L+ . (6.10)
dt dt
For N interacting particles,

N dx 2
a
X
1

L= 2 ma − U {xa }, {ẋa } . (6.11)
dt
a=1
Here, U is the potential energy. Generally, U is of the form

X X
U= U1 (xa ) + v(xa − xa0 ) , (6.12)
a a<a0
however, as we shall see, velocity-dependent potentials appear in the case of charged parti-
cles interacting with electromagnetic fields. In general, though,
L=T −U , (6.13)
where T is the kinetic energy, and U is the potential energy.
6.3 Remarks on the Choice of Generalized Coordinates
Any choice of generalized coordinates will yield an equivalent set of equations of motion.
However, some choices result in an apparently simpler set than others. This is often true
with respect to the form of the potential energy. Additionally, certain constraints that may
be present are more amenable to treatment using a particular set of generalized coordinates.
The kinetic energy T is always simple to write in Cartesian coordinates, and it is good
practice, at least when one is first learning the method, to write T in Cartesian coordinates
and then convert to generalized coordinates. In Cartesian coordinates, the kinetic energy
of a single particle of mass m is
T = 21 m ẋ2 + ẏ 2 + ẋ2 .

(6.14)
If the motion is two-dimensional, and confined to the plane z = const., one of course has
T = 21 m ẋ2 + ẏ 2 .

Two other commonly used coordinate systems are the cylindrical and spherical systems. In
cylindrical coordinates (ρ, φ, z), ρ is the radial coordinate in the (x, y) plane and φ is the
azimuthal angle:
x = ρ cos φ ẋ = cos φ ρ̇ − ρ sin φ φ̇ (6.15)

y = ρ sin φ ẏ = sin φ ρ̇ + ρ cos φ φ̇ , (6.16)
and the third, orthogonal coordinate is of course z. The kinetic energy is
T = 12 m ẋ2 + ẏ 2 + ẋ2

= 12 m ρ̇2 + ρ2 φ̇2 + ż 2 .

(6.17)
6.4. HOW TO SOLVE MECHANICS PROBLEMS 81
When the motion is confined to a plane with z = const., this coordinate system is often
referred to as ‘two-dimensional polar’ coordinates.
In spherical coordinates (r, θ, φ), r is the radius, θ is the polar angle, and φ is the azimuthal
angle. On the globe, θ would be the ‘colatitude’, which is θ = π2 − λ, where λ is the latitude.
I.e. θ = 0 at the north pole. In spherical polar coordinates,
x = r sin θ cos φ ẋ = sin θ cos φ ṙ + r cos θ cos φ θ̇ − r sin θ sin φ φ̇ (6.18)

y = r sin θ sin φ ẏ = sin θ sin φ ṙ + r cos θ sin φ θ̇ + r sin θ cos φ φ̇ (6.19)
z = r cos θ ż = cos θ ṙ − r sin θ θ̇ . (6.20)
The kinetic energy is
T = 21 m ẋ2 + ẏ 2 + ż 2

= 12 m ṙ2 + r2 θ̇2 + r2 sin2 θ φ̇2 .

(6.21)
6.4 How to Solve Mechanics Problems
Here are some simple steps you can follow toward obtaining the equations of motion:
1. Choose a set of generalized coordinates {q1 , . . . , qn }.
2. Find the kinetic energy T (q, q̇, t), the potential energy U (q, t), and the Lagrangian
L(q, q̇, t) = T − U . It is often helpful to first write the kinetic energy in Cartesian
coordinates for each particle before converting to generalized coordinates.
∂L ∂L
3. Find the canonical momenta pσ = ∂ q̇σ and the generalized forces Fσ = ∂qσ .
4. Evaluate the time derivatives ṗσ and write the equations of motion ṗσ = Fσ . Be careful
to differentiate properly, using the chain rule and the Leibniz rule where appropriate.
5. Identify any conserved quantities (more about this later).
6.5 Examples
6.5.1 One-dimensional motion
For a one-dimensional mechanical system with potential energy U (x),
L = T − U = 12 mẋ2 − U (x) . (6.22)
The canonical momentum is

∂L
p= = mẋ (6.23)
∂ ẋ
and the equation of motion is

d ∂L ∂L
= ⇒ mẍ = −U 0 (x) , (6.24)
dt ∂ ẋ ∂x
which is of course F = ma.
Note that we can multiply the equation of motion by ẋ to get

n o d n1 o dE
0 = ẋ mẍ + U 0 (x) = mẋ2
+ U (x) = , (6.25)
dt 2 dt
where E = T + U .
6.5.2 Central force in two dimensions
Consider next a particle of mass m moving in two dimensions under p the influence of a
potential U (ρ) which is a function of the distance from the origin ρ = x2 + y 2 . Clearly
cylindrical (2d polar) coordinates are called for:
L = 21 m ρ̇2 + ρ2 φ̇2 − U (ρ) .

(6.26)

d ∂L ∂L
= ⇒ mρ̈ = mρ φ̇2 − U 0 (ρ) (6.27)
dt ∂ ρ̇ ∂ρ

d ∂L ∂L d
mρ2 φ̇ = 0 .

= ⇒ (6.28)
dt ∂ φ̇ ∂φ dt
Note that the canonical momentum conjugate to φ, which is to say the angular momentum,
is conserved:
pφ = mρ2 φ̇ = const. (6.29)
We can use this to eliminate φ̇ from the first Euler-Lagrange equation, obtaining
p2φ
mρ̈ = − U 0 (ρ) . (6.30)
mρ3
We can also write the total energy as
E = 12 m ρ̇2 + ρ2 φ̇2 + U (ρ)

p2φ
= 12 m ρ̇2 + + U (ρ) , (6.31)
2mρ2
from which it may be shown that E is also a constant:
dE p2φ 0

= m ρ̈ − + U (ρ) ρ̇ = 0 . (6.32)
dt mρ3
We shall discuss this case in much greater detail in the coming weeks.
6.5. EXAMPLES 83
Figure 6.1: A wedge of mass M and opening angle α slides frictionlessly along a horizontal
surface, while a small object of mass m slides frictionlessly along the wedge.
6.5.3 A sliding point mass on a sliding wedge
Consider the situation depicted in Fig. 6.1, in which a point object of mass m slides
frictionlessly along a wedge of opening angle α. The wedge itself slides frictionlessly along a
horizontal surface, and its mass is M . We choose as generalized coordinates the horizontal
position X of the left corner of the wedge, and the horizontal distance x from the left corner
to the sliding point mass. The vertical coordinate of the sliding mass is then y = x tan α,
where the horizontal surface lies at y = 0. With these generalized coordinates, the kinetic
energy is
T = 12 M Ẋ 2 + 12 m (Ẋ + ẋ)2 + 12 mẏ 2

= 12 (M + m)Ẋ 2 + mẊ ẋ + 12 m (1 + tan2 α) ẋ2 . (6.33)
The potential energy is simply
U = mgy = mg x tan α . (6.34)
Thus, the Lagrangian is
L = 21 (M + m)Ẋ 2 + mẊ ẋ + 12 m (1 + tan2 α) ẋ2 − mg x tan α , (6.35)
and the equations of motion are

d ∂L ∂L
= ⇒ (M + m)Ẍ + m ẍ = 0
dt ∂ Ẋ ∂X

d ∂L ∂L
= ⇒ mẌ + m (1 + tan2 α) ẍ = −mg tan α . (6.36)
dt ∂ ẋ ∂x
At this point we can use the first of these equations to write

m
Ẍ = − ẍ . (6.37)
M +m
Figure 6.2: The spring–pendulum system.
Substituting this into the second equation, we obtain the constant accelerations
(M + m)g sin α cos α mg sin α cos α
ẍ = − , Ẍ = . (6.38)
M + m sin2 α M + m sin2 α
6.5.4 A pendulum attached to a mass on a spring
Consider next the system depicted in Fig. 6.2 in which a mass M moves horizontally while
attached to a spring of spring constant k. Hanging from this mass is a pendulum of arm
length ` and bob mass m.
A convenient set of generalized coordinates is (x, θ), where x is the displacement of the mass
M relative to the equilibrium extension a of the spring, and θ is the angle the pendulum
arm makes with respect to the vertical. Let the Cartesian coordinates of the pendulum bob
be (x1 , y1 ). Then
x1 = a + x + ` sin θ , y1 = −l cos θ . (6.39)
T = 21 M ẋ2 + 12 m (ẋ2 + ẏ 2 )
h i
= 12 M ẋ2 + 12 m (ẋ + ` cos θ θ̇)2 + (` sin θ θ̇)2
= 21 (M + m) ẋ2 + 12 m`2 θ̇2 + m` cos θ ẋ θ̇ , (6.40)
and the potential energy is
U = 12 kx2 + mgy1
= 12 kx2 − mg` cos θ . (6.41)
6.5. EXAMPLES 85
Thus,
L = 12 (M + m) ẋ2 + 12 m`2 θ̇2 + m` cos θ ẋ θ̇ − 21 kx2 + mg` cos θ . (6.42)
The canonical momenta are

∂L
px = = (M + m) ẋ + m` cos θ θ̇
∂ ẋ
∂L
pθ = = m` cos θ ẋ + m`2 θ̇ , (6.43)
∂ θ̇
and the canonical forces are
∂L
Fx = = −kx
∂x
∂L
Fθ = = −m` sin θ ẋ θ̇ − mg` sin θ . (6.44)
∂θ
The equations of motion then yield

(M + m) ẍ + m` cos θ θ̈ − m` sin θ θ̇2 = −kx (6.45)
2
m` cos θ ẍ + m` θ̈ = −mg` sin θ . (6.46)
Small Oscillations : If we assume both x and θ are small, we may write sin θ ≈ θ and
cos θ ≈ 1, in which case the equations of motion may be linearized to
(M + m) ẍ + m` θ̈ + kx = 0 (6.47)
2
m` ẍ + m` θ̈ + mg` θ = 0 . (6.48)
If we define
x m k g
u≡ , α≡ , ω02 ≡ , ω12 ≡ , (6.49)
` M M `
then
(1 + α) ü + α θ̈ + ω02 u = 0 (6.50)
ü + θ̈ + ω12 θ =0. (6.51)
We can solve by writing
u(t) a
= e−iωt , (6.52)
θ(t) b
in which case
ω02 − (1 + α) ω 2 −α ω 2

a 0
= . (6.53)
−ω 2 ω12 − ω 2 b 0
In order to have a nontrivial solution (i.e. without a = b = 0), the determinant of the above
2 × 2 matrix must vanish. This gives a condition on ω 2 , with solutions
q 2
2
= 21 ω02 + (1 + α) ω12 ± 12 ω02 − ω12 + 2α (ω02 + ω12 ) ω12 .

ω± (6.54)
Figure 6.3: The double pendulum, with generalized coordinates θ1 and θ2 . All motion is
confined to a single plane.
6.5.5 The double pendulum
As yet another example of the generalized coordinate approach to Lagrangian dynamics,

consider the double pendulum system, sketched in Fig. 6.3. We choose as generalized
coordinates the two angles θ1 and θ2 . In order to evaluate the Lagrangian, we must obtain
the kinetic and potential energies in terms of the generalized coordinates {θ1 , θ2 } and their
corresponding velocities {θ̇1 , θ̇2 }.
In Cartesian coordinates,
T = 21 m1 (ẋ21 + ẏ12 ) + 12 m2 (ẋ22 + ẏ22 ) (6.55)

U = m1 g y1 + m2 g y2 . (6.56)
We therefore express the Cartesian coordinates {x1 , y1 , x2 , y2 } in terms of the generalized

coordinates {θ1 , θ2 }:
x1 = `1 sin θ1 x2 = `1 sin θ1 + `2 sin θ2 (6.57)

y1 = −`1 cos θ1 y2 = −`1 cos θ1 − `2 cos θ2 . (6.58)
Thus, the velocities are
ẋ1 = `1 θ̇1 cos θ1 ẋ2 = `1 θ̇1 cos θ1 + `2 θ̇2 cos θ2 (6.59)

ẏ1 = `1 θ̇1 sin θ1 ẏ2 = `1 θ̇1 sin θ1 + `2 θ̇2 sin θ2 . (6.60)
Thus,
n o
T = 21 m1 `21 θ̇12 + 12 m2 `21 θ̇12 + 2`1 `2 cos(θ1 − θ2 ) θ̇1 θ̇2 + `22 θ̇22 (6.61)
U = −m1 g `1 cos θ1 − m2 g `1 cos θ1 − m2 g `2 cos θ2 , (6.62)

6.5. EXAMPLES 87
and
L = T − U = 12 (m1 + m2 ) `21 θ̇12 + m2 `1 `2 cos(θ1 − θ2 ) θ̇1 θ̇2 + 21 m2 `22 θ̇22
+ (m1 + m2 ) g `1 cos θ1 + m2 g `2 cos θ2 . (6.63)
The generalized (canonical) momenta are

∂L
p1 = = (m1 + m2 ) `21 θ̇1 + m2 `1 `2 cos(θ1 − θ2 ) θ̇2 (6.64)
∂ θ̇1
∂L
p2 = = m2 `1 `2 cos(θ1 − θ2 ) θ̇1 + m2 `22 θ̇2 , (6.65)
∂ θ̇2
and the equations of motion are
ṗ1 = (m1 + m2 ) `21 θ̈1 + m2 `1 `2 cos(θ1 − θ2 ) θ̈2 − m2 `1 `2 sin(θ1 − θ2 ) (θ̇1 − θ̇2 ) θ̇2
∂L
= −(m1 + m2 ) g `1 sin θ1 − m2 `1 `2 sin(θ1 − θ2 ) θ̇1 θ̇2 = (6.66)
∂θ1
and
ṗ2 = m2 `1 `2 cos(θ1 − θ2 ) θ̈1 − m2 `1 `2 sin(θ1 − θ2 ) (θ̇1 − θ̇2 ) θ̇1 + m2 `22 θ̈2

∂L
= −m2 g `2 sin θ2 + m2 `1 `2 sin(θ1 − θ2 ) θ̇1 θ̇2 = . (6.67)
∂θ2
We therefore find
m2 `2 m2 `2
`1 θ̈1 + cos(θ1 − θ2 ) θ̈2 + sin(θ1 − θ2 ) θ̇22 + g sin θ1 = 0 (6.68)
m1 + m2 m1 + m2
`1 cos(θ1 − θ2 ) θ̈1 + `2 θ̈2 − `1 sin(θ1 − θ2 ) θ̇12 + g sin θ2 = 0 . (6.69)
Small Oscillations : The equations of motion are coupled, nonlinear second order ODEs.
When the system is close to equilibrium, the amplitudes of the motion are small, and we
may expand in powers of the θ1 and θ2 . The linearized equations of motion are then
θ̈1 + αβ θ̈2 + ω02 θ1 = 0 (6.70)
θ̈1 + β θ̈2 + ω02 θ2 = 0 , (6.71)
where we have defined

m2 `2 g
α≡ , β≡ , ω02 ≡ . (6.72)
m1 + m2 `1 `1
We can solve this coupled set of equations by a nifty trick. Let’s take a linear combination
of the first equation plus an undetermined coefficient, r, times the second:
(1 + r) θ̈1 + (α + r)β θ̈2 + ω02 (θ1 + r θ2 ) = 0 . (6.73)
We now demand that the ratio of the coefficients of θ2 and θ1 is the same as the ratio of
the coefficients of θ̈2 and θ̈1 :
(α + r) β p
=r ⇒ r± = 12 (β − 1) ± 1
2 (1 − β)2 + 4αβ (6.74)
1+r
When r = r± , the equation of motion may be written
d2 ω02
2
θ 1 + r θ
± 2 = − θ 1 + r± θ 2 (6.75)
dt 1 + r±
and defining the (unnormalized) normal modes

ξ± ≡ θ1 + r± θ2 , (6.76)
we find
ξ¨± + ω±
2
ξ± = 0 , (6.77)
with
ω0
ω± = p . (6.78)
1 + r±
Thus, by switching to the normal coordinates, we decoupled the equations of motion, and
identified the two normal frequencies of oscillation. We shall have much more to say about
small oscillations further below.
For example, with `1 = `2 = ` and m1 = m2 = m, we have α = 21 , and β = 1, in which case
√
q r
g
r± = ± √12 , ξ± = θ1 ± √1
2
θ2 , ω± = 2∓ 2 . (6.79)
`
Note that the oscillation frequency for the ‘in-phase’ mode ξ+ is low, and that for the ‘out
of phase’ mode ξ− is high.
6.5.6 The thingy
Four massless rods of length L are hinged together at their ends to form a rhombus. A
particle of mass M is attached to each vertex. The opposite corners are joined by springs
of spring constant k. In the square configuration, the strings are unstretched. The motion
is confined to a plane, and the particles move only along the diagonals of the rhombus.
Introduce suitable generalized coordinates and find the Lagrangian of the system. Deduce
the equations of motion and find the frequency of small oscillations about equilibrium.
6.5. EXAMPLES 89
Solution
The rhombus is depicted in figure 6.4. Let a be the equilibrium length of the springs; clearly
L = √a2 . Let φ be half of one of the opening angles, as shown. Then the masses are located
at (±X, 0) and (0, ±Y ), with X = √a2 cos φ and Y = √a2 sin φ. The spring extensions are
δX = 2X − a and δY = 2Y − a. The kinetic and potential energies are therefore
Figure 6.4: The thingy: a rhombus with opening angles 2φ and π − 2φ.
T = M Ẋ 2 + Ẏ 2

= 12 M a2 φ̇2
and
2 2
U = 12 k δX + 12 k δY
n √ 2 √ 2 o
= 12 ka2 2 cos φ − 1 + 2 sin φ − 1
n √ o
= 12 ka2 3 − 2 2(cos φ + sin φ) .
Note that minimizing U (φ) gives sin φ = cos φ, i.e. φeq = π4 . The Lagrangian is then
√
L = T − U = 12 M a2 φ̇2 + 2 ka2 cos φ + sin φ + const.


d ∂L ∂L √
= ⇒ M a2 φ̈ = 2 ka2 (cos φ − sin φ)
dt ∂ φ̇ ∂φ
π
It’s always smart to expand about equilibrium, so let’s write φ = 4 + δ, which leads to
δ̈ + ω02 sin δ = 0 ,
p
with ω0 = 2k/M . This is the equation of a pendulum! Linearizing gives δ̈ + ω02 δ = 0, so
the small oscillation frequency is just ω0 .
6.6 Conserved Quantities
A conserved quantity Λ(q, q̇, t) is one which does not vary throughout the motion of the
system. This means
dΛ
= 0. (6.80)
dt

q=q(t)
We shall discuss conserved quantities in detail in the chapter on Noether’s Theorem, which
follows.
6.6.1 Momentum conservation
The simplest case of a conserved quantity occurs when the Lagrangian does not explicitly
depend on one or more of the generalized coordinates, i.e. when
∂L
Fσ = =0. (6.81)
∂qσ
We then say that L is cyclic in the coordinate qσ . In this case, the Euler-Lagrange equations
ṗσ = Fσ say that the conjugate momentum pσ is conserved. Consider, for example, the
motion of a particle of mass m near the surface of the earth. Let (x, y) be coordinates
parallel to the surface and z the height. We then have
T = 12 m ẋ2 + ẏ 2 + ż 2

(6.82)
U = mgz (6.83)
L = T − U = 12 m ẋ2 + ẏ 2 + ż 2

− mgz . (6.84)
Since
∂L ∂L
Fx = =0 and Fy = =0, (6.85)
∂x ∂y
we have that px and py are conserved, with
∂L ∂L
px = = mẋ , py = = mẏ . (6.86)
∂ ẋ ∂ ẏ
6.6. CONSERVED QUANTITIES 91
These first order equations can be integrated to yield

px py
x(t) = x(0) + t , y(t) = y(0) + t. (6.87)
m m
The z equation is of course
ṗz = mz̈ = −mg = Fz , (6.88)
with solution
z(t) = z(0) + ż(0) t − 12 gt2 . (6.89)
As another example, consider

p a particle
moving in the (x, y) plane under the influence of
a potential U (x, y) = U x2 + y 2 which depends only on the particle’s distance from
p
the origin ρ = x2 + y 2 . The Lagrangian, expressed in two-dimensional polar coordinates
(ρ, φ), is
L = 12 m ρ̇2 + ρ2 φ̇2 − U (ρ) .

(6.90)
We see that L is cyclic in the angle φ, hence
∂L
pφ = = mρ2 φ̇ (6.91)
∂ φ̇
is conserved. pφ is the angular momentum of the particle about the ẑ axis. In the language
of the calculus of variations, momentum conservation is what follows when the integrand of
a functional is independent of the independent variable.
6.6.2 Energy conservation
When the integrand of a functional is independent of the dependent variable, another con-
servation law follows. For Lagrangian mechanics, consider the expression
n
X
H(q, q̇, t) = pσ q̇σ − L . (6.92)
σ=1
Now we take the total time derivative of H:

n
dH X ∂L ∂L ∂L
= pσ q̈σ + ṗσ q̇σ − q̇σ − q̈σ − . (6.93)
dt ∂qσ ∂ q̇σ ∂t
σ=1
We evaluate Ḣ along the motion of the system, which entails that the terms in the curly
brackets above cancel for each σ:
∂L ∂L
pσ = , ṗσ = . (6.94)
∂ q̇σ ∂qσ
Thus, we find
dH ∂L
=− , (6.95)
dt ∂t
which means that H is conserved whenever the Lagrangian contains no explicit time depen-
dence. For a Lagrangian of the form
X
1 2
L= 2 ma ṙa − U (r1 , . . . , rN ) , (6.96)
a
we have that pa = ma ṙa , and

X
H =T +U = 1
2 ma ṙa2 + U (r1 , . . . , rN ) . (6.97)
a
However, it is not always the case that H = T + U is the total energy, as we shall see in
the next chapter.
6.7 Appendix : Virial Theorem
The virial theorem is a statement about the time-averaged motion of a mechanical system.
Define the virial , X
G(q, p) = pσ q σ . (6.98)
σ
Then
dG X
= ṗσ qσ + pσ q̇σ
dt σ
X X ∂L
= q σ Fσ + q̇σ . (6.99)
σ σ
∂ q̇σ
Now suppose that T = 21 σ,σ0 Tσσ0 q̇σ q̇σ0 is homogeneous of degree k = 2 in q̇, and that U
P
is homogeneous of degree zero in q̇. Then
X ∂L X ∂T
q̇σ = q̇σ = 2 T, (6.100)
σ
∂ q̇σ σ
∂ q̇σ
which follows from Euler’s theorem on homogeneous functions.
Now consider the time average of Ġ over a period τ :

D dG E Zτ
1 dG
= dt
dt τ dt
0
1h i
= G(τ ) − G(0) . (6.101)
τ
If G(t) is bounded, then in the limit τ → ∞ we must have hĠi = 0. Any bounded motion,
such as the orbit of the earth around the Sun, will result in hĠiτ →∞ = 0. But then
D dG E
X
= 2 hT i + q σ Fσ = 0 , (6.102)
dt σ
6.7. APPENDIX : VIRIAL THEOREM 93
which implies
X
DX E ∂U
hT i = − 21 q σ Fσ = + 2 1
qσ
σ σ
∂qσ
D X E
= 21

ri · ∇i U r1 , . . . , rN (6.103)
i
= 12 k hU i , (6.104)
where the last line pertains to homogeneous potentials of degree k. Finally, since T +U = E
is conserved, we have
kE 2E
hT i = , hU i = . (6.105)
k+2 k+2
Chapter 7
Noether’s Theorem
7.1 Continuous Symmetry Implies Conserved Charges
Consider a particle moving in two dimensions under the influence of an external potential
U (r). The potential is a function only of the magnitude of the vector r. The Lagrangian is
then
L = T − U = 21 m ṙ2 + r2 φ̇2 − U (r) ,

(7.1)
where we have chosen generalized coordinates (r, φ). The momentum conjugate to φ is
pφ = m r2 φ̇. The generalized force Fφ clearly vanishes, since L does not depend on the
coordinate φ. (One says that L is ‘cyclic’ in φ.) Thus, although r = r(t) and φ = φ(t)
will in general be time-dependent, the combination pφ = m r2 φ̇ is constant. This is the
conserved angular momentum about the ẑ axis.
If instead the particle moved in a potential U (y), independent of x, then writing
L = 21 m ẋ2 + ẏ 2 − U (y) ,

(7.2)
we have that the momentum px = ∂L/∂ ẋ = mẋ is conserved, because the generalized force
Fx = ∂L/∂x = 0 vanishes. This situation pertains in a uniform gravitational field, with
U (x, y) = mgy, independent of x. The horizontal component of momentum is conserved.
In general, whenever the system exhibits a continuous symmetry, there is an associated

conserved charge. (The terminology ‘charge’ is from field theory.) Indeed, this is a rigorous
result, known as Noether’s Theorem. Consider a one-parameter family of transformations,
qσ −→ q̃σ (q, ζ) , (7.3)
where ζ is the continuous parameter. Suppose further (without loss of generality) that at
ζ = 0 this transformation is the identity, i.e. q̃σ (q, 0) = qσ . The transformation may be
nonlinear in the generalized coordinates. Suppose further that the Lagrangian L s invariant
95
96 CHAPTER 7. NOETHER’S THEOREM
under the replacement q → q̃. Then we must have

d ∂L ∂ q̃ σ
∂L ∂ ˙σ
q̃
0= ˙ t) =
L(q̃, q̃,

+
dζ ∂qσ ∂ζ ∂ q̇σ ∂ζ

ζ=0 ζ=0 ζ=0

d ∂L ∂ q̃σ ∂L d ∂ q̃σ
= +
dt ∂ q̇σ ∂ζ ∂ q̇σ dt ∂ζ ζ=0

ζ=0

d ∂L ∂ q̃σ
= . (7.4)
dt ∂ q̇σ ∂ζ ζ=0
Thus, there is an associated conserved charge

∂L ∂ q̃σ
Λ= . (7.5)
∂ q̇σ ∂ζ

ζ=0
7.1.1 Examples of one-parameter families of transformations
Consider the Lagrangian

p
L = 21 m(ẋ2 + ẏ 2 ) − U

x2 + y 2 . (7.6)
In two-dimensional polar coordinates, we have
L = 12 m(ṙ2 + r2 φ̇2 ) − U (r) , (7.7)
and we may now define
r̃(ζ) = r (7.8)
φ̃(ζ) = φ + ζ . (7.9)
Note that r̃(0) = r and φ̃(0) = φ, i.e. the transformation is the identity when ζ = 0. We
now have
X ∂L ∂ q̃σ ∂L ∂ r̃ ∂L ∂ φ̃
Λ= = + = mr2 φ̇ . (7.10)
∂ q̇ ∂ζ ∂ ṙ ∂ζ ∂ζ

σ σ
ζ=0
∂ φ̇
ζ=0

ζ=0
Another way to derive the same result which is somewhat instructive is to work out the
transformation in Cartesian coordinates. We then have
x̃(ζ) = x cos ζ − y sin ζ (7.11)

ỹ(ζ) = x sin ζ + y cos ζ . (7.12)
Thus,
∂ x̃ ∂ ỹ
= −y(ζ) , = x(ζ) (7.13)
∂ζ ∂ζ
7.2. CONSERVATION OF LINEAR AND ANGULAR MOMENTUM 97
and
∂L ∂ x̃ ∂L ∂ ỹ
Λ= + = m(xẏ − y ẋ) . (7.14)
∂ ẋ ∂ζ ∂ ẏ ∂ζ

ζ=0 ζ=0
But
m(xẏ − y ẋ) = mẑ · r × ṙ = mr2 φ̇ . (7.15)
As another example, consider the potential
U (ρ, φ, z) = V (ρ, aφ + z) , (7.16)
where (ρ, φ, z) are cylindrical coordinates for a particle of mass m, and where a is a constant
with dimensions of length. The Lagrangian is
1 2 2 2 2

2 m ρ̇ + ρ φ̇ + ẋ − V (ρ, aφ + z) . (7.17)
This model possesses a helical symmetry, with a one-parameter family
ρ̃(ζ) = ρ (7.18)
φ̃(ζ) = φ + ζ (7.19)
z̃(ζ) = z − ζa . (7.20)
Note that
aφ̃ + z̃ = aφ + z , (7.21)
so the potential energy, and the Lagrangian as well, is invariant under this one-parameter
family of transformations. The conserved charge for this symmetry is

∂L ∂ ρ̃ ∂L ∂ φ̃ ∂L ∂ z̃
Λ= + + = mρ2 φ̇ − maż . (7.22)
∂ ρ̇ ∂ζ ∂ φ̇ ∂ζ ∂ ż ∂ζ

ζ=0 ζ=0 ζ=0
We can check explicitly that Λ is conserved, using the equations of motion

d ∂L d ∂L ∂V
mρ2 φ̇ =

= = −a (7.23)
dt ∂ φ̇ dt ∂φ ∂z
d ∂L d ∂L ∂V
= (mż) = =− . (7.24)
dt ∂ φ̇ dt ∂φ ∂z
Thus,
d d
mρ2 φ̇ − a (mż) = 0 .

Λ̇ = (7.25)
dt dt
7.2 Conservation of Linear and Angular Momentum
Suppose that the Lagrangian of a mechanical system is invariant under a uniform translation
of all particles in the n̂ direction. Then our one-parameter family of transformations is given
by
x̃a = xa + ζ n̂ , (7.26)
and the associated conserved Noether charge is

X ∂L
Λ= · n̂ = n̂ · P , (7.27)
a
∂ ẋa
P
where P = a pa is the total momentum of the system.
If the Lagrangian of a mechanical system is invariant under rotations about an axis n̂, then
x̃a = R(ζ, n̂) xa

= xa + ζ n̂ × xa + O(ζ 2 ) , (7.28)
where we have expanded the rotation matrix R(ζ, n̂) in powers of ζ. The conserved Noether
charge associated with this symmetry is
X ∂L X
Λ= · n̂ × xa = n̂ · xa × pa = n̂ · L , (7.29)
a
∂ ẋa a
where L is the total angular momentum of the system.
7.3 Advanced Discussion : Invariance of L vs. Invariance of

S
Observant readers might object that demanding invariance of L is too strict. We should
instead be demanding invariance of the action S 1 . Suppose S is invariant under
t → t̃(q, t, ζ) (7.30)
qσ (t) → q̃σ (q, t, ζ) . (7.31)
Then invariance of S means
Ztb Zt̃b
S= dt L(q, q̇, t) = ˙ t) .
dt L(q̃, q̃, (7.32)
ta t̃a
Note that t is a dummy variable of integration, so it doesn’t matter whether we call it t

or t̃. The endpoints of the integral, however, do change under the transformation.
Now
consider an infinitesimal transformation, for which δt = t̃ − t and δq = q̃ t̃ − q(t) are both
small. Invariance of S means
Ztb tb +δtb
Z n ∂L ∂L o
S= dt L(q, q̇, t) = dt L(q, q̇, t) + δ̄qσ + δ̄ q̇σ + . . . , (7.33)
∂qσ ∂ q̇σ
ta ta +δta
1
Indeed, we should be demanding that S only change by a function of the endpoint values.
7.3. ADVANCED DISCUSSION : INVARIANCE OF L VS. INVARIANCE OF S 99
where
δ̄qσ (t) ≡ q̃σ (t) − qσ (t)

= q̃σ t̃ − q̃σ t̃ + q̃σ (t) − qσ (t)
= δqσ − q̇σ δt + O(δq δt) (7.34)
Subtracting the top line from the bottom, we obtain
tb +δtb ( )
Z
∂L ∂L ∂L d ∂L
0 = Lb δtb − La δta + δ̄q − δ̄q + dt − δ̄q(t)
∂ q̇σ b σ,b ∂ q̇σ a σ,a ∂qσ dt ∂ q̇σ
ta +δta
Ztb ( )
d ∂L ∂L
= dt L− q̇ δt + δq . (7.35)
dt ∂ q̇σ σ ∂ q̇σ σ
ta
Thus, if ζ ≡ δζ is infinitesimal, and
δt = A(q, t) δζ (7.36)
δqσ = Bσ (q, t) δζ , (7.37)
then the conserved charge is

∂L ∂L
Λ= L− q̇σ A(q, t) + B (q, t)
∂ q̇σ ∂ q̇σ σ
= − H(q, p, t) A(q, t) + pσ Bσ (q, t) . (7.38)
Thus, when A = 0, we recover our earlier results, obtained by assuming invariance of L.

Note that conservation of H follows from time translation invariance: t → t + ζ, for which
A = 1 and Bσ = 0. Here we have written
H = pσ q̇σ − L , (7.39)
and expressed it in terms of the momenta pσ , the coordinates qσ , and time t. H is called
the Hamiltonian.
7.3.1 The Hamiltonian
The Lagrangian is a function of generalized coordinates, velocities, and time. The canonical
momentum conjugate to the generalized coordinate qσ is
∂L
pσ = . (7.40)
∂ q̇σ
The Hamiltonian is a function of coordinates, momenta, and time. It is defined as the

Legendre transform of L: X
H(q, p, t) = pσ q̇σ − L . (7.41)
σ
Let’s examine the differential of H:
X ∂L ∂L

∂L
dH = q̇σ dpσ + pσ dq̇σ − dqσ − dq̇σ − dt
σ
∂qσ ∂ q̇σ ∂t
X ∂L

∂L
= q̇σ dpσ − dqσ − dt , (7.42)
σ
∂qσ ∂t
where we have invoked the definition of pσ to cancel the coefficients of dq̇σ . Since ṗσ =
∂L/∂qσ , we have Hamilton’s equations of motion,
∂H ∂H
q̇σ = , ṗσ = − . (7.43)
∂pσ ∂qσ
Thus, we can write
X ∂L
dH = q̇σ dpσ − ṗσ dqσ − dt . (7.44)
σ
∂t
Dividing by dt, we obtain
dH ∂L
=− , (7.45)
dt ∂t
which says that the Hamiltonian is conserved (i.e. it does not change with time) whenever
there is no explicit time dependence to L.
Example #1 : For a simple d = 1 system with L = 12 mẋ2 − U (x), we have p = mẋ and
p2
H = p ẋ − L = 21 mẋ2 + U (x) = + U (x) . (7.46)
2m
Example #2 : Consider now the mass point – wedge system analyzed above, with
L = 21 (M + m)Ẋ 2 + mẊ ẋ + 12 m (1 + tan2 α) ẋ2 − mg x tan α , (7.47)
∂L
P = = (M + m) Ẋ + mẋ (7.48)
∂ Ẋ
∂L
p= = mẊ + m (1 + tan2 α) ẋ . (7.49)
∂ ẋ
The Hamiltonian is given by
H = P Ẋ + p ẋ − L
= 12 (M + m)Ẋ 2 + mẊ ẋ + 12 m (1 + tan2 α) ẋ2 + mg x tan α . (7.50)
However, this is not quite H, since H = H(X, x, P, p, t) must be expressed in terms of the
coordinates and the momenta and not the coordinates and velocities. So we must eliminate
Ẋ and ẋ in favor of P and p. We do this by inverting the relations

P M +m m Ẋ
= (7.51)
p m m (1 + tan2 α) ẋ
to obtain
m (1 + tan2 α)

Ẋ 1 −m P
= . (7.52)
−m

ẋ m M + (M + m) tan2 α M +m p
Substituting into 7.50, we obtain
M + m P 2 cos2 α P p cos2 α p2
H= − + + mg x tan α . (7.53)
2m M + m sin2 α M + m sin2 α 2 (M + m sin2 α)
∂L
Notice that Ṗ = 0 since ∂X = 0. P is the total horizontal momentum of the system (wedge
plus particle) and it is conserved.
7.3.2 Is H = T + U ?
The most general form of the kinetic energy is
T = T2 + T1 + T0
(2)
= 12 Tσσ0 (q, t) q̇σ q̇σ0 + Tσ(1) (q, t) q̇σ + T (0) (q, t) , (7.54)
where T (n) (q, q̇, t) is homogeneous of degree n in the velocities2 . We assume a potential
energy of the form
U = U1 + U0
= Uσ(1) (q, t) q̇σ + U (0) (q, t) , (7.55)
which allows for velocity-dependent forces, as we have with charged particles moving in an
electromagnetic field. The Lagrangian is then
(2)
L = T − U = 21 Tσσ0 (q, t) q̇σ q̇σ0 + Tσ(1) (q, t) q̇σ + T (0) (q, t) − Uσ(1) (q, t) q̇σ − U (0) (q, t) . (7.56)
We have assumed U (q, t) is velocity-independent, but the above form for L = T − U is quite
general. (E.g. any velocity-dependence in U can be absorbed into the Bσ q̇σ term.) The
canonical momentum conjugate to qσ is
∂L (2)
pσ = = Tσσ0 q̇σ0 + Tσ(1) (q, t) − Uσ(1) (q, t) (7.57)
∂ q̇σ
2
A homogeneous
Pnfunction of degree k satisfies f (λx1 , . . . , λxn ) = λk f (x1 , . . . , xn ). It is then easy to prove
∂f
Euler’s theorem, i=1 xi ∂xi = kf .
Figure 7.1: A bead of mass m on a rotating hoop of radius a.
which is inverted to give
(2) −1

(1) (1)
q̇σ = Tσσ0 pσ0 − Tσ0 + Uσ0 . (7.58)
The Hamiltonian is then
H = pσ q̇σ − L
(2) −1

(1) (1)
= 1
2 Tσσ0 pσ − Tσ(1) + Uσ(1) pσ 0 − Tσ 0 + Uσ0 − T0 + U0 (7.59)
= T2 − T0 + U0 . (7.60)
If T0 , T1 , and U1 vanish, i.e. if T (q, q̇, t) is a homogeneous function of degree two in the
generalized velocities, and U (q, t) is velocity-independent, then H = T + U . But if T0 or T1
is nonzero, or the potential is velocity-dependent, then H 6= T + U .
7.3.3 Example: A bead on a rotating hoop
Consider a bead of mass m constrained to move along a hoop of radius a. The hoop is
further constrained to rotate with angular velocity ω about the ẑ-axis, as shown in Fig.
7.1.
The most convenient set of generalized coordinates is spherical polar (r, θ, φ), in which case
T = 12 m ṙ2 + r2 θ̇2 + r2 sin2 θ φ̇2

= 12 ma2 θ̇2 + ω 2 sin2 θ .

(7.61)
Thus, T2 = 21 ma2 θ̇2 and T0 = 12 ma2 ω 2 sin2 θ. The potential energy is U (θ) = mga(1−cos θ).
The momentum conjugate to θ is pθ = ma2 θ̇, and thus
H(θ, p) = T2 − T0 + U
= 12 ma2 θ̇2 − 12 ma2 ω 2 sin2 θ + mga(1 − cos θ)
p2θ
= − 1 ma2 ω 2 sin2 θ + mga(1 − cos θ) . (7.62)
2ma2 2
For this problem, we can define the effective potential
Ueff (θ) ≡ U − T0 = mga(1 − cos θ) − 12 ma2 ω 2 sin2 θ

ω2
= mga 1 − cos θ − 2 sin2 θ , (7.63)
2ω0
where ω0 ≡ g/a2 . The Lagrangian may then be written
L = 12 ma2 θ̇2 − Ueff (θ) , (7.64)
and thus the equations of motion are
∂Ueff
ma2 θ̈ = − . (7.65)
∂θ
0 (θ) = 0, which gives

Equilibrium is achieved when Ueff
∂Ueff n ω2 o
= mga sin θ 1 − 2 cos θ = 0 , (7.66)
∂θ ω0
i.e. θ∗ = 0, θ∗ = π, or θ∗ = ± cos−1 (ω02 /ω 2 ), where the last pair of equilibria are present
only for ω 2 > ω02 . The stability of these equilibria is assessed by examining the sign of
00 (θ ∗ ). We have
Ueff
00
n ω2 o
Ueff (θ) = mga cos θ − 2 2 cos2 θ − 1 . (7.67)
ω0
Thus, 
ω2


mga 1 − ω02
at θ∗ = 0






00 ∗ 2
Ueff (θ ) = −mga 1 + ωω2 at θ∗ = π (7.68)
 0




 2 ω02

mga ω22 − ω02 at θ∗ = ± cos−1

.

ω 0
ω ω2
Thus, θ∗ = 0 is stable for ω 2 < ω02 but becomes unstable when the rotation frequency ω
is sufficiently large, i.e. when ω 2 > ω02 . In this regime, there are two new equilibria, at
θ∗ = ± cos−1 (ω02 /ω 2 ), which are both stable. The equilibrium at θ∗ = π is always unstable,
independent of the value of ω. The situation is depicted in Fig. 7.2.
ω2 2

Figure 7.2: The effective potential Ueff (θ) = mga 1 − cos θ − 2ω 2 sin θ . (The dimensionless
0 √
potential Ũeff (x) = Ueff /mga is shown, where x = θ/π.) Left panels: ω = 12 3 ω0 . Right
√
panels: ω = 3 ω0 .
7.4 Charged Particle in a Magnetic Field
Consider next the case of a charged particle moving in the presence of an electromagnetic
field. The particle’s potential energy is
q
U (r) = q φ(r, t) − A(r, t) · ṙ , (7.69)
c
which is velocity-dependent. The kinetic energy is T = 12 m ṙ 2 , as usual. Here φ(r) is the

scalar potential and A(r) the vector potential. The electric and magnetic fields are given
by
1 ∂A
E = −∇φ − , B =∇×A . (7.70)
c ∂t
The canonical momentum is

∂L q
p= = m ṙ + A , (7.71)
∂ ṙ c
7.4. CHARGED PARTICLE IN A MAGNETIC FIELD 105
and hence the Hamiltonian is
H(r, p, t) = p · ṙ − L
q q
= mṙ 2 + A · ṙ − 12 m ṙ 2 − A · ṙ + q φ
c c
= 12 m ṙ 2 + q φ
1 q 2
= p − A(r, t) + q φ(r, t) . (7.72)
2m c
If A and φ are time-independent, then H(r, p) is conserved.
Let’s work out the equations of motion. We have

!
d ∂L ∂L
= (7.73)
dt ∂ ṙ ∂r
which gives
q dA q
m r̈ + = −q ∇φ + ∇(A · ṙ) , (7.74)
c dt c
or, in component notation,
q ∂Ai q ∂Ai ∂φ q ∂Aj

m ẍi + ẋj + = −q + ẋ , (7.75)
c ∂xj c ∂t ∂xi c ∂xi j
which is to say
∂φ q ∂Ai q ∂Aj ∂Ai
m ẍi = −q − + − ẋj . (7.76)
∂xi c ∂t c ∂xi ∂xj
It is convenient to express the cross product in terms of the completely antisymmetric tensor
of rank three, ijk :
∂Ak
Bi = ijk , (7.77)
∂xj
and using the result
ijk imn = δjm δkn − δjn δkm , (7.78)
we have ijk Bi = ∂j Ak − ∂k Aj , and
∂φ q ∂Ai q
m ẍi = −q − + ijk ẋj Bk , (7.79)
∂xi c ∂t c
or, in vector notation,
q ∂A q
m r̈ = −q ∇φ − + ṙ × (∇ × A)
c ∂t c
q
= q E + ṙ × B , (7.80)
c
which is, of course, the Lorentz force law.
7.5 Fast Perturbations : Rapidly Oscillating Fields
Consider a free particle moving under the influence of an oscillating force,
mq̈ = F sin ωt . (7.81)
The motion of the system is then
F sin ωt
q(t) = qh (t) − , (7.82)
mω 2
where qh (t) = A + Bt is the solution to the homogeneous (unforced) equation of motion.

Note that the amplitude of the response q − qh goes as ω −2 and is therefore small when ω
is large.
Now consider a general n = 1 system, with
H(q, p, t) = H0 (q, p) + V (q) sin(ωt + δ) . (7.83)
We assume that ω is much greater than any natural oscillation frequency associated with
H0 . We separate the motion q(t) and p(t) into slow and fast components:
q(t) = q̄(t) + ζ(t) (7.84)
p(t) = p̄(t) + π(t) , (7.85)
where ζ(t) and π(t) oscillate with the driving frequency ω. Since ζ and π will be small, we
expand Hamilton’s equations in these quantities:
∂H0 ∂ 2H0 ∂ 2H0 1 ∂ 3H0 2 ∂ 3H0 1 ∂ 3H0 2

q̄˙ + ζ̇ = + π + ζ + ζ + ζπ + π + ... (7.86)
∂ p̄ ∂ p̄2 ∂ q̄ ∂ p̄ 2 ∂ q̄ 2 ∂ p̄ ∂ q̄ ∂ p̄2 2 ∂ p̄3
∂H0 ∂ 2H0 ∂ 2H0 1 ∂ 3H0 2 ∂ 3H0 1 ∂ 3H0 2
p̄˙ + π̇ = − − ζ − π − ζ − ζπ − π
∂ q̄ ∂ q̄ 2 ∂ q̄ ∂ p̄ 2 ∂ q̄ 3 ∂ q̄ 2 ∂ p̄ 2 ∂ q̄ ∂ p̄2
∂V ∂ 2V
− sin(ωt + δ) − 2 ζ sin(ωt + δ) − . . . . (7.87)
∂ q̄ ∂ q̄
We now average over the fast degrees of freedom to obtain an equation of motion for the slow
variables q̄ and p̄, which we here carry to lowest nontrivial order in averages of fluctuating
quantities:
∂H0 1 ∂ 3H0
2 ∂ 3H0
1 ∂ 3H0
2
q̄˙ = + ζ + ζπ + π (7.88)
∂ p̄ 2 ∂ q̄ 2 ∂ p̄ ∂ q̄ ∂ p̄2 2 ∂ p̄3
∂H0 1 ∂ 3H0
2 ∂ 3H0
1 ∂ 3H0
2 ∂ 2V

p̄˙ = − − 3
ζ − 2
ζπ − 2
π − 2 ζ sin(ωt + δ) . (7.89)
∂ q̄ 2 ∂ q̄ ∂ q̄ ∂ p̄ 2 ∂ q̄ ∂ p̄ ∂ q̄
7.5. FAST PERTURBATIONS : RAPIDLY OSCILLATING FIELDS 107
The fast degrees of freedom obey

∂ 2H0 ∂ 2H0
ζ̇ = ζ+ π (7.90)
∂ q̄ ∂ p̄ ∂ p̄2
∂ 2H0 ∂ 2H0 ∂V
π̇ = − 2
ζ − π− sin(ωt + δ) . (7.91)
∂ q̄ ∂ q̄ ∂ p̄ ∂q
Let us analyze the coupled equations3

ζ̇ = A ζ + B π (7.92)
π̇ = −C ζ − A π + F e−iωt . (7.93)
The solution is of the form
ζ α
= e−iωt . (7.94)
π β
Plugging in, we find
BF BF
= − 2 + O ω −4

α= 2 2
(7.95)
BC − A − ω ω
(A + iω)F iF −3

β=− = + O ω . (7.96)
BC − A2 − ω 2 ω
Taking the real part, and restoring the phase shift δ, we have
−BF 1 ∂V ∂ 2H0
ζ(t) = sin(ωt + δ) = sin(ωt + δ) (7.97)
ω2 ω 2 ∂ q̄ ∂ p̄2
F 1 ∂V
π(t) = − cos(ωt + δ) = cos(ωt + δ) . (7.98)
ω ω ∂ q̄
The desired averages, to lowest order, are thus
∂V 2 ∂ 2H0 2

2 1
ζ = (7.99)
2ω 4 ∂ q̄ ∂ p̄2
∂V 2

2 1
π = (7.100)
2ω 2 ∂ q̄

1 ∂V ∂ 2H0
ζ sin(ωt + δ) = , (7.101)
2ω 2 ∂ q̄ ∂ p̄2

along with ζπ = 0.
Finally, we substitute the averages into the equations of motion for the slow variables q̄ and
p̄, resulting in the time-independent effective Hamiltonian
1 ∂ 2H0 ∂V 2

K(q̄, p̄) = H0 (q̄, p̄) + 2 , (7.102)
4ω ∂ p̄2 ∂ q̄
3
With real coefficients A, B, and C, one can always take the real part to recover the fast variable equations
of motion.
and the equations of motion
∂K ∂K
q̄˙ = , p̄˙ = − . (7.103)
∂ p̄ ∂ q̄
7.5.1 Example : pendulum with oscillating support
Consider a pendulum with a vertically oscillating point of support. The coordinates of the
pendulum bob are
x = ` sin θ , y = a(t) − ` cos θ . (7.104)
The Lagrangian is easily obtained:
L = 12 m`2 θ̇2 + mà̇ θ̇ sin θ + mg` cos θ + 12 mȧ2 − mga (7.105)

these may be dropped
z }| {
1 2 2 1 2 d
= 2 m` θ̇ + m(g + ä)` cos θ+ 2 mȧ − mga − mà̇ sin θ . (7.106)
dt
Thus we may take the Lagrangian to be
L̄ = 12 m`2 θ̇2 + m(g + ä)` cos θ , (7.107)
from which we derive the Hamiltonian
p2θ
H(θ, pθ , t) = − mg` cos θ − mà̈ cos θ (7.108)
2m`2
= H0 (θ, pθ , t) + V1 (θ) sin ωt . (7.109)
We have assumed a(t) = a0 sin ωt, so
V1 (θ) = mà0 ω 2 cos θ . (7.110)
The effective Hamiltonian, per eqn. 7.102, is

p̄θ
K(θ̄, p̄θ ) = − mg` cos θ̄ + 14 m a20 ω 2 sin2 θ̄ . (7.111)
2m`2
Let’s define the dimensionless parameter
2g`
≡ . (7.112)
ω 2 a20
The slow variable θ̄ executes motion in the effective potential Veff (θ̄) = mg` v(θ̄), with
1
v(θ̄) = − cos θ̄ + sin2 θ̄ . (7.113)
2
7.6. FIELD THEORY: SYSTEMS WITH SEVERAL INDEPENDENT VARIABLES 109
Figure 7.3: Dimensionless potential v(θ) for = 1.5 (black curve) and = 0.5 (blue curve).
Differentiating, and dropping the bar on θ, we find that Veff (θ) is stationary when
v 0 (θ) = 0 ⇒ sin θ cos θ = − sin θ . (7.114)
Thus, θ = 0 and θ = π, where sin θ = 0, are equilibria. When < 1 (note > 0 always),
there are two new solutions, given by the roots of cos θ = −.
To assess stability of these equilibria, we compute the second derivative:

1
v 00 (θ) = cos θ + cos 2θ . (7.115)

From this, we see that θ = 0 is stable (i.e. v 00 (θ = 0) > 0) always, but θ = π is stable for
< 1 and unstable for > 1. When < 1, two new solutions appear, at cos θ = −, for
which
1
v 00 (cos−1 (−)) = − , (7.116)

which is always negative since < 1 in order for these equilibria to exist. The situation is
sketched in fig. 7.3, showing v(θ) for two representative values of the parameter . For > 1,
the equilibrium at θ = π is unstable, but as decreases, a subcritical pitchfork bifurcation is
encountered at = 1, and θ = π becomes stable, while the outlying θ = cos−1 (−) solutions
are unstable.
7.6 Field Theory: Systems with Several Independent Vari-

ables
Suppose φa (x) depends on several independent variables: {x1 , x2 , . . . , xn }. Furthermore,

suppose Z

S {φa (x) = dx L(φa ∂µ φa , x) , (7.117)
Ω
i.e. the Lagrangian density L is a function of the fields φa and their partial derivatives
∂φa /∂xµ . Here Ω is a region in RK . Then the first variation of S is
Z ( )
∂L ∂L ∂ δφa
δS = dx δφ +
∂φa a ∂(∂µ φa ) ∂xµ
Ω
I Z ( )
∂L ∂L ∂ ∂L
= dΣ nµ δφ − dx − δφa , (7.118)
∂(∂µ φa ) a ∂φa ∂xµ ∂(∂µ φa )
∂Ω Ω
where ∂Ω is the (n − 1)-dimensional boundary of Ω,

dΣ is the differential
surface area, and
nµ is the unit normal. If we demand ∂L/∂(∂µ φa )∂Ω = 0 of δφa ∂Ω = 0, the surface term
vanishes, and we conclude

δS ∂L ∂ ∂L
= − . (7.119)
δφa (x) ∂φa ∂xµ ∂(∂µ φa )
As an example, consider the case of a stretched string of linear mass density µ and tension
τ . The action is a functional of the height y(x, t), where the coordinate along the string, x,
and time, t, are the two independent variables. The Lagrangian density is
2 2
∂y ∂y
L = 12 µ − 12 τ , (7.120)
∂t ∂x
whence the Euler-Lagrange equations are

δS ∂ ∂L ∂ ∂L
0= =− −
δy(x, t) ∂x ∂y 0 ∂t ∂ ẏ
∂ 2y ∂ 2y
=τ − µ , (7.121)
∂x2 ∂t2
∂y
where y 0 = ∂x and ẏ = ∂y 00
∂t . Thus, µÿ = τ y , which is the Helmholtz equation. We’ve
assumed boundary conditions where δy(xa , t) = δy(xb , t) = δy(x, ta ) = δy(x, tb ) = 0.
The Lagrangian density for an electromagnetic field with sources is

1
L = − 16π Fµν F µν − 1c jµ Aµ . (7.122)
The equations of motion are then

∂L ∂ ∂L 4π ν
− =0 ⇒ ∂µ F µν = j , (7.123)
∂Aν ∂xν ∂(∂ µ Aν ) c
which are Maxwell’s equations.
Recall the result of Noether’s theorem for mechanical systems:

!
d ∂L ∂ q̃σ
=0, (7.124)
dt ∂ q̇σ ∂ζ
ζ=0
where q̃σ = q̃σ (q, ζ) is a one-parameter (ζ) family of transformations of the generalized
coordinates which leaves L invariant. We generalize to field theory by replacing
qσ (t) −→ φa (x, t) , (7.125)
where {φa (x, t)} are a set of fields, which are functions of the independent variables {x, y, z, t}.
We will adopt covariant relativistic notation and write for four-vector xµ = (ct, x, y, z). The
generalization of dΛ/dt = 0 is
!
∂ ∂L ∂ φ̃a
=0, (7.126)
∂xµ ∂ (∂µ φa ) ∂ζ
ζ=0
where there is an implied sum on both µ and a. We can write this as ∂µ J µ = 0, where

µ ∂L ∂ φ̃a
J ≡ . (7.127)
∂ (∂µ φa ) ∂ζ

ζ=0
We call Λ = J 0 /c the total charge. If we assume J = 0 at the spatial boundaries of our

system, then integrating the conservation law ∂µ J µ over the spatial region Ω gives
Z Z I
dΛ 3 0 3
= d x ∂0 J = − d x ∇ · J = − dΣ n̂ · J = 0 , (7.128)
dt
Ω Ω ∂Ω
assuming J = 0 at the boundary ∂Ω.
As an example, consider the case of a complex scalar field, with Lagrangian density4
L(ψ, , ψ ∗ , ∂µ ψ, ∂µ ψ ∗ ) = 12 K (∂µ ψ ∗ )(∂ µ ψ) − U ψ ∗ ψ .

(7.129)
This is invariant under the transformation ψ → eiζ ψ, ψ ∗ → e−iζ ψ ∗ . Thus,
∂ ψ̃ ∂ ψ̃ ∗
= i eiζ ψ , = −i e−iζ ψ ∗ , (7.130)
∂ζ ∂ζ
and, summing over both ψ and ψ ∗ fields, we have
∂L ∂L
Jµ = · (iψ) + · (−iψ ∗ )
∂ (∂µ ψ) ∂ (∂µ ψ ∗ )
K ∗ µ
ψ ∂ ψ − ψ ∂ µψ∗ .

= (7.131)
2i
The potential, which depends on |ψ|2 , is independent of ζ. Hence, this form of conserved
4-current is valid for an entire class of potentials.
4
We raise and lower indices using the Minkowski metric gµν = diag (+, −, −, −).
7.6.1 Gross-Pitaevskii model
As one final example of a field theory, consider the Gross-Pitaevskii model, with
∂ψ h̄2 2
L = ih̄ ψ ∗ − ∇ψ ∗ · ∇ψ − g |ψ|2 − n0 . (7.132)
∂t 2m
This describes a Bose fluid with repulsive short-ranged interactions. Here ψ(x, t) is again
a complex scalar field, and ψ ∗ is its complex conjugate. Using the Leibniz rule, we have
δS[ψ ∗ , ψ] = S[ψ ∗ + δψ ∗ , ψ + δψ]
h̄2 h̄2
Z Z
∂δψ ∂ψ
= dt d x ih̄ ψ ∗
d
+ ih̄ δψ ∗ − ∇ψ ∗ · ∇δψ − ∇δψ ∗ · ∇ψ
∂t ∂t 2m 2m

2
∗ ∗
− 2g |ψ| − n0 (ψ δψ + ψδψ )
(
∂ψ ∗ h̄2 2 ∗
Z Z
d 2
∗
= dt d x − ih̄ + ∇ ψ − 2g |ψ| − n0 ψ δψ
∂t 2m
)
h̄2 2

∂ψ
∇ ψ − 2g |ψ|2 − n0 ψ δψ ∗ ,

+ ih̄ + (7.133)
∂t 2m
where we have integrated by parts where necessary and discarded the boundary terms.
Extremizing S[ψ ∗ , ψ] therefore results in the nonlinear Schrödinger equation (NLSE),
∂ψ h̄2 2
∇ ψ + 2g |ψ|2 − n0 ψ

ih̄ =− (7.134)
∂t 2m
as well as its complex conjugate,
∂ψ ∗ h̄2 2 ∗
∇ ψ + 2g |ψ|2 − n0 ψ ∗ .

−ih̄ =− (7.135)
∂t 2m
Note that these equations are indeed the Euler-Lagrange equations:

δS ∂L ∂ ∂L
= − (7.136)
δψ ∂ψ ∂xµ ∂ ∂µ ψ

δS ∂L ∂ ∂L
∗
= ∗
− µ , (7.137)
δψ ∂ψ ∂x ∂ ∂µ ψ ∗
with xµ = (t, x)5 Plugging in
∂L ∂L ∂L h̄2
= −2g |ψ|2 − n0 ψ ∗ = ih̄ ψ ∗ ∇ψ ∗

, , =− (7.138)
∂ψ ∂ ∂t ψ ∂ ∇ψ 2m
and
∂L ∂L ∂L h̄2
= ih̄ ψ − 2g |ψ|2 − n0 ψ

∗
, =0 , = − ∇ψ , (7.139)
∂ψ ∂ ∂t ψ ∗ ∂ ∇ψ ∗ 2m
5
In the nonrelativistic case, there is no utility in defining x0 = ct, so we simply define x0 = t.
we recover the NLSE and its conjugate.
The Gross-Pitaevskii model also possesses a U(1) invariance, under
ψ(x, t) → ψ̃(x, t) = eiζ ψ(x, t) , ψ ∗ (x, t) → ψ̃ ∗ (x, t) = e−iζ ψ ∗ (x, t) . (7.140)
Thus, the conserved Noether current is then

∂L ∂ ψ̃ ∂L ∂ ψ̃ ∗
Jµ = +

∂ ∂µ ψ ∂ζ

∗
∂ ∂µ ψ ∂ζ

ζ=0 ζ=0
J 0 = −h̄ |ψ|2 (7.141)
h̄2
ψ ∗ ∇ψ − ψ∇ψ ∗ .

J =− (7.142)
2im
Dividing out by h̄, taking J 0 ≡ −h̄ρ and J ≡ −h̄j, we obtain the continuity equation,
∂ρ
+∇·j =0 , (7.143)
∂t
where
h̄
ρ = |ψ|2 ψ ∗ ∇ψ − ψ∇ψ ∗ .

, j= (7.144)
2im
are the particle density and the particle current, respectively.
Chapter 8
Constraints
A mechanical system of N point particles in d dimensions possesses n = dN degrees of free-

dom1 . To specify these degrees of freedom, we can choose any independent set of generalized
coordinates {q1 , . . . , qK }. Oftentimes, however, not all n coordinates are independent.
Consider, for example, the situation in Fig. 8.1, where a cylinder of radius a rolls over a half-
cylinder of radius R. If there is no slippage, then the angles θ1 and θ2 are not independent,
and they obey the equation of constraint,
R θ1 = a (θ2 − θ1 ) . (8.1)
In this case, we can easily solve the constraint equation and substitute θ2 = 1 + Ra θ1 . In

other cases, though, the equation of constraint might not be so easily solved (e.g. it may be
nonlinear). How then do we proceed?
8.1 Constraints and Variational Calculus
Before addressing the subject of constrained dynamical systems, let’s consider the issue of
constraints in the broader context of variational calculus. Suppose we have a functional
Zxb
F [y(x)] = dx L(y, y 0 , x) , (8.2)
xa
which we want to extremize subject to some constraints. Here y may stand for a set of
functions {yσ (x)}. There are two classes of constraints we will consider:
1
For N rigid bodies, the number of degrees of freedom is n0 = 12 d(d + 1)N , corresponding to d center-
of-mass coordinates and 21 d(d − 1) angles of orientation for each particle. The dimension of the group of
rotations in d dimensions is 12 d(d − 1), corresponding to the number of parameters in a general rank-d
orthogonal matrix (i.e. an element of the group O(d)).
115
116 CHAPTER 8. CONSTRAINTS
Figure 8.1: A cylinder of radius a rolls along a half-cylinder of radius R. When there is no
slippage, the angles θ1 and θ2 obey the constraint equation Rθ1 = a(θ2 − θ1 ).
1. Integral constraints: These are of the form

Zxb
dx Nl (y, y 0 , x) = Cl , (8.3)
xa
where k labels the constraint.
2. Holonomic constraints: These are of the form
Gk (y, x) = 0 . (8.4)
The cylinders system in Fig. 8.1 provides an example of a holonomic constraint. There,
G(θ, t) = R θ1 − a (θ2 − θ1 ) = 0. As an example of a problem with an integral constraint,
suppose we want to know the shape of a hanging rope of fixed length C. This means we
minimize the rope’s potential energy,
Zxb Zxb q
U [y(x)] = λg ds y(x) = λg dx y 1 + y 0 2 , (8.5)
xa xa
where λ is the linear mass density of the rope, subject to the fixed-length constraint
Zxb Zxb q
C = ds = dx 1 + y 0 2 . (8.6)
xa xa
p
Note ds = dx2 + dy 2 is the differential element of arc length along the rope. To solve
problems like these, we turn to Lagrange’s method of undetermined multipliers.
8.2. CONSTRAINED EXTREMIZATION OF FUNCTIONS 117
8.2 Constrained Extremization of Functions
Given F (x1 , . . . , xn ) to be extremized subject to k constraints of the form Gj (x1 , . . . , xn ) = 0

where j = 1, . . . , k, construct
k
X
∗

F x1 , . . . , xn ; λ1 , . . . , λk ≡ F (x1 , . . . , xn ) + λj Gj (x1 , . . . , xn ) (8.7)
j=1

which is a function of the (n + k) variables x1 , . . . , xn ; λ1 , . . . , λk . Now freely extremize
the extended function F ∗ :
n k
X ∂F ∗ X ∂F ∗
dF ∗ = dxσ +dλj (8.8)
∂xσ ∂λj
σ=1 j=1
 
n k k
X
 ∂F +
X ∂Gj  X
= λj dxσ + Gj dλj = 0 (8.9)
∂xσ ∂xσ
σ=1 j=1 j=1
This results in the (n + k) equations

X ∂Gjk
∂F
+ λj =0 (σ = 1, . . . , n) (8.10)
∂xσ ∂xσ
j=1
Gj = 0 (j = 1, . . . , k) . (8.11)
The interpretation of all this is as follows. The n equations in 8.10 can be written in vector
form as
X k
∇F + λj ∇Gj = 0 . (8.12)
j=1
This says that the (n-component) vector ∇F is linearly dependent upon the k vectors
∇Gj . Thus, any movement in the direction of ∇F must necessarily entail movement along
one or more of the directions ∇Gj . This would require violating the constraints, since
movement along ∇Gj takes us off the level set Gj = 0. Were ∇F linearly independent of
the set {∇Gj }, this would mean that we could find a differential displacement dx which
has finite overlap with ∇F but zero overlap with each ∇Gj . Thus x + dx would still satisfy
Gj (x + dx) = 0, but F would change by the finite amount dF = ∇F (x) · dx.
8.3 Extremization of Functionals : Integral Constraints
Given a functional
Zxb
dx L {yσ }, {yσ0 }, x

F {yσ (x)} = (σ = 1, . . . , n) (8.13)
xa
subject to boundary conditions δyσ (xa ) = δyσ (xb ) = 0 and k constraints of the form
Zxb
dx Nl {yσ }, {yσ0 }, x = Cl

(l = 1, . . . , k) , (8.14)
xa
construct the extended functional

Zxb k k
X
∗
0
X 0

F {yσ (x)}; {λj } ≡ dx L {yσ }, {yσ }, x + λl Nl {yσ }, {yσ }, x − λl Cl (8.15)
xa l=1 l=1
and freely extremize over {y1 , . . . , yn ; λ1 , . . . , λk }. This results in (n + k) equations

k
X ( )
∂L d ∂L ∂Nl d ∂Nl
− + λl − =0 (σ = 1, . . . , n) (8.16)
∂yσ dx ∂yσ0 ∂yσ dx ∂yσ0
l=1
Zxb
dx Nl {yσ }, {yσ0 }, x = Cl

(l = 1, . . . , k) . (8.17)
xa
8.4 Extremization of Functionals : Holonomic Constraints
Given a functional
Zxb
dx L {yσ }, {yσ0 }, x

F {yσ (x)} = (σ = 1, . . . , n) (8.18)
xa
subject to boundary conditions δyσ (xa ) = δyσ (xb ) = 0 and k constraints of the form

Gj {yσ (x)}, x = 0 (j = 1, . . . , k) , (8.19)
construct the extended functional
Zxb k
X
F ∗ {yσ (x)}; {λj (x)} ≡ 0

dx L {yσ }, {yσ }, x + λj Gj {yσ } (8.20)
xa j=1

and freely extremize over y1 , . . . , yn ; λ1 , . . . , λk :
Zxb ( Xn X k k
)
∗ ∂L d ∂L ∂Gj X
δF = dx − + λj δyσ + Gj δλj = 0 , (8.21)
∂yσ dx ∂yσ0 ∂yσ
xa σ=1 j=1 j=1
resulting in the (n + k) equations

k
!
d ∂L ∂L X ∂Gj
0
− = λj (σ = 1, . . . , n) (8.22)
dx ∂yσ ∂yσ ∂yσ
j=1

Gj {yσ }, x = 0 (j = 1, . . . , k) . (8.23)
8.4. EXTREMIZATION OF FUNCTIONALS : HOLONOMIC CONSTRAINTS 119
8.4.1 Examples of extremization with constraints
Volume of a cylinder : As a warm-up problem, let’s maximize the volume V = πa2 h of a

cylinder of radius a and height h, subject to the constraint
h2
G(a, h) = 2πa + −`=0 . (8.24)
b
We therefore define
V ∗ (a, h, λ) ≡ V (a, h) + λ G(a, h) , (8.25)
and set
∂V ∗
= 2πah + 2πλ = 0 (8.26)
∂a
∂V ∗ h
= πa2 + 2λ = 0 (8.27)
∂h b
∂V ∗ h2
= 2πa + −`=0 . (8.28)
∂λ b
Solving these three equations simultaneously gives
r
2` b` 2π 4
a= , h= , λ = 3/2 b1/2 `3/2 , V = `5/2 b1/2 . (8.29)
5π 5 5 55/2 π
Hanging rope : We minimize the energy functional
Zx2 q
dx y 1 + y 0 2 ,

E y(x) = µg (8.30)
x1
where µ is the linear mass density, subject to the constraint of fixed total length,
Zx2 q
dx 1 + y 0 2 .

C y(x) = (8.31)
x1
Thus,
Zx2
∗
dx L∗ (y, y 0 , x) ,

E y(x), λ = E y(x) + λC y(x) = (8.32)
x1
with q
∗ 0
L (y, y , x) = (µgy + λ) 1 + y02 . (8.33)
∂L∗
Since ∂x = 0 we have that
∂L∗ µgy + λ
J = y0 0
− L∗ = − p (8.34)
∂y 1 + y02
is constant. Thus,
dy
= ±J −1 (µgy + λ)2 − J 2 ,
p
(8.35)
dx
with solution
λ J µg
y(x) = − + cosh (x − a) . (8.36)
µg µg J
Here, J , a, and λ are constants to be determined by demanding y(xi ) = yi (i = 1, 2), and
that the total length of the rope is C.
Geodesic on a curved surface : Consider next the problem of a geodesic on a curved surface.
Let the equation for the surface be
G(x, y, z) = 0 . (8.37)
We wish to extremize the distance,
Zb Zb p
D = ds = dx2 + dy 2 + dz 2 . (8.38)
a a
We introduce a parameter t defined on the unit interval: t ∈ [0, 1], such that x(0) = xa ,
x(1) = xb , etc. Then D may be regarded as a functional, viz.
Z1 p

D x(t), y(t), z(t) = dt ẋ2 + ẏ 2 + ż 2 . (8.39)
0
We impose the constraint by forming the extended functional, D∗ :
Z1 p
∗

D x(t), y(t), z(t), λ(t) ≡ dt 2 2 2
ẋ + ẏ + ż + λ G(x, y, z) , (8.40)
0
and we demand that the first functional derivatives of D∗ vanish:
δD∗

d ẋ ∂G
=− p +λ =0 (8.41)
δx(t) dt ẋ2 + ẏ 2 + ż 2 ∂x
δD∗

d ẏ ∂G
=− p +λ =0 (8.42)
δy(t) dt 2 2
ẋ + ẏ + ż 2 ∂y
δD∗

d ż ∂G
=− p +λ =0 (8.43)
δz(t) dt 2 2
ẋ + ẏ + ż 2 ∂z
δD∗
= G(x, y, z) = 0 . (8.44)
δλ(t)
Thus,
vẍ − ẋv̇ v ÿ − ẏ v̇ vz̈ − ż v̇
λ(t) = 2
= 2 = 2 , (8.45)
v ∂x G v ∂y G v ∂z G
p ∂
with v = ẋ2 + ẏ 2 + ż 2 and ∂x ≡ ∂x , etc. These three equations are supplemented by
G(x, y, z) = 0, which is the fourth.
8.5. APPLICATION TO MECHANICS 121
8.5 Application to Mechanics
Let us write our system of constraints in the differential form

n
X
gjσ (q, t) dqσ + hj (q, t)dt = 0 (j = 1, . . . , k) . (8.46)
σ=1
If the partial derivatives satisfy

∂gjσ ∂gjσ0 ∂gjσ ∂hj
= , = , (8.47)
∂qσ0 ∂qσ ∂t ∂qσ
then the differential can be integrated to give dG(q, t) = 0, where
∂Gj ∂Gj
gjσ = , hj = . (8.48)
∂qσ ∂t
The action functional is

Ztb

S[{qσ (t)}] = dt L {qσ }, {q̇σ }, t (σ = 1, . . . , n) , (8.49)
ta
subject to boundary conditions δqσ (ta ) = δqσ (tb ) = 0. The first variation of S is given by
Ztb X n
( )
∂L d ∂L
δS = dt − δqσ . (8.50)
∂qσ dt ∂ q̇σ
ta σ=1
Since the {qσ (t)} are no longer independent, we cannot infer that the term in brackets
vanishes for each σ. What are the constraints on the variations δqσ (t)? The constraints are
expressed in terms of virtual displacements which take no time: δt = 0. Thus,
n
X
gjσ (q, t) δqσ (t) = 0 . (8.51)
σ=1
We may now relax the constraint by introducing k undetermined functions λj (t), by adding
integrals of the above equations with undetermined coefficient functions to δS:
n k
( X )
X ∂L d ∂L
− + λj (t) gjσ (q, t) δqσ (t) = 0 . (8.52)
∂qσ dt ∂ q̇σ
σ=1 j=1
Now we can demand that the term in brackets vanish for all σ. Thus, we obtain a set of
(n + k) equations,
k
d ∂L ∂L X
− = λj (t) gjσ (q, t) ≡ Qσ (8.53)
dt ∂ q̇σ ∂qσ
j=1
gjσ (q, t) q̇σ + hj (q, t) = 0 , (8.54)


in (n + k) unknowns q1 , . . . , qn , λ1 , . . . , λk . Here, Qσ is the force of constraint conjugate
to the generalized coordinate qσ . Thus, with
k
∂L ∂L X
pσ = , Fσ = , Qσ = λj gjσ , (8.55)
∂ q̇σ ∂qσ
j=1
we write Newton’s second law as

ṗσ = Fσ + Qσ . (8.56)
Note that we can write

δS ∂L d ∂L
= − (8.57)
δq(t) ∂q dt ∂ q̇
and that the instantaneous constraints may be written
gj · δq = 0 (j = 1, . . . , k) . (8.58)
Thus, by demanding
k
δS X
+ λj gj = 0 (8.59)
δq(t)
j=1
we require that the functional derivative be linearly dependent on the k vectors gj .
8.5.1 Constraints and conservation laws
We have seen how invariance of the Lagrangian with respect to a one-parameter family of
coordinate transformations results in an associated conserved quantity Λ, and how a lack of
explicit time dependence in L results in the conservation of the Hamiltonian H. In deriving
both these results, however, we used the equations of motion ṗσ = Fσ . What happens when
we have constraints, in which case ṗσ = Fσ + Qσ ?
Let’s begin with the Hamiltonian. We have H = q̇σ pσ − L, hence

dH ∂L ∂L ∂L
= pσ − q̈σ + ṗσ − q̇σ −
dt ∂ q̇σ ∂qσ ∂t
∂L
= Qσ q̇σ − . (8.60)
∂t
We now use
Qσ q̇σ = λj gjσ q̇σ = −λj hj (8.61)
to obtain
dH ∂L
= −λj hj − . (8.62)
dt ∂t
We therefore conclude that in a system with constraints of the form gjσ q̇σ + hj = 0, the
Hamiltonian is conserved if each hj = 0 and if L is not explicitly dependent on time. In
8.6. WORKED EXAMPLES 123
∂Gj
the case of holonomic constraints, hj = ∂t , so H is conserved if neither L nor any of the
constraints Gj is explicitly time-dependent.
Next, let us rederive Noether’s theorem when constraints are present. We assume a one-
parameter family of transformations qσ → q̃σ (ζ) leaves L invariant. Then
dL ∂L ∂ q̃σ ∂L ∂ q̃˙σ
0= = +
dζ ∂ q̃σ ∂ζ ∂ q̃˙σ ∂ζ

∂ q̃σ d ∂ q̃σ
˙
= p̃σ − Q̃σ + p̃σ
∂ζ dt ∂ζ

d ∂ q̃σ ∂ q̃σ
= p̃ − λj g̃jσ . (8.63)
dt σ ∂ζ ∂ζ
Now let us write the constraints in differential form as
g̃jσ dq̃σ + h̃j dt + k̃j dζ = 0 . (8.64)
We now have
dΛ
= λj k̃j , (8.65)
dt
which says that if the constraints are independent of ζ then Λ is conserved. For holonomic
constraints, this means that
∂Gj
Gj q̃(ζ), t = 0 ⇒ k̃j = =0, (8.66)
∂ζ
i.e. Gj (q̃, t) has no explicit ζ dependence.
8.6 Worked Examples
Here we consider several example problems of constrained dynamics, and work each out in
full detail.
8.6.1 One cylinder rolling off another
As an example of the constraint formalism, consider the system in Fig. 8.1, where a cylinder
of radius a rolls atop a cylinder of radius R. We have two constraints:
G1 (r, θ1 , θ2 ) = r − R − a = 0 (cylinders in contact) (8.67)
G2 (r, θ1 , θ2 ) = R θ1 − a (θ2 − θ1 ) = 0 (no slipping) , (8.68)
from which we obtain the gjσ :

1 0 0
gjσ = , (8.69)
0 R + a −a
which is to say
∂G1 ∂G1 ∂G1

=1 =0 =0 (8.70)
∂r ∂θ1 ∂θ2
∂G2 ∂G2 ∂G2

=0 =R+a = −a . (8.71)
∂r ∂θ1 ∂θ2
The Lagrangian is
L = T − U = 21 M ṙ2 + r2 θ̇12 + 12 I θ̇22 − M gr cos θ1 ,

(8.72)
where M and I are the mass and rotational inertia of the rolling cylinder, respectively.
Note that the kinetic energy is a sum of center-of-mass translation Ttr = 12 M ṙ2 + r2 θ̇12
and rotation about the center-of-mass, Trot = 12 I θ̇22 . The equations of motion are

d ∂L ∂L
− = M r̈ − M r θ̇12 + M g cos θ1 = λ1 ≡ Qr (8.73)
dt ∂r ∂r

d ∂L ∂L
− = M r2 θ̈1 + 2M rṙ θ̇1 − M gr sin θ1 = (R + a) λ2 ≡ Qθ1 (8.74)
dt ∂θ1 ∂θ1

d ∂L ∂L
− = I θ̈2 = −a λ2 ≡ Qθ2 . (8.75)
dt ∂θ2 ∂θ2
To these three
equations we add the two constraints, resulting in five equations in the five
unknowns r, θ1 , θ2 , λ1 , λ2 .
We solve by first implementing the constraints, which give r = (R + a) a constant (i.e.

ṙ = 0), and θ̇2 = 1 + Ra θ̇1 . Substituting these into the above equations gives

−M (R + a) θ̇12 + M g cos θ1 = λ1 (8.76)
M (R + a)2 θ̈1 − M g(R + a) sin θ1 = (R + a) λ2 (8.77)

R+a
I θ̈1 = −aλ2 . (8.78)
a
From eqn. 8.78 we obtain
I R+a
λ2 = − θ̈2 = − 2 I θ̈1 , (8.79)
a a
which we substitute into eqn. 8.77 to obtain

I
M + 2 (R + a)2 θ̈1 − M g(R + a) sin θ1 = 0 . (8.80)
a
Multiplying by θ̇1 , we obtain an exact differential, which may be integrated to yield

1 I 2 Mg Mg ◦
2 M 1 + M a2 θ̇1 + R + a cos θ1 = R + a cos θ1 . (8.81)
Figure 8.2: Frictionless motion under gravity along a curved surface. The skier flies off the
surface when the normal force vanishes.
Here, we have assumed that θ̇1 = 0 when θ1 = θ1◦ , i.e. the rolling cylinder is released from
rest at θ1 = θ1◦ . Finally, inserting this result into eqn. 8.76, we obtain the radial force of
constraint,
Mg n o
Qr = (3 + α) cos θ1 − 2 cos θ1◦ , (8.82)
1+α
where α = I/M a2 is a dimensionless parameter (0 ≤ α ≤ 1). This is the radial component
of the normal force between the two cylinders. When Qr vanishes, the cylinders lose contact
– the rolling cylinder flies off. Clearly this occurs at an angle θ1 = θ1∗ , where
◦

∗ −1 2 cos θ1
θ1 = cos . (8.83)
3+α
The detachment angle θ1∗ is an increasing function of α, which means that larger I delays
detachment. This makes good sense, since when I is larger the gain in kinetic energy is
split between translational and rotational motion of the rolling cylinder.
8.6.2 Frictionless motion along a curve
Consider the situation in Fig. 8.2 where a skier moves frictionlessly under the influence of
gravity along a general curve y = h(x). The Lagrangian for this problem is
L = 12 m(ẋ2 + ẏ 2 ) − mgy (8.84)
and the (holonomic) constraint is
G(x, y) = y − h(x) = 0 . (8.85)
Accordingly, the Euler-Lagrange equations are

d ∂L ∂L ∂G
− =λ , (8.86)
dt ∂ q̇σ ∂qσ ∂qσ
where q1 = x and q2 = y. Thus, we obtain
mẍ = −λ h0 (x) = Qx (8.87)

mÿ + mg = λ = Qy . (8.88)
We eliminate y in favor of x by invoking the constraint. Since we need ÿ, we must differen-
tiate the constraint, which gives
ẏ = h0 (x) ẋ , ÿ = h0 (x) ẍ + h00 (x) ẋ2 . (8.89)
Using the second Euler-Lagrange equation, we then obtain

λ
= g + h0 (x) ẍ + h00 (x) ẋ2 . (8.90)
m
Finally, we substitute this into the first E-L equation to obtain an equation for x alone:
2
1 + h0 (x) ẍ + h0 (x) h00 (x) ẋ2 + g h0 (x) = 0 .

(8.91)
Had we started by eliminating y = h(x) at the outset, writing

2 2
L(x, ẋ) = 12 m 1 + h0 (x)

ẋ − mg h(x) , (8.92)
we would also have obtained this equation of motion.
The skier flies off the curve when the vertical force of constraint Qy = λ starts to become
negative, because the curve can only supply a positive normal force. Suppose the skier
starts from rest at a height y0 . We may then determine the point x at which the skier
detaches from the curve by setting λ(x) = 0. To do so, we must eliminate ẋ and ẍ in terms
of x. For ẍ, we may use the equation of motion to write
0
gh + h0 h00 ẋ2

ẍ = − , (8.93)
1 + h0 2
which allows us to write
g + h00 ẋ2

λ=m . (8.94)
1 + h0 2
To eliminate ẋ, we use conservation of energy,
2
E = mgy0 = 21 m 1 + h0 ẋ2 + mgh , (8.95)
which fixes
2 y0 − h
ẋ = 2g . (8.96)
1 + h0 2
Putting it all together, we have

mg n 02 00
o
λ(x) = 1 + h + 2(y 0 − h) h . (8.97)
1 + h0 2

Figure 8.3: Finding the local radius of curvature: z = η 2 /2R.
The skier detaches from the curve when λ(x) = 0, i.e. when
2
1 + h0 + 2(y0 − h) h00 = 0 . (8.98)
There is a somewhat easier way of arriving at the same answer. This is to note that the
skier must fly off when the local centripetal force equals the gravitational force normal to
the curve, i.e.
m v 2 (x)
= mg cos θ(x) , (8.99)
R(x)
−1/2
where R(x) is the local radius of curvature. Now tan θ = h0 , so cos θ = 1 + h0 2 . The
2 2 2 0 2 2
square of the velocity is v = ẋ + ẏ = 1 + h ẋ . What is the local radius of curvature
R(x)? This can be determined from the following argument, and from the sketch in Fig.
8.3. Writing x = x∗ + , we have
y = h(x∗ ) + h0 (x∗ ) + 21 h00 (x∗ ) 2 + . . . . (8.100)
We now drop a perpendicular segment of length z from the point (x, y) to the line which is
tangent to the curve at x∗ , h(x∗ ) . According to Fig. 8.3, this means

0
1 1 1 −h
= η · √ 02 0 − z · √ 02 . (8.101)
y 1+h h 1+h 1
Thus, we have
y = h0 + 21 h00 2
η + z h0 η + z h0 2

0 1 00
=h p + 2h p
1 + h0 2 1 + h0 2
η h0 + z h0 2 h00 η 2
= p + + O(ηz)
1 + h0 2 2 1 + h0 2
η h0 − z
=p , (8.102)
1 + h0 2
from which we obtain

h00 η 2
z=− 3/2 + O(η 3 ) (8.103)
2 1+ h0 2
and therefore
1 0 2 3/2
R(x) = − · 1 + h (x) . (8.104)
h00 (x)
Thus, the detachment condition,
mv 2 m h00 ẋ2 mg
= −p =p = mg cos θ (8.105)
R 1+h 0 2
1 + h0 2
reproduces the result from eqn. 8.94.
8.6.3 Disk rolling down an inclined plane
A hoop of mass m and radius R rolls without slipping down an inclined plane. The inclined
plane has opening angle α and mass M , and itself slides frictionlessly along a horizontal
surface. Find the motion of the system.
Figure 8.4: A hoop rolling down an inclined plane lying on a frictionless surface.
Solution : Referring to the sketch in Fig. 8.4, the center of the hoop is located at
x = X + s cos α − a sin α
y = s sin α + a cos α ,
where X is the location of the lower left corner of the wedge, and s is the distance
along the wedge to the bottom of the hoop. If the hoop rotates through an angle θ,
the no-slip condition is a θ̇ + ṡ = 0. Thus,
L = 12 M Ẋ 2 + 12 m ẋ2 + ẏ 2 + 12 I θ̇2 − mgy

I
= 21 m + 2 ṡ2 + 12 (M + m)Ẋ 2 + m cos α Ẋ ṡ − mgs sin α − mga cos α .
a
Since X is cyclic in L, the momentum
PX = (M + m)Ẋ + m cos α ṡ ,
is preserved: ṖX = 0. The second equation of motion, corresponding to the generalized

coordinate s, is
I
1+ s̈ + cos α Ẍ = −g sin α .
ma2
Using conservation of PX , we eliminate s̈ in favor of Ẍ, and immediately obtain
g sin α cos α
Ẍ = ≡ aX .
1+ M
m 1 + ma
I
2 − cos 2α
The result
g 1+ M m sin α
s̈ = − ≡ as
1+ M
m 1 + I
ma2
− cos 2α
follows immediately. Thus,
X(t) = X(0) + Ẋ(0) t + 12 aX t2

s(t) = s(0) + ṡ(0) t + 12 as t2 .
Note that as < 0 while aX > 0, i.e. the hoop rolls down and to the left as the wedge
slides to the right. Note that I = ma2 for a hoop; we’ve computed the answer here
for general I.
8.6.4 Pendulum with nonrigid support
A particle of mass m is suspended from a flexible string of length ` in a uniform gravitational

field. While hanging motionless in equilibrium, it is struck a horizontal blow resulting in
an initial angular velocity ω0 . Treating the system as one with two degrees of freedom and
a constraint, answer the following:
(a) Compute the Lagrangian, the equation of constraint, and the equations of motion.
Solution : The Lagrangian is
L = 12 m ṙ2 + r2 θ̇2 + mgr cos θ .

The constraint is r = `. The equations of motion are

mr̈ − mr θ̇2 − mg cos θ = λ
mr2 θ̈ + 2mr ṙ θ̇ − mg sin θ = 0 .
(b) Compute the tension in the string as a function of angle θ.

Solution : Energy is conserved, hence
2 2
1
2 m` θ̇ − mg` cos θ = 12 m`2 θ̇02 − mg` cos θ0 .
We take θ0 = 0 and θ̇0 = ω0 . Thus,

θ̇2 = ω02 − 2 Ω 2 1 − cos θ ,

p
with Ω = g/`. Substituting this into the equation for λ, we obtain
ω02

λ = mg 2 − 3 cos θ − 2 .
Ω
(c) Show that if ω02 < 2g/` then the particle’s motion is confined below the horizontal
and that the tension in the string is always positive (defined such that positive means
exerting a pulling force and negative means exerting a pushing force). Note that the
difference between a string and a rigid rod is that the string can only pull but the rod
can pull or push. Thus, the string tension must always be positive or else the string
goes “slack”.
Solution : Since θ̇2 ≥ 0, we must have
ω02
≥ 1 − cos θ .
2Ω 2
The condition for slackness is λ = 0, or
ω02
= 1 − 32 cos θ .
2Ω 2
Thus, if ω02 < 2Ω 2 , we have
ω02
1> > 1 − cos θ > 1 − 32 cos θ ,
2Ω 2
and the string never goes slack. Note the last equality follows from cos θ > 0. The
string rises to a maximum angle
ω2
θmax = cos−1 1 − 02 .
2Ω
(d) Show that if 2g/` < ω02 < 5g/` the particle rises above the horizontal and the string
becomes slack (the tension vanishes) at an angle θ∗ . Compute θ∗ .
Solution : When ω 2 > 2Ω 2 , the string rises above the horizontal and goes slack at an
angle
ω2
θ∗ = cos−1 23 − 02 .
3Ω
This solution craps out when the string is still taut at θ = π, which means ω02 = 5Ω 2 .
(e) Show that if ω02 > 5g/` the tension is always positive and the particle executes circular
motion.
Solution : For ω02 > 5Ω 2 , the string never goes slack. Furthermore, θ̇ never vanishes.
Therefore, the pendulum undergoes circular motion, albeit not with constant angular
velocity.
8.6.5 Falling ladder
A uniform ladder of length ` and mass m has one end on a smooth horizontal floor and the
other end against a smooth vertical wall. The ladder is initially at rest and makes an angle
θ0 with respect to the horizontal.
Figure 8.5: A ladder sliding down a wall and across a floor.
(a) Make a convenient choice of generalized coordinates and find the Lagrangian.
Solution : I choose as generalized coordinates the Cartesian coordinates (x, y) of the

ladder’s center of mass, and the angle θ it makes with respect to the floor. The
Lagrangian is then
L = 12 m (ẋ2 + ẏ 2 ) + 12 I θ̇2 + mgy .
There are two constraints: one enforcing contact along the wall, and the other enforc-
ing contact along the floor. These are written
G1 (x, y, θ) = x − 12 ` cos θ = 0
G2 (x, y, θ) = y − 21 ` sin θ = 0 .
(b) Prove that the ladder leaves the wall when its upper end has fallen to a height 23 L sin θ0 .
X ∂Gj
d ∂L ∂L
− = λj .
j
Thus, we have
m ẍ = λ1 = Qx
m ÿ + mg = λ2 = Qy
I θ̈ = 12 ` λ1 sin θ − λ2 cos θ = Qθ .

We now implement the constraints to eliminate x and y in terms of θ. We have
ẋ = − 21 ` sin θ θ̇ ẍ = − 12 ` cos θ θ̇2 − 12 ` sin θ θ̈

ẏ = 1
2 ` cos θ θ̇ ÿ = − 21 ` sin θ θ̇2 + 12 ` cos θ θ̈ .
We can now obtain the forces of constraint in terms of the function θ(t):
λ1 = − 12 m` sin θ θ̈ + cos θ θ̇2

λ2 = + 12 m` cos θ θ̈ − sin θ θ̇2 + mg .

We substitute these into the last equation of motion to obtain the result
I θ̈ = −I0 θ̈ − 12 mg` cos θ ,
or
(1 + α) θ̈ = −2ω02 cos θ ,
with I0 = 41 m`2 , α ≡ I/I0 and ω0 = g/`. This may be integrated once (multiply by
p
θ̇ to convert to a total derivative) to yield

1
2 (1 + α) θ̇2 + 2 ω02 sin θ = 2 ω02 sin θ0 ,
which is of course a statement of energy conservation. This,

4 ω02 (sin θ0 − sin θ)
θ̇2 =
1+α
2 ω02 cos θ
θ̈ = − .
1+α
We may now obtain λ1 (θ) and λ2 (θ):

mg
λ1 (θ) = − 3 sin θ − 2 sin θ0 cos θ
1+α
mg n o
λ2 (θ) = (3 sin θ − 2 sin θ0 sin θ + α .
1+α
Demanding λ1 (θ) = 0 gives the detachment angle θ = θd , where

2
sin θd = 3 sin θ0 .
Note that λ2 (θd ) = mgα/(1 + α) > 0, so the normal force from the floor is always
positive for θ > θd . The time to detachment is
Z √ Zθ0
dθ 1+α dθ
T1 (θ0 ) = = √ .
θ̇ 2 ω0 sin θ0 − sin θ
θd
(c) Show that the subsequent motion can be reduced to quadratures (i.e. explicit inte-
grals).
Solution : After the detachment, there is no longer a constraint G1 . The equations of

motion are
m ẍ = 0 (conservation of x-momentum)
m ÿ + m g = λ
I θ̈ = − 12 ` λ cos θ ,
1
along with the constraint y = 2 ` sin θ. Eliminating y in favor of θ using the constraint,
the second equation yields
λ = mg − 12 m` sin θ θ̇2 + 12 m` cos θ θ̈ .
Plugging this into the third equation of motion, we find
I θ̈ = −2 I0 ω02 cos θ + I0 sin θ cos θ θ̇2 − I0 cos2 θ θ̈ .
Multiplying by θ̇ one again obtains a total time derivative, which is equivalent to

rediscovering energy conservation:
E = 21 I + I0 cos2 θ θ̇2 + 2 I0 ω02 sin θ .

By continuity with the first phase of the motion, we obtain the initial conditions for
this second phase:
θ = sin−1 23 sin θ0

s
sin θ0
θ̇ = −2 ω0 .
3 (1 + α)
Figure 8.6: Plot of time to fall for the slipping ladder. Here x = sin θ0 .
Thus,
4 ω02 sin θ0 1
E= 1
2 I + I0 − 49 I0 sin2 θ0 · + 3 mg` sin θ0
3 (1 + α)
2
2 4 sin θ0
= 2 I0 ω0 · 1 + 27 sin θ0 .
1+α
(d) Find an expression for the time T (θ0 ) it takes the

p ladder to smack against the floor.
Note that, expressed in units of the time scale L/g, T is a dimensionless function
of θ0 . Numerically integrate this expression and plot T versus θ0 .
Solution : The time from detachment to smack is

Zθd s
1 + α cos2 θ
Z
dθ 1
T2 (θ0 ) = = dθ .
2 ω0 4 sin2 θ0

θ̇ 1 − 27 1+α sin θ0 − sin θ
0
The total time is then T (θ0 ) = T1 (θ0 ) + T2 (θ0 ). For a uniformly dense ladder, I =
1 2 1 1
12 m` = 3 I0 , so α = 3 .
(e) What is the horizontal velocity of the ladder at long times?
Solution : From the moment of detachment, and thereafter,

s
g`
ẋ = − 12 ` sin θ θ̇ = sin3/2 θ0 .
3 (1 + α)
(f) Describe in words the motion of the ladder subsequent to it slapping against the floor.
Solution : Only a fraction of the ladder’s initial potential energy is converted into
kinetic energy of horizontal motion. The rest is converted into kinetic energy of
vertical motion and of rotation. The slapping of the ladder against the floor is an
elastic collision. After the collision, the ladder must rise again, and continue to rise
and fall ad infinitum, as it slides along with constant horizontal velocity.
8.6.6 Point mass inside rolling hoop
Consider the point mass m inside the hoop of radius R, depicted in Fig. 8.7. We choose
as generalized coordinates the Cartesian coordinates (X, Y ) of the center of the hoop, the
Cartesian coordinates (x, y) for the point mass, the angle φ through which the hoop turns,
and the angle θ which the point mass makes with respect to the vertical. These six coordi-
nates are not all independent. Indeed, there are only two independent coordinates for this
system, which can be taken to be θ and φ. Thus, there are four constraints:
X − Rφ ≡ G1 = 0 (8.106)
Y − R ≡ G2 = 0 (8.107)
x − X − R sin θ ≡ G3 = 0 (8.108)
y − Y + R cos θ ≡ G4 = 0 . (8.109)
Figure 8.7: A point mass m inside a hoop of mass M , radius R, and moment of inertia I.
The kinetic and potential energies are easily expressed in terms of the Cartesian coordinates,
aside from the energy of rotation of the hoop about its CM, which is expressed in terms of
φ̇:
T = 12 M (Ẋ 2 + Ẏ 2 ) + 12 m(ẋ2 + ẏ 2 ) + 12 I φ̇2 (8.110)

U = M gY + mgy . (8.111)
The moment of inertia of the hoop about its CM is I = M R2 , but we could imagine a
situation in which I were different. For example, we could instead place the point mass
inside a very short cylinder with two solid end caps, in which case I = 21 M R2 . The
Lagrangian is then
L = 12 M (Ẋ 2 + Ẏ 2 ) + 12 m(ẋ2 + ẏ 2 ) + 12 I φ̇2 − M gY − mgy . (8.112)
Note that L as written is completely independent of θ and θ̇!
Continuous symmetry
Note that there is an continuous symmetry to L which is satisfied by all the constraints,
under
X̃(ζ) = X + ζ Ỹ (ζ) = Y (8.113)
x̃(ζ) = x + ζ ỹ(ζ) = y (8.114)
ζ
φ̃(ζ) = φ + θ̃(ζ) = θ . (8.115)
R
Thus, according to Noether’s theorem, there is a conserved quantity
∂L ∂L 1 ∂L
Λ= + +
∂ Ẋ ∂ ẋ R ∂ φ̇
I
= M Ẋ + mẋ + φ̇ . (8.116)
R
This means Λ̇ = 0. This reflects the overall conservation of momentum in the x-direction.
Energy conservation
Since neither L nor any of the constraints are explicitly time-dependent, the Hamiltonian
is conserved. And since T is homogeneous of degree two in the generalized velocities, we
have H = E = T + U :
E = 12 M (Ẋ 2 + Ẏ 2 ) + 12 m(ẋ2 + ẏ 2 ) + 12 I φ̇2 + M gY + mgy . (8.117)
Equations of motion
We have n = 6 generalized coordinates and k = 4 constraints. Thus, there are four un-
determined multipliers {λ1 , λ2 , λ3 , λ4 } used to impose the constraints. This makes for ten
unknowns:
X , Y , x , y , φ , θ , λ 1 , λ2 , λ3 , λ4 . (8.118)
Accordingly, we have ten equations: six equations of motion plus the four equations of
constraint. The equations of motion are obtained from
k
d ∂L ∂L X ∂Gj
= + λj . (8.119)
j=1
Taking each generalized coordinate in turn, the equations of motion are thus
M Ẍ = λ1 − λ3 (8.120)
M Ÿ = −M g + λ2 − λ4 (8.121)
mẍ = λ3 (8.122)
mÿ = −mg + λ4 (8.123)
I φ̈ = −R λ1 (8.124)
0 = −R cos θ λ3 − R sin θ λ4 . (8.125)
Along with the four constraint equations, these determine the motion of the system. Note
that the last of the equations of motion, for the generalized coordinate qσ = θ, says that
Qθ = 0, which means that the force of constraint on the point mass is radial. Were the point
mass replaced by a rolling object, there would be an angular component to this constraint
in order that there be no slippage.
Implementation of constraints
We now use the constraint equations to eliminate X, Y , x, and y in terms of θ and φ:
X = Rφ , Y =R , x = Rφ + R sin θ , y = R(1 − cos θ) . (8.126)
We also need the derivatives:
ẋ = R φ̇ + R cos θ θ̇ , ẍ = R φ̈ + R cos θ θ̈ − R sin θ θ̇2 , (8.127)
and
ẏ = R sin θ θ̇ , ẍ = R sin θ θ̈ + R cos θ θ̇2 , (8.128)
as well as
Ẋ = R φ̇ , Ẍ = R φ̈ , Ẏ = 0 , Ÿ = 0 . (8.129)
We now may write the conserved charge as
1
Λ= (I + M R2 + mR2 ) φ̇ + mR cos θ θ̇ . (8.130)
R
This, in turn, allows us to eliminate φ̇ in terms of θ̇ and the constant Λ:

γ Λ
φ̇ = − θ̇ cos θ , (8.131)
1 + γ mR
where
mR2
γ= . (8.132)
I + M R2
The energy is then
E = 12 (I + M R2 ) φ̇2 + 12 m R2 φ̇2 + R2 θ̇2 + 2R2 cos θ φ̇ θ̇ + M gR + mgR(1 − cos θ)

( )
1 + γ sin2 θ Λ 2 2M g

1 2 2 2g γ
= 2 mR θ̇ + (1 − cos θ) + + . (8.133)
1+γ R 1 + γ mR mR
The last two terms inside the big bracket are constant, so we can write this as
1 + γ sin2 θ

2g 4gk
θ̇2 + (1 − cos θ) = . (8.134)
1+γ R R
Here, k is a dimensionless measure of the energy of the system, after subtracting the afore-
mentioned constants. If k > 1, then θ̇2 > 0 for all θ, which would result in ‘loop-the-loop’
motion of the point mass inside the hoop – provided, that is, the normal force of the hoop
doesn’t vanish and the point mass doesn’t detach from the hoop’s surface.
Equation motion for θ(t)
The equation of motion for θ obtained by eliminating all other variables from the original
set of ten equations is the same as Ė = 0, and may be written
1 + γ sin2 θ

γ sin θ cos θ g
θ̈ + θ̇2 = − . (8.135)
1+γ 1+γ R
We can use this to write θ̈ in terms of θ̇2 , and, after invoking eqn. 22.51, in terms of θ itself.
We find

2 4g 1+γ
k − sin2 12 θ

θ̇ = · 2 (8.136)
R 1 + γ sin θ
g (1 + γ) sin θ h 21
2
i
θ̈ = − · 4γ k − sin 2 θ cos θ + 1 + γ sin θ . (8.137)
R 1 + γ sin2 θ 2

Forces of constraint
P ∂Gj
We can solve for the λj , and thus obtain the forces of constraint Qσ = j λj ∂qσ .
λ3 = mẍ = mR φ̈ + mR cos θ θ̈ − mR sin θ θ̇2

mR h i
= θ̈ cos θ − θ̇2 sin θ (8.138)
1+γ
λ4 = mÿ + mg = mg + mR sin θ θ̈ + mR cos θ θ̇2

h gi
= mR θ̈ sin θ + θ̇2 sin θ + (8.139)
R
I (1 + γ)I
λ1 = − φ̈ = λ3 (8.140)
R mR2
λ2 = (M + m)g + mÿ = λ4 + M g . (8.141)
One can check that λ3 cos θ + λ4 sin θ = 0.
The condition that the normal force of the hoop on the point mass vanish is λ3 = 0, which
entails λ4 = 0. This gives
−(1 + γ sin2 θ) cos θ = 4(1 + γ) k − sin2 12 θ .

(8.142)
Note that this requires cos θ < 0, i.e. the point of detachment lies above the horizontal
diameter of the hoop. Clearly if k is sufficiently large, the equality cannot be satisfied, and
the point mass executes a periodic ‘loop-the-loop’ motion. In particular, setting θ = π, we
find that
1
kc = 1 + . (8.143)
4(1 + γ)
If k > kc , then there is periodic ‘loop-the-loop’ motion. If k < kc , then the point mass may
detach at a critical angle θ∗ , but only if the motion allows for cos θ < 0. From the energy
conservation equation, we have that the maximum value of θ achieved occurs when θ̇ = 0,
which means
cos θmax = 1 − 2k . (8.144)
If 12 < k < kc , then, we have the possibility of detachment. This means the energy must be
large enough but not too large.
Chapter 9
Central Forces and Orbital

Mechanics
9.1 Reduction to a one-body problem

Consider two particles interacting via a potential U (r1 , r2 ) = U |r1 − r2 | . Such a poten-
tial, which depends only on the relative distance between the particles, is called a central
potential. The Lagrangian of this system is then
L = T − U = 12 m1 ṙ12 + 12 m2 ṙ22 − U |r1 − r2 | .

(9.1)
9.1.1 Center-of-mass (CM) and relative coordinates
The two-body central force problem may always be reduced to two independent one-body
problems, by transforming to center-of-mass (R) and relative (r) coordinates (see Fig. 9.1),
viz.
m1 r1 + m2 r2 m2
R= r1 = R + r (9.2)
m1 + m2 m1 + m2
m1
r = r1 − r2 r2 = R − r (9.3)
m1 + m2
We then have
L = 21 m1 ṙ1 2 + 12 m2 ṙ2 2 − U |r1 − r2 |

(9.4)
= 21 M Ṙ2 + 12 µṙ 2 − U (r) . (9.5)
141
142 CHAPTER 9. CENTRAL FORCES AND ORBITAL MECHANICS
Figure 9.1: Center-of-mass (R) and relative (r) coordinates.
where
M = m1 + m2 (total mass) (9.6)
m1 m2
µ= (reduced mass) . (9.7)
m1 + m2
9.1.2 Solution to the CM problem
We have ∂L/∂R = 0, which gives R̈ = 0 and hence

R(t) = R(0) + Ṙ(0) t . (9.8)
Thus, the CM problem is trivial. The center-of-mass moves at constant velocity.
9.1.3 Solution to the relative coordinate problem
Angular momentum conservation: We have that ` = r × p = µr × ṙ is a constant of the

motion. This means that the motion r(t) is confined to a plane perpendicular to `. It is
convenient to adopt two-dimensional polar coordinates (r, φ). The magnitude of ` is
` = µr2 φ̇ = 2µȦ (9.9)
where dA = 21 r2 dφ is the differential element of area subtended relative to the force center.
The relative coordinate vector for a central force problem subtends equal areas in equal times.
This is known as Kepler’s Second Law.
Energy conservation: The equation of motion for the relative coordinate is

d ∂L ∂L ∂U
= ⇒ µr̈ = − . (9.10)
dt ∂ ṙ ∂r ∂r
9.1. REDUCTION TO A ONE-BODY PROBLEM 143
Taking the dot product with ṙ, we have
∂U
0 = µr̈ · ṙ + · ṙ
∂r
d 1 2
n o dE
= µṙ + U (r) = . (9.11)
dt 2 dt
Thus, the relative coordinate contribution to the total energy is itself conserved. The total
energy is of course Etot = E + 12 M Ṙ2 .
Since ` is conserved, and since r · ` = 0, all motion is confined to a plane perpendicular to

`. Choosing coordinates such that ẑ = ˆ`, we have
`2
E = 12 µṙ 2 + U (r) = 12 µṙ2 + + U (r)
2µr2
= 12 µṙ2 + Ueff (r) (9.12)
`2
Ueff (r) = + U (r) . (9.13)
2µr2
Integration of the Equations of Motion, Step I: The second order equation for r(t)
is
dE `2 dU (r) dUeff (r)
= 0 ⇒ µr̈ = 3 − =− . (9.14)
dt µr dr dr
However, conservation of energy reduces this to a first order equation, via
q
µ
2 dr
r
2
ṙ = ± E − Ueff (r) ⇒ dt = ± q . (9.15)
µ 2
E − ` − U (r) 2µr 2
This gives t(r), which must be inverted to obtain r(t). In principle this is possible. Note
that a constant of integration also appears at this stage – call it r0 = r(t = 0).
Integration of the Equations of Motion, Step II: After finding r(t) one can inte-
grate to find φ(t) using the conservation of `:
` `
φ̇ = ⇒ dφ = dt . (9.16)
µr2 µr2 (t)
This gives φ(t), and introduces another constant of integration – call it φ0 = φ(t = 0).
Pause to Reflect on the Number of Constants: Confined to the plane perpendicular

to `, the relative coordinate vector has two degrees of freedom. The equations of motion
are second order in time, leading to four constants of integration. Our four constants are
E, `, r0 , and φ0 .
The original problem involves two particles, hence six positions and six velocities, making
for 12 initial conditions. Six constants are associated with the CM system: R(0) and Ṙ(0).
The six remaining constants associated with the relative coordinate system are ` (three
components), E, r0 , and φ0 .
Geometric Equation of the Orbit: From ` = µr2 φ̇, we have

d ` d
= 2 , (9.17)
dt µr dφ
leading to
2
d2r µr4

2 dr
2
− = F (r) + r (9.18)
dφ r dφ `2
where F (r) = −dU (r)/dr is the magnitude of the central force. This second order equation
may be reduced to a first order one using energy conservation:
E = 12 µṙ2 + Ueff (r)
2
`2 dr
= 4
+ Ueff (r) . (9.19)
2µr dφ
Thus,
` dr
dφ = ± √ · p , (9.20)
2µ r2 E − Ueff (r)
which can be integrated to yield φ(r), and then inverted to yield r(φ). Note that only
one integration need be performed to obtain the geometric shape of the orbit, while two
integrations – one for r(t) and one for φ(t) – must be performed to obtain the full motion
of the system.
It is sometimes convenient to rewrite this equation in terms of the variable s = 1/r:

d2s µ −1

+ s = − F s . (9.21)
dφ2 `2 s2
As an example, suppose the geometric orbit is r(φ) = k eαφ , known as a logarithmic spiral.
What is the force? We invoke (9.18), with s00 (φ) = α2 s, yielding
`2 C
F s−1 = −(1 + α2 ) s3

⇒ F (r) = − (9.22)
µ r3
with
µC
α2 = −1 . (9.23)
`2
The general solution for s(φ) for this force law is

A cosh(αφ) + B sinh(−αφ)
 if `2 > µC
s(φ) = (9.24)

 0
A cos |α|φ + B 0 sin |α|φ `2

if < µC .
The logarithmic spiral shape is a special case of the first kind of orbit.
9.2. ALMOST CIRCULAR ORBITS 145
Figure 9.2: Stable and unstable circular orbits. Left panel: U (r) = −k/r produces a stable
circular orbit. Right panel: U (r) = −k/r4 produces an unstable circular orbit.
9.2 Almost Circular Orbits
A circular orbit with r(t) = r0 satisfies r̈ = 0, which means that Ueff 0 (r ) = 0, which says
0
that F (r0 ) = −`2 /µr03 . This is negative, indicating that a circular orbit is possible only if
the force is attractive over some range of distances. Since ṙ = 0 as well, we must also have
E = Ueff (r0 ). An almost circular orbit has r(t) = r0 + η(t), where |η/r0 | 1. To lowest
order in η, one derives the equations
d2η 1 00
= −ω 2 η , ω2 = U (r ) . (9.25)
dt2 µ eff 0
If ω 2 > 0, the circular orbit is stable and the perturbation oscillates harmonically. If ω 2 < 0,
the circular orbit is unstable and the perturbation grows exponentially. For the geometric
shape of the perturbed orbit, we write r = r0 + η, and from (9.18) we obtain
d2 η µr04 0

= F (r 0 ) − 3 η = −β 2 η , (9.26)
dφ2 `2
with
2 d ln F (r)
β =3+ . (9.27)
d ln r

r0
The solution here is

η(φ) = η0 cos β(φ − δ0 ) , (9.28)
where η0 and δ0 are initial conditions. Setting η = η0 , we obtain the sequence of φ values
2πn
φn = δ0 + , (9.29)
β
at which η(φ) is a local maximum, i.e. at apoapsis, where r = r0 + η0 . Setting r = r0 − η0

is the condition for closest approach, i.e. periapsis. This yields the identical set if angles,
just shifted by π. The difference,
∆φ = φn+1 − φn − 2π = 2π β −1 − 1 ,

(9.30)
is the amount by which the apsides (i.e. periapsis and apoapsis) precess during each cycle.
If β > 1, the apsides advance, i.e. it takes less than a complete revolution ∆φ = 2π between
successive periapses. If β < 1, the apsides retreat, and it takes longer than a complete
revolution between successive periapses. The situation is depicted in Fig. 9.3 for the case
β = 1.1. Below, we will exhibit a soluble model in which the precessing orbit may be
determined exactly. Finally, note that if β = p/q is a rational number, then the orbit is
closed , i.e. it eventually retraces itself, after every q revolutions.
As an example, let F (r) = −kr−α . Solving for a circular orbit, we write
0 k `2
Ueff (r) = − =0, (9.31)
rα µr3
which has a solution only for k > 0, corresponding to an attractive potential. We then find
2 1/(3−α)
`
r0 = , (9.32)
µk
and β 2 = 3 − α. The shape of the perturbed orbits follows from η 00 = −β 2 η. Thus, while
circular orbits exist whenever k > 0, small perturbations about these orbits are stable only
for β 2 > 0, i.e. for α < 3. One then has η(φ) = A cos β(φ − φ0 ). The perturbed orbits are
closed, at least to lowest order in η, for α = 3 − (p/q)2 , i.e. for β = p/q. The situation is
depicted in Fig. 9.2, for the potentials U (r) = −k/r (α = 2) and U (r) = −k/r4 (α = 5).
9.3 Precession in a Soluble Model
Let’s start with the answer and work backwards. Consider the geometrical orbit,
r0
r(φ) = . (9.33)
1 − cos βφ
Our interest is in bound orbits, for which 0 ≤ < 1 (see Fig. 9.3). What sort of potential
gives rise to this orbit? Writing s = 1/r as before, we have
s(φ) = s0 (1 − ε cos βφ) . (9.34)

Substituting into (9.21), we have
µ −1
d2s
− F s = +s
`2 s2 dφ2
= β 2 s0 cos βφ + s
= (1 − β 2 ) s + β 2 s0 , (9.35)
9.3. PRECESSION IN A SOLUBLE MODEL 147
Figure 9.3: Precession in a soluble model, with geometric orbit r(φ) = r0 /(1 − ε cos βφ),
shown here with β = 1.1. Periapsis and apoapsis advance by ∆φ = 2π(1 − β −1 ) per cycle.
from which we conclude

k C
F (r) = − 2
+ 3 , (9.36)
r r
with
`2 `2
k = β 2 s0 , C = (β 2 − 1) . (9.37)
µ µ
The corresponding potential is
k C
U (r) = − + 2 + U∞ , (9.38)
r 2r
where U∞ is an arbitrary constant, conveniently set to zero. If µ and C are given, we have
r
`2 C µC
r0 = + , β = 1+ 2 . (9.39)
µk k `
When C = 0, these expressions recapitulate those from the Kepler problem. Note that
when `2 + µC < 0 that the effective potential is monotonically increasing as a function of
r. In this case, the angular momentum barrier is overwhelmed by the (attractive, C < 0)
inverse square part of the potential, and Ueff (r) is monotonically increasing. The orbit then
passes through the force center. It is a useful exercise to derive the total energy for the
orbit,
µk 2 2E(`2 + µC)
E = (ε2 − 1) ⇐⇒ ε 2
= 1 + . (9.40)
2(`2 + µC) µk 2
Figure 9.4: The effective potential for the Kepler problem, and associated phase curves.
The orbits are geometrically described as conic sections: hyperbolae (E > 0), parabolae
(E = 0), ellipses (Emin < E < 0), and circles (E = Emin ).
9.4 The Kepler Problem: U (r) = −k r−1
9.4.1 Geometric shape of orbits
The force is F (r) = −kr−2 , hence the equation for the geometric shape of the orbit is
d2 s µ µk
2
+ s = − 2 2 F (s−1 ) = 2 , (9.41)
dφ ` s `
with s = 1/r. Thus, the most general solution is
s(φ) = s0 − C cos(φ − φ0 ) , (9.42)
where C and φ0 are constants. Thus,
r0
r(φ) = , (9.43)
1 − ε cos(φ − φ0 )
where r0 = `2 /µk and where we have defined a new constant ε ≡ Cr0 .
9.4.2 Laplace-Runge-Lenz vector
Consider the vector

A = p × ` − µk r̂ , (9.44)
9.4. THE KEPLER PROBLEM: U (R) = −K R−1 149
where r̂ = r/|r| is the unit vector pointing in the direction of r. We may now show that A
is conserved:
dA dn ro
= p × ` − µk
dt dt r
rṙ − r ṙ
= ṗ × ` + p × `˙ − µk
r2
kr ṙ ṙr
= − 3 × (µr × ṙ) − µk + µk 2
r r r
r(r · ṙ) ṙ(r · r) ṙ ṙr
= −µk 3
+ µk 3
− µk + µk 2 = 0 . (9.45)
r r r r
So A is a conserved vector which clearly lies in the plane of the motion. A points toward
periapsis, i.e. toward the point of closest approach to the force center.
Let’s assume apoapsis occurs at φ = φ0 . Then
A · r = −Ar cos(φ − φ0 ) = `2 − µkr (9.46)
giving
`2 a(1 − ε2 )
r(φ) = = , (9.47)
µk − A cos(φ − φ0 ) 1 − ε cos(φ − φ0 )
where
A `2
ε= , a(1 − ε2 ) = . (9.48)
µk µk
The orbit is a conic section with eccentricity ε. Squaring A, one finds
A2 = (p × `)2 − 2µk r̂ · p × ` + µ2 k 2
k
= p2 `2 − 2µ`2
+ µ2 k 2
r
2
k µk 2 µk 2

2 p 2
= 2µ` − + 2 = 2µ` E + 2 (9.49)
2µ r 2` 2`
and thus
k 2E`2
a=− , ε2 = 1 + . (9.50)
2E µk 2
9.4.3 Kepler orbits are conic sections
There are four classes of conic sections:
• Circle: ε = 0, E = −µk 2 /2`2 , radius a = `2 /µk. The force center lies at the center of
circle.
• Ellipse: 2 2
√ 0 < ε < 1, −µk /2` < E < 0, semimajor axis a = −k/2E, semiminor axis
b = a 1 − ε2 . The force center is at one of the foci.
Figure 9.5: Keplerian orbits are conic sections, classified according to eccentricity: hyper-
bola ( > 1), parabola ( = 1), ellipse (0 < < 1), and circle ( = 0). The Laplace-Runge-
Lenz vector, A, points toward periapsis.
• Parabola: ε = 1, E = 0, force center is the focus.
• Hyperbola: ε > 1, E > 0, force center is closest focus (attractive) or farthest focus
(repulsive).
To see that the Keplerian orbits are indeed conic sections, consider the ellipse of Fig. 9.6.
The law of cosines gives
ρ2 = r2 + 4f 2 − 4rf cos φ , (9.51)
where f = εa is the focal distance. Now for any point on an ellipse, the sum of the distances
to the left and right foci is a constant, and taking φ = 0 we see that this constant is 2a.
Thus, ρ = 2a − r, and we have
(2a − r)2 = 4a2 − 4ar + r2 = r2 + 4ε2 a2 − 4εr cos φ

⇒ r(1 − ε cos φ) = a(1 − ε2 ) . (9.52)
Thus, we obtain
a (1 − ε2 )
r(φ) = , (9.53)
1 − ε cos φ
and we therefore conclude that
`2
r0 = = a (1 − ε2 ) . (9.54)
µk
Figure 9.6: The Keplerian ellipse, with the force center at the left focus. The focal distance
is f √= εa, where a is the semimajor axis length. The length of the semiminor axis is
b = 1 − ε2 a.
Next let us examine the energy,
E = 12 µṙ2 + Ueff (r)

` dr 2 `2

1 k
= 2µ 2
+ 2
−
µr dφ 2µr r
`2 ds 2 `2 2

= + s − ks , (9.55)
2µ dφ 2µ
with
1 µk
s= = 2 1 − ε cos φ . (9.56)
r `
Thus,
ds µk
= 2 ε sin φ , (9.57)
dφ `
and
2
µ2 k 2 2

ds
= ε sin2 φ
dφ `4
2
µ2 k 2 ε2

µk
= − −s
`4 `2
2µk µ2 k 2
= −s2 + 2 s + 4 ε2 − 1 .

(9.58)
` `
Substituting this into eqn. 9.55, we obtain
µk 2 2
E= ε − 1 . (9.59)
2`2
Figure 9.7: The Keplerian hyperbolae, with the force center at the left focus. The left (blue)
branch corresponds to an attractive potential, while the right (red) branch corresponds to
a repulsive potential. The equations of these branches are r = ρ = ∓2a, where the top sign
corresponds to the left branch and the bottom sign to the right branch.
For the hyperbolic orbit, depicted in Fig. 9.7, we have r − ρ = ∓2a, depending on whether
we are on the attractive or repulsive branch, respectively. We then have
(r ± 2a)2 = 4a2 ± 4ar + r2 = r2 + 4ε2 a2 − 4εr cos φ
⇒ r(±1 + ε cos φ) = a(ε2 − 1) . (9.60)
This yields
a (ε2 − 1)
r(φ) = . (9.61)
±1 + ε cos φ
9.4.4 Period of bound Kepler orbits

√
From ` = µr2 φ̇ = 2µȦ, the period is τ = 2µA/`, where A = πa2 1 − ε2 is the area enclosed
by the orbit. This gives
3 1/2 3 1/2
µa a
τ = 2π = 2π (9.62)
k GM
as well as
a3 GM
2
= , (9.63)
τ 4π 2
where k = Gm1 m2 and M = m1 + m2 is the total mass. For planetary orbits, m1 = M is
the solar mass and m2 = mp is the planetary mass. We then have
a3 mp GM GM
= 1 + ≈ , (9.64)
τ2 M 4π 2 4π 2
which is to an excellent approximation independent of the planetary mass. (Note that

mp /M ≈ 10−3 even for Jupiter.) This analysis also holds, mutatis mutandis, for the
case of satellites orbiting the earth, and indeed in any case where the masses are grossly
disproportionate in magnitude.
9.4.5 Escape velocity
The threshold for escape from a gravitational potential occurs at E = 0. Since E = T + U

is conserved, we determine the escape velocity for a body a distance r from the force center
by setting r
1 2 GM m 2G(M + m)
E = 0 = 2 µvesc (t) − ⇒ vesc (r) = . (9.65)
r r
p
When √ M m, vesc (r) = 2GM/r. Thus, for an object at the surface of the earth,
vesc = 2gRE = 11.2 km/s.
9.4.6 Satellites and spacecraft
A satellite in a circular orbit a distance h above the earth’s surface has an orbital period
2π
τ=√ (RE + h)3/2 , (9.66)
GME
msatellite ME . For low earth orbit (LEO), h RE = 6.37 × 106 m, in which

where we take p
case τLEO = 2π RE /g = 1.4 hr.
Consider a weather satellite in an elliptical orbit whose closest approach to the earth
(perigee) is 200 km above the earth’s surface and whose farthest distance (apogee) is 7200
km above the earth’s surface. What is the satellite’s orbital period? From Fig. 9.6, we see
that
dapogee = RE + 7200 km = 13571 km

dperigee = RE + 200 km = 6971 km
a = 12 (dapogee + dperigee ) = 10071 km . (9.67)
We then have a 3/2

τ= · τLEO ≈ 2.65 hr . (9.68)
RE
What happens if a spacecraft in orbit about the earth fires its rockets? Clearly the energy
and angular momentum of the orbit will change, and this means the shape will change. If
the rockets are fired (in the direction of motion) at perigee, then perigee itself is unchanged,
because v ·qr = 0 is left unchanged at this point. However, E is increased, hence the eccen-
2E`2
tricity ε = 1+ µk2
increases. This is the most efficient way of boosting a satellite into an
Figure 9.8: At perigee of an elliptical orbit ri (φ), a radial impulse ∆p is applied. The shape
of the resulting orbit rf (φ) is shown.
orbit with higher eccentricity. Conversely, and somewhat paradoxically, when a satellite in
LEO loses energy due to frictional drag of the atmosphere, the energy E decreases. Initially,
because the drag is weak and the atmosphere is isotropic, the orbit remains circular. Since
E decreases, hT i = −E must increase, which means that the frictional forces cause the
satellite to speed up!
9.4.7 Two examples of orbital mechanics
• Problem #1: At perigee of an elliptical Keplerian orbit, a satellite receives an impulse

∆p = p0 r̂. Describe the resulting orbit.
◦ Solution #1: Since the impulse is radial, the angular momentum ` = r × p is un-
changed. The energy, however, does change, with ∆E = p20 /2µ. Thus,
2Ef `2 `p0 2

2 2
εf = 1 + = εi + . (9.69)
µk 2 µk
The new semimajor axis length is
`2 /µk 1 − ε2i
af = = ai ·
1 − ε2f 1 − ε2f
ai
= . (9.70)
1 − (ai p20 /µk)
The shape of the final orbit must also be a Keplerian ellipse, described by
`2 1
rf (φ) = · , (9.71)
µk 1 − ε cos(φ + δ)
f
Figure 9.9: The larger circular orbit represents the orbit of the earth. The elliptical orbit
represents that for an object orbiting the Sun with distance at perihelion equal to the Sun’s
radius.
where the phase shift δ is determined by setting
`2 1
ri (π) = rf (π) = · . (9.72)
µk 1 + ε
i
Solving for δ, we obtain

δ = cos−1 εi /εf .

(9.73)
The situation is depicted in Fig. 9.8.
• Problem #2: Which is more energy efficient – to send nuclear waste outside the solar
system, or to send it into the Sun?
p
◦ Solution #2: Escape velocity for the solar system is vesc, (r) = GM /r. At a
√ p
distance aE , we then have vesc, (aE ) = 2 vE , where vE = GM /aE = 2πaE /τE =
29.9 km/s is the velocity of the earth in its orbit. The satellite is launched from earth,
and clearly the most energy efficient launch will be one in the direction of√the earth’s

motion, in which case the velocity after escape from earth must be u = 2 − 1 vE =
12.4 km/s. The speed just above the earth’s atmosphere must then be ũ, where
2 GME m
1
2 mũ − = 12 mu2 , (9.74)
RE
or, in other words,
ũ2 = u2 + vesc,
2
E . (9.75)
We compute ũ = 16.7 km/s.
The second method is to place the trash ship in an elliptical orbit whose perihelion
is the Sun’s radius, R = 6.98 × 108 m, and whose aphelion is aE . Using the general
equation r(φ) = (`2 /µk)/(1 − ε cos φ) for a Keplerian ellipse, we therefore solve the
two equations
1 `2
r(φ = π) = R = · (9.76)
1 + ε µk
1 `2
r(φ = 0) = aE = · . (9.77)
1 − ε µk
We thereby obtain
aE − R
ε= = 0.991 , (9.78)
aE + R
which is a very eccentric ellipse, and
`2 a2E v 2 v2
= ≈ aE · 2
µk G(M + m) vE
2aE R
= (1 − ε) aE = . (9.79)
aE + R
Hence,
2R
v2 = v2 , (9.80)
aE + R E
and the necessary velocity relative to earth is
s !
2R
u= − 1 vE ≈ −0.904 vE , (9.81)
aE + R
i.e. u = −27.0 km/s. Launch is in the opposite direction from the earth’s orbital mo-
tion, and from ũ2 = u2 + vesc,
2
E we find ũ = −29.2 km/s, which is larger (in magnitude)
than in the first scenario. Thus, it is cheaper to ship the trash out of the solar system
than to send it crashing into the Sun, by a factor ũ2I /ũ2II = 0.327.
9.5 Appendix I : Mission to Neptune
Four earth-launched spacecraft have escaped the solar system: Pioneer 10 (launch 3/3/72),
Pioneer 11 (launch 4/6/73), Voyager 1 (launch 9/5/77), and Voyager 2 (launch 8/20/77).1
The latter two are still functioning, and each are moving away from the Sun at a velocity
of roughly 3.5 AU/yr.
As the first objects of earthly origin to leave our solar system, both Pioneer spacecraft
featured a graphic message in the form of a 6” x 9” gold anodized plaque affixed to the
spacecrafts’ frame. This plaque was designed in part by the late astronomer and popular
science writer Carl Sagan. The humorist Dave Barry, in an essay entitled Bring Back Carl’s
Plaque, remarks,
1
There is a very nice discussion in the Barger and Olsson book on ‘Grand Tours of the Outer Planets’.
Here I reconstruct and extend their discussion.
9.5. APPENDIX I : MISSION TO NEPTUNE 157
Figure 9.10: The unforgivably dorky Pioneer 10 and Pioneer 11 plaque.
But the really bad part is what they put on the plaque. I mean, if we’re going to
have a plaque, it ought to at least show the aliens what we’re really like, right?
Maybe a picture of people eating cheeseburgers and watching “The Dukes of
Hazzard.” Then if aliens found it, they’d say, “Ah. Just plain folks.”
But no. Carl came up with this incredible science-fair-wimp plaque that features
drawings of – you are not going to believe this – a hydrogen atom and naked
people. To represent the entire Earth! This is crazy! Walk the streets of any
town on this planet, and the two things you will almost never see are hydrogen
atoms and naked people.
During August, 1989, Voyager 2 investigated the planet Neptune. A direct trip to Neptune
along a Keplerian ellipse with rp = aE = 1 AU and ra = aN = 30.06 AU would take 30.6
years. To see this, note that rp = a (1 − ε) and ra = a (1 + ε) yield
1
aN − aE
a= 2 aE + aN = 15.53 AU , ε= = 0.9356 . (9.82)
aN + aE
Thus,
a 3/2
1
τ= 2 τE · = 30.6 yr . (9.83)
aE
The energy cost per kilogram of such a mission is computed as follows. Let the speed of
the probe after its escape from earth be vp = λvE , and the speed just above the atmosphere
(i.e. neglecting atmospheric friction) is v0 . For the most efficient launch possible, the probe
is shot in the direction of earth’s instantaneous motion about the Sun. Then we must have
2 GME m
1
2 m v0 − = 12 m (λ − 1)2 vE2 , (9.84)
RE
since the speed of the probe in the frame of the earth is vp − vE = (λ − 1) vE . Thus,
E h i
= 12 v02 = 12 (λ − 1)2 + h vE2 (9.85)
m
GM
vE2 = = 6.24 × 107 J/kg ,
aE
where
ME aE
h≡ · = 7.050 × 10−2 . (9.86)
M R E
Therefore, a convenient dimensionless measure of the energy is
2E v02
η≡ = = (λ − 1)2 + 2h . (9.87)
mvE2 vE2
As we shall derive below, a direct mission to Neptune requires
r
2aN
λ≥ = 1.3913 , (9.88)
aN + aE
√
which is close to the criterion for escape from the solar system, λesc = 2. Note that about
52% of the energy is expended after the probe escapes the Earth’s pull, and 48% is expended
in liberating the probe from Earth itself.
This mission can be done much more economically by taking advantage of a Jupiter flyby, as
shown in Fig. 9.11. The idea of a flyby is to steal some of Jupiter’s momentum and then fly
away very fast before Jupiter realizes and gets angry. The CM frame of the probe-Jupiter
system is of course the rest frame of Jupiter, and in this frame conservation of energy means
that the final velocity uf is of the same magnitude as the initial velocity ui . However, in
the frame of the Sun, the initial and final velocities are vJ + ui and vJ + uf , respectively,
where vJ is the velocity of Jupiter in the rest frame of the Sun. If, as shown in the inset
to Fig. 9.11, uf is roughly parallel to vJ , the probe’s velocity in the Sun’s frame will be
enhanced. Thus, the motion of the probe is broken up into three segments:
I: Earth to Jupiter
II : Scatter off Jupiter’s gravitational pull
III : Jupiter to Neptune
We now analyze each of these segments in detail. In so doing, it is useful to recall that the
general form of a Keplerian orbit is
d `2
= ε2 − 1 a .

r(φ) = , d= (9.89)
1 − ε cos φ µk
Figure 9.11: Mission to Neptune. The figure at the lower right shows the orbits of Earth,
Jupiter, and Neptune in black. The cheapest (in terms of energy) direct flight to Neptune,
shown in blue, would take 30.6 years. By swinging past the planet Jupiter, the satellite can
pick up great speed and with even less energy the mission time can be cut to 8.5 years (red
curve). The inset in the upper left shows the scattering event with Jupiter.
The energy is
µk 2
E = (ε2 − 1)
, (9.90)
2`2
with k = GM m, where M is the mass of either the Sun or a planet. In either case, M
dominates, and µ = M m/(M + m) ' m to extremely high accuracy. The time for the
trajectory to pass from φ = φ1 to φ = φ2 is
Zφ2 Zφ2 Zφ2
`3
Z
dφ µ 2 dφ
T = dt = = dφ r (φ) = 2 2 . (9.91)
` µk

φ̇ 1 − ε cos φ
φ1 φ1 φ1
For reference,
aE = 1 AU aJ = 5.20 AU aN = 30.06 AU
24 27
ME = 5.972 × 10 kg MJ = 1.900 × 10 kg M = 1.989 × 1030 kg
with 1 AU = 1.496 × 108 km. Here aE,J,N and ME,J, are the orbital radii and masses of
Earth, Jupiter, and Neptune, and the Sun. The last thing we need to know is the radius of
Jupiter,
RJ = 9.558 × 10−4 AU .
We need RJ because the distance of closest approach to Jupiter, or perijove, must be RJ or
greater, or else the probe crashes into Jupiter!
9.5.1 I. Earth to Jupiter
The probe’s velocity at perihelion is vp = λvE . The angular momentum is ` = µaE · λvE ,
whence
(aE λvE )2
d= = λ2 aE . (9.92)
GM
From r(π) = aE , we obtain
εI = λ2 − 1 . (9.93)
This orbit will intersect the orbit of Jupiter if ra ≥ aJ , which means
r
d 2aJ
≥ aJ ⇒ λ ≥ = 1.2952 . (9.94)
1 − εI aJ + aE
If this inequality holds, then intersection of Jupiter’s orbit will occur for
aJ − λ2 aE

−1
φJ = 2π − cos . (9.95)
(λ2 − 1) aJ
Finally, the time for this portion of the trajectory is
ZφJ
3 dφ 1
τEJ = τE · λ . (9.96)
2π 1 − (λ2 − 1) cos φ 2

π
9.5.2 II. Encounter with Jupiter
We are interested in the final speed vf of the probe after its encounter with Jupiter. We
will determine the speed vf and the angle δ which the probe makes with respect to Jupiter
after its encounter. According to the geometry of Fig. 9.11,
vf2 = vJ2 + u2 − 2uvJ cos(χ + γ) (9.97)

v2 J + −vf2 u2
cos δ = (9.98)
2vf vJ
Note that
GM aE 2
vJ2 = = ·v . (9.99)
aJ aJ E
But what are u, χ, and γ?
To determine u, we invoke
u2 = vJ2 + vi2 − 2vJ vi cos β . (9.100)
The initial velocity (in the frame of the Sun) when the probe crosses Jupiter’s orbit is given
by energy conservation:
2 GM m GM m
1
2 m(λvE ) − = 12 mvi2 − , (9.101)
aE aJ
which yields 2aE 2
vi2 = λ2 − 2 + vE . (9.102)
aJ
As for β, we invoke conservation of angular momentum:
aE
µ(vi cos β)aJ = µ(λvE )aE ⇒ vi cos β = λ vE . (9.103)
aJ
The angle γ is determined from
vJ = vi cos β + u cos γ . (9.104)
Putting all this together, we obtain

p
vi = vE λ2 − 2 + 2x (9.105)
p
u = vE λ2 − 2 + 3x − 2λx3/2 (9.106)
√
x − λx
cos γ = p , (9.107)
λ2 − 2 + 3x − 2λx3/2
where
aE
x≡ = 0.1923 . (9.108)
aJ
We next consider the scattering of the probe by the planet Jupiter. In the Jovian frame,
we may write
κRJ (1 + εJ )
r(φ) = , (9.109)
1 + εJ cos φ
where perijove occurs at
r(0) = κRJ . (9.110)
Here, κ is a dimensionless quantity, which is simply perijove in units of the Jovian radius.
Clearly we require κ > 1 or else the probe crashes into Jupiter! The probe’s energy in this
frame is simply E = 12 mu2 , which means the probe enters into a hyperbolic orbit about
Jupiter. Next, from
k ε2 − 1
E= (9.111)
2 `2 /µk
`2
= (1 + ε) κRJ (9.112)
µk
Figure 9.12: Total time for Earth-Neptune mission as a function of dimensionless velocity
at perihelion, λ = vp /vE . Six different values of κ, the value of perijove in units of the
Jovian radius, are shown: κ = 1.0 (thick blue), κ = 5.0 (red), κ = 20 (green), κ = 50
(blue), κ = 100 (magenta), and κ = ∞ (thick black).
we find 2
RJ M u
εJ = 1 + κ . (9.113)
aE MJ vE
The opening angle of the Keplerian hyperbola is then φc = cos−1 ε−1

J , and the angle χ is
related to φc through
1
χ = π − 2φc = π − 2 cos−1 . (9.114)
εJ
Therefore, we may finally write
√
q
vf = x vE2 + u2 + 2 u vE x cos(2φc − γ) (9.115)
x vE2 + vf2 − u2
cos δ = √ . (9.116)
2 vf vE x
9.5.3 III. Jupiter to Neptune
Immediately after undergoing gravitational scattering off Jupiter, the energy and angular
momentum of the probe are
GM m
E = 12 mvf2 − (9.117)
aJ
9.6. APPENDIX II : RESTRICTED THREE-BODY PROBLEM 163
and
` = µ vf aJ cos δ . (9.118)
We write the geometric equation for the probe’s orbit as
d
r(φ) = , (9.119)
1 + ε cos(φ − φJ − α)
where 2
`2

vf aJ cos δ
d= = aE . (9.120)
µk vE aE
Setting E = (µk 2 /2`2 )(ε2 − 1), we obtain the eccentricity
v !
u
u vf2 2aE d
ε= 1+
t − . (9.121)
vE2 aJ aE
√
Note that the orbit is hyperbolic – the probe will escape the Sun – if vf > vE · 2x. The
condition that this orbit intersect Jupiter at φ = φJ yields

1 d
cos α = −1 , (9.122)
ε aJ
which determines the angle α. Interception of Neptune occurs at

d 1 d
= aN ⇒ φN = φJ + α + cos−1 −1 . (9.123)
1 + ε cos(φN − φJ − α) ε aN
We then have
3 ZφN
d dφ 1
τJN = τE · . (9.124)
aE 2π 1 + ε cos(φ − φJ − α) 2

φJ
The total time to Neptune is then the sum,
τEN = τEJ + τJN . (9.125)
In Fig. 9.12, we plot the mission time τEN versus the velocity at perihelion, vp = λvE , for
various values of κ. The value κ = ∞ corresponds to the case of no Jovian encounter at all.
9.6 Appendix II : Restricted Three-Body Problem
Problem : Consider the ‘restricted three body problem’ in which a light object of mass m
(e.g. a satellite) moves in the presence of two celestial bodies of masses m1 and m2 (e.g. the
sun and the earth, or the earth and the moon). Suppose m1 and m2 execute stable circular
motion about their common center of mass. You may assume m m2 ≤ m1 .
Figure 9.13: The Lagrange points for the earth-sun system. Credit: WMAP project.
(a) Show that the angular frequency for the motion of masses 1 and 2 is related to their
(constant) relative separation, by
GM
ω02 = 3 , (9.126)
r0
where M = m1 + m2 is the total mass.
Solution : For a Kepler potential U = −k/r, the circular orbit lies at r0 = `2 /µk, where
` = µr2 φ̇ is the angular momentum and k = Gm1 m2 . This gives
`2 k GM
ω02 = = 3 = 3 , (9.127)
µ2 r04 µr0 r0
with ω0 = φ̇.
(b) The satellite moves in the combined gravitational field of the two large bodies; the
satellite itself is of course much too small to affect their motion. In deriving the motion
for the satellilte, it is convenient to choose a reference frame whose origin is the CM and
which rotates with angular velocity ω0 . In the rotating frame the masses m1 and m2 lie,
respectively, at x1 = −αr0 and x2 = βr0 , with
m2 m1
α= , β= (9.128)
M M
and with y1 = y2 = 0. Note α + β = 1.
Show that the Lagrangian for the satellite in this rotating frame may be written
2 2 G m1 m G m2 m
L = 12 m ẋ − ω0 y + 12 m ẏ + ω0 x + q +q . (9.129)
(x + αr0 )2 + y 2 (x − βr0 )2 + y 2
Solution : Let the original (inertial) coordinates be (x0 , y0 ). Then let us define the rotated
coordinates (x, y) as
x = cos(ω0 t) x0 + sin(ω0 t) y0 (9.130)

y = − sin(ω0 t) x0 + cos(ω0 t) y0 . (9.131)
Therefore,
ẋ = cos(ω0 t) ẋ0 + sin(ω0 t) ẏ0 + ω0 y (9.132)

ẏ = − sin(ω0 t) x0 + cos(ω0 t) y0 − ω0 x . (9.133)
Therefore
(ẋ − ω0 y)2 + (ẏ + ω0 x)2 = ẋ20 + ẏ02 , (9.134)
The Lagrangian is then
2 2 G m1 m G m2 m
L = 21 m ẋ − ω0 y + 12 m ẏ + ω0 x + p +p , (9.135)
(x − x1 )2 + y 2 (x − x2 )2 + y 2
which, with x1 ≡ −αr0 and x2 ≡ βr0 , agrees with eqn. 9.129
(c) Lagrange discovered that there are five special points where the satellite remains fixed
in the rotating frame. These are called the Lagrange points {L1, L2, L3, L4, L5}. A sketch
of the Lagrange points for the earth-sun system is provided in Fig. 9.13. Observation:
In working out the rest of this problem, I found it convenient to measure all distances in
units of r0 and times in units of ω0−1 , and to eliminate G by writing Gm1 = β ω02 r03 and
Gm2 = α ω02 r03 .
Assuming the satellite is stationary in the rotating frame, derive the equations for the
positions of the Lagrange points.
Solution : At this stage it is convenient to measure all distances in units of r0 and times
in units of ω0−1 to factor out a term m r02 ω02 from L, writing the dimensionless Lagrangian
e ≡ L/(m r2 ω 2 ). Using as well the definition of ω 2 to eliminate G, we have
L 0 0 0
β α
L
e= 1
2 (ξ˙ − η)2 + 12 (η̇ + ξ)2 + p +p , (9.136)
(ξ + α)2 + η2 (ξ − β)2 + η 2
with
x y 1 dx 1 dy
ξ≡ , η≡ , ξ˙ ≡ , η̇ ≡ . (9.137)
r0 r0 ω0 r0 dt ω0 r0 dt
β(ξ + α) α(ξ − β)
ξ¨ − 2η̇ = ξ − − (9.138)
d31 d32
βη αη
η̈ + 2ξ˙ = η − 3 − 3 , (9.139)
d1 d2
where p p
d1 = (ξ + α)2 + η 2 , d2 = (ξ − β)2 + η 2 . (9.140)
Here, ξ ≡ x/r0 , ξ = y/r0 , etc. Recall that α + β = 1. Setting the time derivatives to zero
yields the static equations for the Lagrange points:
β(ξ + α) α(ξ − β)
ξ= + (9.141)
d31 d32
βη αη
η= 3 + 3 , (9.142)
d1 d2
(d) Show that the Lagrange points with y = 0 are determined by a single nonlinear equation.
Show graphically that this equation always has three solutions, one with x < x1 , a second
with x1 < x < x2 , and a third with x > x2 . These solutions correspond to the points L3,
L1, and L2, respectively.
Solution : If η = 0 the second equation is automatically satisfied. The first equation then
gives
ξ+α ξ−β
ξ=β· 3 + α · . (9.143)
ξ + α ξ − β 3
The RHS of the above equation diverges to +∞ for ξ = −α + 0+ and ξ = β + 0+ , and

diverges to −∞ for ξ = −α − 0+ and ξ = β − 0+ , where 0+ is a positive infinitesimal. The
situation is depicted in Fig. 9.14. Clearly there are three solutions, one with ξ < −α, one
with −α < ξ < β, and one with ξ > β.
(e) Show that the remaining two Lagrange points, L4 and L5, lie along equilateral triangles
with the two masses at the other vertices.
Solution : If η 6= 0, then dividing the second equation by η yields
β α
1= 3 + 3 . (9.144)
d1 d2
Substituting this into the first equation,

β α 1 1
ξ= + ξ + − αβ , (9.145)
d31 d32 d31 d32
gives
d1 = d2 . (9.146)
Figure 9.14: Graphical solution for the Lagrange points L1, L2, and L3.
Reinserting this into the previous equation then gives the remarkable result,
d1 = d1 = 1 , (9.147)
which says that each of L4 and L5 lies on an equilateral triangle whose two other vertices
are the masses m1 and m2 . The side length of this equilateral triangle is r0 . Thus, the
dimensionless coordinates of L4 and L5 are
√ √
ξL4 , ηL4 = 21 − α, 23 , ξL5 , ηL5 = 21 − α, − 23 . (9.148)
It turns out that L1, L2, and L3 are always unstable. Satellites placed in these positions
must undergo periodic course corrections in order to remain approximately fixed. The SOlar
and Heliopheric Observation satellite, SOHO, is located at L1, which affords a continuous
unobstructed view of the Sun.
(f ) Show that the Lagrange points L4 and L5 are stable (obviously you need only consider
one of them) provided that the mass ratio m1 /m2 is sufficiently large. Determine this
critical ratio. Also find the frequency of small oscillations for motion in the vicinity of L4
and L5.
Solution : Now we write
ξ = ξL4 + δξ , η = ηL4 + δη , (9.149)

and derive the linearized dynamics. Expanding the equations of motion to lowest order in
δξ and δη, we have
! !
∂d 1 ∂d 2 ∂d 1 ∂d 2
δ ξ¨ − 2δ η̇ = 1 − β + 2 β
3 3 3 3

−α− α δξ + 2 β − α δη
∂ξ L4 2 ∂ξ L4
∂η L4
2 ∂η L4
√
3 3 3
= 4 δξ + 4 ε δη (9.150)
and
√
√
! √
√
!
∂d1 ∂d2 ∂d1 ∂d2
δ η̈ + 2δ ξ˙ = 3 3
2 β + 3 3
2 α δξ + 3 3
2 β + 3 3
2 α δη
∂ξ L4 ∂ξ L4 ∂η L4 ∂η L4
√
3 3
= 4 ε δξ + 94 δη , (9.151)

m1 − m2
ε≡β−α= . (9.152)
m1 + m2
As defined, ε ∈ [0, 1].
Fourier transforming the differential equation, we replace each time derivative by (−iν),
and thereby obtain √
ν2 +√ 3
−2iν + 34 3 ε δ ξˆ

4 =0. (9.153)
3 2 9
2iν + 4 3 ε ν +4 δ η̂
Nontrivial solutions exist only when the determinant D vanishes. One easily finds
D(ν 2 ) = ν 4 − ν 2 + 27 2

16 1 − ε , (9.154)
which yields a quadratic equation in ν 2 , with roots

p
ν 2 = 12 ± 41 27 ε2 − 23 . (9.155)
These frequencies are dimensionless. To convert to dimensionful units, we simply multiply

the solutions for ν by ω0 , since we have rescaled time by ω0−1 .
Note that the L4 and L5 points are stable only if ε2 > 2327 . If we define the mass ratio
γ ≡ m1 /m2 , the stability condition is equivalent to
√ √
m1 27 + 23
γ= >√ √ = 24.960 , (9.156)
m2 27 − 23
which is satisfied for both the Sun-Jupiter system (γ = 1047) – and hence for the Sun and
any planet – and also for the Earth-Moon system (γ = 81.2).
Objects found at the L4 and L5 points are called Trojans, after the three large asteroids
Agamemnon, Achilles, and Hector found orbiting in the L4 and L5 points of the Sun-Jupiter
system. No large asteroids have been found in the L4 and L5 points of the Sun-Earth system.
Personal aside : David T. Wilkinson
The image in fig. 9.13 comes from the education and outreach program of the Wilkinson
Microwave Anisotropy Probe (WMAP) project, a NASA mission, launched in 2001, which
has produced some of the most important recent data in cosmology. The project is named
in honor of David T. Wilkinson, who was a leading cosmologist at Princeton, and a founder
of the Cosmic Background Explorer (COBE) satellite (launched in 1989). WMAP was sent
to the L2 Lagrange point, on the night side of the earth, where it can constantly scan the
cosmos with an ultra-sensitive microwave detector, shielded by the earth from interfering
solar electromagnetic radiation. The L2 point is of course unstable, with a time scale of
about 23 days. Satellites located at such points must undergo regular course and attitude
corrections to remain situated.
During the summer of 1981, as an undergraduate at Princeton, I was a member of Wilkin-

son’s “gravity group,” working under Jeff Kuhn and Ken Libbrecht. It was a pretty big
group and Dave – everyone would call him Dave – used to throw wonderful parties at his
home, where we’d always play volleyball. I was very fortunate to get to know David Wilkin-
son a bit – after working in his group that summer I took a class from him the following
year. He was a wonderful person, a superb teacher, and a world class physicist.
Chapter 10
Small Oscillations
10.1 Coupled Coordinates
We assume, for a set of n generalized coordinates {q1 , . . . , qn }, that the kinetic energy is a
quadratic function of the velocities,
1
T = 2 Tσσ0 (q1 , . . . , qn ) q̇σ q̇σ0 , (10.1)
where the sum on σ and σ 0 from 1 to n is implied. For example, expressed in terms of polar
coordinates (r, θ, φ), the matrix Tij is
 
1 0 0
r2 0  =⇒ T = 12 m ṙ2 + r2 θ̇2 + r2 sin2 θ φ̇2 .

Tσσ0 = m 0 (10.2)
0 0 r2 sin2 θ
The potential U (q1 , . . . , qn ) is assumed to be a function of the generalized coordinates alone:

U = U (q). A more general formulation of the problem of small oscillations is given in the
appendix, section 10.8.
The generalized momenta are

∂L
pσ = = Tσσ0 q̇σ0 , (10.3)
∂ q̇σ
and the generalized forces are
∂L 1 ∂Tσ0 σ00 ∂U
Fσ = = q̇ 0 q̇ 00 − . (10.4)
∂qσ 2 ∂qσ σ σ ∂qσ
The Euler-Lagrange equations are then ṗσ = Fσ , or

∂Tσσ0 1 ∂Tσ0 σ00 ∂U
Tσσ0 q̈σ0 + − q̇σ0 q̇σ00 = − (10.5)
∂qσ00 2 ∂qσ ∂qσ
which is a set of coupled nonlinear second order ODEs. Here we are using the Einstein
‘summation convention’, where we automatically sum over any and all repeated indices.
171
172 CHAPTER 10. SMALL OSCILLATIONS
10.2 Expansion about Static Equilibrium
Small oscillation theory begins with the identification of a static equilibrium {q̄1 , . . . , q̄n },
which satisfies the n nonlinear equations

∂U
=0. (10.6)
∂qσ q=q̄
Once an equilibrium is found (note that there may be more than one static equilibrium),
we expand about this equilibrium, writing
qσ ≡ q̄σ + ησ . (10.7)
The coordinates {η1 , . . . , ηn } represent the displacements relative to equilibrium.
We next expand the Lagrangian to quadratic order in the generalized displacements, yielding
L = 21 Tσσ0 η̇σ η̇σ0 − 12 Vσσ0 ησ ησ0 , (10.8)
where
∂2T ∂2U
Tσσ0 = , Vσσ0 = . (10.9)

∂ q̇σ ∂ q̇σ0 ∂qσ ∂qσ0

q=q̄ q=q̄
Writing η t for the row-vector (η1 , . . . , ηn ), we may suppress indices and write
L= 1
2 η̇ t T η̇ − 12 η t V η , (10.10)
where T and V are the constant matrices of eqn. 10.9.
10.3 Method of Small Oscillations
The idea behind the method of small oscillations is to effect a coordinate transformation
from the generalized displacements η to a new set of coordinates ξ, which render the
Lagrangian particularly simple. All that is required is a linear transformation,
ησ = Aσi ξi , (10.11)
where both σ and i run from 1 to n. The n × n matrix Aσi is known as the modal matrix.
With the substitution η = A ξ (hence η t = ξ t At , where Atiσ = Aσi is the matrix transpose),
we have
L = 12 ξ̇ t At T A ξ̇ − 12 ξ t At V A ξ . (10.12)
We now choose the matrix A such that
At T A = 1 (10.13)
t
ω12 , ωn2

A V A = diag ... , . (10.14)
10.3. METHOD OF SMALL OSCILLATIONS 173
With this choice of A, the Lagrangian decouples:

n
X
L = 12 ξ˙i2 − ωi2 ξi2 , (10.15)
i=1
with the solution

ξi (t) = Ci cos(ωi t) + Di sin(ωi t) , (10.16)
where {C1 , . . . , Cn } and {D1 , . . . , Dn } are 2n constants of integration, and where no sum is
implied on i. Note that
ξ = A−1 η = At T η . (10.17)
In terms of the original generalized displacements, the solution is

n
X n o
ησ (t) = Aσi Ci cos(ωi t) + Di sin(ωi t) , (10.18)
i=1
and the constants of integration are linearly related to the initial generalized displacements
and generalized velocities:
Ci = Atiσ Tσσ0 ησ0 (0) (10.19)
Di = ωi−1 Atiσ Tσσ0 η̇σ0 (0) , (10.20)
again with no implied sum on i on the RHS of the second equation, and where we have
used A−1 = At T, from eqn. 10.13. (The implied sums in eqn. 10.20 are over σ and σ 0 .)
√
Note that the normal coordinates have unusual dimensions: [ξ] = M · L, where L is length
and M is mass.
10.3.1 Can you really just choose an A so that both these wonderful
things happen in 10.13 and 10.14?
Yes.
10.3.2 Er...care to elaborate?
Both T and V are symmetric matrices. Aside from that, there is no special relation between
them. In particular, they need not commute, hence they do not necessarily share any
eigenvectors. Nevertheless, they may be simultaneously diagonalized as per 10.13 and 10.14.
Here’s why:
• Since T is symmetric, it can be diagonalized by an orthogonal transformation. That

is, there exists a matrix O1 ∈ O(n) such that
O1t T O1 = Td , (10.21)
where Td is diagonal.
• We may safely assume that T is positive definite. Otherwise the kinetic energy can
become arbitrarily negative, which is unphysical. Therefore, one may form the matrix
−1/2
Td which is the diagonal matrix whose entries are the inverse square roots of the
−1/2
corresponding entries of Td . Consider the linear transformation O1 Td . Its effect
on T is
−1/2 t −1/2
Td O1 T O 1 T d =1. (10.22)
• Since O1 and Td are wholly derived from T, the only thing we know about
e ≡ T−1/2 Ot V O T−1/2
V (10.23)
d 1 1 d
is that it is explicitly a symmetric matrix. Therefore, it may be diagonalized by some

orthogonal matrix O2 ∈ O(n). As T has already been transformed to the identity, the
additional orthogonal transformation has no effect there. Thus, we have shown that
there exist orthogonal matrices O1 and O2 such that
−1/2 −1/2
O2t Td O1t T O1 Td O2 =1 (10.24)
−1/2 −1/2
O2t Td O1t V O1 Td O2 = diag (ω12 , . . . , ωn2 ) . (10.25)
−1/2
All that remains is to identify the modal matrix A = O1 Td O2 .
Note that it is not possible to simultaneously diagonalize three symmetric matrices in gen-
eral.
10.3.3 Finding the modal matrix
While the above proof allows one to construct A by finding the two orthogonal matrices O1
and O2 , such a procedure is extremely cumbersome. It would be much more convenient if
A could be determined in one fell swoop. Fortunately, this is possible.
We start with the equations of motion, T η̈ + V η = 0. In component notation, we have
Tσσ0 η̈σ0 + Vσσ0 ησ0 = 0 . (10.26)
We now assume that η(t) oscillates with a single frequency ω, i.e. ησ (t) = ψσ e−iωt . This
results in a set of linear algebraic equations for the components ψσ :
ω 2 Tσσ0 − Vσσ0 ψσ0 = 0 .

(10.27)
These are n equations in n unknowns: one for each value of σ = 1, . . . , n. Because the
equations are homogeneous and linear, there is always a trivial solution ψ = 0. In fact one
might think this is the only solution, since
? −1
ω2 T − V ψ = 0 ψ = ω2 T − V

=⇒ 0=0. (10.28)
10.3. METHOD OF SMALL OSCILLATIONS 175
However, this fails when the matrix ω 2 T − V is defective1 , i.e. when

det ω 2 T − V = 0 .

(10.29)
Since T and V are of rank n, the above determinant yields an nth order polynomial in ω 2 ,
whose n roots are the desired squared eigenfrequencies {ω12 , . . . , ωn2 }.
Once the n eigenfrequencies are obtained, the modal matrix is constructed as follows. Solve
the equations
n
X (i)
ωi2 Tσσ0 − Vσσ0 ψσ0 = 0 (10.30)
σ 0 =1
which are a set of (n − 1) linearly independent equations among the n components of the
eigenvector ψ (i) . That is, there are n equations (σ = 1, . . . , n), but one linear dependency
since det (ωi2 T − V) = 0. The eigenvectors may be chosen to satisfy a generalized orthogo-
nality relationship,
(j)
ψσ(i) Tσσ0 ψσ0 = δσσ0 . (10.31)
To see this, let us duplicate eqn. 10.30, replacing i with j, and multiply both equations as
follows:
(i)
ψσ(j) × ωi2 Tσσ0 − Vσσ0 ψσ0 = 0 (10.32)
(j)
ψσ(i) × ωj2 Tσσ0 − Vσσ0 ψσ0 = 0 .

(10.33)
Using the symmetry of T and V, upon subtracting these equations we obtain
n
(j)
X
(ωi2 − ωj2 ) ψσ(i) Tσσ0 ψσ0 = 0 , (10.34)
σ,σ 0 =1
where the sums on i and j have been made explicit. This establishes that eigenvectors ψ (i)
and ψ (j) corresponding to distinct eigenvalues ωi2 6= ωj2 are orthogonal: (ψ (i) )t T ψ (j) = 0.
For degenerate eigenvalues, the eigenvectors are not a priori orthogonal, but they may be
orthogonalized via application of the Gram-Schmidt procedure. The remaining degrees of
freedom - one for each eigenvector – are fixed by imposing the condition of normalization:
q
(i) (i) (i) (i) (j)
ψσ → ψσ ψµ Tµµ0 ψµ0 =⇒ ψσ(i) Tσσ0 ψσ0 = δij . (10.35)
(i)
The modal matrix is just the matrix of eigenvectors: Aσi = ψσ .
(i)
With the eigenvectors ψσ thusly normalized, we have
(j)
0 = ψσ(i) ωj2 Tσσ0 − Vσσ0 ψσ0
(j)
= ωj2 δij − ψσ(i) Vσσ0 ψσ0 , (10.36)
with no sum on j. This establishes the result
At V A = diag ω12 , . . . , ωn2 .

(10.37)
1
The label defective has a distastefully negative connotation. In modern parlance, we should instead refer
to such a matrix as determinantally challenged .
10.4 Example: Masses and Springs
Two blocks and three springs are configured as in Fig. 22.6. All motion is horizontal. When
the blocks are at rest, all springs are unstretched.
Figure 10.1: A system of masses and springs.
(a) Choose as generalized coordinates the displacement of each block from its equilibrium
position, and write the Lagrangian.
(b) Find the T and V matrices.
(c) Suppose
m1 = 2m , m2 = m , k1 = 4k , k2 = k , k3 = 2k ,
Find the frequencies of small oscillations.
(d) Find the normal modes of oscillation.
(e) At time t = 0, mass #1 is displaced by a distance b relative to its equilibrium position.

I.e. x1 (0) = b. The other initial conditions are x2 (0) = 0, ẋ1 (0) = 0, and ẋ2 (0) = 0.
Find t∗ , the next time at which x2 vanishes.
Solution
(a) The Lagrangian is
L = 12 m1 x21 + 12 m2 x22 − 12 k1 x21 − 12 k2 (x2 − x1 )2 − 12 k3 x22
(b) The T and V matrices are

! !
∂2T m1 0 ∂2U k1 + k2 −k2
Tij = = , Vij = =
∂ ẋi ∂ ẋj 0 m2 ∂xi ∂xj −k2 k2 + k3
10.4. EXAMPLE: MASSES AND SPRINGS 177
(c) We have m1 = 2m, m2 = m, k1 = 4k, k2 = k, and k3 = 2k. Let us write ω 2 ≡ λ ω02 ,

p
where ω0 ≡ k/m. Then

2 2λ − 5 1
ω T−V=k .
1 λ−3
The determinant is
det (ω 2 T − V) = (2λ2 − 11λ + 14) k 2

= (2λ − 7) (λ − 2) k 2 .
There are two roots: λ− = 2 and λ+ = 72 , corresponding to the eigenfrequencies

r r
2k 7k
ω− = , ω+ =
m 2m
~ (a) = 0. Plugging in λ = 2 we have

(d) The normal modes are determined from (ωa2 T − V) ψ
~ (−)
for the normal mode ψ
(−)
−1 1 ψ1 ~ (−) 1
=0 ⇒ ψ = C−
1 −1 ψ2(−) 1
Plugging in λ = 7 ~ (+)
we have for the normal mode ψ
2
(+)
2 1 ψ1 ~ (+) = C 1
=0 ⇒ ψ
1 12 ψ2(+) + −2
(a) (b)
The standard normalization ψi Tij ψj = δab gives
1 1
C− = √ , C2 = √ . (10.38)
3m 6m
(e) The general solution is

!
x1 1 1 1 1
=A cos(ω− t) + B cos(ω+ t) + C sin(ω− t) + D sin(ω+ t) .
x2 1 −2 1 −2
The initial conditions x1 (0) = b, x2 (0) = ẋ1 (0) = ẋ2 (0) = 0 yield
A = 23 b , B = 13 b , C=0 , D=0.
Thus,

x1 (t) = 13 b · 2 cos(ω− t) + cos(ω+ t)

x2 (t) = 23 b · cos(ω− t) − cos(ω+ t) .
Figure 10.2: The double pendulum.
Setting x2 (t∗ ) = 0, we find
2π
cos(ω− t∗ ) = cos(ω+ t∗ ) ⇒ π − ω− t = ω+ t − π ⇒ t∗ =
ω− + ω+
10.5 Example: Double Pendulum
As a second example, consider the double pendulum, with m1 = m2 = m and `1 = `2 = `.

The kinetic and potential energies are
T = m`2 θ̇12 + m`2 cos(θ1 − θ1 ) θ̇1 θ̇2 + 21 m`2 θ̇22 (10.39)

V = −2mg` cos θ1 − mg` cos θ2 , (10.40)
leading to
2m`2 m`2

2mg` 0
T= , V= . (10.41)
m`2 m`2 0 mg`
Then
2ω 2 − 2ω02 ω2

2 2
ω T − V = m` 2 , (10.42)
ω ω − ω02
2
p
with ω0 = g/`. Setting the determinant to zero gives
√
2(ω 2 − ω02 )2 − ω 4 = 0 ⇒ ω 2 = (2 ± 2) ω02 . (10.43)
We find the unnormalized eigenvectors by setting (ωi2 T − V ) ψ (i) = 0. This gives

+ 1
√ − 1
√
ψ = C+ , ψ = C− , (10.44)
− 2 + 2
10.6. ZERO MODES 179
(i) (j)
where C± are constants. One can check Tσσ0 ψσ ψσ0 vanishes for i 6= j. We then normalize
(i) (i)
by demanding Tσσ0 ψσ ψσ0 = 1 (no sum on i), which determines the coefficients C± =
1
q √
2 (2 ± 2)/m`2 . Thus, the modal matrix is
 p √ p √ 
ψ1+ ψ1−
 
2+ 2 2− 2
A=  = √1 
 .

(10.45)
√ √
2
 p
ψ2+ ψ2− 2 m` p
− 4+2 2 + 4−2 2
10.6 Zero Modes
Recall Noether’s theorem, which says that for every continuous one-parameter family of
coordinate transformations,
qσ −→ q̃σ (q, ζ) , q̃σ (q, ζ = 0) = qσ , (10.46)
which leaves the Lagrangian invariant, i.e. dL/dζ = 0, there is an associated conserved
quantity,
X ∂L ∂ q̃σ dΛ
Λ= satisfies =0. (10.47)
∂ q̇ ∂ζ dt

σ σ
ζ=0
For small oscillations, we write qσ = q̄σ + ησ , hence

X
Λk = Ckσ η̇σ , (10.48)
σ
where k labels the one-parameter families (in the event there is more than one continuous
symmetry), and where
X ∂ q̃σ0
Ckσ = Tσσ0 . (10.49)
∂ζk

0 σ ζ=0
Therefore, we can define the (unnormalized) normal mode

X
ξk = Ckσ ησ , (10.50)
σ
which satisfies ξ¨k = 0. Thus, in systems with continuous symmetries, to each such contin-
uous symmetry there is an associated zero mode of the small oscillations problem, i.e. a
mode with ωk2 = 0.
10.6.1 Example of zero mode oscillations
The simplest example of a zero mode would be a pair of masses m1 and m2 moving fric-
tionlessly along a line and connected by a spring of force constant k. We know from our
Figure 10.3: Coupled oscillations of three masses on a frictionless hoop of radius R. All
three springs have the same force constant k, but the masses are all distinct.
study of central forces that the Lagrangian may be written
L = 12 m1 ẋ21 + 12 m2 ẋ22 − 12 k(x1 − x2 )2

= 12 M Ẋ 2 + 12 µẋ2 − 12 kx2 , (10.51)
where X = (m1 x1 + m2 x2 )/(m1 + m2 ) is the center of mass position, x = x1 − x2 is the

relative coordinate, M = m1 + m2 is the total mass, and µ = m1 m2 /(m1 + m2 ) is the
2
reducedp mass. The relative coordinate obeys ẍ = −ω0 x, where the oscillation frequency is
ω0 = k/µ. The center of mass coordinate obeys Ẍ = 0, i.e. its oscillation frequency is
zero. The center of mass motion is a zero mode.
Another example is furnished by the system depicted in fig. 10.3, where three distinct masses
m1 , m2 , and m3 move around a frictionless hoop of radius R. The masses are connected to
their neighbors by identical springs of force constant k. We choose as generalized coordinates
the angles φσ (σ = 1, 2, 3), with the convention that
φ1 ≤ φ2 ≤ φ3 ≤ 2π + φ1 . (10.52)
Let Rχ be the equilibrium length for each of the springs. Then the potential energy is
n o
U = 21 kR2 (φ2 − φ1 − χ)2 + (φ3 − φ2 − χ)2 + (2π + φ1 − φ3 − χ)2
n o
= 12 kR2 (φ2 − φ1 )2 + (φ3 − φ2 )2 + (2π + φ1 − φ3 )2 + 3χ2 − 4π χ . (10.53)
Note that the equilibrium angle χ enters only in an additive constant to the potential
energy. Thus, for the calculation of the equations of motion, it is irrelevant. It doesn’t
10.6. ZERO MODES 181
matter whether or not the equilibrium configuration is unstretched (χ = 2π/3) or not

(χ 6= 2π/3).
The kinetic energy is simple:

1 2
T = 2R m1 φ̇21 + m2 φ̇22 + m3 φ̇23 . (10.54)
The T and V matrices are then

 
m1 R2 2kR2 −kR2 −kR2
 
0 0
T= 0 m2 R2 0  , V = −kR2 2kR2 −kR2  . (10.55)
 
0 0 m3 R2 −kR2 −kR2 2kR2
We then have  ω2 
Ω12
−2 1 1
ω2
ω 2 T − V = kR2 
 
 1 Ω22
−2 1  .
 (10.56)
ω2
1 1 Ω32
−2
We compute the determinant to find the characteristic polynomial:
P (ω) = det(ω 2 T − V) (10.57)

ω6

1 1 1 1 1 1
= −2 + + ω4 + 3 + 2+ 2 ω2 ,
Ω12 Ω22 Ω32 Ω12 Ω22 Ω22 Ω32 Ω12 Ω32 2
Ω1 Ω2 Ω3
where Ωi2 ≡ k/mi . The equation P (ω) = 0 yields a cubic equation in ω 2 , but clearly ω 2 is
a factor, and when we divide this out we obtain a quadratic equation. One root obviously
is ω12 = 0. The other two roots are solutions to the quadratic equation:
q 2 2 2
2 2 2 2
ω2,3 = Ω1 + Ω2 + Ω3 ± 12 Ω12 − Ω22 + 12 Ω22 − Ω32 + 12 Ω12 − Ω32 . (10.58)
To find the eigenvectors and the modal matrix, we set

 ω2 
j
2 − 2 1 1  (j) 
 Ω1 ψ1
ωj2

  (j) 
 1 −2 1  ψ 2  = 0 , (10.59)

Ω22

ωj2
 (j)
1 1 − 2 ψ3
Ω 2
3
Writing down the three coupled equations for the components of ψ (j) , we find
2 2 2
ωj ωj ωj

(j) (j) (j)
− 3 ψ1 = − 3 ψ2 = − 3 ψ3 . (10.60)
Ω12 Ω22 Ω32
We therefore conclude  −1 
ωj2
 Ω12 −3
 ω2 −1 
(j)

ψ = Cj  j
 Ω22 −3  .
 (10.61)
 ω 2 −1 
j
Ω32
−3
(i) (j)
The normalization condition ψσ Tσσ0 ψσ0 = δij then fixes the constants Cj :
" −2 −2 −2 #
ωj2
2 2
ωj ωj 2
m1 2 −3 + m2 2 −3 + m3 2 −3 C = 1 .
j (10.62)
Ω1 Ω2 Ω3
The Lagrangian is invariant under the one-parameter family of transformations

φσ −→ φσ + ζ (10.63)
for all σ = 1, 2, 3. The associated conserved quantity is
X ∂L ∂ φ̃σ
Λ=
σ ∂ φ̇σ
∂ζ
= R2 m1 φ̇1 + m2 φ̇2 + m3 φ̇3 ,

(10.64)
which is, of course, the total angular momentum relative to the center of the ring. Thus,
from Λ̇ = 0 we identify the zero mode as ξ1 , where

ξ1 = C m1 φ 1 + m2 φ2 + m3 φ 3 , (10.65)
where C is a constant. Recall the relation ησ = Aσi ξi between the generalized displacements
ησ and the normal coordinates ξi . We can invert this relation to obtain
ξi = A−1 t
iσ ησ = Aiσ Tσσ 0 ησ 0 . (10.66)
Here we have used the result At T A = 1 to write
A−1 = At T . (10.67)
This is a convenient result, because it means that if we ever need to express the normal
coordinates in terms of the generalized displacements, we don’t have to invert any matrices
– we just need to do one matrix multiplication. In our case here, the T matrix is diagonal,
so the multiplication is trivial. From eqns. 10.65 and 10.66, we conclude that the matrix
At T must have a first row which is proportional to (m1 , m2 , m3 ). Since these are the very
diagonal entries of T, we conclude that At itself must have a first row which is proportional
to (1, 1, 1), which means that the first column of A is proportional to (1, 1, 1). But this is
2 (1) (1) (1)
confirmed by eqn. 10.60 when we take j = 1, since ωj=1 = 0: ψ1 = ψ2 = ψ3 .
10.7 Chain of Mass Points
Next consider an infinite chain of identical masses, connected by identical springs of spring
constant k and equilibrium length a. The Lagrangian is
X X
L = 12 m ẋ2n − 12 k (xn+1 − xn − a)2
n n
X X
= 1
2m u̇2n − 1
2k (un+1 − un )2 , (10.68)
n n
10.7. CHAIN OF MASS POINTS 183
where un ≡ xn − na − b is the displacement from equilibrium of the nth mass. The constant
b is arbitrary. The Euler-Lagrange equations are

d ∂L ∂L
= mün =
dt ∂ u̇n ∂un
= k(un+1 − un ) − k(un − un−1 )
= k(un+1 + un−1 − 2un ) . (10.69)
Now let us assume that the system is placed on a large ring of circumference N a, where
N 1. Then un+N = un and we may shift to Fourier coefficients,
1 X iqan
un = √ e ûq (10.70)
N q
1 X −iqan
ûq = √ e un , (10.71)
N n
where qj = 2πj/N a, and both sums are over the set j, n ∈ {1, . . . , N }. Expressed in terms
of the {ûq }, the equations of motion become
¨ = √1
X
û q e−iqna ün
N n
k 1 X −iqan
= √ e (un+1 + un−1 − 2un )
m N n
k 1 X −iqan −iqa
= √ e (e + e+iqa − 2) un
m N n
2k
sin2 12 qa ûq

=− (10.72)
m
Thus, the {ûq } are the normal modes of the system (up to a normalization constant), and
the eigenfrequencies are
2k
sin 12 qa .

ωq = (10.73)
m
This means that the modal matrix is
1
Anq = √ eiqan , (10.74)
Nm
where we’ve included the √1 factor for a proper normalization. (The normal modes them-
m
√
selves are then ξq = A†qn Tnn0 un0 = m ûq . For complex A, the normalizations are A† TA = 1
and A† VA = diag(ω12 , . . . , ωN
2 ).
Note that
Tnn0 = m δn,n0 (10.75)

Vnn0 = 2k δn,n0 − k δn,n0 +1 − k δn,n0 −1 (10.76)
and that
N X
X N
(A† TA)qq0 = A∗nq Tnn0 An0 q0
n=1 n0 =1
N N
1 X X −iqan 0 0
= e m δnn0 eiq an
Nm 0 n=1 n =1
N
1 X 0
= ei(q −q)an = δqq0 , (10.77)
N
n=1
and
N X
X N
(A† VA)qq0 = A∗nq Tnn0 An0 q0
n=1 n0 =1
N N
1 X X −iqan 0 0
= e 2k δn,n0 − k δn,n0 +1 − k δn,n0 −1 eiq an
Nm 0 n=1 n =1
N
k 1 X
i(q 0 −q)an 0 0
2 − e−iq a − eiq a

= e
mN
n=1
4k
sin2 1
δqq0 = ωq2 δqq0

= 2 qa (10.78)
m
Since x̂q+G = x̂q , where G = 2π a , we may choose any set of q values such that no two are
separated by an integer multiple of G. The set of points {jG} with j ∈ Z is called the
reciprocal lattice. For a linear chain, the reciprocal lattice is itself a linear chain2 . One
natural set to choose is q ∈ − πa , πa . This is known as the first Brillouin zone of the
reciprocal lattice.
Finally, we can write the Lagrangian itself in terms of the {uq }. One easily finds
X ∗ X
L = 12 m û˙ q û˙ q − k (1 − cos qa) û∗q ûq , (10.79)
q q
where the sum is over q in the first Brillouin zone. Note that
û−q = û−q+G = û∗q . (10.80)
This means that we can restrict the sum to half the Brillouin zone:
X ∗ 4k ∗

1
L = 2m ˙ ˙
ûq ûq − 2 1
sin 2 qa ûq ûq . (10.81)
π
m
q∈[0, a ]
2
For higher dimensional Bravais lattices, the reciprocal lattice is often different than the real space
(“direct”) lattice. For example, the reciprocal lattice of a face-centered cubic structure is a body-centered
cubic lattice.
10.7. CHAIN OF MASS POINTS 185
Now ûq and û∗q may be regarded as linearly independent, as one regards complex variables
z and z ∗ . The Euler-Lagrange equation for û∗q gives

d ∂L ∂L ¨ = −ω 2 û .
∗ = ⇒ û q q q (10.82)
dt ∂ û˙ q ∂ û∗q
Extremizing with respect to ûq gives the complex conjugate equation.
10.7.1 Continuum limit
Let us take N → ∞, a → 0, with L0 = N a fixed. We’ll write
un (t) −→ u(x = na, t) (10.83)
in which case
dx ∂u 2
X Z
T = 12 m u̇2n −→ 1
2m (10.84)
n
a ∂t
dx u(x + a) − u(x) 2 2
X Z
V = 21 k (un+1 − un )2 −→ 1
2k a (10.85)
n
a a
Recognizing the spatial derivative above, we finally obtain

Z
L = dx L(u, ∂t u, ∂x u)
2 2
1 ∂u 1 ∂u
L= 2µ −2τ , (10.86)
∂t ∂x
where µ = m/a is the linear mass density and τ = ka is the tension3 . The quantity L is
the Lagrangian density; it depends on the field u(x, t) as well as its partial derivatives ∂t u
and ∂x u4 . The action is
Ztb Zxb

S u(x, t) = dt dx L(u, ∂t u, ∂x u) , (10.87)
ta xa
where {xa , xb } are the limits on the x coordinate. Setting δS = 0 gives the Euler-Lagrange
equations
∂L ∂ ∂L ∂ ∂L
− − =0. (10.88)
∂u ∂t ∂ (∂t u) ∂x ∂ (∂x u)
For our system, this yields the Helmholtz equation,
1 ∂ 2u ∂ 2u
= , (10.89)
c2 ∂t2 ∂x2
3
For a proper limit, we demand µ and τ be neither infinite nor infinitesimal.
4
L may also depend explicitly on x and t.
p
where c = τ /µ is the velocity of wave propagation. This is a linear equation, solutions of
which are of the form
u(x, t) = C eiqx e−iωt , (10.90)
where
ω = cq . (10.91)
Note that in the continuum limit a → 0, the dispersion relation derived for the chain
becomes
4k ka2 2
ωq2 = sin2 12 qa −→ q = c2 q 2 ,

(10.92)
m m
and so the results agree.
10.8 Appendix I : General Formulation
In the development in section 10.1, we assumed that the kinetic energy T is a homogeneous
function of degree 2, and the potential energy U a homogeneous function of degree 0, in
the generalized velocities q̇σ . However, we’ve encountered situations where this is not so:
problems with time-dependent holonomic constraints, such as the mass point on a rotating
hoop, and problems involving charged particles moving in magnetic fields. The general
Lagrangian is of the form
1
L= 2 T2 σσ0 (q) q̇σ q̇σ0 + T1 σ (q) q̇σ + T0 (q) − U1 σ (q) q̇σ − U0 (q) , (10.93)
where the subscript 0, 1, or 2 labels the degree of homogeneity of each term in the generalized
velocities. The generalized momenta are then
∂L
pσ = = T2 σσ0 q̇σ0 + T1 σ − U1 σ (10.94)
∂ q̇σ
and the generalized forces are
∂L ∂(T0 − U0 ) ∂(T1 σ0 − U1 σ0 ) 1 ∂T2 σ0 σ00

Fσ = = + q̇σ0 + q̇σ0 q̇σ00 , (10.95)
∂qσ ∂qσ ∂qσ 2 ∂qσ
and the equations of motion are again ṗσ = Fσ . Once we solve
In equilibrium, we seek a time-independent solution of the form qσ (t) = q̄σ . This entails

∂
U0 (q) − T 0 (q) =0, (10.96)
∂qσ

q=q̄
which give us n equations in the n unknowns (q1 , . . . , qn ). We then write qσ = q̄σ + ησ and
expand in the notionally small quantities ησ . It is important to understand that we assume
η and all of its time derivatives as well are small. Thus, we can expand L to quadratic order
in (η, η̇) to obtain
1
L= 2 Tσσ0 η̇σ η̇σ0 − 12 Bσσ0 ησ η̇σ0 − 12 Vσσ0 ησ ησ0 , (10.97)
10.8. APPENDIX I : GENERAL FORMULATION 187
where

∂ 2 U0 − T0 ∂ U1 σ0 − T1 σ0
Tσσ0 = T2 σσ0 (q̄) , Vσσ0 = , Bσσ0 =2 . (10.98)

∂qσ ∂qσ0 ∂qσ

q=q̄ q=q̄
Note that the T and V matrices are symmetric. The Bσσ0 term is new.
Now we can always write B = 12 (Bs +Ba ) as a sum over symmetric and antisymmetric parts,
with Bs = B + Bt and Ba = B − Bt . Since,
d 1 s
Bsσσ0 ησ η̇σ0 = B 0 η η 0 , (10.99)
dt 2 σσ σ σ
any symmetric part to B contributes a total time derivative to L, and thus has no effect on
the equations of motion. Therefore, we can project B onto its antisymmetric part, writing
!
∂ U1 σ0 − T1 σ0 ∂ U1 σ − T1 σ
Bσσ0 = − . (10.100)
∂qσ ∂qσ0
q=q̄
We now have
∂L
pσ = = Tσσ0 η̇σ0 + 12 Bσσ0 ησ0 , (10.101)
∂ η̇σ
and
∂L
Fσ = = − 12 Bσσ0 η̇σ0 − Vσσ0 ησ0 . (10.102)
∂ησ
The equations of motion, ṗσ = Fσ , then yield
Tσσ0 η̈σ0 + Bσσ0 η̇σ0 + Vσσ0 ησ0 = 0 . (10.103)
Let us write η(t) = η e−iωt . We then have
ω 2 T + iω B − V η = 0 .

(10.104)

To solve eqn. 10.104, we set P (ω) = 0, where P (ω) = det Q(ω) , with
Q(ω) ≡ ω 2 T + iω B − V . (10.105)
Since T, B, and V are real-valued matrices, and since det(M ) = det(M t ) for any matrix
∗
M , we can use Bt = −B to obtain P (−ω) = P (ω) and P (ω ∗ ) = P (ω) . This establishes

that if P (ω) = 0, i.e. if ω is an eigenfrequency, then P (−ω) = 0 and P (ω ∗ ) = 0, i.e. −ω

and ω ∗ are also eigenfrequencies (and hence −ω ∗ as well).
10.9 Appendix II : Additional Examples
10.9.1 Right Triatomic Molecule
A molecule consists of three identical atoms located at the vertices of a 45◦ right triangle.
Each pair of atoms interacts by an effective spring potential, with all spring constants equal
to k. Consider only planar motion of this molecule.
(a) Find three ‘zero modes’ for this system (i.e. normal modes whose associated eigenfre-
quencies vanish).
(b) Find the remaining three normal modes.
Solution
It is useful to choose the following coordinates:

(X1 , Y1 ) = (x1 , y1 ) (10.106)
(X2 , Y2 ) = (a + x2 , y2 ) (10.107)
(X3 , Y3 ) = (x3 , a + y3 ) . (10.108)

The three separations are then
q
d12 = (a + x2 − x1 )2 + (y2 − y1 )2
= a + x2 − x1 + . . . (10.109)
q
d23 = (−a + x3 − x2 )2 + (a + y3 − y2 )2
√
2 a − √12 x3 − x2 + √12 y3 − y2 + . . .

= (10.110)
q
d13 = (x3 − x1 )2 + (a + y3 − y1 )2
= a + y3 − y1 + . . . . (10.111)
The potential is then
2 √ 2 2
U = 12 k d12 − a + 12 k d23 − 2 a + 12 k d13 − a (10.112)
2 2 2
= 12 k x2 − x1 + 14 k x3 − x2 + 14 k y3 − y2
2
− 12 k x3 − x2 y3 − y2 + 12 k y3 − y1

(10.113)
10.9. APPENDIX II : ADDITIONAL EXAMPLES 189
Defining the row vector
η t ≡ x1 , y1 , x2 , y2 , x3 , y3 ,

(10.114)
we have that U is a quadratic form:
U = 21 ησ Vσσ0 ησ0 = 12 η t V η, (10.115)
with  
1 0 −1 0 0 0
 
 
0
 1 0 0 0 −1  
 
 
3
−1 0 − 12 − 12 1 

∂ 2 U

2 2 
V = Vσσ0 = = k (10.116)
 
∂qσ ∂qσ0 eq.
 
0 0 − 12 1 1
− 12 
 
 2 2 
 
0 − 12 1 1
− 12 
 
0 2 2
 
 
0 −1 12 − 12 − 12 3
2
The kinetic energy is simply
T = 21 m ẋ21 + ẏ12 + ẋ22 + ẏ22 + ẋ23 + ẏ32 ,

(10.117)
which entails
Tσσ0 = m δσσ0 . (10.118)
(b) The three zero modes correspond to x-translation, y-translation, and rotation. Their
eigenvectors, respectively, are
     
1 0 1
0 1 −1
     
1  1 1  0 1  1 .

ψ1 = √   , ψ 2 = √   , ψ 3 = √ (10.119)
3m 0 1
3m  2 3m  2
  
1 0 −2
0 1 −1
To find the unnormalized rotation vector, we find the CM of the triangle, located at a3 , a3 ,

and sketch orthogonal displacements ẑ × (Ri − RCM ) at the position of mass point i.
(c) The remaining modes may be determined by symmetry, and are given by
−1 −1
     
1
−1 −1 −1
     
1  0  1 −1 1 2
ψ4 = √   , ψ = √   , ψ = √
5 6
  , (10.120)
2 m 1  2 m 0  2 3m −1
    
1 0 −1
0 1 2
Figure 10.4: Normal modes of the 45◦ right triangle. The yellow circle is the location of the
CM of the triangle.
with r r r
k 2k 3k
ω1 = , ω2 = , ω3 = . (10.121)
m m m
Since T = m·1 is a multiple of the unit matrix, the orthogonormality relation ψia Tij ψjb = δ ab
entails that the eigenvectors are mutually orthogonal in the usual dot product sense, with
ψa · ψb = m−1 δab . One can check that the eigenvectors listed here satisfy this condition.
The simplest of the set {ψ4 , ψ5 , ψ6 } to find is the uniform dilation ψ6 , sometimes called
the ‘breathing’ mode. This must keep the triangle in the same shape, which means that
the deviations at each mass point are proportional to the distance to the CM. Next, it is
simplest to find ψ4 , in which the long and short sides of the triangle oscillate out of phase.
Finally, the mode ψ5 must be orthogonal to all the remaining modes. No heavy lifting (e.g.
Mathematica) is required!
10.9.2 Triple Pendulum
Consider a triple pendulum consisting of three identical masses m and three identical rigid
massless rods of length `, as depicted in Fig. 10.5.
(a) Find the T and V matrices.

(b) Find the equation for the eigenfrequencies.

p
(c) Numerically solve the eigenvalue equation for ratios ωa2 /ω02 , where ω0 = g/`. Find the
three normal modes.
Solution
The Cartesian coordinates for the three masses are
x1 = ` sin θ1 y1 = −` cos θ1
x2 = ` sin θ1 + ` sin θ2 y2 = −` cos θ1 − ` cos θ2
x3 = ` sin θ1 + ` sin θ2 + ` sin θ3 y3 = −` cos θ1 − ` cos θ2 − ` cos θ3 .
By inspection, we can write down the kinetic energy:
T = 12 m ẋ21 + ẏ12 + ẋ22 + ẏ22 + ẋ33 + ẏ32

n
= 21 m `2 3 θ̇12 + 2 θ̇22 + θ̇32 + 4 cos(θ1 − θ2 ) θ̇1 θ̇2
o
+ 2 cos(θ1 − θ3 ) θ̇1 θ̇3 + 2 cos(θ2 − θ3 ) θ̇2 θ̇3
The potential energy is

n o
U = −mg` 3 cos θ1 + 2 cos θ2 + cos θ3 ,
and the Lagrangian is L = T − U :

n
L = 21 m `2 3 θ̇12 + 2 θ̇22 + θ̇32 + 4 cos(θ1 − θ2 ) θ̇1 θ̇2 + 2 cos(θ1 − θ3 ) θ̇1 θ̇3
o n o
+ 2 cos(θ2 − θ3 ) θ̇2 θ̇3 + mg` 3 cos θ1 + 2 cos θ2 + cos θ3 .
The Cartesian coordinates for the three masses are
x1 = ` sin θ1 y1 = −` cos θ1
x2 = ` sin θ1 + ` sin θ2 y2 = −` cos θ1 − ` cos θ2
x3 = ` sin θ1 + ` sin θ2 + ` sin θ3 y3 = −` cos θ1 − ` cos θ2 − ` cos θ3 .
By inspection, we can write down the kinetic energy:
T = 21 m ẋ21 + ẏ12 + ẋ22 + ẏ22 + ẋ33 + ẏ32

n
= 21 m `2 3 θ̇12 + 2 θ̇22 + θ̇32 + 4 cos(θ1 − θ2 ) θ̇1 θ̇2
o
+ 2 cos(θ1 − θ3 ) θ̇1 θ̇3 + 2 cos(θ2 − θ3 ) θ̇2 θ̇3
Figure 10.5: The triple pendulum.

n o
U = −mg` 3 cos θ1 + 2 cos θ2 + cos θ3 ,
and the Lagrangian is L = T − U :

n
L = 12 m `2 3 θ̇12 + 2 θ̇22 + θ̇32 + 4 cos(θ1 − θ2 ) θ̇1 θ̇2 + 2 cos(θ1 − θ3 ) θ̇1 θ̇3
o n o
+ 2 cos(θ2 − θ3 ) θ̇2 θ̇3 + mg` 3 cos θ1 + 2 cos θ2 + cos θ3 .
Write down expressions for the conjugate momenta. The momenta are given by
∂L n o
π1 = = m `2 3 θ̇1 + 2 θ̇2 cos(θ1 − θ2 ) + θ̇3 cos(θ1 − θ3 )
∂ θ̇1
∂L n o
π2 = = m `2 2 θ̇2 + 2 θ̇1 cos(θ1 − θ2 ) + θ̇3 cos(θ2 − θ3 )
∂ θ̇2
∂L n o
π3 = = m `2 θ̇3 + θ̇1 cos(θ1 − θ3 ) + θ̇2 cos(θ2 − θ3 ) .
∂ θ̇2
The only conserved quantity is the total energy, E = T + U .
(a) As for the T and V matrices, we have

 
3 2 1
∂2T

= m`2 2 2 1

Tσσ0 =
∂θσ ∂θσ0 θ=0
1 1 1
and  
3 0 0
∂2U

Vσσ0 = = mg` 0 2 0 .
∂θσ ∂θσ0 θ=0
0 0 1
p
(b) The eigenfrequencies are roots of the equation det (ω 2 T − V) = 0. Defining ω0 ≡ g/`,
we have
3(ω 2 − ω02 ) 2 ω2 ω2
 
ω 2 T − V = m`2  2 ω 2 2(ω 2 − ω02 ) ω2 

ω 2 ω 2 2 2
(ω − ω0 )
and hence
h i h i
det (ω 2 T − V) = 3(ω 2 − ω02 ) · 2(ω 2 − ω02 )2 − ω 4 − 2 ω 2 · 2 ω 2 (ω 2 − ω02 ) − ω 4
h i
+ ω 2 · 2 ω 4 − 2 ω 2 (ω 2 − ω02 )
= 6 (ω 2 − ω02 )3 − 9 ω 4 (ω 2 − ω02 ) + 4ω 6
= ω 6 − 9 ω02 ω 4 + 18 ω04 ω 2 − 6 ω06 .
(c) The equation for the eigenfrequencies is
λ3 − 9 λ2 + 18 λ − 6 = 0 , (10.122)
where ω 2 = λ ω02 . This is a cubic equation in λ. Numerically solving for the roots, one finds
ω12 = 0.415774 ω02 , ω22 = 2.29428 ω02 , ω32 = 6.28995 ω02 . (10.123)
I find the (unnormalized) eigenvectors to be

     
1 1 1
ψ1 = 1.2921 , ψ2 =  0.35286  , ψ3 = −1.6450 . (10.124)
1.6312 −2.3981 0.76690
10.9.3 Equilateral Linear Triatomic Molecule
Consider the vibrations of an equilateral triangle of mass points, depicted in figure 10.6 .
The system is confined to the (x, y) plane, and in equilibrium all the strings are unstretched
and of length a.
(a) Choose as generalized coordinates the Cartesian displacements (xi , yi ) with respect to
equilibrium. Write down the exact potential energy.
(c) There are three normal modes of oscillation for which the corresponding eigenfrequencies
all vanish: ωa = 0. Write down these modes explicitly, and provide a physical interpretation
for why ωa = 0. Since this triplet is degenerate, there is no unique answer – any linear
combination will also serve as a valid ‘zero mode’. However, if you think physically, a
natural set should emerge.
Figure 10.6: An equilateral triangle of identical mass points and springs.
(d) The three remaining modes all have finite oscillation frequencies. They correspond to
distortions of the triangular shape. One such mode is the “breathing mode” in which the
triangle uniformly expands and contracts. Write down the eigenvector associated with this
normal mode and compute its associated oscillation frequency.
(e) The fifth and sixth modes are degenerate. They must be orthogonal (with respect to
the inner product defined by T) to all the other modes. See if you can figure out what these
modes are, and compute their oscillation frequencies. As in (a), any linear combination of
these modes will also be an eigenmode.
(f) Write down your full expression for the modal matrix Aai , and check that it is correct
by using Mathematica.
Solution
Choosing as generalized coordinates the Cartesian displacements relative to equilibrium, we

have the following:

#1 : x1 , y1

#2 : a + x2 , y2
√
#3 : 12 a + x3 , 23 a + y3 .

Let dij be the separation of particles i and j. The potential energy of the spring connecting
1
them is then 2 k (dij − a)2 .
2 2
d212 = a + x2 − x1 + y2 − y1
2 √ 2
d223 = − 21 a + x3 − x2 + 23 a + y3 − y2
2 √ 2
d213 = 12 a + x3 − x1 + 23 a + y3 − y1 .
Figure 10.7: Zero modes of the mass-spring triangle.
The full potential energy is

1
2 2 2
U= 2 k d12 − a + 12 k d23 − a + 12 k d13 − a . (10.125)
This is a cumbersome expression, involving square roots.
To find T and V, we need to write T and V as quadratic forms, neglecting higher order
terms. Therefore, we must expand dij − a to linear order in the generalized coordinates.
This results in the following:

d12 = a + x2 − x1 + . . .
√
d23 = a − 12 x3 − x2 + 23 y3 − y2 + . . .

√
d13 = a + 12 x3 − x1 + 23 y3 − y1 + . . . .

Thus,
1
2 √ √ 2
U= 2 k x2 − x1 + 18 k x2 − x3 − 3 y2 + 3 y3
√ √ 2
+ 18 k x3 − x1 + 3 y3 − 3 y1 + higher order terms .
Defining
q1 , q2 , q3 , q4 , q5 , q6 = x1 , y1 , x2 , y2 , x3 , y3 ,
we may now read off
√ √
−1/4 − 3/4
 
5/4 3/4 −1 0
√ √
 3/4 3/40 0 − 3/4 −3/4 
√ √
∂2U
 
 −1 0 5/4 − 3/4 −1/4 3/4 
Vσσ0 = = k √ √ 
∂qσ ∂qσ0 q̄  0 0 − 3/4 3/4 3/4 −3/4 
 −1/4 −√3/4 −1/4 √
 
3/4 1/2 0 
√ √
− 3/4 −3/4 3/4 −3/4 0 3/2
Figure 10.8: Finite oscillation frequency modes of the mass-spring triangle.
The T matrix is trivial. From
T = 12 m ẋ21 + ẏ12 + ẋ22 + ẏ22 + ẋ23 + ẏ32 .

we obtain
∂2T
Tij = = m δij ,
∂ q̇i ∂ q̇j
and T = m · 1 is a multiple of the unit matrix.
The zero modes are depicted graphically in figure 10.7. Explicitly, we have
   
1 0 1/2
0  1 −√3/2
   
1 1 
 1 0
 1  1/2 
ξx = √ , ξy = √ , ξrot =√  √  .
3m 
0 
 3m 
1
 3m 
 3/2 

1  0  −1 
0 1 0
That these are indeed zero modes may be verified by direct multiplication: V ξx.y = V ξrot =
0.
The three modes with finite oscillation frequency are depicted graphically in figure 10.8.
Explicitly, we have
   √   √ 
−1/2 − 3/2 − 3/2
−√3/2  1/2   −1/2 
   √   √ 
1  −1/2  1  3/2  1  3/2 
ξA = √ √  , ξB = √   , ξdil =√   .
3m 
 3/2 
 3m 
 1/2 
 3m 
 −1/2 

 1   0   0 
0 −1 1
10.10. ASIDE : CHRISTOFFEL SYMBOLS 197
Figure 10.9: John Henry, statue by Charles O. Cooper (1972). “Now the man that invented
the steam drill, he thought he was mighty fine. But John Henry drove fifteen feet, and the
steam drill only made nine.” - from The Ballad of John Henry.
The oscillation frequencies of these modes are easily checked by multiplying the eigenvectors
by the matrix V. Since T = m · 1 is diagonal, we have V ξa = mωa2 ξa . One finds
r r
3k 3k
ωA = ωB = , ωdil = .
2m m
Mathematica? I don’t need no stinking Mathematica.
10.10 Aside : Christoffel Symbols
The coupled equations in eqn. 10.5 may be written in the form
q̈σ + Γσµν q̇µ q̇ν = Fσ , (10.126)
with
−1 ∂Tαµ ∂Tαν ∂Tµν
Γσµν = 1
2 Tσα + − (10.127)
∂qν ∂qµ ∂qα
and
−1 ∂U
Fσ = −Tσα . (10.128)
∂qα
The components of the rank-three tensor Γσαβ are known as Christoffel symbols, in the case
where Tµν (q) defines a metric on the space of generalized coordinates.
Chapter 11
Elastic Collisions
11.1 Center of Mass Frame
A collision or ‘scattering event’ is said to be elastic if it results in no change in the internal

state of any of the particles involved. Thus, no internal energy is liberated or captured in
an elastic process.
Consider the elastic scattering of two particles. Recall the relation between laboratory
coordinates {r1 , r2 } and the CM and relative coordinates {R, r}:
m1 r1 + m2 r2 m2
R= r1 = R + r (11.1)
m1 + m2 m1 + m2
m1
r = r1 − r2 r2 = R − r (11.2)
m1 + m2
If external forces are negligible, the CM momentum P = M Ṙ is constant, and therefore the
frame of reference whose origin is tied to the CM position is an inertial frame of reference.
In this frame,
m2 v m1 v
v1CM = , v2CM = − , (11.3)
m1 + m2 m1 + m2
where v = v1 − v2 = v1CM − v2CM is the relative velocity, which is the same in both L and
CM frames. Note that the CM momenta satisfy
pCM CM
1 = m1 v1 = µv (11.4)
pCM CM
2 = m2 v2 = −µv , (11.5)
where µ = m1 m2 /(m1 + m2 ) is the reduced mass. Thus, pCM CM

1 + p2 = 0 and the total
momentum in the CM frame is zero. We may then write
p20 p2 p2
pCM
1 ≡ p0 n̂ , pCM
2 ≡ −p0 n̂ ⇒ E CM = + 0 = 0 . (11.6)
2m1 2m2 2µ
199
200 CHAPTER 11. ELASTIC COLLISIONS
Figure 11.1: The scattering of two hard spheres of radii a and b The scattering angle is χ.
The energy is evaluated when the particles are asymptotically far from each other, in which
case the potential energy is assumed to be negligible. After the collision, energy and mo-
mentum conservation require
p20
p01 ≡ p0 n̂0 p02 ≡ −p0 n̂0 E0
CM CM CM
, ⇒ = E CM = . (11.7)
2µ
The angle between n and n0 is the scattering angle χ:
n · n0 ≡ cos χ . (11.8)
The value of χ depends on the details of the scattering process, i.e. on the interaction
potential U (r). As an example, consider the scattering of two hard spheres, depicted in
Fig. 11.1. The potential is
(
∞ if r ≤ a + b
U (r) = (11.9)
0 if r > a + b .
Clearly the scattering angle is χ = π − 2φ0 , where φ0 is the angle between the initial
momentum of either sphere and a line containing their two centers at the moment of contact.
There is a simple geometric interpretation of these results, depicted in Fig. 11.2. We have
p1 = m1 V + p0 n̂ p01 = m1 V + p0 n̂0 (11.10)

p2 = m2 V − p0 n̂ p02 = m2 V − p0 n̂0 . (11.11)
So draw a circle of radius p0 whose center is the origin. The vectors p0 n̂ and p0 n̂0 must
both lie along this circle. We define the angle ψ between V and n:
V̂ · n = cos ψ . (11.12)
11.1. CENTER OF MASS FRAME 201
Figure 11.2: Scattering of two particles of masses m1 and m2 . The scattering angle χ is the
angle between n̂ and n̂0 .
It is now an exercise in geometry, using the law of cosines, to determine everything of

interest in terms of the quantities V , v, ψ, and χ. For example, the momenta are
q
p1 = m21 V 2 + µ2 v 2 + 2m1 µV v cos ψ (11.13)
q
p01 = m21 V 2 + µ2 v 2 + 2m1 µV v cos(χ − ψ) (11.14)
q
p2 = m22 V 2 + µ2 v 2 − 2m2 µV v cos ψ (11.15)
q
p02 = m22 V 2 + µ2 v 2 − 2m2 µV v cos(χ − ψ) , (11.16)
and the scattering angles are

! !
−1 µv sin ψ −1 µv sin(χ − ψ)
θ1 = tan + tan (11.17)
µv cos ψ + m1V µv cos(χ − ψ) + m1V
! !
µv sin ψ µv sin(χ − ψ)
θ2 = tan−1 + tan−1 . (11.18)
µv cos ψ − m2 V µv cos(χ − ψ) − m2 V
Figure 11.3: Scattering when particle 2 is initially at rest.
If particle 2, say, is initially at rest, the situation is somewhat simpler. In this case, V =
m1 V /(m1 + m2 ) and m2 V = µv, which means the point B lies on the circle in Fig. 11.3
(m1 6= m2 ) and Fig. 11.4 (m1 = m2 ). Let ϑ1,2 be the angles between the directions of
motion after the collision and the direction V of impact. The scattering angle χ is the
angle through which particle 1 turns in the CM frame. Clearly
sin χ
tan ϑ1 = m1 , ϑ2 = 12 (π − χ) . (11.19)
m2 + cos χ
We can also find the speeds v10 and v20 in terms of v and χ, from
2 2
p01 = p20 + m m2 p0
1
−2m 1 2
m2 p0 cos(π − χ) (11.20)
and
p22 = 2 p20 (1 − cos χ) . (11.21)
These equations yield
p
0 m21 + m22 + 2m1 m2 cos χ 2m1 v
v1 = v , v20 = sin( 12 χ) . (11.22)
m1 + m2 m1 + m2
Figure 11.4: Scattering of identical mass particles when particle 2 is initially at rest.
11.2. CENTRAL FORCE SCATTERING 203
Figure 11.5: Repulsive (A,C) and attractive (B,D) scattering in the lab (A,B) and CM
(C,D) frames, assuming particle 2 starts from rest in the lab frame. (From Barger and
Olsson.)
The angle ϑmax from Fig. 11.3(b) is given by sin ϑmax = m m1 . Note that when m1 = m2
2
we have ϑ1 + ϑ2 = π. A sketch of the orbits in the cases of both repulsive and attractive
scattering, in both the laboratory and CM frames, in shown in Fig. 11.5.
11.2 Central Force Scattering
Consider a single particle of mass µ movng in a central potential U (r), or a two body central
force problem in which µ is the reduced mass. Recall that
dr dφ dr ` dr
= · = 2· , (11.23)
dt dt dφ µr dφ
and therefore
`2
E = 12 µṙ2 + + U (r)
2µr2
2
`2 dr `2
= + + U (r) . (11.24)
2µr4 dφ 2µr2
dr
Solving for dφ , we obtain
r
dr 2µr4
=± E − U (r) − r2 , (11.25)
dφ `2
Figure 11.6: Scattering in the CM frame. O is the force center and P is the point of
periapsis. The impact parameter is b, and χ is the scattering angle. φ0 is the angle through
which the relative coordinate moves between periapsis and infinity.
Consulting Fig. 11.6, we have that
Z∞
` dr
φ0 = √ p , (11.26)
2µ r2 E − Ueff (r)
rp
where rp is the radial distance at periapsis, and where
`2
Ueff (r) = + U (r) (11.27)
2µr2
is the effective potential, as before. From Fig. 11.6, we conclude that the scattering angle
is

χ = π − 2φ0 . (11.28)
It is convenient to define the impact parameter b as the distance of the asymptotic trajectory
from a parallel line containing the force center. The geometry is shown again in Fig. 11.6.
Note that the energy and angular momentum, which are conserved, can be evaluated at
infinity using the impact parameter:
2
E = 12 µv∞ , ` = µv∞ b . (11.29)
Substituting for `(b), we have
Z∞
dr b
φ0 (E, b) = , (11.30)
r2
q
b2 U (r)
rp 1− r2
− E
Figure 11.7: Geometry of hard sphere scattering.
In physical applications, we are often interested in the deflection of a beam of incident

particles by a scattering center. We define the differential scattering cross section dσ by
# of particles scattered into solid angle dΩ per unit time
dσ = . (11.31)
incident flux
Now for particles of a given energy E there is a unique relationship between the scattering
angle χ and the impact parameter b, as we have just derived in eqn. 11.30. The differential
solid angle is given by dΩ = 2π sin χ dχ, hence
b db d ( 12 b2 )

dσ
= = . (11.32)
dΩ sinχ dχ d cos χ
dσ dσ
Note that dΩ has dimensions of area. The integral of dΩ over all solid angle is the total
scattering cross section,
Zπ
dσ
σT = 2π dχ sinχ . (11.33)
dΩ
0
11.2.1 Hard sphere scattering
Consider a point particle scattering off a hard sphere of radius a, or two hard spheres of
radii a1 and a2 scattering off each other, with a ≡ a1 + a2 . From the geometry of Fig. 11.7,
we have b = a sin φ0 and φ0 = 12 (π − χ), so
b2 = a2 sin2 1
2π − 12 χ) = 12 a2 (1 + cos χ) . (11.34)
We therefore have
dσ d ( 12 b2 )
= = 1
4 a2 (11.35)
dΩ d cos χ
and σT = πa2 . The total scattering cross section is simply the area of a sphere of radius a
projected onto a plane perpendicular to the incident flux.
11.2.2 Rutherford scattering
Consider scattering by the Kepler potential U (r) = − kr . We assume that the orbits are
unbound, i.e. they are Keplerian hyperbolae with E > 0, described by the equation
a (ε2 − 1) 1
r(φ) = ⇒ cos φ0 = ± . (11.36)
±1 + ε cos φ ε
Recall that the eccentricity is given by
2
2E`2

2 µbv∞
ε =1+ =1+ . (11.37)
µk 2 k
We then have
2
µbv∞
= ε2 − 1
k
= sec2 φ0 − 1 = tan2 φ0 = ctn2 1

2χ . (11.38)
Therefore
k 1

b(χ) = 2
ctn 2χ (11.39)
µv∞
We finally obtain
d ( 12 b2 ) k 2 d ctn2 12 χ

dσ 1
= = 2
dΩ d cos χ 2 µv∞ d cos χ
2
1 k d 1 + cos χ
= 2
2 µv∞ d cos χ 1 − cos χ
2
k
csc4 12 χ ,

= 2
(11.40)
2µv∞
which is the same as 2
dσ k
csc4 1

= 2χ . (11.41)
dΩ 4E
dσ
Since dΩ ∝ χ−4 as χ → 0, the total cross section σT diverges! This is a consequence of
the long-ranged nature of the Kepler/Coulomb potential. In electron-atom scattering, the
Coulomb potential of the nucleus is screened by the electrons of the atom, and the 1/r
behavior is cut off at large distances.
11.2.3 Transformation to laboratory coordinates
We previously derived the relation

sin χ
tan ϑ = , (11.42)
γ + cos χ
where ϑ ≡ ϑ1 is the scattering angle for particle 1 in the laboratory frame, and γ = m 1
m2
is the ratio of the masses. We now derive the differential scattering cross section in the
laboratory frame. To do so, we note that particle conservation requires

dσ dσ
· 2π sin ϑ dϑ = · 2π sin χ dχ , (11.43)
dΩ L dΩ CM
which says
dσ dσ d cos χ
= · . (11.44)
dΩ L dΩ CM d cos ϑ
From
1
cos ϑ = √
1 + tan2 ϑ
γ + cos χ
=p , (11.45)
1 + γ 2 + 2γ cos χ
we derive
d cos ϑ 1 + γ cos χ
= 3/2 (11.46)
d cos χ 1 + γ 2 + 2γ cos χ
and, accordingly,
3/2
1 + γ 2 + 2γ cos χ

dσ dσ
= · . (11.47)
dΩ L 1 + γ cos χ dΩ CM
Chapter 12
Noninertial Reference Frames
12.1 Accelerated Coordinate Systems
A reference frame which is fixed with respect to a rotating rigid body is not inertial. The
parade example of this is an observer fixed on the surface of the earth. Due to the rotation of
the earth, such an observer is in a noninertial frame, and there are corresponding corrections
to Newton’s laws of motion which must be accounted for in order to correctly describe
mechanical motion in the observer’s frame. As is well known, these corrections involve
fictitious centrifugal and Coriolis forces.
Consider an inertial frame with a fixed set of coordinate axes êµ , where µ runs from 1 to
d, the dimension of space. Any vector A may be written in either basis:
X X
A= Aµ êµ = A0µ ê0µ , (12.1)
µ µ
where Aµ = A · êµ and A0µ = A · ê0µ are projections onto the different coordinate axes. We
may now write
X dAµ
dA
= ê
dt inertial µ
dt µ
X dA0µ X dê0µ
= ê0µ + A0µ . (12.2)
dt dt
i i
The first term on the RHS is (dA/dt)body , the time derivative of A along body-fixed axes,
i.e. as seen by an observer rotating with the body. But what is dê0i /dt? Well, we can always
expand it in the {ê0i } basis:
X
dê0µ = dΩµν ê0ν ⇐⇒ dΩµν ≡ dê0µ · ê0ν . (12.3)
j
209
210 CHAPTER 12. NONINERTIAL REFERENCE FRAMES
Figure 12.1: Reference frames related by both translation and rotation.
Note that dΩµν = −dΩνµ is antisymmetric, because
0 = d ê0µ · ê0ν = dΩνµ + dΩµν ,

(12.4)
because ê0µ · ê0ν = δµν is a constant. Now we may define dΩ12 ≡ dΩ3 , et cyc., so that
X dΩσ
dΩµν = µνσ dΩσ , ωσ ≡ , (12.5)
σ
dt
which yields
dê0µ
= ω × ê0µ . (12.6)
dt
Finally, we obtain the important result

dA dA
= +ω×A (12.7)
dt inertial dt body
which is valid for any vector A.
Applying this result to the position vector r, we have

dr dr
= +ω×r . (12.8)
dt inertial dt body
Applying it twice,
! !
d2 r

d d
= +ω× +ω× r (12.9)
dt2 inertial dt body dt body
2
d r dω dr
= + × r + 2 ω × + ω × (ω × r) .
dt2 body dt dt body
12.1. ACCELERATED COORDINATE SYSTEMS 211
Note that dω/dt appears with no “inertial” or “body” label. This is because, upon invoking
eq. 12.7,
dω dω
= +ω×ω , (12.10)
dt inertial dt body
and since ω × ω = 0, inertial and body-fixed observers will agree on the value of ω̇inertial =
ω̇body ≡ ω̇.
12.1.1 Translations
Suppose that frame K moves with respect to an inertial frame K 0 , such that the origin of
K lies at R(t). Suppose further that frame K 0 rotates with respect to K, but shares the
same origin (see Fig. 12.1). Consider the motion of an object lying at position ρ relative
to the origin of K 0 , and r relative to the origin of K/K 0 . Thus,
ρ=R+r , (12.11)
and

dρ dR dr
= + +ω×r (12.12)
dt inertial dt
inertial dt body
d2ρ
2 2
dR d r dω
= + + ×r (12.13)
dt2 inertial
2
dt inertial 2
dt body dt

dr
+ 2ω × + ω × (ω × r) .
dt body
Here, ω is the angular velocity in the frame K or K 0 .
12.1.2 Motion on the surface of the earth
The earth both rotates about its axis and orbits the Sun. If we add the infinitesimal effects
of the two rotations,
dr1 = ω1 × r dt
dr2 = ω2 × (r + dr1 ) dt
dr = dr1 + dr2
= (ω1 + ω2 ) dt × r + O (dt)2 .

(12.14)
Thus, infinitesimal rotations add. Dividing by dt, this means that
X
ω= ωi , (12.15)
i
where the sum is over all the rotations. For the earth, ω = ωrot + ωorb .
• The rotation about earth’s axis, ωrot has magnitude ωrot = 2π/(1 day) = 7.29 ×
10−5 s−1 . The radius of the earth is Re = 6.37 × 103 km.
• The orbital rotation about the Sun, ωorb has magnitude ωorb = 2π/(1 yr) = 1.99 ×
10−7 s−1 . The radius of the earth is ae = 1.50 × 108 km.
Thus, ωrot /ωorb = Torb /Trot = 365.25, which is of course the number of days (i.e. rotational
periods) in a year (i.e. orbital period). There is also a very slow precession of the earth’s
axis of rotation, the period of which is about 25,000 years, which we will ignore. Note ω̇ = 0
for the earth. Thus, applying Newton’s second law and then invoking eq. 12.14, we arrive
at
d2 r d2 R

(tot) dr
m =F −m − 2m ω × − mω × (ω × r) , (12.16)
dt2 earth dt2 Sun dt earth
where ω = ωrot + ωorb , and where R̈Sun is the acceleration of the center of the earth around
the Sun, assuming the Sun-fixed frame to be inertial. The force F (tot) is the total force on
the object, and arises from three parts: (i) gravitational pull of the Sun, (ii) gravitational
pull of the earth, and (iii) other earthly forces, such as springs, rods, surfaces, electric fields,
etc.
On the earth’s surface, the ratio of the Sun’s gravity to the earth’s is
2
F GM m GMe m M Re
= = ≈ 6.02 × 10−4 . (12.17)
Fe a2e Re2 Me ae
In fact, it is clear that the Sun’s field precisely cancels with the term m R̈Sun at the earth’s
center, leaving only gradient contributions of even lower order, i.e. multiplied by Re /ae ≈
4.25 × 10−5 . Thus, to an excellent approximation, we may neglect the Sun entirely and
write
d2 r F0 dr
2
= + g − 2ω × − ω × (ω × r) (12.18)
dt m dt
Note that we’ve dropped the ‘earth’ label here and henceforth. We define g = −GMe r̂/r2 ,
the acceleration due to gravity; F 0 is the sum of all earthly forces other than the earth’s
gravity. The last two terms on the RHS are corrections to mr̈ = F due to the noninertial
frame of the earth, and are recognized as the Coriolis and centrifugal acceleration terms,
respectively.
12.2. SPHERICAL POLAR COORDINATES 213
Figure 12.2: The locally orthonormal triad {r̂, θ̂, φ̂}.
12.2 Spherical Polar Coordinates
The locally orthonormal triad {r̂, θ̂, φ̂} varies with position. In terms of the body-fixed
triad {x̂, ŷ, ẑ}, we have
r̂ = sin θ cos φ x̂ + sin θ sin φ ŷ + cos θ ẑ (12.19)
θ̂ = cos θ cos φ x̂ + cos θ sin φ ŷ − sin θ ẑ (12.20)
φ̂ = − sin φ x̂ + cos φ ŷ . (12.21)
Inverting the relation between the triads {r̂, θ̂, φ̂} and {x̂, ŷ, ẑ}, we obtain
x̂ = sin θ cos φ r̂ + cos θ cos φ θ̂ − sin φ φ̂ (12.22)
ŷ = sin θ sin φ x̂ + cos θ sin φ ŷ + cos φ φ̂ (12.23)
ẑ = cos θ r̂ − sin θ θ̂ . (12.24)
The differentials of these unit vectors are
dr̂ = θ̂ dθ + sin θ φ̂ dφ (12.25)
dθ̂ = −r̂ dθ + cos θ φ̂ dφ (12.26)
dφ̂ = − sin θ r̂ dφ − cos θ θ̂ dφ . (12.27)
Thus,
d
r r̂ = ṙ r̂ + r r̂˙

ṙ =
dt
= ṙ r̂ + rθ̇ θ̂ + r sin θ φ̇ φ̂ . (12.28)
If we differentiate a second time, we find, after some tedious accounting,
r̈ = r̈ − r θ̇2 − r sin2 θ φ̇2 r̂ + 2 ṙ θ̇ + r θ̈ − r sin θ cos θ φ̇2 θ̂

+ 2 ṙ φ̇ sin θ + 2 r θ̇ φ̇ cos θ + r sin θ φ̈ φ̂ . (12.29)
12.3 Centrifugal Force
One major distinction between the Coriolis and centrifugal forces is that the Coriolis force
acts only on moving particles, whereas the centrifugal force is present even when ṙ = 0.
Thus, the equation for stationary equilibrium on the earth’s surface is
mg + F 0 − mω × (ω × r) = 0 , (12.30)
involves the centrifugal term. We can write this as F 0 + me

g = 0, where
GMe r̂
ge = − − ω × (ω × r) (12.31)
r2
= − g0 − ω 2 Re sin2 θ r̂ + ω 2 Re sin θ cos θ θ̂ ,

(12.32)
where g0 = GMe /Re2 = 980 cm/s2 . Thus, on the equator, g̃ = − g0 − ω 2 Re r̂, with

ω 2 Re ≈ 3.39 cm/s2 , a small but significant correction. Thus, you weigh less on the equator.
Note also the term in ge along θ̂. This means that a plumb bob suspended from a general
point above the earth’s surface won’t point exactly toward the earth’s center. Moreover, if
the earth were replaced by an equivalent mass of fluid, the fluid would rearrange itself so
as to make its surface locally perpendicular to ge. Indeed, the earth (and Sun) do exhibit
quadrupolar distortions in their mass distributions – both are oblate spheroids. In fact, the
observed difference g̃(θ = π2 ) − g̃(θ = 0) ≈ 5.2 cm/s2 , which is 53% greater than the naı̈vely
expected value of 3.39 cm/s2 . The earth’s oblateness enhances the effect.
12.3.1 Rotating tube of fluid
Consider a cylinder filled with a liquid, rotating with angular frequency ω about its sym-
metry axis ẑ. In steady state, the fluid is stationary in the rotating frame, and we may
write, for any given element of fluid
0 = f 0 + g − ω 2 ẑ × (ẑ × r) , (12.33)
where f 0 is the force per unit mass on the fluid element. Now consider a fluid element on
the surface. Since there is no static friction to the fluid, any component of f 0 parallel to
the fluid’s surface will cause the fluid to flow in that direction. This contradicts the steady
state assumption. Therefore, we must have f 0 = f 0 n̂, where n̂ is the local unit normal
to the fluid surface. We write the equation for the fluid’s surface as z = z(ρ). Thus, with
r = ρ ρ̂ + z(ρ) ẑ, Newton’s second law yields
f 0 n̂ = g ẑ − ω 2 ρ ρ̂ , (12.34)
where g = −g ẑ is assumed. From this, we conclude that the unit normal to the fluid surface
and the force per unit mass are given by
g ẑ − ω 2 ρ ρ̂ p
n̂(ρ) = p , f 0 (ρ) = g 2 + ω 4 ρ2 . (12.35)
g 2 + ω 4 ρ2
12.4. THE CORIOLIS FORCE 215
Figure 12.3: A rotating cylinder of fluid.
Now suppose r(ρ, φ) = ρ ρ̂ + z(ρ) ẑ is a point on the surface of the fluid. We have that
dr = ρ̂ dρ + z 0 (ρ) ẑ dρ + ρ φ̂ dφ , (12.36)
where z 0 = dρ
dz
, and where we have used dρ̂ = φ̂ dφ, which follows from eqn. 12.25 after
π
setting θ = 2 . Now dr must lie along the surface, therefore n̂ · dr = 0, which says
dz
g = ω2 ρ . (12.37)
dρ
Integrating this equation, we obtain the shape of the surface:
ω 2 ρ2
z(ρ) = z0 + . (12.38)
2g
12.4 The Coriolis Force
The Coriolis force is given by FCor = −2m ω × ṙ. According to (12.18), the acceleration
of a free particle (F 0 = 0) isn’t along ge – an orthogonal component is generated by the
Coriolis force. To actually solve the coupled equations of motion is difficult because the
unit vectors {r̂, θ̂, φ̂} change with position, and hence with time. The following standard
problem highlights some of the effects of the Coriolis and centrifugal forces.
PROBLEM: A cannonball is dropped from the top of a tower of height h located at a northerly
latitude of λ. Assuming the cannonball is initially at rest with respect to the tower, and
neglecting air resistance, calculate its deflection (magnitude and direction) due to (a) cen-
trifugal and (b) Coriolis forces by the time it hits the ground. Evaluate for the case h = 100
m, λ = 45◦ . The radius of the earth is Re = 6.4 × 106 m.
SOLUTION: The equation of motion for a particle near the earth’s surface is
r̈ = −2 ω × ṙ − g0 r̂ − ω × (ω × r) , (12.39)
where ω = ω ẑ, with ω = 2π/(24 hrs) = 7.3 × 10−5 rad/s. Here, g0 = GMe /Re2 = 980 cm/s2 .
We use a locally orthonormal coordinate system {r̂, θ̂, φ̂} and write
r = x θ̂ + y φ̂ + (Re + z) r̂ , (12.40)
where Re = 6.4 × 106 m is the radius of the earth. Expressing ẑ in terms of our chosen
orthonormal triad,
ẑ = cos θ r̂ − sin θ θ̂ , (12.41)
where θ = π2 − λ is the polar angle, or ‘colatitude’. Since the height of the tower and
the deflections are all very small on the scale of Re , we may regard the orthonormal triad
as fixed and time-independent. (In general, these unit vectors change as a function of r.)
Thus, we have ṙ ' ẋ θ̂ + ẏ φ̂ + ż r̂, and we find
ẑ × ṙ = −ẏ cos θ θ̂ + (ẋ cos θ + ż sin θ) φ̂ − ẏ sin θ r̂ (12.42)

2 2 2
ω × (ω × r) = −ω Re sin θ cos θ θ̂ − ω Re sin θ r̂ , (12.43)
where we neglect the O(z) term in the second equation, since z Re .
The equation of motion, written in components, is then
ẍ = 2ω cos θ ẏ + ω 2 Re sin θ cos θ (12.44)

ÿ = −2ω cos θ ẋ − 2ω sin θ ż (12.45)
2 2
z̈ = −g0 + 2ω sin θ ẏ + ω Re sin θ . (12.46)
While these (inhomogeneous) equations are linear, they also are coupled, so an exact an-
alytical solution is not trivial to obtain (but see below). Fortunately, the deflections are
small, so we can solve this perturbatively. We write x = x(0) + δx, etc., and solve to lowest
order by including only the g0 term on the RHS. This gives z (0) (t) = z0 − 12 g0 t2 , along
with x(0) (t) = y (0) (t) = 0. We then substitute this solution on the RHS and solve for the
deflections, obtaining
δx(t) = 21 ω 2 Re sin θ cos θ t2 (12.47)

1 3
δy(t) = 3 ωg0 sin θ t (12.48)
1 2 2 2
δz(t) = 2 ω Re sin θ t . (12.49)
The deflection along θ̂ and r̂ is due to the centrifugal term, while that along φ̂ is due to
the Coriolis term. (At higher order, the two terms interact and the deflection in any given
direction can’t uniquely be associated to a single fictitious force.) To p
find the deflection of
(0) ∗ ∗
an object dropped from a height h, solve z (t ) = 0 to obtain t = 2h/g0 for the drop
time, and substitute. For h = 100 m and λ = π2 , find δx(t∗ ) = 17 cm south (centrifugal)
and δy(t∗ ) = 1.6 cm east (Coriolis).
In fact, an exact solution to (12.46) is readily obtained, via the following analysis. The
equations of motion may be written v̇ = 2iωJ v + b, or
J b
z }| z }| {
  {    
v̇x 0 −i cos θ 0 vx g1 sin θ cos θ
v̇y  = 2i ω i cos θ 0 i sin θ vy  +  0 (12.50)
    
0 −i sin θ 0 2
v̇x vx −g0 + g1 sin θ
with g1 ≡ ω 2 Re . Note that J † = J , i.e. J is a Hermitian matrix. The formal solution is

2iωJ t
e −1
v(t) = e2iωJ t
v(0) + J −1 b . (12.51)
2iω
When working with matrices, it is convenient to work in an eigenbasis. The characteristic
polynomial for J is P (λ) = det (λ · 1 − J ) = λ (λ2 − 1), hence the eigenvalues are λ1 = 0,
λ2 = +1, and λ3 = −1. The corresponding eigenvectors are easily found to be
     
sin θ cos θ cos θ
1 1
ψ1 =  0  , ψ2 = √  i  , ψ3 = √  −i  . (12.52)
− cos θ 2 sin θ 2 sin θ
Note that ψa† · ψa0 = δaa0 .

∗ v and
Expanding v and b in this eigenbasis, we have v̇a = 2iωλa va + ba , where va = ψia i
∗ b . The solution is
ba = ψia i
2iλ ωt
e a −1
va (t) = va (0) e2iλa ωt + ba , (12.53)
2iλa ω
which entails !
e2iλa ωt − 1
X
∗
vi (t) = ψia ψja bj , (12.54)
a
2iλa ω
where we have taken v(0) = 0, i.e. the object is released from rest. Doing the requisite
matrix multiplications,
sin2 ωt
   
t sin2 θ + sin2ω2ωt cos2 θ − 1 sin 2ωt 
vx (t) ω cos θ 2 t sin 2θ + 4ω sin 2θ g1 sin θ cos θ
2 2
vy (t) =  − sinω ωt cos θ sin 2ωt
− sinω ωt sin θ 0 ,
   
2ω 
1 sin 2ωt 2
sin ωt sin 2ωt 2
2 2 −g0 + g1 sin θ
vz (t) − 2 t sin 2θ + 4ω sin 2θ ω sin θ t cos θ + 2ω sin θ
(12.55)
which says

vx (t) = 12 sin 2θ + sin4ωt
2ωt
sin 2θ · g0 t + sin4ωt
2ωt
sin 2θ · g1 t
sin2 ωt sin2 ωt
vy (t) = ωt · g0 t − ωt sin θ · g1 t (12.56)

sin2 ωt
vz (t) = − cos2 θ + sin2ωt
2ωt sin2 θ · g0 t + 2ωt · g1 t .
Figure 12.4: Foucault’s pendulum.
Why is the deflection always to the east? The earth rotates eastward, and an object starting
from rest in the earth’s frame has initial angular velocity equal to that of the earth. To
conserve angular momentum, the object must speed up as it falls.
12.4.1 Foucault’s pendulum
A pendulum swinging over one of the poles moves in a fixed inertial plane while the earth
rotates underneath. Relative to the earth, the plane of motion of the pendulum makes
one revolution every day. What happens at a general latitude? Assume the pendulum is
located at colatitude θ and longitude φ. Assuming the length scale of the pendulum is small
compared to Re , we can regard the local triad {θ̂, φ̂, r̂} as fixed. The situation is depicted
in Fig. 12.4. We write
r = x θ̂ + y φ̂ + z r̂ , (12.57)
with
x = ` sin ψ cos α , y = ` sin ψ sin α , z = ` (1 − cos ψ) . (12.58)
In our analysis we will ignore centrifugal effects, which are of higher order in ω, and we
take g = −g r̂. We also idealize the pendulum, and consider the suspension rod to be of
negligible mass.
The total force on the mass m is due to gravity and tension:
F = mg + T

= − T sin ψ cos α, −T sin ψ sin α, T cos ψ − mg

= − T x/`, −T y/`, T − M g − T z/` . (12.59)
The Coriolis term is
FCor = −2m ω × ṙ (12.60)

= −2m ω cos θ r̂ − sin θ θ̂ × ẋ θ̂ + ẏ φ̂ + ż r̂

= 2mω ẏ cos θ, −ẋ cos θ − ż sin θ, ẏ sin θ . (12.61)
The equations of motion are m r̈ = F + FCor :
mẍ = −T x/` + 2mω cos θ ẏ (12.62)
mÿ = −T y/` − 2mω cos θ ẋ − 2mω sin θ ż (12.63)
mz̈ = T − mg − T z/` + 2mω sin θ ẏ . (12.64)
These three equations are to be solved for the three unknowns x, y, and T . Note that
x2 + y 2 + (` − z)2 = `2 , (12.65)
so z = z(x, y) is not an independent degree of freedom. This equation may be recast in the
form z = (x2 + y 2 + z 2 )/2` which shows that if x and y are both small, then z is at least of
second order in smallness. Therefore, we will approximate z ' 0, in which case ż may be
neglected from the second equation of motion. The third equation is used to solve for T :
T ' mg − 2mω sin θ ẏ . (12.66)
Adding the first plus i times the second then gives the complexified equation
T
ξ¨ = − ξ − 2iω cos θ ξ˙
m`
≈ −ω02 ξ − 2iω cos θ ξ˙ (12.67)
p
where ξ ≡ x + iy, and where ω0 = g/`. Note that we have approximated T ≈ mg in
deriving the second line.
It is now a trivial matter to solve the homogeneous linear ODE of eq. 12.67. Writing
ξ = ξ0 e−iΩt (12.68)
and plugging in to find Ω, we obtain
Ω 2 − 2ω⊥ Ω − ω02 = 0 , (12.69)
with ω⊥ ≡ ω cos θ. The roots are

q
Ω± = ω⊥ ± ω02 + ω⊥
2 , (12.70)
hence the most general solution is
ξ(t) = A+ e−iΩ+ t + A− e−iΩ− t . (12.71)

Finally, if we take as initial conditions x(0) = a, y(0) = 0, ẋ(0) = 0, and ẏ(0) = 0, we

obtain
a n o
x(t) = · ω⊥ sin(ω⊥ t) sin(νt) + ν cos(ω⊥ t) cos(νt) (12.72)
ν
a n o
y(t) = · ω⊥ cos(ω⊥ t) sin(νt) − ν sin(ω⊥ t) cos(νt) , (12.73)
ν
q
with ν = 2 . Typically ω ω , since ω = 7.3 × 10−5 s−1 . In the limit ω ω ,
ω02 + ω⊥ 0 ⊥ ⊥ 0
then, we have ν ≈ ω0 and
x(t) ' a cos(ω⊥ t) cos(ω0 t) , y(t) ' −a sin(ω⊥ t) cos(ω0 t) , (12.74)
and the plane of motion rotates with angular frequency −ω⊥ , i.e. the period is | sec θ | days.
Viewed from above, the rotation is clockwise in the northern hemisphere, where cos θ > 0
and counterclockwise in the southern hemisphere, where cos θ < 0.
Chapter 13
Rigid Body Motion and Rotational

Dynamics
13.1 Rigid Bodies
A rigid body consists of a group of particles whose separations are all fixed in magnitude. Six
independent coordinates are required to completely specify the position and orientation of a
rigid body. For example, the location of the first particle is specified by three coordinates. A
second particle requires only two coordinates since the distance to the first is fixed. Finally,
a third particle requires only one coordinate, since its distance to the first two particles
is fixed (think about the intersection of two spheres). The positions of all the remaining
particles are then determined by their distances from the first three. Usually, one takes
these six coordinates to be the center-of-mass position R = (X, Y, Z) and three angles
specifying the orientation of the body (e.g. the Euler angles).
As derived previously, the equations of motion are

X
P = mi ṙi , Ṗ = F (ext) (13.1)
i
X
L= mi ri × ṙi , L̇ = N (ext) . (13.2)
i
These equations determine the motion of a rigid body.
13.1.1 Examples of rigid bodies
Our first example of a rigid body is of a wheel rolling with constant angular velocity φ̇ = ω,
and without slipping, This is shown in Fig. 13.1. The no-slip condition is dx = R dφ, so
ẋ = VCM = Rω. The velocity of a point within the wheel is
v = VCM + ω × r , (13.3)
221
222 CHAPTER 13. RIGID BODY MOTION AND ROTATIONAL DYNAMICS
Figure 13.1: A wheel rolling to the right without slipping.
where r is measured from the center of the disk. The velocity of a point on the surface is
then given by v = ωR x̂ + ω̂ × r̂).
As a second example, consider a bicycle wheel of mass M and radius R affixed to a light, firm
rod of length d, as shown in Fig. 13.2. Assuming L lies in the (x, y) plane, one computes
the gravitational torque N = r × (M g) = M gd φ̂. The angular momentum vector then
rotates with angular frequency φ̇. Thus,
dL M gd
dφ = =⇒ φ̇ = . (13.4)
L L
But L = M R2 ω, so the precession frequency is
gd
ωp = φ̇ = . (13.5)
ωR2
For R = d = 30 cm and ω/2π = 200 rpm, find ωp /2π ≈ 15 rpm. Note that we have here
ignored the contribution to L from the precession itself, which lies along ẑ, resulting in the
nutation of the wheel. This is justified if Lp /L = (d2 /R2 ) · (ωp /ω) 1.
13.2 The Inertia Tensor
Suppose first that a point within the body itself is fixed. This eliminates the translational
degrees of freedom from consideration. We now have

dr
=ω×r , (13.6)
dt inertial
13.2. THE INERTIA TENSOR 223
Figure 13.2: Precession of a spinning bicycle wheel.
since ṙbody = 0. The kinetic energy is then
2
1
X dri 1
X
T = 2 mi = 2 mi (ω × ri ) · (ω × ri )
dt inertial
i i
X h i
= 1
2 mi ω 2 ri2 − (ω · ri )2 ≡ 12 Iαβ ωα ωβ , (13.7)
i
where ωα is the component of ω along the body-fixed axis eα . The quantity Iαβ is the
inertia tensor,
X
Iαβ = mi ri2 δαβ − ri,α ri,β (13.8)
Zi
= dd r %(r) r 2 δαβ − rα rβ (continuous media) . (13.9)
The angular momentum is

X dri
L= mi ri ×
dt inertial
i
X
= mi ri × (ω × ri ) = Iαβ ωβ . (13.10)
i
The diagonal elements of Iαβ are called the moments of inertia, while the off-diagonal
elements are called the products of inertia.
13.2.1 Coordinate transformations
Consider the basis transformation

ê0α = Rαβ êβ . (13.11)
We demand ê0α · ê0β = δαβ , which means R ∈ O(d) is an orthogonal matrix, i.e. Rt = R−1 .
Thus the inverse transformation is eα = Rtαβ e0β . Consider next a general vector A = Aβ êβ .
Expressed in terms of the new basis {ê0α }, we have
êβ A0
z }| { z }|α {
A = Aβ Rtβα ê0α = Rαβ Aβ ê0α (13.12)
Thus, the components of A transform as A0α = Rαβ Aβ . This is true for any vector.
Under a rotation, the density ρ(r) must satisfy ρ0 (r 0 ) = ρ(r). This is the transformation
rule for scalars. The inertia tensor therefore obeys
Z h i
2
Iαβ = d3 r0 ρ0 (r 0 ) r 0 δαβ − rα0 rβ0
0
Z h i
d3 r ρ(r) r 2 δαβ − Rαµ rµ Rβν rν

=
= Rαµ Iµν Rtνβ . (13.13)

I.e. =I0 RIRt , the transformation rule for tensors. The angular frequency ω is a vector, so
0
ωα = Rαµ ωµ . The angular momentum L also transforms as a vector. The kinetic energy
is T = 12 ω t · I · ω, which transforms as a scalar.
13.2.2 The case of no fixed point
If there is no fixed point, we can let r 0 denote the distance from the center-of-mass (CM),
which will serve as the instantaneous origin in the body-fixed frame. We then adopt the
notation where R is the CM position of the rotating body, as observed in an inertial frame,
and is computed from the expression
Z
1 X 1
R= mi ρi = d3 r ρ(r) , (13.14)
M M
i
where the total mass is of course
X Z
M= mi = d3 r ρ(r) . (13.15)
i
The kinetic energy and angular momentum are then
T = 12 M Ṙ2 + 12 Iαβ ωα ωβ (13.16)
Lα = αβγ M Rβ Ṙγ + Iαβ ωβ , (13.17)
where Iαβ is given in eqs. 13.8 and 13.9, where the origin is the CM.
13.3. PARALLEL AXIS THEOREM 225
Figure 13.3: Application of the parallel axis theorem to a cylindrically symmetric mass
distribution.
13.3 Parallel Axis Theorem
Suppose Iαβ is given in a body-fixed frame. If we displace the origin in the body-fixed frame
by d, then let Iαβ (d) be the inertial tensor with respect to the new origin. If, relative to
the origin at 0 a mass element lies at position r, then relative to an origin at d it will lie at
r − d. We then have
X n o
Iαβ (d) = mi (ri2 − 2d · ri + d2 ) δαβ − (ri,α − dα )(ri,β − dβ ) . (13.18)
i
If ri is measured with respect to the CM, then

X
mi ri = 0 (13.19)
i
and
Iαβ (d) = Iαβ (0) + M d2 δαβ − dα dβ ,

(13.20)
a result known as the parallel axis theorem.
As an example of the theorem, consider the situation depicted in Fig. 13.3, where a cylin-
drically symmetric mass distribution is rotated about is symmetry axis, and about an axis
tangent to its side. The component Izz of the inertia tensor is easily computed when the
origin lies along the symmetry axis:
Z Za
Izz = d3 r ρ(r) (r 2 − z 2 ) = ρL · 2π dr⊥ r⊥
3
0
= π2 ρLa4 = 12 M a2 , (13.21)
where M = πa2 Lρ is the total mass. If we compute Izz about a vertical axis which is
tangent to the cylinder, the parallel axis theorem tells us that
0
Izz = Izz + M a2 = 23 M a2 . (13.22)
2 would be tedious!
R
Doing this calculation by explicit integration of dm r⊥
13.3.1 Example
Problem: Compute the CM and the inertia tensor for the planar right triangle of Fig.
13.4, assuming it to be of uniform two-dimensional mass density ρ.
Solution: The total mass is M = 12 ρ ab. The x-coordinate of the CM is then

x
Za b(1−
Z a) Za
1 ρ
dx b 1 − xa x

X= dx dy ρ x =
M M
0 0 0
Z1
ρ a2 b ρ a2 b 1
= du u(1 − u) = = 3 a. (13.23)
M 6M
0
1
Clearly we must then have Y = 3 b, which may be verified by explicit integration.
We now compute the inertia tensor, with the origin at (0, 0, 0). Since the figure is planar,
z = 0 everywhere, hence Ixz = Izx = 0, Iyz = Izy = 0, and also Izz = Ixx + Iyy . We now
compute the remaining independent elements:
x
Z a)
Za b(1− Za
3
Ixx = ρ dx dy y = ρ dx 13 b3 1 − xa
2
0 0 0
Z1
= 3 ρ ab du (1 − u)3 =
1 3 1
12 ρ ab
3
= 16 M b2 (13.24)
0
and
x
Za b(1−
Z a) Za
2
2
Ixy = −ρ dx dy x y = − 2 ρ b dx x 1 − xa
1
0 0 0
Z1
= − 12 ρ a2 b2 du u (1 − u)2 = − 24
1
ρ a2 b2 = − 12
1
M ab . (13.25)
0
Thus,
b2 − 21 ab
 
0
M 1
I= − 2 ab a2 0  . (13.26)
6
0 0 a + b2
2
13.3. PARALLEL AXIS THEOREM 227
Figure 13.4: A planar mass distribution in the shape of a triangle.
Suppose we wanted the inertia tensor relative in a coordinate system where the CM lies at
the origin. What we computed in eqn. 13.26 is I(d), with d = − 3a x̂ − 3b ŷ. Thus,
 2 
b −ab 0
1
d2 δαβ − dα dβ = −ab a2 0  . (13.27)
9 2 2
0 0 a +b
Since
I(d) = I CM + M d2 δαβ − dα dβ , (13.28)
we have that

I CM = I(d) − M d2 δαβ − dα dβ (13.29)
 2 1 
b 2 ab 0
M 1
= ab a2 0  . (13.30)
18 2
0 0 a + b2
2
13.3.2 General planar mass distribution
For a general planar mass distribution,
ρ(x, y, z) = σ(x, y) δ(z) , (13.31)
which is confined to the plane z = 0, we have

Z Z
Ixx = dx dy σ(x, y) y 2 (13.32)
Z Z
Iyy = dx dy σ(x, y) x2 (13.33)
Z Z
Ixy = − dx dy σ(x, y) xy (13.34)
and Izz = Ixx + Iyy , regardless of the two-dimensional mass distribution σ(x, y).
13.4 Principal Axes of Inertia
We found that an orthogonal transformation to a new set of axes ê0α = Rαβ êβ entails
I 0 = RIRt for the inertia tensor. Since I = I t is manifestly a symmetric matrix, it can
be brought to diagonal form by such an orthogonal transformation. To find R, follow this
recipe:
1. Find the diagonal elements of I 0 by setting P (λ) = 0, where

P (λ) = det λ · 1 − I , (13.35)
is the characteristic polynomial for I, and 1 is the unit matrix.
2. For each eigenvalue λa , solve the d equations
X
Iµν ψνa = λa ψµa . (13.36)
ν
Here, ψµa is the µth component of the ath eigenvector. Since (λ · 1 − I) is degenerate,
these equations are linearly dependent, which means that the first d − 1 components
may be determined in terms of the dth component.
3. Because I = I t , eigenvectors corresponding to different eigenvalues are orthogonal.
In cases of degeneracy, the eigenvectors may be chosen to be orthogonal, e.g. via the
Gram-Schmidt procedure.
4. Due to the underdetermined aspect to step 2, we may choose an arbitrary normaliza-
tion for
P each eigenvector. It is conventional to choose the eigenvectors to be orthonor-
mal: µ ψµa ψµb = δ ab .
5. The matrix R is explicitly given by Raµ = ψµa , the matrix whose row vectors are the
eigenvectors ψ a . Of course Rt is then the corresponding matrix of column vectors.
6. The eigenvectors form a complete basis. The resolution of unity may be expressed as
X
ψµa ψνa = δµν . (13.37)
a
As an example, consider the inertia tensor for a general planar mass distribution, which is
of the form  
Ixx Ixy 0
I = Iyx Iyy 0  , (13.38)
0 0 Izz
where Iyx = Ixy and Izz = Ixx + Iyy . Define
A = 12 Ixx + Iyy

(13.39)
q 2
B = 1/4 Ixx − Iyy + Ixy 2 (13.40)

−1 2Ixy
ϑ = tan , (13.41)
Ixx − Iyy
13.5. EULER’S EQUATIONS 229
so that  
A + B cos ϑ B sin ϑ 0
I =  B sin ϑ A − B cos ϑ 0  , (13.42)
0 0 2A
The characteristic polynomial is found to be
h i
P (λ) = (λ − 2A) (λ − A)2 − B 2 , (13.43)
which gives λ1 = A, λ2,3 = A ± B. The corresponding normalized eigenvectors are
cos 12 ϑ − sin 12 ϑ
     
0
ψ 1 = 0 , ψ 2 =  sin 12 ϑ  , ψ 3 =  cos 12 ϑ  (13.44)
1 0 0
and therefore  
0 0 1
R =  cos 12 ϑ sin 12 ϑ 0 . (13.45)
− sin 12 ϑ cos 12 ϑ 0
13.5 Euler’s Equations
Let us now choose our coordinate axes to be the principal axes of inertia, with the CM at
the origin. We may then write
     
ω1 I1 0 0 I1 ω1
ω = ω2
  , I= 0  I2 0  =⇒ L = I2 ω2  . (13.46)
ω3 0 0 I3 I3 ω3

ext dL
N =
dt inertial

dL
= +ω×L
dt body
= I ω̇ + ω × (I ω) .
Thus, we arrive at Euler’s equations:
I1 ω̇1 = (I2 − I3 ) ω2 ω3 + N1ext (13.47)

I2 ω̇2 = (I3 − I1 ) ω3 ω1 + N2ext (13.48)
I3 ω̇3 = (I1 − I2 ) ω1 ω2 + N3ext . (13.49)
These are coupled and nonlinear. Also note the fact that the external torque must be
evaluated along body-fixed principal axes. We can however make progress in the case
Figure 13.5: Wobbling of a torque-free symmetric top.
where N ext = 0, i.e. when there are no external torques. This is true for a body in free
space, or in a uniform gravitational field. In the latter case,
X X
N ext = ri × (mi g) = mi ri × g , (13.50)
i i
where g is theP
uniform gravitational acceleration. In a body-fixed frame whose origin is the
CM, we have i mi ri = 0, and the external torque vanishes!
Precession of torque-free symmetric tops: Consider a body which has a symme-

try axis ê3 . This guarantees I1 = I2 , but in general we still have I1 6= I3 . In the absence
of external torques, the last of Euler’s equations says ω̇3 = 0, so ω3 is a constant. The
remaining two equations are then

I1 − I3 I3 − I1
ω̇1 = ω3 ω2 , ω̇2 = ω3 ω1 . (13.51)
I1 I1
I.e.ω̇1 = −Ω ω2 and ω̇2 = +Ω ω1 , with

I3 − I1
Ω= ω3 , (13.52)
I1
which are the equations of a harmonic oscillator. The solution is easily obtained:

ω1 (t) = ω⊥ cos Ωt + δ , ω2 (t) = ω⊥ sin Ωt + δ , ω3 (t) = ω3 , (13.53)
where ω⊥ and δ are constants of integration, and where |ω| = (ω⊥ 2 + ω 2 )1/2 . This motion is
3
sketched in Fig. 13.5. Note that the perpendicular components of ω oscillate harmonically,
and that the angle ω makes with respect to ê3 is λ = tan−1 (ω⊥ /ω3 ).
1
For the earth, (I3 − I1 )/I1 ≈ 305 , so ω3 ≈ ω, and Ω ≈ ω/305, yielding a precession period
of 305 days, or roughly 10 months. Astronomical observations reveal such a precession,
13.5. EULER’S EQUATIONS 231
known as the Chandler wobble. For the earth, the precession angle is λChandler ' 6 × 10−7
rad, which means that the North Pole moves by about 4 meters during the wobble. The
Chandler wobble has a period of about 14 months, so the naı̈ve prediction of 305 days is
off by a substantial amount. This discrepancy is attributed to the mechanical properties of
the earth: elasticity and fluidity. The earth is not solid!1
Asymmetric tops: Next, consider the torque-free motion of an asymmetric top, where
I1 6= I2 6= I3 6= I1 . Unlike the symmetric case, there is no conserved component of ω. True,
we can invoke conservation of energy and angular momentum,
E = 21 I1 ω12 + 12 I2 ω22 + 12 I3 ω32 (13.54)
L2 = I12 ω12 + I22 ω22 + I32 ω32 , (13.55)
and, in principle, solve for ω1 and ω2 in terms of ω3 , and then invoke Euler’s equations
(which must honor these conservation laws). However, the nonlinearity greatly complicates
matters and in general this approach is a dead end.
We can, however, find a particular solution quite easily – one in which the rotation is about
a single axis. Thus, ω1 = ω2 = 0 and ω3 = ω0 is indeed a solution for all time, according to
Euler’s equations. Let us now perturb about this solution, to explore its stability. We write
ω = ω0 ê3 + δω , (13.56)
and we invoke Euler’s equations, linearizing by dropping terms quadratic in δω. This yield
I1 δ ω̇1 = (I2 − I3 ) ω0 δω2 + O(δω2 δω3 ) (13.57)

I2 δ ω̇2 = (I3 − I1 ) ω0 δω1 + O(δω3 δω1 ) (13.58)
I3 δ ω̇3 = 0 + O(δω1 δω2 ) . (13.59)
Taking the time derivative of the first equation and invoking the second, and vice versa,
yields
δ ω̈1 = −Ω 2 δω1 , δ ω̈2 = −Ω 2 δω2 , (13.60)
with
(I3 − I2 )(I3 − I1 ) 2
Ω2 = · ω0 . (13.61)
I1 I2
The solution is then δω1 (t) = C cos(Ωt + δ).
If Ω 2 > 0, then Ω is real, and the deviation results in a harmonic precession. This occurs
if I3 is either the largest or the smallest of the moments of inertia. If, however, I3 is the
middle moment, then Ω 2 < 0, and Ω is purely imaginary. The perturbation will in general
increase exponentially with time, which means that the initial solution to Euler’s equations
is unstable with respect to small perturbations. This result can be vividly realized using a
tennis racket, and sometimes goes by the name of the “tennis racket theorem.”
1
The earth is a layered like a Mozartkugel, with a solid outer shell, an inner fluid shell, and a solid (iron)
core.
13.5.1 Example
PROBLEM: A unsuspecting solid spherical planet of mass M0 rotates with angular velocity
ω0 . Suddenly, a giant asteroid of mass αM0 smashes into and sticks to the planet at a
location which is at polar angle θ relative to the initial rotational axis. The new mass
distribution is no longer spherically symmetric, and the rotational axis will precess. Recall
Euler’s equation
dL
+ ω × L = N ext (13.62)
dt
for rotations in a body-fixed frame.
(a) What is the new inertia tensor Iαβ along principle center-of-mass frame axes? Don’t
forget that the CM is no longer at the center of the sphere! Recall I = 25 M R2 for a solid
sphere.
(b) What is the period of precession of the rotational axis in terms of the original length
of the day 2π/ω0 ?
SOLUTION: Let’s choose body-fixed axes with ẑ pointing from the center of the planet to
the smoldering asteroid. The CM lies a distance
αM0 · R + M0 · 0 α
d= = R (13.63)
(1 + α)M0 1+α
from the center of the sphere. Thus, relative to the center of the sphere, we have
   
1 0 0 1 0 0
I = 25 M0 R2 0 1 0 + αM0 R2 0 1 0 . (13.64)
0 0 1 0 0 0
Now we shift to a frame with the CM at the origin, using the parallel axis theorem,
+ M d2 δαβ − dα dβ .
CM

Iαβ (d) = Iαβ (13.65)
Thus, with d = dẑ,

     
1 0 0 1 0 0 1 0 0
2 2
CM
Iαβ 2
= 5 M0 R 0 1 0 + αM0 R 0
 1 0 − (1 + α)M0 d2 0 1 0 (13.66)
0 0 1 0 0 0 0 0 0
2 α 
5 + 1+α 0 0
2  2 α
= M0 R 0 5 + 1+α 0
 . (13.67)
2
0 0 5
13.6. EULER’S ANGLES 233
In the absence of external torques, Euler’s equations along principal axes read
dω1
I1 = (I2 − I3 ) ω2 ω3
dt
dω2
I2 = (I3 − I1 ) ω3 ω1
dt
dω3
I3 = (I1 − I2 ) ω1 ω2
dt
(13.68)
Since I1 = I2 , ω3 (t) = ω3 (0) = ω0 cos θ is a constant. We then obtain ω̇1 = Ωω2 , and
ω̇2 = −Ωω1 , with
I2 − I3 5α
Ω= ω3 = ω . (13.69)
I1 7α + 2 3
The period of precession τ in units of the pre-cataclysmic day is
τ ω 7α + 2
= = . (13.70)
T Ω 5α cos θ
13.6 Euler’s Angles
In d dimensions, an orthogonal matrix R ∈ O(d) has 21 d(d − 1) independent parameters.

To see this, consider the constraint Rt R = 1. The matrix Rt R is manifestly symmetric,
so it has 12 d(d + 1) independent entries (e.g. on the diagonal and above the diagonal).
This amounts to 12 d(d + 1) constraints on the d2 components of R, resulting in 21 d(d − 1)
freedoms. Thus, in d = 3 rotations are specified by three parameters. The Euler angles
{φ, θ, ψ} provide one such convenient parameterization.
A general rotation R(φ, θ, ψ) is built up in three steps. We start with an orthonormal triad
ê0µ of body-fixed axes. The first step is a rotation by an angle φ about ê03 :
 
cos φ sin φ 0
ê0µ = Rµν ê03 , φ ê0ν , R ê03 , φ = − sin φ cos φ 0

(13.71)
0 0 1
This step is shown in panel (a) of Fig. 13.6. The second step is a rotation by θ about the
new axis ê01 :  
1 0 0
ê00µ = Rµν ê01 , θ ê0ν , R ê01 , θ = 0 cos θ sin θ 

(13.72)
0 − sin θ cos θ
This step is shown in panel (b) of Fig. 13.6. The third and final step is a rotation by ψ
about the new axis ê003 :
 
cos ψ sin ψ 0
ê000 00 00
R ê003 , ψ = − sin ψ cos ψ 0

µ = Rµν ê3 , ψ êν , (13.73)
0 0 1
Figure 13.6: A general rotation, defined in terms of the Euler angles {φ, θ, ψ}. Three
successive steps of the transformation are shown.
This step is shown in panel (c) of Fig. 13.6. Putting this all together,
R(φ, θ, ψ) = R ê003 , φ R ê01 , θ R ê03 , ψ

(13.74)
   
cos ψ sin ψ 0 1 0 0 cos φ sin φ 0
= − sin ψ cos ψ 0 0 cos θ sin θ  − sin φ cos φ 0
0 0 1 0 − sin θ cos θ 0 0 1
 
cos ψ cos φ − sin ψ cos θ sin φ cos ψ sin φ + sin ψ cos θ cos φ sin ψ sin θ
= − sin ψ cos φ − cos ψ cos θ sin φ − sin ψ sin φ + cos ψ cos θ cos φ cos ψ sin θ .
sin θ sin φ − sin θ cos φ cos θ
Next, we’d like to relate the components ωµ = ω · êµ (with êµ ≡ ê000
µ ) of the rotation in the
body-fixed frame to the derivatives φ̇, θ̇, and ψ̇. To do this, we write
ω = φ̇ êφ + θ̇ êθ + ψ̇ êψ , (13.75)
where
ê03 = êφ = sin θ sin ψ ê1 + sin θ cos ψ ê2 + cos θ ê3 (13.76)
êθ = cos ψ ê1 − sin ψ ê2 (“line of nodes”) (13.77)
êψ = ê3 . (13.78)
This gives
ω1 = ω · ê1 = φ̇ sin θ sin ψ + θ̇ cos ψ (13.79)

ω2 = ω · ê2 = φ̇ sin θ cos ψ − θ̇ sin ψ (13.80)
ω3 = ω · ê3 = φ̇ cos θ + ψ̇ . (13.81)
Note that
φ̇ ↔ precession , θ̇ ↔ nutation , ψ̇ ↔ axial rotation . (13.82)
The general form of the kinetic energy is then

2
T = 12 I1 φ̇ sin θ sin ψ + θ̇ cos ψ
2 2
+ 12 I2 φ̇ sin θ cos ψ − θ̇ sin ψ + 21 I3 φ̇ cos θ + ψ̇ . (13.83)
Note that
L = pφ êφ + pθ êθ + pψ êψ , (13.84)
which may be verified by explicit computation.
13.6.1 Torque-free symmetric top
A body falling in a gravitational field experiences no net torque about its CM:
X X
N ext = ri × (−mi g) = g × mi ri = 0 . (13.85)
i i
For a symmetric top with I1 = I2 , we have

2
T = 12 I1 θ̇2 + φ̇2 sin2 θ + 12 I3 φ̇ cos θ + ψ̇ .

(13.86)
The potential is cyclic in the Euler angles, hence the equations of motion are
d ∂T ∂T
= . (13.87)
dt ∂(φ̇, θ̇, ψ̇) ∂(φ, θ, ψ)
Since φ and ψ are cyclic in T , their conjugate momenta are conserved:
∂L
pφ = = I1 φ̇ sin2 θ + I3 (φ̇ cos θ + ψ̇) cos θ (13.88)
∂ φ̇
∂L
pψ = = I3 (φ̇ cos θ + ψ̇) . (13.89)
∂ ψ̇
Note that pψ = I3 ω3 , hence ω3 is constant, as we have already seen.
To solve for the motion, we first note that L is conserved in the inertial frame. We are
therefore permitted to define L̂ = ê03 = êφ . Thus, pφ = L. Since êφ · êψ = cos θ, we have
Figure 13.7: A dreidl is a symmetric top. The four-fold symmetry axis guarantees I1 = I2 .
The blue diamond represents the center-of-mass.
that pψ = L · êψ = L cos θ. Finally, êφ · êθ = 0, which means pθ = L · êθ = 0. From the
equations of motion,
ṗθ = I1 θ̈ = I1 φ̇ cos θ − pψ φ̇ sin θ , (13.90)
hence we must have
pψ
θ̇ = 0 , φ̇ = . (13.91)
I1 cos θ
Note that θ̇ = 0 follows from conservation of pψ = L cos θ. From the equation for pψ , we
may now conclude
pψ pψ

I3 − I1
ψ̇ = − = ω3 , (13.92)
I3 I1 I3
which recapitulates (13.52), with ψ̇ = Ω.
13.6.2 Symmetric top with one point fixed
Consider the case of a symmetric top with one point fixed, as depicted in Fig. 13.7. The
Lagrangian is
2
L = 12 I1 θ̇2 + φ̇2 sin2 θ + 12 I3 φ̇ cos θ + ψ̇ − M g` cos θ .

(13.93)
Here, ` is the distance from the fixed point to the CM, and the inertia tensor is defined along
principal axes whose origin lies at the fixed point (not the CM!). Gravity now supplies a
torque, but as in the torque-free case, the Lagrangian is still cyclic in φ and ψ, so
pφ = (I1 sin2 θ + I3 cos2 θ) φ̇ + I3 cos θ ψ̇ (13.94)

pψ = I3 cos θ φ̇ + I3 ψ̇ (13.95)
Figure 13.8: The effective potential of eq. 13.102.
are each conserved. We can invert these relations to obtain φ̇ and ψ̇ in terms of {pφ , pψ , θ}:
pφ − pψ cos θ pψ (pφ − pψ cos θ) cos θ

φ̇ = , ψ̇ = − . (13.96)
I1 sin2 θ I3 I1 sin2 θ
In addition, since ∂L/∂t = 0, the total energy is conserved:
Ueff (θ)
z }| {
2 p 2
(p φ − p ψ cos θ) ψ
E = T + U = 12 I1 θ̇2 + + + M g` cos θ , (13.97)
2I1 sin2 θ 2I3
where the term under the brace is the effective potential Ueff (θ).
The problem thus reduces to the one-dimensional dynamics of θ(t), i.e.

∂Ueff
I1 θ̈ = − , (13.98)
∂θ
with
(pφ − pψ cos θ)2 p2ψ
Ueff (θ) = + + M g` cos θ . (13.99)
2I1 sin2 θ 2I3
Using energy conservation, we may write
r
I1 dθ
dt = ± p . (13.100)
2 E − Ueff (θ)
and thus the problem is reduced to quadratures:
r Zθ
I1 1
t(θ) = t(θ0 ) ± dϑ p . (13.101)
2 E − Ueff (ϑ)
θ0
Figure 13.9: Precession and nutation of the symmetry axis of a symmetric top.
We can gain physical insight into the motion by examining the shape of the effective po-
tential,
Ueff (θ) = + M g` cos θ + , (13.102)
2I1 sin2 θ 2I3
over the interval θ ∈ [0, π]. Clearly Ueff (0) = Ueff (π) = ∞, so the motion must be bounded.
What is not yet clear, but what is nonetheless revealed by some additional analysis, is that
Ueff (θ) has a single minimum on this interval, at θ = θ0 . The turning points for the θ motion
are at θ = θa and θ = θb , where Ueff (θa ) = Ueff (θb ) = E. Clearly if we expand about θ0 and
write θ = θ0 + η, the η motion will be harmonic, with
s
00 (θ )
Ueff 0
η(t) = η0 cos(Ωt + δ) , Ω = . (13.103)
I1
To prove that Ueff (θ) has these features, let us define u ≡ cos θ. Then u̇ = − θ̇ sin θ, and
from E = 21 I1 θ̇2 + Ueff (θ) we derive
p2ψ pφ − pψ u 2

2 2E 2 2M g` 2
u̇ = − (1 − u ) − (1 − u ) u − ≡ f (u) . (13.104)
I1 I1 I3 I1 I1
The turning points occur at f (u) = 0. The function f (u) is cubic, and the coefficient of the
cubic term is 2M g`/I1 , which is positive. Clearly f (u = ±1) = −(pφ ∓ pψ )2 /I12 is negative,
so there must be at least one solution to f (u) = 0 on the interval u ∈ (1, ∞). Clearly there
can be at most three real roots for f (u), since the function is cubic in u, hence there are at
most two turning points on the interval u ∈ [−1, 1]. Thus, Ueff (θ) has the form depicted in
fig. 13.8.
To apprehend the full motion of the top in an inertial frame, let us follow the symmetry
axis ê3 :
ê3 = sin θ sin φ ê01 − sin θ cos φ ê02 + cos θ ê03 . (13.105)
Once we know θ(t) and φ(t) we’re done. The motion θ(t) is described above: θ oscillates
between turning points at θa and θb . As for φ(t), we have already derived the result
pφ − pψ cos θ
φ̇ = . (13.106)
I1 sin2 θ
13.7. ROLLING AND SKIDDING MOTION OF REAL TOPS 239
Figure 13.10: A top with a peg end. The frictional forces f and fskid are shown. When the
top rolls without skidding, fskid = 0.
Thus, if pφ > pψ cos θa , then φ̇ will remain positive throughout the motion. If, on the other
hand, we have
pψ cos θb < pφ < pψ cos θa , (13.107)
then φ̇ changes sign at an angle θ∗ = cos−1 pφ /pψ . The motion is depicted in Fig. 13.9.

An extensive discussion of this problem is given in H. Goldstein, Classical Mechanics.
13.7 Rolling and Skidding Motion of Real Tops
The material in this section is based on the corresponding sections from V. Barger and
M. Olsson, Classical Mechanics: A Modern Perspective. This is an excellent book which
contains many interesting applications and examples.
13.7.1 Rolling tops
In most tops, the point of contact rolls or skids along the surface. Consider the peg end top
of Fig. 13.10, executing a circular rolling motion, as sketched in Fig. 13.11. There are three
components to the force acting on the top: gravity, the normal force from the surface, and
friction. The frictional force is perpendicular to the CM velocity, and results in centripetal
acceleration of the top:
f = M Ω 2 ρ ≤ µM g , (13.108)
where Ω is the frequency of the CM motion and µ is the coefficient of friction. If the above
inequality is violated, the top starts to slip.
The frictional and normal forces combine to produce a torque N = M g` sin θ − f ` cos θ
Figure 13.11: Circular rolling motion of the peg top.
about the CM2 . This torque is tangent to the circular path of the CM, and causes L to
precess. We assume that the top is spinning rapidly, so that L very nearly points along
the symmetry axis of the top itself. (As we’ll see, this is true for slow precession but not
for fast precession, where the precession frequency is proportional to ω3 .) The precession is
then governed by the equation
N = M g` sin θ − f ` cos θ

= L̇ = Ω × L ≈ Ω I3 ω3 sin θ , (13.109)
where ê3 is the instantaneous symmetry axis of the top. Substituting f = M Ω 2 ρ,
Ω2ρ

M g`
1− ctn θ = Ω , (13.110)
I3 ω3 g
which is a quadratic equation for Ω. We supplement this with the ‘no slip’ condition,

ω3 δ = Ω ρ + ` sin θ , (13.111)
resulting in two equations for the two unknowns Ω and ρ.
Substituting for ρ(Ω) and solving for Ω, we obtain

( s 2 )
I3 ω3 M g`δ M g`δ 4M `2 M g`
Ω= 1+ ctn θ ± 1+ ctn θ − · . (13.112)
2M `2 cos θ I3 I3 I3 I3 ω32
This in order to have a real solution we must have

2M `2 sin θ
r
g
ω3 ≥ . (13.113)
I3 sin θ + M g`δ cos θ `
2
Gravity of course produces no net torque about the CM.
If the inequality is satisfied, there are two possible solutions for Ω, corresponding to fast
and slow precession. Usually one observes slow precession. Note that it is possible that
ρ < 0, in which case the CM and the peg end lie on opposite sides of a circle from each
other.
13.7.2 Skidding tops
A skidding top experiences a frictional force which opposes the skidding velocity, until
vskid = 0 and a pure rolling motion sets in. This force provides a torque which makes the
top rise:
Nskid µM g`
θ̇ = − =− . (13.114)
L I3 ω3
Suppose δ ≈ 0, in which case ρ + ` sin θ = 0, from eqn. 13.111, and the point of contact
remains fixed. Now recall the effective potential for a symmetric top with one point fixed:

Ueff (θ) = + + M g` cos θ . (13.115)
2I1 sin2 θ 2I3
0 (θ ) = 0, which yields
We demand Ueff 0
cos θ0 · β 2 − pψ sin2 θ0 · β + M gÌ1 sin4 θ0 = 0 , (13.116)
where
β ≡ pφ − pψ cos θ0 = I1 sin2 θ0 φ̇ . (13.117)
Solving the quadratic equation for β, we find
s !
I3 ω3 4M gÌ1 cos θ0
φ̇ = 1± 1− . (13.118)
2I1 cos θ0 I32 ω32
This is simply a recapitulation of eqn. 13.112, with δ = 0 and with M `2 replaced by I1 .

Note I1 = M `2 by the parallel axis theorem if I1CM = 0. But to the extent that I1CM 6= 0, our
treatment of the peg top was incorrect. It turns out to be OK, however, if the precession is
slow, i.e. if Ω/ω3 1.
On a level surface, cos θ0 > 0, and therefore we must have
2
q
ω3 ≥ M gÌ1 cos θ0 . (13.119)
I3
Thus, if the top spins too slowly, it cannot maintain precession. Eqn. 13.118 says that there
are two possible precession frequencies. When ω3 is large, we have
M g` I3 ω3
φ̇slow = + O(ω3−1 ) , φ̇fast = + O(ω3−3 ) . (13.120)
I3 ω3 I1 cos θ0
Again, one usually observes slow precession.
Figure 13.12: The tippie-top behaves in a counterintuitive way. Once started spinning with
the peg end up, the peg axis rotates downward. Eventually the peg scrapes the surface and
the top rises to the vertical in an inverted orientation.
2√
A top with ω3 > I3 M gÌ1 may ‘sleep’ in the vertical position with θ0 = 0. Due to the
constant action of frictional forces, ω3 will eventually drop below this value, at which time
the vertical position is no longer stable. The top continues to slow down and eventually
falls.
13.7.3 Tippie-top
A particularly nice example from the Barger and Olsson book is that of the tippie-top, a
truncated sphere with a peg end, sketched in Fig. 13.12 The CM is close to the center
of curvature, which means that there is almost no gravitational torque acting on the top.
The frictional force f opposes slipping, but as the top spins f rotates with it, and hence
the time-averaged frictional force hf i ≈ 0 has almost no effect on the motion of the CM.
A similar argument shows that the frictional torque, which is nearly horizontal, also time
averages to zero:
dL
≈ 0. (13.121)
dt inertial
In the body-fixed frame, however, N is roughly constant, with magnitude N ≈ µM gR,

where R is the radius of curvature and µ the coefficient of sliding friction. Now we invoke

dL
N= + ω×L . (13.122)
dt body
The second term on the RHS is very small, because the tippie-top is almost spherical, hence
inertia tensor is very nearly diagonal, and this means
ω × L ≈ ω × Iω = 0 . (13.123)
Thus, L̇body ≈ N , and taking the dot product of this equation with the unit vector k̂, we
obtain
d
−N sin θ = k̂ · N = k̂ · Lbody = −L sin θ θ̇ . (13.124)
dt
Thus,
N µM gR
θ̇ = ≈ . (13.125)
L Iω
Once the stem scrapes the table, the tippie-top rises to the vertical just like any other rising
top.
Chapter 14
Continuum Mechanics
14.1 Strings
Consider a string of linear mass density µ(x) under tension τ (x).1 Let the string move in a
plane, such that its shape is described by a smooth function y(x), the vertical displacement
of the string at horizontal position x, as depicted in fig. 14.1. The action is a functional
of the height y(x, t), where the coordinate along the string, x, and time, t, are the two
independent variables. Consider a differential element of the string extending from x to
x + dx. The change in length relative to the unstretched (y = 0) configuration is
2
p 1 ∂y
dx + O dx2 .

d` = dx2 + dy 2 − dx = (14.1)
2 ∂x
The differential potential energy is then
2
1 ∂y
dU = τ (x) d` = 2 τ (x) dx . (14.2)
∂x
The differential kinetic energy is simply
2
1 ∂y
dT = 2 µ(x) dx . (14.3)
∂t
We can then write Z
L= dx L , (14.4)
where the Lagrangian density L is

2 2
0 1 ∂y 1 ∂y
L(y, ẏ, y ; x, t) = 2 µ(x) − 2 τ (x) . (14.5)
∂t ∂x
1
As an example of a string with a position-dependent tension, consider a string of length ` freely suspended
from one end at z = 0 in a gravitational field. The tension is then τ (z) = µg (` − z).
245
246 CHAPTER 14. CONTINUUM MECHANICS
Figure 14.1: A string is described by the vertical displacement field y(x, t).
The action for the string is now a double integral,
Ztb Zxb
S = dt dx L(y, ẏ, y 0 ; x, t) , (14.6)
ta xa
where y(x, t) is the vertical displacement field. Typically, we have L = 12 µẏ 2 − 12 τ y 0 2 . The
first variation of S is
Zxb Ztb " #

∂L ∂ ∂L ∂ ∂L
δS = dx dt − − δy (14.7)
∂y ∂x ∂y 0 ∂t ∂ ẏ
xa ta
Zxb t=t Ztb x=xa

∂L b ∂L
+ dx δy + dt δy , (14.8)
∂ ẏ t=ta ∂y 0 x=x
xa ta b
which simply recapitulates the general result from eqn. 14.201. There are two boundary
terms, one of which is an integral over time and the other an integral over space. The first
boundary term vanishes provided δy(x, ta ) = δy(x, tb ) = 0. The second boundary term
vanishes provided τ (x) y 0 (x) δy(x) = 0 at x = xa and x = xb , for all t. Assuming τ (x) does
not vanish, this can happen in one of two ways: at each endpoint either y(x) is fixed or
y 0 (x) vanishes.
Assuming that either y(x) is fixed or y 0 (x) = 0 at the endpoints x = xa and x = xb , the
Euler-Lagrange equations for the string are obtained by setting δS = 0:

δS ∂ ∂L ∂ ∂L
0= =− −
δy(x, t) ∂x ∂y 0 ∂t ∂ ẏ
∂2y

∂ ∂y
= τ (x) − µ(x) 2 , (14.9)
∂x ∂x ∂t
∂y ∂y
where y 0 = ∂x and ẏ = ∂t . When τ (x) = τ and µ(x) = µ are both constants, we obtain the
14.2. D’ALEMBERT’S SOLUTION TO THE WAVE EQUATION 247
Helmholtz equation,
1 ∂2y ∂2y
− =0, (14.10)
c2 ∂t2 ∂x2
p
which is the wave equation for the string, where c = τ /µ has dimensions of velocity. We
will now see that c is the speed of wave propagation on the string.
14.2 d’Alembert’s Solution to the Wave Equation
Let us define two new variables,
u ≡ x − ct , v ≡ x + ct . (14.11)
We then have
∂ ∂u ∂ ∂v ∂ ∂ ∂
= + = + (14.12)
∂x ∂x ∂u ∂x ∂v ∂u ∂v
1 ∂ 1 ∂u ∂ 1 ∂v ∂ ∂ ∂
= + =− + . (14.13)
c ∂t c ∂t ∂u c ∂t ∂v ∂u ∂v
Thus,
1 ∂2 ∂2 ∂2
− = −4 . (14.14)
c2 ∂t2 ∂x2 ∂u ∂v
Thus, the wave equation may be solved:
∂2y
=0 =⇒ y(u, v) = f (u) + g(v) , (14.15)
∂u ∂v
where f (u) and g(v) are arbitrary functions. For the moment, we work with an infinite
string, so we have no spatial boundary conditions to satisfy. Note that f (u) describes a
right-moving disturbance, and g(v) describes a left-moving disturbance:
y(x, t) = f (x − ct) + g(x + ct) . (14.16)
We do, however, have boundary conditions in time. At t = 0, the configuration of the string
is given by y(x, 0), and its instantaneous vertical velocity is ẏ(x, 0). We then have
y(x, 0) = f (x) + g(x) (14.17)
ẏ(x, 0) = −c f 0 (x) + c g 0 (x) , (14.18)
hence
f 0 (x) = 1
2 y 0 (x, 0) − 1
2c ẏ(x, 0) (14.19)
g 0 (x) = 1
2 y 0 (x, 0) + 1
2c ẏ(x, 0) , (14.20)
and integrating we obtain the right and left moving components
Zξ
f (ξ) = 1
2 y(ξ, 0) − dξ 0 ẏ(ξ 0 , 0) − C
1
2c (14.21)
0
Zξ
g(ξ) = 12 y(ξ, 0) + 2c
1
dξ 0 ẏ(ξ 0 , 0) + C , (14.22)
0
where C is an arbitrary constant. Adding these together, we obtain the full solution
x+ct
Z
h i
1 1
y(x, t) = 2 y(x − ct, 0) + y(x + ct, 0) + 2c dξ ẏ(ξ, 0) , (14.23)
x−ct
valid for all times.
14.2.1 Energy density and energy current
The Hamiltonian density for a string is
H = Π ẏ − L , (14.24)
where
∂L
Π= = µ ẏ (14.25)
∂ ẏ
is the momentum density. Thus,
Π 2 1 02
H= + 2τ y . (14.26)
2µ
Expressed in terms of ẏ rather than Π, this is the energy density E,
2
E = 12 µ ẏ 2 + 21 τ y 0 . (14.27)
We now evaluate Ė for a solution to the equations of motion:

∂E ∂y ∂ 2 y ∂y ∂ 2 y
=µ + τ
∂t ∂t ∂t2 ∂x ∂t ∂x
∂y ∂ 2 y

∂y ∂ ∂y
=τ τ +τ
∂t ∂x ∂x ∂x ∂t ∂x

∂ ∂y ∂y ∂j
= τ ≡− E , (14.28)
∂x ∂x ∂t ∂x
where the energy current density (or energy flux) is
∂y ∂y
jE = −τ . (14.29)
∂x ∂t
Figure 14.2: Reflection of a pulse at an interface at x = 0, with y(0, t) = 0.
We therefore have that solutions of the equation of motion also obey the energy continuity
equation
∂E ∂j
+ E =0. (14.30)
∂t ∂x
Let us integrate the above equation between points x1 and x2 . We obtain
Zx2 Zx2
∂ ∂j (x, t)
dx E(x, t) = − dx E = jE (x1 , t) − jE (x2 , t) , (14.31)
∂t ∂x
x1 x1

which says that the time rate of change of the energy contained in the interval x1 , x2 is
equal to the difference between the entering and exiting energy flux.
When τ (x) = τ and µ(x) = µ, we have
y(x, t) = f (x − ct) + g(x + ct) (14.32)
and we find
2 2
E(x, t) = τ [f 0 (x − ct) + τ g 0 (x + ct)

(14.33)
2 2
jE (x, t) = cτ f 0 (x − ct) − cτ g 0 (x + ct) ,

(14.34)
which are each sums over right-moving and left-moving contributions.
14.2.2 Reflection at an interface

Consider a semi-infinite string on the interval 0, ∞ , with y(0, t) = 0. We can still invoke
d’Alembert’s solution, y(x, t) = f (x − ct) + g(x + ct), but we must demand
y(0, t) = f (−ct) + g(ct) = 0 ⇒ f (ξ) = −g(−ξ) . (14.35)
Thus,
y(x, t) = g(ct + x) − g(ct − x) . (14.36)
Figure 14.3: Reflection of a pulse at an interface at x = 0, with y 0 (0, t) = 0.
Now suppose g(ξ) describes a pulse, and is nonzero only within a neighborhood of ξ = 0.
For large negative values of t, the right-moving part, −g(ct − x), is negligible everywhere,
since x > 0 means that the argument ct − x is always large and negative. On the other
hand, the left moving part g(ct + x) is nonzero for x ≈ −ct > 0. Thus, for t < 0 we have
a left-moving pulse incident from the right. For t > 0, the situation is reversed, and the
left-moving component is negligible, and we have a right moving reflected wave. However,
the minus sign in eqn. 14.35 means that the reflected wave is inverted.
If instead of fixing the endpoint at x = 0 we attach this end of the string to a massless ring
which frictionlessly slides up and down a vertical post, then we must have y 0 (0, t) = 0, else
there is a finite vertical force on the massless ring, resulting in infinite acceleration. We
again write y(x, t) = f (x − ct) + g(x + ct), and we invoke
y 0 (0, t) = f 0 (−ct) + g 0 (ct) ⇒ f 0 (ξ) = −g 0 (−ξ) , (14.37)
which, upon integration, yields f (ξ) = g(−ξ), and therefore
y(x, t) = g(ct + x) + g(ct − x) . (14.38)
The reflected pulse is now ‘right-side up’, in contrast to the situation with a fixed endpoint.
14.2.3 Mass point on a string
Next, consider the case depicted in Fig. 14.4, where a point mass m is affixed to an infinite
string at x = 0. Let us suppose that at large negative values of t, a right moving wave
f (ct − x) is incident from the left. The full solution may then be written as a sum of
incident, reflected, and transmitted waves:
x<0 : y(x, t) = f (ct − x) + g(ct + x) (14.39)
x>0 : y(x, t) = h(ct − x) . (14.40)
At x = 0, we invoke Newton’s second Law, F = ma:
m ÿ(0, t) = τ y 0 (0+ , t) − τ y 0 (0− , t) . (14.41)

Figure 14.4: Reflection and transmission at an impurity. A point mass m is affixed to an

infinite string at x = 0.
Any discontinuity in the derivative y 0 (x, t) at x = 0 results in an acceleration of the point

mass. Note that
y 0 (0− , t) = −f 0 (ct) + g 0 (ct) , y 0 (0+ , t) = −h0 (ct) . (14.42)
Further invoking continuity at x = 0, i.e. y(0− , t) = y ( 0+ , t), we have
h(ξ) = f (ξ) + g(ξ) , (14.43)
and eqn. 14.41 becomes

2τ 0
g 00 (ξ) + g (ξ) = −f 00 (ξ) . (14.44)
mc2
We solve this equation by Fourier analysis:
Z∞ Z∞
dk ˆ
f (ξ) = f (k) eikξ , fˆ(k) = dξ f (ξ) e−ikξ . (14.45)
2π
−∞ −∞
Defining κ ≡ 2τ /mc2 = 2µ/m, we have
− k 2 + iκk ĝ(k) = k 2 fˆ(k) .

(14.46)
We then have
k
ĝ(k) = − fˆ(k) ≡ r(k) fˆ(k) (14.47)
k − iκ
−iκ ˆ
ĥ(k) = f (k) ≡ t(k) fˆ(k) , (14.48)
k − iκ
where r(k) and t(k) are the reflection and transmission amplitudes, respectively. Note that
t(k) = 1 + r(k) . (14.49)

In real space, we have

Z∞
dk
h(ξ) = t(k) fˆ(k) eikξ (14.50)
2π
−∞
Z∞ " Z∞ #
dk 0
= dξ 0 t(k) eik(ξ−ξ ) f (ξ 0 ) (14.51)
2π
−∞ −∞
Z∞
≡ dξ 0 T (ξ − ξ 0 ) f (ξ 0 ) , (14.52)
−∞
where
Z∞
0 dk 0
T (ξ − ξ ) = t(k) eik(ξ−ξ ) , (14.53)
2π
−∞
is the transmission kernel in real space. For our example with r(k) = −iκ/(k − iκ), the
integral is done easily using the method of contour integration:
Z∞
dk −iκ ik(ξ−ξ0 ) 0
T (ξ − ξ 0 ) = e = κ e−κ(ξ−ξ ) Θ(ξ − ξ 0 ) . (14.54)
2π k − iκ
−∞
Therefore,
Zξ
0
h(ξ) = κ dξ 0 e−κ(ξ−ξ ) f (ξ 0 ) , (14.55)
−∞
and of course g(ξ) = h(ξ) − f (ξ). Note that m = ∞ means κ = 0, in which case r(k) = −1
and t(k) = 0. Thus we recover the inversion of the pulse shape under reflection found
earlier.

For example, let the incident pulse shape be f (ξ) = b Θ a − |ξ| . Then
Zξ
0
h(ξ) = κ dξ 0 e−κ(ξ−ξ ) b Θ(a − ξ 0 ) Θ(a + ξ 0 )
−∞
h i
= b e−κξ eκ min(a,ξ) − e−κa Θ(ξ + a) . (14.56)
Taking cases, 
0 if ξ < −a


h(ξ) = b 1 − e −κ(a+ξ) if − a < ξ < a (14.57)

2b e−κξ sinh(κa) if ξ > a .

In Fig. 14.5 we show the reflection and transmission of this square pulse for two different
values of κa.
Figure 14.5: Reflection and transmission of a square wave pulse by a point mass at x = 0
The configuration of the string is shown for six different times, for κa = 0.5 (left panel)
and κa = 5.0 (right panel). Note that the κa = 0.5 case, which corresponds to a large mass
m = 2µ/κ, results in strong reflection with inversion, and weak transmission. For large κ,
corresponding to small mass m, the reflection is weak and the transmission is strong.
14.2.4 Interface between strings of different mass density
Consider the situation in fig. 14.6, where the string for x < 0 is of density µL and for x > 0
is of density µR . The d’Alembert solution in the two regions, with an incoming wave from
the left, is
x < 0: y(x, t) = f (cL t − x) + g(cL t + x) (14.58)
x > 0: y(x, t) = h(cR t − x) . (14.59)

At x = 0 we have
f (cL t) + g(cL t) = h(cR t) (14.60)
−f 0 (cL t) + g 0 (cL t) = −h0 (cR t) , (14.61)
where the second equation follows from τ y 0 (0+ , t) = τ y 0 (0− , t), so there is no finite verti-
cal force on the infinitesimal interval bounding x = 0, which contains infinitesimal mass.
Defining α ≡ cR /cL , we integrate the second of these equations and have
f (ξ) + g(ξ) = h(α ξ) (14.62)
f (ξ) − g(ξ) = α−1 h(α ξ) . (14.63)

Figure 14.6: String formed from two semi-infinite regions of different densities..
Note that y(±∞, 0) = 0 fixes the constant of integration. The solution is then
α−1
g(ξ) = f (ξ) (14.64)
α+1
2α
h(ξ) = f (ξ/α) . (14.65)
α+1
Thus,

α−1
x < 0: y(x, t) = f cL t − x + f cL t + x (14.66)
α+1
2α
x > 0: y(x, t) = f (cR t − x)/α . (14.67)
α+1
It is instructive to compute the total energy in the string. For large negative values of the
time t, the entire disturbance is confined to the region x < 0. The energy is
Z∞
2
E(−∞) = τ dξ f 0 (ξ) .

(14.68)
−∞
For large positive times, the wave consists of the left-moving reflected g(ξ) component in
the region x < 0 and the right-moving transmitted component h(ξ) in the region x > 0.
The energy in the reflected wave is
2 Z∞
α−1 2
dξ f 0 (ξ) .

EL (+∞) = τ (14.69)
α+1
−∞
For the transmitted portion, we use
2
y 0 (x > 0, t) = f 0 (cR t − x)/α

(14.70)
α+1
14.3. FINITE STRINGS : BERNOULLI’S SOLUTION 255
to obtain
Z∞
4τ 2
dξ f 0 (ξ/α)

ER (∞) =
(α + 1)2
−∞
Z∞
4ατ 2
dξ f 0 (ξ) .

= (14.71)
(α + 1)2
−∞
Thus, EL (∞) + ER (∞) = E(−∞), and energy is conserved.
14.3 Finite Strings : Bernoulli’s Solution
Suppose xa = 0 and xb = L are the boundaries of the string, where y(0, t) = y(L, t) = 0.
Again we write
y(x, t) = f (x − ct) + g(x + ct) . (14.72)
Applying the boundary condition at xa = 0 gives, as earlier,
y(x, t) = g(ct + x) − g(ct − x) . (14.73)
Next, we apply the boundary condition at xb = L, which results in
g(ct + L) − g(ct − L) = 0 =⇒ g(ξ) = g(ξ + 2L) . (14.74)
Thus, g(ξ) is periodic, with period 2L. Any such function may be written as a Fourier sum,
∞
( )
X nπξ nπξ
g(ξ) = An cos + Bn sin . (14.75)
L L
n=1
The full solution for y(x, t) is then
y(x, t) = g(ct + x) − g(ct − x)
∞
1/2 X ( )
2 nπx nπct nπct
= sin An cos + Bn sin , (14.76)
µL L L L
n=1
√ √
where An = 2µL Bn and Bn = − 2µL An . This is known as Bernoulli’s solution.
We define the functions 1/2

2 nπx
ψn (x) ≡ sin . (14.77)
µL L
We also write
nπx nπc
kn ≡ , ωn ≡ , n = 1, 2, 3, . . . , ∞ . (14.78)
L L
p
Thus, ψn (x) = 2/µL sin(kn x) has (n + 1) nodes at x = jL/n, for j ∈ {0, . . . , n}. Note
that
ZL

ψm ψn ≡ dx µ ψm (x) ψn (x) = δmn . (14.79)
0
Furthermore, this basis is complete:
∞
X
µ ψn (x) ψn (x0 ) = δ(x − x0 ) . (14.80)
n=1
Our general solution is thus equivalent to

∞
X
y(x, 0) = An ψn (x) (14.81)
n=1
∞
X nπc
ẏ(x, 0) = Bn ψn (x) . (14.82)
L
n=1
The Fourier coefficients {An , Bn } may be extracted from the initial data using the orthonor-
mality of the basis functions and their associated resolution of unity:
ZL
An = dx µ ψn (x) y(x, 0) (14.83)
0
ZL
L
Bn = dx µ ψn (x) ẏ(x, 0) . (14.84)
nπc
0
As an example, suppose our initial configuration is a triangle, with


2b
L x
 if 0 ≤ x ≤ 12 L
y(x, 0) = (14.85)

 2b 1
L (L − x) if 2 L ≤ x ≤ L ,
and ẏ(x, 0) = 0. Then Bn = 0 for all n, while

1/2 ( ZL/2 ZL )
2µ 2b nπx nπx
An = · dx x sin + dx (L − x) sin
L L L L
0 L/2
4b
= (2µL)1/2 · 1

sin 2 nπ δn,odd , (14.86)
n2 π 2

after changing variables to x = Lθ/nπ and using θ sin θ dθ = d sin θ − θ cos θ . Another
way to write this is to separately give the results for even and odd coefficients:
4b 1/2 (−1)k
A2k = 0 , A2k+1 = (2µL) · . (14.87)
π2 (2k + 1)2
14.4. STURM-LIOUVILLE THEORY 257
Figure 14.7: Evolution of a string with fixed ends starting from an isosceles triangle shape.
Note that each ψ2k (x) = −ψ2k (L − x) is antisymmetric about the midpoint x = 21 L, for
all k. Since our initial conditions are that y(x, 0) is symmetric about x = 12 L, none of the
even order eigenfunctions can enter into the expansion, precisely as we have found. The
d’Alembert solution to this problem is particularly simple and is shown in Fig. 14.7. Note
that g(x) = 12 y(x, 0) must be extended to the entire real line. We know that g(x) = g(x+2L)

is periodic with
spatial
period 2L, but how to we extend g(x) from the interval 0, L to
the interval − L, 0 ? To do this, we use y(x, 0) = g(x) − g(−x), which says that g(x) must
be antisymmetric, i.e. g(x) = −g(−x). Equivalently, ẏ(x, 0) = cg 0 (x) − cg 0 (−x) = 0, which
integrates to g(x) = −g(−x).
14.4 Sturm-Liouville Theory
Consider the Lagrangian density

2
L= 1
2 µ(x) ẏ 2 − 12 τ (x) y 0 − 12 v(x) y 2 . (14.88)
The last term is new and has the physical interpretation of a harmonic potential which
attracts the string to the line y = 0. The Euler-Lagrange equations are then
∂2y

∂ ∂y
− τ (x) + v(x) y = −µ(x) 2 . (14.89)
∂x ∂x ∂t
This equation is invariant under time translation. Thus, if y(x, t) is a solution, then so is
y(x, t + t0 ), for any t0 . This means that the solutions can be chosen to be eigenstates of
the operator ∂t , which is to say y(x, t) = ψ(x) e−iωt . Because the coefficients are real, both
y and y ∗ are solutions, and taking linear combinations we have
y(x, t) = ψ(x) cos(ωt + φ) . (14.90)
Plugging this into eqn. 14.89, we obtain
dh i
− τ (x) ψ 0 (x) + v(x) ψ(x) = ω 2 µ(x) ψ(x) . (14.91)
dx
This is the Sturm-Liouville equation. There are four types of boundary conditions that we
shall consider:
1. Fixed endpoint: ψ(x) = 0, where x = xa,b .
2. Natural: τ (x) ψ 0 (x) = 0, where x = xa,b .
3. Periodic: ψ(x) = ψ(x + L), where L = xb − xa .
4. Mixed homogeneous: α ψ(x) + β ψ 0 (x) = 0, where x = xa,b .
The Sturm-Liouville equation is an eigenvalue equation. The eigenfunctions {ψn (x)} satisfy
dh i
− τ (x) ψn0 (x) + v(x) ψn (x) = ωn2 µ(x) ψn (x) . (14.92)
dx
Now suppose we a second solution ψm (x), satisfying
dh 0
i
2
− τ (x) ψm (x) + v(x) ψm (x) = ωm µ(x) ψm (x) . (14.93)
dx
Now multiply (14.92)∗ by ψm (x) and (14.93) by ψn∗ (x) and subtract, yielding
dh 0 i d h 0∗ i
ψn∗ τ ψ n = ωn∗ 2 − ωm
2
µ ψm ψn∗

τ ψm − ψm (14.94)
dx dx
dh ∗ 0 ∗
i
= τ ψn ψm − τ ψm ψ 0 n . (14.95)
dx
We integrate this equation over the length of the string, to get
Zxb h ix=xa
∗
ωn∗ 2 2
dx µ(x) ψn∗ (x) ψm (x) = τ (x) ψn∗ (x) ψm
0
(x) − τ (x) ψm (x) ψ 0 n (x)

− ωm
x=xb
xa
=0. (14.96)
The RHS vanishes for any of the four types of boundary conditions articulated above.
Thus, we have
ωn∗ 2 − ωm
2

ψn ψm = 0 , (14.97)
where the inner product is defined as
Zxb
dx µ(x) ψ ∗ (x) φ(x) .

ψ φ ≡ (14.98)
xa

Note that the distribution µ(x) is non-negative definite. Setting m = n, we have ψn ψn ≥
0, and hence ωn∗ 2 = ωn2 , which says that ωn2 ∈ R. When ωm 2 6= ω 2 , the eigenfunctions
n
are orthogonal with respect to the above inner product. In the case of degeneracies, we
may invoke the Gram-Schmidt procedure, which orthogonalizes the eigenfunctions within a
14.4. STURM-LIOUVILLE THEORY 259
given degenerate subspace. Since the Sturm-Liouville equation is linear, we may normalize
the eigenfunctions, taking

ψm ψn = δmn . (14.99)
Finally, since the coefficients in the Sturm-Liouville equation are all real, we can and hence-
forth do choose the eigenfunctions themselves to be real.
14.4.1 Variational method
Consider the functional

Rxb n o
1
2 dx τ (x) ψ 0 2 (x) + v(x) ψ 2 (x)
xa N
ω 2 ψ(x) =

≡ . (14.100)
1
Rxb D
2 dx µ(x) ψ 2 (x)
xa
The variation is
δN N δD
δω 2 = −
D D2
δN − ω 2 δD
= . (14.101)
D
Thus,
δω 2 = 0 =⇒ δN = ω 2 δD , (14.102)
which says
d dψ(x)
− τ (x) + v(x) ψ(x) = ω 2 µ(x) ψ(x) , (14.103)
dx dx
which is the Sturm-Lioiuville equation. In obtaining this equation, we have dropped a
boundary term, which is correct provided
h ix=x
b
τ (x) ψ 0 (x) ψ(x) =0. (14.104)
x=xa
This condition is satisfied for any of the first three classes of boundary conditions: ψ = 0
(fixed endpoint), τ ψ 0 = 0 (natural), or ψ(xa ) = ψ(xb ), ψ 0 (xa ) = ψ 0 (xb ) (periodic). For
the fourth class of boundary conditions, αψ + βψ 0 = 0 (mixed homogeneous), the Sturm-
Liouville equation may still be derived, provided one uses a slightly different functional,
N
e = N + α τ x ψ2 x − τ x ψ2 x
e h i
ω 2 ψ(x) =

with N b b a a , (14.105)
D 2β
since then
Zxb ( )
d dψ(x)
δNe −Ne δD = dx − τ (x) + v(x) ψ(x) − ω 2 µ(x) ψ(x) δψ(x)
dx dx
xa
" #x=x
b
0 α
+ τ (x) ψ (x) + ψ(x) δψ(x) , (14.106)
β
x=xa
and the last term vanishes as a result of the boundary conditions.
For all four classes of boundary conditions we may write

K
z { }|
Rxb d d h i
dx ψ(x) − τ (x) + v(x) ψ(x)
2
xa dx dx
ω ψ(x) = (14.107)
Rxb
dx µ(x) ψ 2 (x)
xa
If we expand ψ(x) in the basis of eigenfunctions of the Sturm-Liouville operator K,

∞
X
ψ(x) = Cn ψn (x) , (14.108)
n=1
we obtain P∞
|Cj |2 ωj2
ω ψ(x) = ω 2 (C1 , . . . , C∞ ) =
2 Pj=1

∞ 2
. (14.109)
k=1 |Ck |
If ω12 ≤ ω22 ≤ . . ., then we see that ω 2 ≥ ω12 , so an arbitrary function ψ(x) will always yield
an upper bound to the lowest eigenvalue.
As an example, consider a violin string (v = 0) with a mass m affixed in the center. We

write µ(x) = µ + m δ(x − 21 L), hence
RL
τ dx ψ 0 2 (x)
0
ω 2 ψ(x) =

(14.110)
RL
m ψ 2 ( 12 L) + µ dx ψ 2 (x)
0
Now consider a trial function


A x
 α if 0 ≤ x ≤ 21 L
ψ(x) = (14.111)
 1
A (L − x)α if 2L ≤x≤L.

The functional ω 2 ψ(x) now becomes an ordinary function of the trial parameter α, with

R L/2 2 2α−2 2
α2 (2α + 1)

2 2τ 0 dx α x 2c
ω (α) = = · m ,
(14.112)
2α L/2 L (2α − 1) 1 + (2α + 1) M
1
R
m 2L + 2µ dx s2α
0
where M = µL is the mass of the string alone. We minimize ω 2 (α) to obtain the optimal
solution of this form:
dω 2 m
=0 =⇒ 4α2 − 2α − 1 + (2α + 1)2 (α − 1) =0. (14.113)
dα M
14.5. CONTINUA IN HIGHER DIMENSIONS 261
Figure 14.8: One-parameter variational solution for a string with a mass m affixed at
x = 12 L.
√
For m/M → 0, we obtain α = 41 1 + 5 ≈ 0.809. The variational estimate for the
eigenvalue is then 6.00% larger than the exact answer ω10 = πc/L. In the opposite limit,
m/M → ∞, the inertia of the string may be neglected. The normal mode is then piecewise
linear, in the shape of an isosceles triangle with base L and height y. Thep equation of
motion is then mÿ = −2τ · (y/ 12 L), assuming |y/L| 1. Thus, ω1 = (2c/L) M/m. This
is reproduced exactly by the variational solution, for which α → 1 as m/M → ∞.
14.5 Continua in Higher Dimensions
In higher dimensions, we generalize the operator K as follows:

∂ ∂
K=− τ (x) β + v(x) . (14.114)
∂xα αβ ∂x
The eigenvalue equation is again
Kψ(x) = ω 2 µ(x) ψ(x) , (14.115)
and the Green’s function again satisfies
h i
K − ω 2 µ(x) Gω (x, x0 ) = δ(x − x0 ) , (14.116)
and has the eigenfunction expansion,

∞
X ψn (x) ψn (x0 )
Gω (x, x0 ) = . (14.117)
ωn2 − ω 2
n=1
The eigenfunctions form a complete and orthonormal basis:

∞
X
µ(x) ψn (x) ψn (x0 ) = δ(x − x0 ) (14.118)
n=1
Z
dx µ(x) ψm (x) ψn (x) = δmn , (14.119)
Ω
where Ω is the region of space in which the continuous medium exists. For purposes of
simplicity, we consider here fixed boundary conditions u(x, t)∂Ω = 0, where ∂Ω is the
boundary of Ω. The general solution to the wave equation
∂2

∂ ∂
µ(x) 2 − α ταβ (x) β + v(x) u(x, t) = 0 (14.120)
∂t ∂x ∂x
is
∞
X
u(x, t) = Cn ψn (x) cos(ωn t + δn ) . (14.121)
n=1
The variational approach generalizes as well. We define

Z
∂ψ ∂ψ 2
N ψ(x) = dx ταβ +vψ (14.122)
∂xα ∂xβ
Ω
Z
D ψ(x) = dx µ ψ 2 ,

(14.123)
Ω
and
2
N ψ(x)
ω ψ(x) = . (14.124)
D ψ(x)
Setting the variation δω 2 = 0 recovers the eigenvalue equation Kψ = ω 2 µ ψ.
14.5.1 Membranes
Consider a surface where the height z is a function of the lateral coordinates x and y:
z = u(x, y) . (14.125)
The equation of the surface is then
F (x, y, z) = z − u(x, y) = 0 . (14.126)

Let the differential element of surface area be dS. The projection of this element onto the
(x, y) plane is
dA = dx dy
= n̂ · ẑ dS . (14.127)
The unit normal n̂ is given by

∇F ẑ − ∇u
n̂ = =p . (14.128)
∇F 1 + (∇u)2
Thus,
dx dy p
dS = = 1 + (∇u)2 dx dy . (14.129)
n̂ · ẑ
The potential energy for a deformed surface can take many forms. In the case we shall
consider here, we consider only the effect of surface tension σ, and we write the potential
energy functional as
Z

U u(x, y, t) = σ dS
Z
= U0 + 2 dA (∇u)2 + . . . .
1
(14.130)
The kinetic energy functional is

Z
1
dA µ(x) (∂t u)2 .

T u(x, y, t) = 2 (14.131)
Thus, the action is Z

d2x L(u, ∇u, ∂t u, x) ,

S u(x, t) = (14.132)
where the Lagrangian density is
L = 21 µ(x) (∂t u)2 − 12 σ(x) (∇u)2 , (14.133)
where here we have allowed both µ(x) and σ(x) to depend on the spatial coordinates. The
equations of motion are
∂ ∂L ∂L ∂L
0= +∇· − (14.134)
∂t ∂ ∂t u ∂ ∇u ∂u
∂ 2u n o
= µ(x) 2 − ∇ · σ(x) ∇u . (14.135)
∂t
14.5.2 Helmholtz equation
When µ and σ are each constant, we obtain the Helmholtz equation:

1 ∂2

2
∇ − 2 2 u(x, t) = 0 , (14.136)
c ∂t
p
with c = σ/µ. The d’Alembert solution still works – waves of arbitrary shape can
propagate in a fixed direction k̂:
u(x, t) = f (k̂ · x − ct) . (14.137)
This is called a plane wave because the three dimensional generalization of this wave has
wavefronts which are planes. In our case, it might better be called a line wave, but people
will look at you funny if you say that, so we’ll stick with plane wave. Note that the locus
of points of constant f satisfies
φ(x, t) = k̂ · x − ct = constant , (14.138)
and setting dφ = 0 gives

dx
k̂ · =c, (14.139)
dt
which means that the velocity along k̂ is c. The component of x perpendicular to k̂ is
arbitrary, hence the regions of constant φ correspond to lines which are orthogonal to k̂.
Owing to the linearity of the wave equation, we can construct arbitrary superpositions of
plane waves. The most general solution is written
Z 2 h
dk i(k·x−ckt) i(k·x+ckt)
i
u(x, t) = A(k) e + B(k) e . (14.140)
(2π)2
The first term in the bracket on the RHS corresponds to a plane wave moving in the +k̂
direction, and the second term to a plane wave moving in the −k̂ direction.
14.5.3 Rectangles
Consider a rectangular membrane where x ∈ [0, a] and y ∈ [0, b], and subject to the bound-
ary conditions u(0, y) = u(a, y) = u(x, 0) = u(x, b) = 0. We try a solution of the form
u(x, y, t) = X(x) Y (y) T (t) . (14.141)
This technique is known as separation of variables. Dividing the Helmholtz equation by u

then gives
1 ∂ 2X 1 ∂ 2Y 1 1 ∂ 2T
+ = . (14.142)
X ∂x2 Y ∂y 2 c2 T ∂t2
The first term on the LHS depends only on x. The second term on the LHS depends only
on y. The RHS depends only on t. Therefore, each of these terms must individually be
constant. We write
1 ∂ 2X 1 ∂ 2Y 1 ∂ 2T
= −kx2 , = −ky2 , = −ω 2 , (14.143)
X ∂x2 Y ∂y 2 T ∂t2
with
ω2
kx2 + ky2 = . (14.144)
c2
Thus, ω = ±c|k|. The most general solution is then
X(x) = A cos(kx x) + B sin(kx x) (14.145)

Y (y) = C cos(ky y) + D sin(ky y) (14.146)
T (t) = E cos(ωt) + B sin(ωt) . (14.147)
The boundary conditions now demand
A=0 , C=0 , sin(kx a) = 0 , sin(ky b) = 0 . (14.148)
Thus, the most general solution subject to the boundary conditions is

∞ X∞
X mπx nπy
u(x, y, t) = Amn sin sin cos ωmn t + δmn , (14.149)
a b
m=1 n=1
where s 2 2
mπc nπc
ωmn = + . (14.150)
a b
14.5.4 Circles
For a circular membrane, such as a drumhead, it is convenient to work in two-dimensional

polar coordinates (r, ϕ). The Laplacian is then
1 ∂ ∂ 1 ∂2
∇2 = r + 2 . (14.151)
r ∂r ∂r r ∂ϕ2
We seek a solution to the Helmholtz equation which satisfies the boundary conditions u(r =
a, ϕ, t) = 0. Once again, we invoke the separation of variables method, writing
u(r, ϕ, t) = R(r) Φ(ϕ) T (t) , (14.152)
resulting in
1 1 ∂ 2Φ 1 1 ∂2T

1 1 ∂ ∂R
r + 2 = . (14.153)
R r ∂r ∂r r Φ ∂ϕ2 c2 T ∂t2
The azimuthal and temporal functions are
Φ(ϕ) = eimϕ , T (t) = cos(ωt + δ) , (14.154)
where m is an integer in order that the function u(r, ϕ, t) be single-valued. The radial
equation is then
∂ 2R 1 ∂R
2
m2

ω
+ + − R=0. (14.155)
∂r2 r ∂r c2 r2
This is Bessel’s equation, with solution
ωr ωr
R(r) = A Jm + B Nm , (14.156)
c c
where Jm (z) and Nm (z) are the Bessel and Neumann functions of order m, respectively.
Since the Neumann functions diverge at r = 0, we must exclude them, setting B = 0 for
each m.
We now invoke the boundary condition u(r = a, ϕ, t) = 0. This requires

ωa c
Jm =0 =⇒ ω = ωm` = xm` , (14.157)
c a
where Jm (xm` ) = 0, i.e. xm` is the `th zero of Jm (x). The mose general solution is therefore
∞ X
X ∞

u(r, ϕ, t) = Am` Jm xm` r/a cos mϕ + βm` cos(ωm` t + δm` . (14.158)
m=0 `=1
14.5.5 Sound in fluids
Let %(x, t) and v(x, t) be the density and velocity fields in a fluid. Mass conservation
requires
∂%
+ ∇ · (% v) = 0 . (14.159)
∂t
This is the continuity equation for mass.
Focus now on a small packet of fluid

of infinitesimal volume dV . The total force on this
fluid element is dF = −∇p + % g dV . By Newton’s Second Law,
dv
dF = % dV (14.160)
dt
Note that the chain rule gives
dv ∂v
= + v·∇ v . (14.161)
dt ∂t
Thus, dividing eqn, 14.160 by dV , we obtain

∂v
% + v · ∇ v = −∇p + % g . (14.162)
∂t
This is the inviscid (i.e. zero viscosity) form of the Navier-Stokes equation.
Locally the fluid can also be described in terms of thermodynamic variables p(x, t) (pressure)
and T (x, t) (temperature). For a one-component fluid there is necessarily an equation of
state of the form p = p(%, T ). Thus, we may write

∂p ∂p
dp = d% + dT . (14.163)
∂% T ∂T %
We now make the following approximations. First, we assume that the fluid is close to
equilibrium at v = 0, meaning we write p = p̄ + δp and % = %̄ + δ%, and assume that δp, δ%,
and v are small. The smallness of v means we can neglect the nonlinear term (v · ∇)v in
eqn. 14.162. Second, we neglect gravity (more on this later). The continuity equation then
takes the form
∂ δ%
+ %̄ ∇ · v = 0 , (14.164)
∂t
and the Navier-Stokes equation becomes
∂v
%̄ = −∇δp . (14.165)
∂t
Taking the time derivative of the former, and then invoking the latter of these equations
yields
∂ 2 δ%

2 ∂p
2
=∇ p= ∇2 δ% ≡ c2 ∇2 δ% . (14.166)
∂t ∂%
The speed of wave propagation, i.e. the speed of sound, is given by
s
∂p
c= . (14.167)
∂%
Finally, we must make an assumption regarding the conditions under which the derivative
∂p/∂% is computed. If the fluid is an excellent conductor of heat, then the temperature will
equilibrate quickly and it is a good approximation to take the derivative at fixed temper-
ature. The resulting value of c is called the isothermal sound speed cT . If, on the other
hand, the fluid is a poor conductor of heat, as is the case for air, then it is more appropriate
to take the derivative at constant entropy, yielding the adiabatic sound speed. Thus,
s s
∂p ∂p
cT = , cS = . (14.168)
∂% T ∂% S
√
In an ideal gas, cS /cT = γ, where γ = cp /cV is the ratio of the specific heat at constant
pressure to that at constant volume. For a (mostly) diatomic gas like air (comprised of N2
and O2 and just a little Ar), γ = 75 . Note that one can write c2 = 1/%κ, where

1 ∂%
κ= (14.169)
% ∂p
is the compressibility, which is the inverse of the bulk modulus. Again, one must specify
whether one is talking about κT or κS . For reference in air at T = 293 K, using M =
28.8 g/mol, one obtains cT = 290.8 m/s and cS = 344.0 m/s. In H2 O at 293 K, c = 1482 m/s.
In Al at 273 K, c = 6420 m/s.
If we retain gravity, the wave equation becomes

∂ 2 δ%
= c2 ∇2 δ% − g · ∇δ% . (14.170)
∂t2
The dispersion relation is then
p
ω(k) = c2 k 2 − ig · k . (14.171)
We are permitted to ignore the effects of gravity so long as c2 k 2 gk. In terms of the
wavelength λ = 2π/k, this requires
2πc2
λ = 75.9 km (at T = 293 K) . (14.172)
g
14.6 Dispersion
The one-dimensional Helmholtz equation ∂x2 y = c−2 ∂t2 y is solved by a plane wave
y(x, t) = A eikx e−iωt , (14.173)
provided ω = ±ck. We say that there are two branches to the dispersion relation ω(k)
for this equation. In general, we may add solutions, due to the linearity of the Helmholtz
equation. The most general solution is then
Z∞
dk h ˆ i
y(x, t) = f (k) eik(x−ct) + ĝ(k) eik(x+ct)
2π
−∞
= f (x − ct) + g(x + ct) , (14.174)
which is consistent with d’Alembert’s solution.
Consider now the free particle Schrödinger equation in one space dimension,
∂ψ h̄2 ∂ 2ψ
ih̄ =− . (14.175)
∂t 2m ∂x2
The function ψ(x, t) is the quantum mechanical wavefunction for a particle of mass m
moving freely along a one-dimensional line. The probability density for finding the particle
at position x at time t is 2
ρ(x, t) = ψ(x, t) . (14.176)
Conservation of probability therefore requires
Z∞
2
dx ψ(x, t) = 1 . (14.177)
−∞
This condition must hold at all times t.
As is the case with the Helmholtz equation, the Schrödinger equation is solved by a plane
wave of the form
ψ(x, t) = A eikx e−iωt , (14.178)
where the dispersion relation now only has one branch, and is given by
h̄k 2
ω(k) = . (14.179)
2m
14.6. DISPERSION 269
The most general solution is then

Z∞
dk 2
ψ(x, t) = ψ̂(k) eikx e−ih̄k t/2m . (14.180)
2π
−∞
Let’s suppose we start at time t = 0 with a Gaussian wavepacket,

−1/4 2 /2`2
ψ(x, 0) = π`20 e−x 0 eik0 x . (14.181)
To find the amplitude ψ̂(k), we perform the Fourier transform:

Z∞
ψ̂(k) = dx ψ(x, 0) e−ikx
−∞
√ −1/4 2 `2 /2
= 2 π`20 e−(k−k0 ) 0 . (14.182)
We now compute ψ(x, t) valid for all times t:
√ Z∞
dk ikx −(k−k0 )2 `20 /2 ikx −ih̄k2 t/2m
−1/4
ψ(x, t) = 2 π`20 e e e e (14.183)
2π
−∞
" 2 #
2 −1/4
−1/2 x − h̄k0 t/m
= π`0 1 + it/τ exp − 2
2 `0 1 + t2 /τ 2
" #
i 2k0 `20 x + x2 t/τ − k02 `40 t/τ
× exp , (14.184)
2 `20 1 + t2 /τ 2

where τ ≡ m`20 /h̄. The probability density is then the normalized Gaussian
1 2 /`2 (t)
ρ(x, t) = p e−(x−v0 t) , (14.185)
π `2 (t)
where v0 = h̄k0 /m and p

`(t) = `0 1 + t2 /τ 2 . (14.186)
Note that `(t) gives the width of the wavepacket, and that this width increases as a function
of time, with `(t τ ) ' `0 t/τ .
Unlike the case of the Helmholtz equation, the solution to the Schrödinger equation does not
retain its shape as it moves. This phenomenon is known as the spreading of the wavepacket.
In fig. 14.9, we show the motion and spreading of the wavepacket.
For a given plane wave eikx e−iω(k)t , the wavefronts move at the phase velocity
ω(k)
vp (k) = . (14.187)
k
Figure 14.9: Wavepacket spreading for k0 `0 = 2 with t/τ = 0, 2, 4, 6, and 8.
The center of the wavepacket, however, travels at the group velocity

dω
vg (k) = , (14.188)
dk k
0
2
where k = k0 is the maximum of ψ̂(k) .
14.7 Appendix I : Three Strings
Problem: Three identical strings are connected to a ring of mass m as shown in fig. 14.10.
The linear mass density of each string is σ and each string is under identical tension τ . In
equilibrium, all strings are coplanar. All motion on the string is in the ẑ-direction, which is
perpendicular to the equilibrium plane. The ring slides frictionlessly along a vertical pole.
It is convenient to describe each string as a half line [−∞, 0]. We can choose coordinates
x1 , x2 , and x3 for the three strings, respectively. For each string, the ring lies at xi = 0.
A pulse is sent down the first string. After a time, the pulse arrives at the ring. Transmitted
waves are sent down the other two strings, and a reflected wave down the first string. The
solution to the wave equation in the strings can be written as follows. In string #1, we have
z = f (ct − x1 ) + g(ct + x1 ) . (14.189)
In the other two strings, we may write z = hA (ct + x2 ) and z = hB (ct + x3 ), as indicated in
the figure.
14.7. APPENDIX I : THREE STRINGS 271
Figure 14.10: Three identical strings arranged symmetrically in a plane, attached to a

common end. All motion is in the direction perpendicular to this plane. The red ring,
whose mass is m, slides frictionlessly in this direction along a pole.
(a) Write the wave equation in string #1. Define all constants.
(b) Write the equation of motion for the ring.
(c) Solve for the reflected wave g(ξ) in terms of the incident wave f (ξ). You may write this
relation in terms of the Fourier transforms fˆ(k) and ĝ(k).
(d) Suppose a very long wavelength pulse of maximum amplitude A is incident on the ring.
What is the maximum amplitude of the reflected pulse? What do we mean by “very long
wavelength”?
Solution:
(a) The wave equation is

∂2z 1 ∂2z
= , (14.190)
∂x2 c2 ∂t2
p
where x is the coordinate along the string, and c = τ /σ is the speed of wave propagation.
(b) Let Z be the vertical coordinate of the ring. Newton’s second law says mZ̈ = F , where
the force on the ring is the sum of the vertical components of the tension in the three strings
at x = 0: h i
F = −τ − f 0 (ct) + g 0 (ct) + h0A (ct) + h0B (ct) , (14.191)
where prime denotes differentiation with respect to argument.
(c) To solve for the reflected wave, we must eliminate the unknown functions hA,B and
then obtain g in terms of f . This is much easier than it might at first seem. We start by
demanding continuity at the ring. This means
Z(t) = f (ct) + g(ct) = hA (ct) = hB (ct) (14.192)
for all t. We can immediately eliminate hA,B :
hA (ξ) = hB (ξ) = f (ξ) + g(ξ) , (14.193)
for all ξ. Newton’s second law from part (b) may now be written as
mc2 f 00 (ξ) + g 00 (ξ) = −τ f 0 (ξ) + 3g 0 (ξ) .

(14.194)
This linear ODE becomes a simple linear algebraic equation for the Fourier transforms,
Z∞
dk ˆ
f (ξ) = f (k) eikξ , (14.195)
2π
−∞
etc. We readily obtain

k − iQ ˆ
ĝ(k) = − f (k) , (14.196)
k − 3iQ
where Q ≡ τ /mc2 has dimensions of inverse length. Since hA,B = f + g, we have

2iQ
ĥA (k) = ĥB (k) = − fˆ(k) . (14.197)
k − 3iQ
(d) For a very long wavelength pulse, composed of plane waves for which |k| Q, we
have ĝ(k) ≈ − 31 fˆ(k). Thus, the reflected pulse is inverted, and is reduced by a factor 13 in
amplitude. Note that for a very short wavelength pulse, for which k Q, we have perfect
reflection with inversion, and no transmission. This is due to the inertia of the ring.
It is straightforward to generalize this problem to one with n strings. The transmission

into each of the (n − 1) channels is of course identical (by symmetry). One then finds the
reflection and transmission amplitudes

k − i(n − 2)Q 2iQ
r(k) = − , t(k) = − . (14.198)
k − inQ k − inQ
Conservation of energy means that the sum of the squares of the reflection amplitude and
all the (n − 1) transmission amplitudes must be unity:
r(k)2 + (n − 1) t(k)2 = 1 .

(14.199)
14.8. APPENDIX II : GENERAL FIELD THEORETIC FORMULATION 273
14.8 Appendix II : General Field Theoretic Formulation
Continuous systems possess an infinite number of degrees of freedom. They are described
by a set of fields φa (x, t) which depend on space and time. These fields may represent
local displacement, pressure, velocity, etc. The equations of motion of the fields are again
determined by extremizing the action, which, in turn, is an integral of the Lagrangian
density over all space and time. Extremization yields a set of (generally coupled) partial
differential equations.
14.8.1 Euler-Lagrange equations for classical field theories
Suppose φa (x) depends on n independent variables, {x1 , x2 , . . . , xn }. Consider the func-

tional Z

S {φa (x) = dx L(φa ∂µ φa , x) , (14.200)
Ω
i.e. the Lagrangian density L is a function of the fields φa and their partial derivatives
∂φa /∂xµ . Here Ω is a region in Rn . Then the first variation of S is
Z ( )
∂L ∂L ∂ δφa
δS = dx δφ +
∂φa a ∂(∂µ φa ) ∂xµ
Ω
I Z ( )
∂L ∂L ∂ ∂L
= dΣ nµ δφ + dx − δφa , (14.201)
∂(∂µ φa ) a ∂φa ∂xµ ∂(∂µ φa )
∂Ω Ω
where ∂Ω is the (n − 1)-dimensional boundary of Ω,

dΣ is the differential
surface area, and
µ
n is the unit normal. If we demand ∂L/∂(∂µ φa ) ∂Ω = 0 of δφa ∂Ω = 0, the surface term

vanishes, and we conclude
" #
δS ∂L ∂ ∂L
= − , (14.202)
δφa (x) ∂φa ∂xµ ∂(∂µ φa )
x
where the subscript means we are to evaluate the term in brackets at x. In a mechanical
system, one of the n independent variables (usually x0 ), is the time t. However, we may
be interested in a time-independent context in which we wish to extremize the energy
functional, for example. In any case, setting the first variation of S to zero yields the
Euler-Lagrange equations,

∂L ∂ ∂L
δS = 0 ⇒ − =0 (14.203)
∂φa ∂xµ ∂(∂µ φa )
The Lagrangian density for an electromagnetic field with sources is

1
L = − 16π Fµν F µν − Jµ Aµ . (14.204)

∂L ∂ ∂L
− ν =0 ⇒ ∂µ F µν = 4πJ ν , (14.205)
∂Aν ∂x ∂(∂ µ Aν )
which are Maxwell’s equations.
14.8.2 Conserved currents in field theory
Recall the result of Noether’s theorem for mechanical systems:

!
d ∂L ∂ η̃σ
=0, (14.206)
dt ∂ q̇σ ∂ζ
ζ=0
where η̃σ = η̃σ (q, ζ) is a one-parameter (ζ) family of transformations of the generalized
coordinates which leaves L invariant. We generalize to field theory by replacing
qσ (t) −→ φa (x, t) , (14.207)
where {φa (x, t)} are a set of fields, which are functions of the independent variables {x, y, z, t}.
We will adopt covariant relativistic notation and write for four-vector xµ = (ct, x, y, z). The
generalization of dQ/dt = 0 is
!
∂ ∂L ∂ φ̃a
=0, (14.208)
∂xµ ∂ (∂µ φa ) ∂ζ
ζ=0
where there is an implied sum on both µ and a. We can write this as ∂µ J µ = 0, where

∂L ∂ φ̃
a
Jµ ≡ . (14.209)
∂ (∂µ φa ) ∂ζ

ζ=0
We call Q = J 0 /c the total charge. If we assume J = 0 at the spatial boundaries of our

system, then integrating the conservation law ∂µ J µ over the spatial region Ω gives
Z Z I
dQ 3 0 3
= d x ∂0 J = − d x ∇ · J = − dΣ n̂ · J = 0 , (14.210)
dt
Ω Ω ∂Ω
assuming J = 0 at the boundary ∂Ω.
As an example, consider the case of a complex scalar field, with Lagrangian density2
L(ψ, , ψ ∗ , ∂µ ψ, ∂µ ψ ∗ ) = 12 K (∂µ ψ ∗ )(∂ µ ψ) − U ψ ∗ ψ .

(14.211)
2
We raise and lower indices using the Minkowski metric gµν = diag (+, −, −, −).
14.8. APPENDIX II : GENERAL FIELD THEORETIC FORMULATION 275
This is invariant under the transformation ψ → eiζ ψ, ψ ∗ → e−iζ ψ ∗ . Thus,
∂ ψ̃ ∂ ψ̃ ∗
= i eiζ ψ , = −i e−iζ ψ ∗ , (14.212)
∂ζ ∂ζ
and, summing over both ψ and ψ ∗ fields, we have
∂L ∂L
Jµ = · (iψ) + · (−iψ ∗ )
∂ (∂µ ψ) ∂ (∂µ ψ ∗ )
K ∗ µ
ψ ∂ ψ − ψ ∂ µψ∗ .

= (14.213)
2i
The potential, which depends on |ψ|2 , is independent of ζ. Hence, this form of conserved
4-current is valid for an entire class of potentials.
14.8.3 Gross-Pitaevskii model
As one final example of a field theory, consider the Gross-Pitaevskii model, with
∂ψ h̄2 2
L = ih̄ ψ ∗ − ∇ψ ∗ · ∇ψ − g |ψ|2 − n0 . (14.214)
∂t 2m
This describes a Bose fluid with repulsive short-ranged interactions. Here ψ(x, t) is again
a complex scalar field, and ψ ∗ is its complex conjugate. Using the Leibniz rule, we have
δS[ψ ∗ , ψ] = S[ψ ∗ + δψ ∗ , ψ + δψ]

h̄2 h̄2
Z Z
∂δψ ∂ψ
= dt d x ih̄ ψ ∗
d
+ ih̄ δψ ∗ − ∇ψ ∗ · ∇δψ − ∇δψ ∗ · ∇ψ
∂t ∂t 2m 2m

2
∗ ∗
− 2g |ψ| − n0 (ψ δψ + ψδψ )
(
∂ψ ∗ h̄2 2 ∗
Z Z
d 2
∗
= dt d x − ih̄ + ∇ ψ − 2g |ψ| − n0 ψ δψ
∂t 2m
)
h̄2 2

∂ψ 2
∗
+ ih̄ + ∇ ψ − 2g |ψ| − n0 ψ δψ , (14.215)
∂t 2m
where we have integrated by parts where necessary and discarded the boundary terms.
Extremizing S[ψ ∗ , ψ] therefore results in the nonlinear Schrödinger equation (NLSE),
∂ψ h̄2 2
∇ ψ + 2g |ψ|2 − n0 ψ

ih̄ =− (14.216)
∂t 2m
as well as its complex conjugate,
∂ψ ∗ h̄2 2 ∗
∇ ψ + 2g |ψ|2 − n0 ψ ∗ .

−ih̄ =− (14.217)
∂t 2m
Note that these equations are indeed the Euler-Lagrange equations:

δS ∂L ∂ ∂L
= − (14.218)
δψ ∂ψ ∂xµ ∂ ∂µ ψ

δS ∂L ∂ ∂L
= − , (14.219)
δψ ∗ ∂ψ ∗ ∂xµ ∂ ∂µ ψ ∗
with xµ = (t, x)3 Plugging in
∂L ∂L ∂L h̄2
= −2g |ψ|2 − n0 ψ ∗ = ih̄ ψ ∗ ∇ψ ∗

, , =− (14.220)
∂ψ ∂ ∂t ψ ∂ ∇ψ 2m
and
∂L ∂L ∂L h̄2
= ih̄ ψ − 2g |ψ|2 − n0 ψ

∗
, =0 , = − ∇ψ , (14.221)
∂ψ ∂ ∂t ψ ∗ ∂ ∇ψ ∗ 2m
we recover the NLSE and its conjugate.
The Gross-Pitaevskii model also possesses a U(1) invariance, under
ψ(x, t) → ψ̃(x, t) = eiζ ψ(x, t) , ψ ∗ (x, t) → ψ̃ ∗ (x, t) = e−iζ ψ ∗ (x, t) . (14.222)
Thus, the conserved Noether current is then

∂L ∂ ψ̃ ∂L ∂ ψ̃ ∗
Jµ = +

∂ ∂µ ψ ∂ζ ∂ ∂µ ψ ∗ ∂ζ

ζ=0 ζ=0
J 0 = −h̄ |ψ|2 (14.223)
h̄2
ψ ∗ ∇ψ − ψ∇ψ ∗ .

J =− (14.224)
2im
Dividing out by h̄, taking J 0 ≡ −h̄ρ and J ≡ −h̄j, we obtain the continuity equation,
∂ρ
+∇·j =0 , (14.225)
∂t
where
h̄
ρ = |ψ|2 ψ ∗ ∇ψ − ψ∇ψ ∗ .

, j= (14.226)
2im
are the particle density and the particle current, respectively.
3
In the nonrelativistic case, there is no utility in defining x0 = ct, so we simply define x0 = t.
14.9. APPENDIX III : GREEN’S FUNCTIONS 277
14.9 Appendix III : Green’s Functions
Suppose we add a forcing term,

∂2y

∂ ∂y h i
µ(x) 2 − τ (x) + v(x) y = Re µ(x) f (x) e−iωt . (14.227)
∂t ∂x ∂x
We write the solution as h i
−iωt
y(x, t) = Re y(x) e , (14.228)
where
d dy(x)
− τ (x) + v(x) y(x) − ω 2 µ(x) y(x) = µ(x) f (x) , (14.229)
dx dx
or h i
K − ω 2 µ(x) y(x) = µ(x) f (x) , (14.230)
where K is a differential operator,
d d
K≡− τ (x) + v(x) . (14.231)
dx dx
Note that the eigenfunctions of K are the {ψn (x)}:
K ψn (x) = ωn2 µ(x) ψn (x) . (14.232)
The formal solution to equation 14.230 is then

h i−1
y(x) = K − ω 2 µ 0
µ(x0 ) f (x0 ) (14.233)
x,x
Zxb
= dx0 µ(x0 ) Gω (x, x0 ) f (x0 ). (14.234)
xa
What do we mean by the term in brackets? If we define the Green’s function

h i−1
Gω (x, x0 ) ≡ K − ω 2 µ 0
, (14.235)
x,x
what this means is h i

K − ω 2 µ(x) Gω (x, x0 ) = δ(x − x0 ) . (14.236)
Note that the Green’s function may be expanded in terms of the (real) eigenfunctions, as
X ψn (x) ψn (x0 )
Gω (x, x0 ) = , (14.237)
n
ωn2 − ω 2
which follows from completeness of the eigenfunctions:

∞
X
µ(x) ψn (x) ψn (x0 ) = δ(x − x0 ) . (14.238)
n=1
The expansion in eqn. 14.237 is formally exact, but difficult to implement, since it requires
summing over an infinite set of eigenfunctions. It is more practical to construct the Green’s
function from solutions to the homogeneous Sturm Liouville equation, as follows. When
x 6= x0 , we have that (K − ω 2 µ) Gω (x, x0 ) = 0, which is a homogeneous ODE of degree two.
Consider first the interval x ∈ [xa , x0 ]. A second order homogeneous ODE has two solutions,
and further invoking the boundary condition at x = xa , there is a unique solution, up to
a multiplicative constant. Call this solution y1 (x). Next, consider the interval x ∈ [x0 , xb ].
Once again, there is a unique solution to the homogeneous Sturm-Liouville equation, up
to a multiplicative constant, which satisfies the boundary condition at x = xb . Call this
solution y2 (x). We then can write

0 0
A(x ) y1 (x) if xa ≤ x < x

Gω (x, x0 ) = (14.239)

B(x0 ) y2 (x) if x0 < x ≤ xb .

Here, A(x0 ) and B(x0 ) are undetermined functions. We now invoke the inhomogeneous
Sturm-Liouville equation,
dGω (x, x0 )

d
− τ (x) + v(x) Gω (x, x0 ) − ω 2 µ(x) , Gω (x, x0 ) = δ(x − x0 ) . (14.240)
dx dx
We integrate this from x = x0 − to x = x0 + , where is a positive infinitesimal. This
yields h i
τ (x0 ) A(x0 ) y10 (x0 ) − B(x0 ) y20 (x0 ) = 1 . (14.241)
Continuity of Gω (x, x0 ) itself demands
A(x0 ) y1 (x0 ) = B(x0 ) y2 (x0 ) . (14.242)
Solving these two equations for A(x0 ) and B(x0 ), we obtain

y2 (x0 ) y1 (x0 )
A(x0 ) = − , B(x0 ) = − , (14.243)
τ (x0 ) Wy1 ,y2 (x0 ) τ (x0 ) Wy1 ,y2 (x0 )
where Wy1 ,y2 (x) is the Wronskian

 
y1 (x) y2 (x)
Wy1 ,y2 (x) = det 
0 0
y1 (x) y2 (x)
= y1 (x) y20 (x) − y2 (x) y10 (x) . (14.244)
Now it is easy to show that Wy1 ,y2 (x) τ (x) = W τ is a constant. This follows from the fact
that
0 = y2 K y1 − y2 K y1
i
d h
0 0
= τ (x) y1 y2 − y2 y1 . (14.245)
dx
Thus, we have 
0 if xa ≤ x < x0
−y1 (x) y2 (x )/W

Gω (x, x0 ) = (14.246)

−y1 (x0 ) y2 (x)/W if x0 < x ≤ xb ,

or, in compact form,

y1 (x< ) y2 (x> )
Gω (x, x0 ) = − , (14.247)
Wτ
where x< = min(x, x0 ) and x> = max(x, x0 ).
As an example, consider a uniform string (i.e. µ and τ constant, v = 0) with fixed endpoints
at xa = 0 and xb = L. The normalized eigenfunctions are
r
2 nπx
ψn (x) = sin , (14.248)
µL L
and the eigenvalues are ωn = nπc/L. The Green’s function is

∞
0 2 X sin(nπx/L) sin(nπx0 /L)
Gω (x, x ) = . (14.249)
µL (nπc/L)2 − ω 2
n=1
Now construct the homogeneous solutions:

2 ωx
(K − ω µ) y1 = 0 , y1 (0) = 0 =⇒ y1 (x) = sin (14.250)
c

ω(L − x)
(K − ω 2 µ) y2 = 0 , y2 (L) = 0 =⇒ y2 (x) = sin . (14.251)
c
The Wronskian is
ω ωL
W= y1 y20 − y2 y10 = − sin . (14.252)
c c
Therefore, the Green’s function is

0sin ωx< /c sin ω(L − x> )/c
Gω (x, x ) = . (14.253)
(ωτ /c) sin(ωL/c)
14.9.1 Perturbation theory
Suppose we have solved for the Green’s function for the linear operator K0 and mass density
µ0 (x). I.e. we have
K0 − ω 2 µ0 (x) G0ω (x, x0 ) = δ(x − x0 ) .

(14.254)
We now imagine perturbing τ0 → τ0 + λτ1 , v0 → v0 + λv2 , µ0 → µ0 + λµ1 . What is the new
Green’s function Gω (x, x0 )? We must solve
L0 + λL1 Gω (x, x0 ) = δ(x − x0 ) ,

(14.255)
Figure 14.11: Diagrammatic representation of the perturbation expansion in eqn. 14.258..
where
L0ω ≡ K0 − ω 2 µ0 (14.256)
L1ω 2
≡ K1 − ω µ1 . (14.257)
Dropping the ω subscript for simplicity, the full Green’s function is then given by
h i−1
Gω = L0ω + λL1ω
h −1 i−1
= G0ω + λL1ω
h i−1
= 1 + λ G0ω L1ω G0ω
= G0ω − λ G0ω L1ω G0ω + λ2 G0ω L1ω G0ω L1ω G0ω + . . . . (14.258)
The ‘matrix multiplication’ is of course a convolution, i.e.

Zxb
0
G0ω (x, x0 ) − λ dx1 G0ω (x, x1 ) L1ω x1 , dx Gω (x1 , x0 ) + . . . .
d
0
Gω (x, x ) = (14.259)
1
xa
Each term in the perturbation expansion of eqn. 14.258 may be represented by a diagram,
as depicted in Fig. 14.11.
As an example, consider a string with xa = 0 and xb = L with a mass point m affixed at

the point x = d. Thus, µ1 (x) = m δ(x − d), and L1ω = −mω 2 δ(x − d), with λ = 1. The
perturbation expansion gives
Gω (x, x0 ) = G0ω (x, x0 ) + mω 2 G0ω (x, d) G0ω (d, x0 ) + m2 ω 4 G0ω (x, d) G0ω (d, d) G0ω (d, x0 ) + . . .
mω 2 G0ω (x, d) G0ω (d, x0 )
= G0ω (x, x0 ) + . (14.260)
1 − mω 2 G0ω (d, d)
Note that the eigenfunction expansion,
X ψn (x) ψn (x0 )
Gω (x, x0 ) = , (14.261)
n
ωn2 − ω 2
says that the exact eigenfrequencies are poles of Gω (x, x0 ), and furthermore the residue at
each pole is
1
Res Gω (x, x0 ) = − ψn (x) ψn (x0 ) . (14.262)
ω=ωn 2ωn
According to eqn. 14.260, the poles of Gω (x, x0 ) are located at solutions to4
mω 2 G0ω (d, d) = 1 . (14.263)
1
For simplicity let us set d = 2 L, so the mass point is in the middle of the string. Then
according to eqn. 14.253,
sin2 (ωL/2c)
G0ω 1 1

2 L, 2 L =
(ωτ /c) sin(ωL/c)

c ωL
= tan . (14.264)
2ωτ 2c
The eigenvalue equation is therefore

ωL 2τ
tan = , (14.265)
2c mωc
which can be manipulated to yield

m
λ = ctn λ , (14.266)
M
where λ = ωL/2c and M = µL is the total mass of the string. When m = 0, the LHS
vanishes, and the roots lie at λ = (n + 21 )π, which gives ω = ω2n+1 . Why don’t we see the
poles at the even mode eigenfrequencies ω2n ? The answer is that these poles are present
in the Green’s function. They do not cancel for d = 12 L because the perturbation does
not couple to the even modes, which all have ψ2n ( 12 L) = 0. The case of general d may be
instructive in this regard. One finds the eigenvalue equation
sin(2λ) m
= , (14.267)
2λ sin 2λ sin 2(1 − )λ M
where = d/L. Now setting m = 0 we recover 2λ = nπ, which says ω = ωn , and all the
modes are recovered.
Note in particular that there is no longer any divergence at the location of the original poles of G0ω (x, x0 ).
4
These poles are cancelled.

14.9.2 Perturbation theory for eigenvalues and eigenfunctions
We wish to solve
K0 + λK1 ψ = ω 2 µ0 + λµ1 ψ ,

(14.268)
which is equivalent to
L0ω ψ = −λL1ω ψ . (14.269)
−1
Multiplying by L0ω = G0ω on the left, we have
Zxb
ψ(x) = −λ dx0 Gω (x, x0 ) L1ω ψ(x0 ) (14.270)
xa
∞ Zxb
X ψm (x)
=λ dx0 ψm (x0 ) L1ω ψ(x0 ) . (14.271)
ω 2 − ωm
2
m=1 xa
We are free to choose any normalization we like for ψ(x). We choose

Zxb

ψ ψn = dx µ0 (x) ψn (x) ψ(x) = 1 , (14.272)
xa
which entails
Zxb
2
ω − ωn2 = λ dx ψn (x) L1ω ψ(x) (14.273)
xa
as well as
Zxb
X ψk (x)
ψ(x) = ψn (x) + λ dx0 ψk (x0 ) L1ω ψ(x0 ) . (14.274)
k
ω 2 − ωk2
(k6=n) xa
By expanding ψ and ω 2 in powers of λ, we can develop an order by order perturbation

series.
To lowest order, we have

Zxb
2
ω = ωn2 + λ dx ψn (x) L1ωn ψn (x) . (14.275)
xa
For the case L1ω = −m ω 2 δ(x − d), we have

δωn 2
= − 12 m ψn (d)

ωn
m 2 nπd

=− sin . (14.276)
M L
For d = 21 L, only the odd n modes are affected, as the even n modes have a node at x = 12 L.
Carried out to second order, one obtains for the eigenvalues,

Zxb
2
ω = ωn2 + λ dx ψn (x) L1ωn ψn (x)
xa
R 2
xb 1 ψ (x)
X xa dx ψ k (x) Lωn n
+ λ2 + O(λ3 )
k
ωn2 − ωk2
(k6=n)
Zxb Zxb
2
− λ dx ψn (x) Lωn ψn (x) · dx0 µ1 (x0 ) ψn (x0 ) + O(λ3 ) .
2 1

(14.277)
xa xa
Chapter 15
Special Relativity
For an extraordinarily lucid, if characteristically brief, discussion, see chs. 1 and 2 of L. D.

Landau and E. M. Lifshitz, The Classical Theory of Fields (Course of Theoretical Physics,
vol. 2).
15.1 Introduction
All distances are relative in physics. They are measured with respect to a fixed frame of
reference. Frames of reference in which free particles move with constant velocity are called
inertial frames. The principle of relativity states that the laws of Nature are identical in
all inertial frames.
15.1.1 Michelson-Morley experiment
We learned how sound waves in a fluid, such as air, obey the Helmholtz equation. Let us
restrict our attention for the moment to solutions of the form φ(x, t) which do not depend
on y or z. We then have a one-dimensional wave equation,
∂ 2φ 1 ∂ 2φ
= . (15.1)
∂x2 c2 ∂t2
The fluid in which the sound propagates is assumed to be at rest. But suppose the fluid
is not at rest. We can investigate this by shifting to a moving frame, defining x0 = x − ut,
with y 0 = y, z 0 = z and of course t0 = t. This is a Galilean transformation. In terms of the
new variables, we have
∂ ∂ ∂ ∂ ∂
= , = −u 0 + 0 . (15.2)
∂x ∂x0 ∂t ∂x ∂t
285
286 CHAPTER 15. SPECIAL RELATIVITY
The wave equation is then
u2 ∂ 2φ 1 ∂ 2φ 2u ∂ 2φ

1− 2 = − . (15.3)
c ∂x0 2 c2 ∂t0 2 c2 ∂x0 ∂t0
Clearly the wave equation acquires a different form when expressed in the new variables
(x0 , t0 ), i.e. in a frame in which the fluid is not at rest. The general solution is then of the
modified d’Alembert form,
φ(x0 , t0 ) = f (x0 − cR t0 ) + g(x0 + cL t0 ) , (15.4)
where cR = c − u and cL = c + u are the speeds of rightward and leftward propagating

disturbances, respectively. Thus, there is a preferred frame of reference – the frame in
which the fluid is at rest. In the rest frame of the fluid, sound waves travel with velocity c
in either direction.
Light, as we know, is a wave phenomenon in classical physics. The propagation of light is

described by Maxwell’s equations,
1 ∂B
∇ · E = 4πρ ∇×E =− (15.5)
c ∂t
4π 1 ∂E
∇·B =0 ∇×B = j+ , (15.6)
c c ∂t
where ρ and j are the local charge and current density, respectively. Taking the curl of
Faraday’s law, and restricting to free space where ρ = j = 0, we once again have (using a
Cartesian system for the fields) the wave equation,
1 ∂ 2E
∇2E = . (15.7)
c2 ∂t2
(We shall discuss below, in section 15.8, the beautiful properties of Maxwell’s equations
under general coordinate transformations.)
In analogy with the theory of sound, it was assumed prior to Einstein that there was in
fact a preferred reference frame for electromagnetic radiation – one in which the medium
which was excited during the EM wave propagation was at rest. This notional medium was
called the lumineferous ether . Indeed, it was generally assumed during the 19th century
that light, electricity, magnetism, and heat (which was not understood until Boltzmann’s
work in the late 19th century) all had separate ethers. It was Maxwell who realized that
light, electricity, and magnetism were all unified phenomena, and accordingly he proposed
a single ether for electromagnetism. It was believed at the time that the earth’s motion
through the ether would result in a drag on the earth.
In 1887, Michelson and Morley set out to measure the changes in the speed of light on earth
due to the earth’s movement through the ether (which was generally assumed to be at rest
in the frame of the Sun). The Michelson interferometer is shown in fig. 15.1, and works as
follows. Suppose the apparatus is moving with velocity u x̂ through the ether. Then the
15.1. INTRODUCTION 287
Figure 15.1: The Michelson-Morley experiment (1887) used an interferometer to effectively

measure the time difference for light to travel along two different paths. Inset: analysis for
the y-directed path.
time it takes a light ray to travel from the half-silvered mirror to the mirror on the right
and back again is
` ` 2`c
tx = + = 2 . (15.8)
c+u c−u c − u2
For motion qalong the other arm of the interferometer, the geometry in the inset of fig. 15.1
0
shows ` = `2 + 1/4u2 t2y , hence
2`0 2q 2 2`
ty = = ` + 1/4u2 t2y ⇒ ty = √ . (15.9)
c c c2− u2
Thus, the difference in times along these two paths is
2`c 2` ` u2
∆t = tx − ty = − √ ≈ · . (15.10)
c2 c2 − u2 c c2
Thus, the difference in phase between the two paths is
∆φ ` u2
= ν ∆t ≈ · 2 , (15.11)
2π λ c
where λ is the wavelength of the light. We take u ≈ 30 km/s, which is the earth’s orbital
velocity, and λ ≈ 5000 Å. From this we find that ∆φ ≈ 0.02 × 2π if ` = 1 m. Michelson
Figure 15.2: Experimental setup of Alvager et al. (1964), who used the decay of high energy
neutral pions to test the source velocity dependence of the speed of light.
and Morley found that the observed fringe shift ∆φ/2π was approximately 0.02 times the
expected value. The inescapable conclusion was that the speed of light did not depend on
the motion of the source. This was very counterintuitive!
The history of the development of special relativity is quite interesting, but we shall not have
time to dwell here on the many streams of scientific thought during those exciting times.
Suffice it to say that the Michelson-Morley experiment, while a landmark result, was not the
last word. It had been proposed that the ether could be dragged, either entirely or partially,
by moving bodies. If the earth dragged the ether along with it, then there would be no
ground-level ‘ether wind’ for the MM experiment to detect. Other experiments, however,
such as stellar aberration, in which the apparent position of a distant star varies due to the
earth’s orbital velocity, rendered the “ether drag” theory untenable – the notional ‘ether
bubble’ dragged by the earth could not reasonably be expected to extend to the distant
stars.
A more recent test of the effect of a moving source on the speed of light was performed
by T. Alvåger et al., Phys. Lett. 12, 260 (1964), who measured the velocity of γ-rays
(photons) emitted from the decay of highly energetic neutral pions (π 0 ). The pion energies
were in excess of 6 GeV, which translates to a velocity of v = 0.99975 c, according to special
relativity. Thus, photons emitted in the direction of the pions should be traveling at close
to 2c, if the source and photon velocities were to add. Instead, the velocity of the photons
was found to be c = 2.9977 ± 0.0004 × 1010 cm/s, which is within experimental error of the
best accepted value.
Figure 15.3: Two reference frames.
15.1.2 Einsteinian and Galilean relativity
The Principle of Relativity states that the laws of nature are the same when expressed in
any inertial frame. This principle can further be refined into two classes, depending on
whether one takes the velocity of the propagation of interactions to be finite or infinite.
The interaction of matter in classical mechanics is described by a potential function U (r1 , . . . , rN ).

P
Typically, one has two-body interactions in which case one writes U = i<j U (ri , rj ). These
interactions are thus assumed to be instantaneous, which is unphysical. The interaction of
particles is mediated by the exchange of gauge bosons, such as the photon (for electro-
magnetic interactions), gluons (for the strong interaction, at least on scales much smaller
than the ‘confinement length’), or the graviton (for gravity). Their velocity of propagation,
according to the principle of relativity, is the same in all reference frames, and is given by
the speed of light, c = 2.998 × 108 m/s.
Since c is so large in comparison with terrestrial velocities, and since d/c is much shorter
than all other relevant time scales for typical interparticle separations d, the assumption
of an instantaneous interaction is usually quite accurate. The combination of the principle
of relativity with finiteness of c is known as Einsteinian relativity. When c = ∞, the
combination comprises Galilean relativity:
c<∞ : Einsteinian relativity

c=∞ : Galilean relativity .
Consider a train moving at speed u. In the rest frame of the train track, the speed of
the light beam emanating from the train’s headlight is c + u. This would contradict the
principle of relativity. This leads to some very peculiar consequences, foremost among them
being the fact that events which are simultaneous in one inertial frame will not in general
be simultaneous in another. In Newtonian mechanics, on the other hand, time is absolute,
and is independent of the frame of reference. If two events are simultaneous in one frame
then they are simultaneous in all frames. This is not the case in Einsteinian relativity!
We can begin to apprehend this curious feature of simultaneity by the following Gedanken-
experiment (a long German word meaning “thought experiment”)1 . Consider the case in
fig. 15.3 in which frame K 0 moves with velocity u x̂ with respect to frame K. Let a source
at S emit a signal (a light pulse) at t = 0. In the frame K 0 the signal’s arrival at equidistant
locations A and B is simultaneous. In frame K, however, A moves toward left-propagating
emitted wavefront, and B moves away from the right-propagating wavefront. For classical
sound, the speed of the left-moving and right-moving wavefronts is c∓u, taking into account
the motion of the source, and thus the relative velocities of the signal and the detectors
remain at c. But according to the principle of relativity, the speed of light is c in all frames,
and is so in frame K for both the left-propagating and right-propagating signals. Therefore,
the relative velocity of A and the left-moving signal is c + u and the relative velocity of B
and the right-moving signal is c − u. Therefore, A ‘closes in’ on the signal and receives it
before B, which is moving away from the signal. We might expect the arrival times to be
t∗A = d/(c + u) and t∗B = d/(c − u), where d is the distance between the source S and either
detector A or B in the K 0 frame. Later on we shall analyze this problem and show that
r r
∗ c−u d ∗ c+u d
tA = · , tB = · . (15.12)
c+u c c−u c
Our naı̈ve analysis has omitted an important detail – the Lorentz contraction of the distance
d as seen by an observer in the K frame.
15.2 Intervals
Now let us express mathematically the constancy of c in all frames. An event is specified
by the time and place where it occurs. Thus, an event is specified by four coordinates,
(t, x, y, z). The four-dimensional space spanned by these coordinates is called spacetime.
The interval between two events in spacetime at (t1 , x1 , y1 , z1 ) and (t2 , x2 , y2 , z2 ) is defined
to be q
s12 = c2 (t1 − t2 )2 − (x1 − x2 )2 − (y1 − y2 )2 − (z1 − z2 )2 . (15.13)
For two events separated by an infinitesimal amount, the interval ds is infinitesimal, with
ds2 = c2 dt2 − dx2 − dy 2 − dz 2 . (15.14)
Now when the two events denote the emission and reception of an electromagnetic signal,
we have ds2 = 0. This must be true in any frame, owing to the invariance of c, hence since
ds and ds0 are differentials of the same order, we must have ds0 2 = ds2 . This last result
requires homogeneity and isotropy of space as well. Finally, if infinitesimal intervals are
invariant, then integrating we obtain s = s0 , and we conclude that the interval between two
space-time events is the same in all inertial frames.
When s212 > 0, the interval is said to be time-like. For timelike intervals, we can always
find a reference frame in which the two events occur at the same locations. As an example,
1
Unfortunately, many important physicists were German and we have to put up with a legacy of long
German words like Gedankenexperiment, Zitterbewegung, Brehmsstrahlung, Stosszahlansatz , Kartoffelsalat,
etc.
15.2. INTERVALS 291
Figure 15.4: A (1 + 1)–dimensional light cone. The forward light cone consists of timelike
events with ∆t > 0. The backward light cone consists of timelike events with ∆t < 0. The
causally disconnected regions are time-like, and intervals connecting the origin to any point
on the light cone itself are light-like.
consider a passenger sitting on a train. Event #1 is the passenger yawning at time t1 . Event
#2 is the passenger yawning again at some later time t2 . To an observer sitting in the train
station, the two events take place at different locations, but in the frame of the passenger,
they occur at the same location.
p
When s212 < 0, the interval is said to be space-like. Note that s12 = s212 ∈ iR is pure
imaginary, so one says that imaginary intervals are spacelike. As an example, at this
moment, in the frame of the reader, the North and South poles of the earth are separated
by a space-like interval. If the interval between two events is space-like, a reference frame
can always be found in which the events are simultaneous.
An interval with s12 = 0 is said to be light-like.
This leads to the concept of the light cone, depicted in fig. 15.4. Consider an event E. In the
frame of an inertial observer, all events with s2 > 0 and ∆t > 0 are in E’s forward light cone
and are part of his absolute future. Events with s2 > 0 and ∆t < 0 lie in E’s backward light
cone are are part of his absolute past. Events with spacelike separations s2 < 0 are causally
disconnected from E. Two events which are causally disconnected can not possible influence
each other. Uniform rectilinear motion is represented by a line t = x/v with constant slope.
If v < c, this line is contained within E’s light cone. E is potentially influenced by all events
in its backward light cone, i.e. its absolute past. It is impossible to find a frame of reference
which will transform past into future, or spacelike into timelike intervals.
15.2.1 Proper time
Proper time is the time read on a clock traveling with a moving observer. Consider two
observers, one at rest and one in motion. If dt is the differential time elapsed in the rest
frame, then
ds2 = c2 dt2 − dx2 − dy 2 − dz 2 (15.15)
02
= c2 dt , (15.16)
where dt0 is the differential time elapsed on the moving clock. Thus,
r
0 v2
dt = dt 1 − 2 , (15.17)
c
and the time elapsed on the moving observer’s clock is
Zt2 r
v 2 (t)
t02 − t01 = dt 1 − 2 . (15.18)
c
t1
Thus, moving clocks run slower . This is an essential feature which is key to understanding
many important aspects of particle physics. A particle with a brief lifetime can, by moving
at speeds close to c, appear to an observer in our frame to be long-lived. It is customary to
define two dimensionless measures of a particle’s velocity:
v 1
β≡ , γ≡p . (15.19)
c 1 − β2
As v → c, we have β → 1 and γ → ∞.
Suppose we wish to compare the elapsed time on two clocks. We keep one clock at rest in
an inertial frame, while the other executes a closed path in space, returning to its initial
location after some interval of time. When the clocks are compared, the moving clock will
show a smaller elapsed time. This is often stated as the “twin paradox.” The total elapsed
time on a moving clock is given by
Zb
1
τ= ds , (15.20)
c
a
where the integral is taken over the world line of the moving clock. The elapsed time τ
takes on a minimum
value when the path from a to b is a straight line. To see this, one can
express τ x(t) as a functional of the path x(t) and extremize. This results in ẍ = 0.
15.2. INTERVALS 293
15.2.2 Irreverent problem from Spring 2002 final exam
Flowers for Algernon – Bob’s beloved hamster, Algernon, is very ill. He has only three hours
to live. The veterinarian tells Bob that Algernon can be saved only through a gallbadder
transplant. A suitable donor gallbladder is available from a hamster recently pronounced
brain dead after a blender accident in New York (miraculously, the gallbladder was un-
scathed), but it will take Life Flight five hours to bring the precious rodent organ to San
Diego.
Bob embarks on a bold plan to save Algernon’s life. He places him in a cage, ties the cage to
the end of a strong meter-long rope, and whirls the cage above his head while the Life Flight
team is en route. Bob reasons that if he can make time pass more slowly for Algernon, the
gallbladder will arrive in time to save his life.
(a) At how many revolutions per second must Bob rotate the cage in order that the gall-
bladder arrive in time for the life-saving surgery? What is Algernon’s speed v0 ?
Solution : We have β(t) = ω0 R/c is constant, therefore, from eqn. 15.18,
∆t = γ ∆t0 . (15.21)
p
Setting ∆t0 = 3 hr and ∆t = 5 hr, we have γ = 53 , which entails β = 1 − γ −2 = 45 . Thus,
v0 = 45 c, which requires a rotation frequency of ω0 /2π = 38.2 MHz.
(b) Bob finds that he cannot keep up the pace! Assume Algernon’s speed is given by
r
t
v(t) = v0 1− (15.22)
T
where v0 is the speed from part (a), and T = 5 h. As the plane lands at the pet hospital’s
emergency runway, Bob peers into the cage to discover that Algernon is dead! In order to
fill out his death report, the veterinarian needs to know: when did Algernon die? Assuming
he died after his own hamster watch registered three hours, derive an expression for the
elapsed time on the veterinarian’s clock at the moment of Algernon’s death.
1/2
Solution : hSnifflei. We have β(t) = 45 1 − Tt . We set
ZT ∗ p
0
T = dt 1 − β 2 (t) (15.23)
0
where T 0 = 3 hr and T ∗ is the time of death in Bob’s frame. We write β0 = 4

5 and
γ0 = (1 − β02 )−1/2 = 53 . Note that T 0 /T = 1 − β02 = γ0−1 .
p
Rescaling by writing ζ = t/T , we have

TZ∗ /T
T0
q
= γ0−1 = dζ 1 − β02 + β02 ζ
T
0
" 3/2 #
2 T∗
= 1− β02 + β02 − (1 − β02 )3/2
3β02 T
" 3/2 #
2 1 T∗
= · 1+ (γ02 − 1) −1 . (15.24)
3γ0 γ02 − 1 T
Solving for T ∗ /T we have

2/3
3 1
T∗ 2 γ02 − 2 −1
= . (15.25)
T γ02 − 1
5
With γ0 = 3 we obtain
T∗ 9
h
11 2/3
i
= 16 3 − 1 = 0.77502 . . . (15.26)
T
Thus, T ∗ = 3.875 hr = 3 hr 52 min 50.5 sec after Bob starts swinging.
(c) Identify at least three practical problems with Bob’s scheme.
Solution : As you can imagine, student responses to this part were varied and generally
sarcastic. E.g. “the atmosphere would ignite,” or “Bob’s arm would fall off,” or “Algernon’s
remains would be found on the inside of the far wall of the cage, squashed flatter than a
coat of semi-gloss paint,” etc.
15.3 Four-Vectors and Lorentz Transformations
We have spoken thus far about different reference frames. So how precisely do the coordi-
nates (t, x, y, z) transform between frames K and K 0 ? In classical mechanics, we have t = t0
and x = x0 + u t, according to fig. 15.3. This yields the Galilean transformation,
    
t 1 0 0 0 t0
x ux 1 0 0 x0 
 
 =   . (15.27)
 y   uy 0 1 0  y 0 
z uz 0 0 1 z0
Such a transformation does not leave intervals invariant.
Let us define the four-vector xµ as

 
ct
µ
 x ct
x = ≡
  . (15.28)
y x
z
15.3. FOUR-VECTORS AND LORENTZ TRANSFORMATIONS 295
Thus, x0 = ct, x1 = x, x2 = y, and x3 = z. In order for intervals to be invariant, the

transformation between xµ in frame K and x0 µ in frame K 0 must be linear:
ν
xµ = Lµν x0 , (15.29)
where we are using the Einstein convention of summing over repeated indices. We define
the Minkowski metric tensor gµν as follows:
 
1 0 0 0
0 −1 0 0
gµν = g µν = 0 0 −1 0  .
 (15.30)
0 0 0 −1
Clearly g = g t is a symmetric matrix.
Note that the matrix Lαβ has one raised index and one lowered index. For the notation we
are about to develop, it is very important to distinguish raised from lowered indices. To
raise or lower an index, we use the metric tensor. For example,
 
ct
−x
xµ = gµν xν = 
 
−y  . (15.31)

−z
The act of summing over an identical raised and lowered index is called index contraction.
Note that  
1 0 0 0
0 1 0 0
g µν = g µρ gρν = δ µν = 
0 0 1 0 .
 (15.32)
0 0 0 1
Now let’s investigate the invariance of the interval. We must have x0 µ x0µ = xµ xµ . Note
that
α
xµ xµ = Lµα x0 Lµβ x0β
α β
= Lµα gµν Lνβ x0 x0 , (15.33)
from which we conclude

Lµα gµν Lνβ = gαβ . (15.34)
This result also may be written in other ways:
Lµα gµν Lνβ = g αβ , Ltαµ gµν Lνβ = gαβ (15.35)
Another way to write this equation is Lt g L = g. A rank-4 matrix which satisfies this
constraint, with g = diag(+, −, −, −) is an element of the group O(3, 1), known as the
Lorentz group.
Let us now count the freedoms in L. As a 4 × 4 real matrix, it contains 16 elements. The
matrix Lt g L is a symmetric 4 × 4 matrix, which contains 10 independent elements: 4 along
the diagonal and 6 above the diagonal. Thus, there are 10 constraints on 16 elements of L,
and we conclude that the group O(3, 1) is 6-dimensional. This is also the dimension of the
four-dimensional orthogonal group O(4), by the way. Three of these six parameters may
be taken to be the Euler angles. That is, the group O(3) constitutes a three-dimensional
subgroup of the Lorentz group O(3, 1), with elements
 
1 0 0 0
0 R
11 R12 R13 

Lµν =   , (15.36)

0 R21 R22 R23 
0 R31 R32 R33
where Rt R = 1, i.e. R ∈ O(3) is a rank-3 orthogonal matrix, parameterized by the three

Euler angles (φ, θ, ψ). The remaining three parameters form a vector β = (βx , βy , βz ) and
define a second class of Lorentz transformations, called boosts:2
 
γ γ βx γ βy γ βz
γ β 1 + (γ − 1) β̂x β̂x (γ − 1) β̂x β̂y (γ − 1) β̂x β̂z 
 
Lµν =  x  , (15.37)
γ βy (γ − 1) β̂x β̂y 1 + (γ − 1) β̂y β̂y (γ − 1) β̂y β̂z 
γ βz (γ − 1) β̂x β̂z (γ − 1) β̂y β̂z 1 + (γ − 1) β̂z β̂z
where
β −1/2
βb = , γ = 1 − β2 . (15.38)
|β|
IMPORTANT : Since the components of β are not the spatial components of a four
vector, we will only write these components with a lowered index, as βi , with i = 1, 2, 3. We
will not write β i with a raised index, but if we did, we’d mean the same thing, i.e. β i = βi .
Note that for the spatial components of a 4-vector like xµ , we have xi = −xi .
Let’s look at a simple example, where βx = β and βy = βz = 0. Then

 
γ γβ 0 0
γ β γ 0 0
Lµν =
 0
 . (15.39)
0 1 0
0 0 0 1
The effect of this Lorentz transformation xµ = Lµν x0 ν is thus
ct = γct0 + γβx0 (15.40)

0 0
x = γβct + γx . (15.41)
How fast is the origin of K 0 moving in the K frame? We have dx0 = 0 and thus
1 dx γβ c dt0
= =β . (15.42)
c dt γ c dt0
2
Unlike rotations, the boosts do not themselves define a subgroup of O(3, 1).
Thus, u = βc, i.e. β = u/c.
It is convenient to take advantage of the fact that Pβij ≡ β̂i β̂j is a projection operator , which
2
satisfies Pβ = Pβ . The action of Pβij on any vector ξ is to project that vector onto the β̂
direction:
Pβ ξ = (β̂ · ξ) β̂ . (15.43)
We may now write the general Lorentz boost, with β = u/c, as
γβ t

γ
L= , (15.44)
γβ I + (γ − 1) Pβ
where I is the 3 × 3 unit matrix, and where we write column and row vectors
 
βx
β = βy  , β t = βx βy βz (15.45)
 
βz
as a mnemonic to help with matrix multiplications. We now have

0
γβ t γct0 + γβ · x0

ct γ ct
= = . (15.46)
x γβ I + (γ − 1) Pβ x0 γβct0 + x0 + (γ − 1) Pβ x0
Thus,
ct = γct0 + γβ·x0 (15.47)

0 0 0
x = γβct + x + (γ − 1) (β̂·x ) β̂ . (15.48)
If we resolve x and x0 into components parallel and perpendicular to β, writing
xk = β̂·x , x⊥ = x − (β̂·x) β̂ , (15.49)
with corresponding definitions for x0k and x0⊥ , the general Lorentz boost may be written as
ct = γct0 + γβx0k (15.50)

xk = γβct0 + γx0k (15.51)
x⊥ = x0⊥ . (15.52)
Thus, the components of x and x0 which are parallel to β enter into a one-dimensional
Lorentz boost along with t and t0 , as described by eqn. 15.41. The components of x and
x0 which are perpendicular to β are unaffected by the boost.
Finally, the Lorentz group O(3, 1) is a group under multiplication, which means that if La
and Lb are elements, then so is the product La Lb . Explicitly, we have
(La Lb )t g La Lb = Ltb (Lta g La ) Lb = Ltb g Lb = g . (15.53)

15.3.1 Covariance and contravariance
Note that
   
γ γβ 0 0 1 0 0 0 γ γβ 0 0
γ β γ 0 0 0 −1 0 0 γ β γ 0 0
Ltαµ gµν Lνβ =   
 0 0 1 0 0 0 −1 0   0 0 1 0
0 0 0 1 0 0 0 −1 0 0 0 1
 
1 0 0 0
0 −1 0 0
=
0 0 −1
=g ,
αβ (15.54)
0
0 0 0 −1
since γ 2 (1 − β 2 ) = 1. This is in fact the general way that tensors transform under a Lorentz
transformation:
ν
covariant vectors : xµ = Lµν x0 (15.55)
αβ αβ
covariant tensors : F µν = Lµα Lνβ F 0 = Lµα F 0 Ltβν (15.56)
Note how index contractions always involve one raised index and one lowered index. Raised
indices are called contravariant indices and lowered indiced are called covariant indices.
The transformation rules for contravariant vectors and tensors are
contravariant vectors : xµ = Lµν x0ν (15.57)

contravariant tensors : Fµν = Lµα Lνβ F 0 αβ = Lµα F 0 αβ Ltβν (15.58)
A Lorentz scalar has no indices at all. For example,
ds2 = gµν dxµ dxν , (15.59)
is a Lorentz scalar. In this case, we have contracted a tensor with two four-vectors. The
dot product of two four-vectors is also a Lorentz scalar:
a · b ≡ aµ bµ = gµν aµ bν
= a0 b0 − a1 b1 − a2 b2 − a3 b3
= a0 b0 − a · b . (15.60)
Note that the dot product a · b of four-vectors is invariant under a simultaneous Lorentz
transformation of both aµ and bµ , i.e. a · b = a0 · b0 . Indeed, this invariance is the very
definition of what it means for something to be a Lorentz scalar. Derivatives with respect
to covariant vectors yield contravariant vectors:
∂f ∂Aµ ∂Aµν
≡ ∂µ f , = ∂ν Aµ ≡ B µν , = ∂λ B µν ≡ C µνλ
∂xµ ∂xν ∂xλ
et cetera. Note that differentiation with respect to the covariant vector xµ is expressed by
the contravariant differential operator ∂µ :

∂ 1∂ ∂ ∂ ∂
≡ ∂µ = , , , (15.61)
∂xµ c ∂t ∂x ∂y ∂z

∂ µ 1∂ ∂ ∂ ∂
≡∂ = ,− ,− ,− . (15.62)
∂xµ c ∂t ∂x ∂y ∂z
The contraction ≡ ∂ µ ∂µ is a Lorentz scalar differential operator, called the D’Alembertian:
1 ∂2 ∂2 ∂2 ∂2
= − − − . (15.63)
c2 ∂t2 ∂x2 ∂y 2 ∂z 2
The Helmholtz equation for scalar waves propagating with speed c can thus be written in
compact form as φ = 0.
15.3.2 What to do if you hate raised and lowered indices
Admittedly, this covariant and contravariant business takes some getting used to. Ulti-
mately, it helps to keep straight which indices transform according to L (covariantly) and
which transform according to Lt (contravariantly). If you find all this irksome, the raising
and lowering can be safely ignored. We define the position four-vector as before, but with no
difference between raised and lowered indices. In fact, we can just represent all vectors and
tensors with lowered indices exclusively, writing e.g. xµ = (ct, x, y, z). The metric tensor is
g = diag(+, −, −, −) as before. The dot product of two four-vectors is
x · y = gµν xµ yν . (15.64)
The Lorentz transformation is

xµ = Lµν x0ν . (15.65)
Since this preserves intervals, we must have
gµν xµ yν = gµν Lµα x0α Lνβ yβ0

= Ltαµ gµν Lνβ x0α yβ0 ,

(15.66)
which entails
Ltαµ gµν Lνβ = gαβ . (15.67)
In terms of the quantity Lµν defined above, we have Lµν = Lµν . In this convention, we could
completely avoid raised indices, or we could simply make no distinction, taking xµ = xµ
and Lµν = Lµν = Lµν , etc.
15.3.3 Comparing frames
Suppose in the K frame we have a measuring rod which is at rest. What is its length as
measured in the K 0 frame? Recall K 0 moves with velocity u = u x̂ with respect to K. From
the Lorentz transformation in eqn. 15.41, we have
x1 = γ(x01 + βc t01 ) (15.68)

x2 = γ(x02 + βc t02 ) , (15.69)
where x1,2 are the positions of the ends of the rod in frame K. The rod’s length in any
frame is the instantaneous spatial separation of its ends. Thus, we set t01 = t02 and compute
the separation ∆x0 = x02 − x01 :
1/2
∆x = γ ∆x0 =⇒ ∆x0 = γ −1 ∆x = 1 − β 2 ∆x . (15.70)
The proper length `0 of a rod is its instantaneous end-to-end separation in its rest frame.
We see that
1/2
`(β) = 1 − β 2 `0 , (15.71)
so the length is always greatest in the rest frame. This is an example of a Lorentz-Fitzgerald
contraction. Note that the transverse dimensions do not contract:
∆y 0 = ∆y , ∆z 0 = ∆z . (15.72)
Thus, the volume contraction of a bulk object is given by its length contraction: V 0 = γ −1 V.
A striking example of relativistic issues of length, time, and simultaneity is the famous
‘pole and the barn’ paradox, described in the Appendix (section ). Here we illustrate some
essential features via two examples.
15.3.4 Example I
Next, let’s analyze the situation depicted in fig. 15.3. In the K 0 frame, we’ll denote the
following spacetime points:
ct0 ct0 ct0 ct0

0 0 0 0
A = , B = , S− = , S− = . (15.73)
−d +d −ct0 +ct0
Note that the origin in K 0 is given by O0 = (ct0 , 0). Here we are setting y = y 0 = z = z 0 = 0
and dealing only with one spatial dimension. The points S± 0 denote the left-moving (S 0 )
−
and right-moving (S+ 0 ) wavefronts. We now use the Lorentz transformation

γ γβ
Lµν = (15.74)
γβ γ
to transform to the K frame. Thus,

0 0
0 1 β ct ct
S− = LS− = γ 0 = γ (1 − β) (15.75)
β 1 −ct −ct0
0 0
0 1 β ct ct
S+ = LS+ =γ = γ (1 + β) . (15.76)
β 1 +ct0 +ct0
We also have
0 0
0 1 β ct ct − βd
A = LA = γ =γ (15.77)
β 1 −d βct0 − d
0 0
0 1 β ct ct + βd
B = LB = γ =γ . (15.78)
β 1 +d βct0 + d
The signal arrives at A in the K frame when A = S− . The solution is

∗
ctA γ (1 − β) d
S− = A = = . (15.79)
x∗A −(1 − β) d
For the signal to arrive at B, we set
ct∗B

γ (1 + β) d
S+ = B = = . (15.80)
x∗B (1 + β) d
Thus, t∗A = γ(1 − β)d/c and t∗B = γ(1 + β)d/c. Thus, the two events are not simultaneous
in K. The arrival at A is first.
15.3.5 Example II
Consider a rod of length `0 extending from the origin to the point `0 x̂ at rest in frame K.
In the frame K, the two ends of the rod are located at spacetime coordinates

ct ct
A= and B = , (15.81)
0 `0
respectively. Now consider the origin in frame K 0 . Its spacetime coordinates are
0
0 ct
C = . (15.82)
0
To an observer in the K frame, we have
0
γct0

γ γβ ct
C= = . (15.83)
γβ γ 0 γβct0
Now consider two events. The first event is the coincidence of A with C, i.e. the origin of
K 0 instantaneously coincides with the origin of K. Setting A = C we obtain t = t0 = 0.
The second event is the coincidence of B with C. Setting B = C we obtain t = l0 /βc and
t0 = `0 /γβc. Note that t = `(β)/βc, i.e. due to the Lorentz-Fitzgerald contraction of the
rod as seen in the K 0 frame, where `(β) = `0 /γ.
Figure 15.5: A rectangular plate moving at velocity V = V x̂.
15.3.6 Deformation of a rectangular plate
Problem: A rectangular plate of dimensions a × b moves at relativistic velocity V = V x̂

as shown in fig. 15.5. In the rest frame of the rectangle, the a side makes an angle θ with
respect to the x̂ axis. Describe in detail and sketch the shape of the plate as measured by
an observer in the laboratory frame. Indicate the lengths of all sides andqthe values of all
2
interior angles. Evaluate your expressions for the case θ = 1/4π and V = 3 c.
Solution: An observer in the laboratory frame will measure lengths parallel to x̂ to be

Lorentz contracted by a factor γ −1 , where γ = (1 − β 2 )−1/2 and β = V /c. Lengths perpen-
dicular to x̂ remain unaffected. Thus, we have the situation depicted in fig. 15.6. Simple
trigonometry then says
tan φ = γ tan θ , tan φ̃ = γ −1 tan θ ,
as well as
q p
0
a = a γ −2 cos2 θ + sin2 θ = a 1 − β 2 cos2 θ
q q
b0 = b γ −2 sin2 θ + cos2 θ = b 1 − β 2 sin2 θ .
The plate deforms to a parallelogram, with internal angles
χ = 21 π + tan−1 (γ tan θ) − tan−1 (γ −1 tan θ)

χ̃ = 12 π − tan−1 (γ tan θ) + tan−1 (γ −1 tan θ) .
Note that the area of the plate as measured in the laboratory frame is
Ω 0 = a0 b0 sin χ = a0 b0 cos(φ − φ̃)

= γ −1 Ω ,
Figure 15.6: Relativistic deformation of the rectangular plate.
where Ω = ab is the proper area. The area contraction factor is γ −1 and not γ −2 (or γ −3
in a three-dimensional system) because only the parallel dimension gets contracted.
q
2
√
Setting V = 3 c gives γ = 3, and with θ = 1/4π we have φ = 13 π and φ̃ = 16 π. The
q q
interior angles are then χ = 32 π and χ̃ = 13 π. The side lengths are a0 = 23 a and b0 = 23 b.
15.3.7 Transformation of velocities
Let K 0 move at velocity u = cβ relative to K. The transformation from K 0 to K is given

by the Lorentz boost,
 
γ γ βx γ βy γ βz
γ β 1 + (γ − 1) β̂x β̂x (γ − 1) β̂x β̂y (γ − 1) β̂x β̂z 
 
Lµν =  x  . (15.84)
γ βy (γ − 1) β̂x β̂y 1 + (γ − 1) β̂y β̂y (γ − 1) β̂y β̂z 
γ βz (γ − 1) β̂x β̂z (γ − 1) β̂y β̂z 1 + (γ − 1) β̂z β̂z
Applying this, we have

dxµ = Lµν dx0 ν . (15.85)
This yields
dx0 = γ dx0 0 + γ β · dx0 (15.86)

00 0 0
dx = γ β dx + dx + (γ − 1) β̂ β̂·dx . (15.87)
We then have
dx c γ β dx0 0 + c dx0 + c (γ − 1) β̂ β̂·dx0
V =c =
dx0 γ dx0 0 + γ β·dx0
u + γ −1 V 0 + (1 − γ −1 ) û û·V 0
= . (15.88)
1 + u·V 0 /c2
The second line is obtained by dividing both numerator and denominator by dx0 0 , and then
writing V 0 = dx0 /dx0 0 . There are two special limiting cases:
(u + V 0 ) û
velocities parallel û· V̂ 0 = 1) =⇒ V = (15.89)
1 + u V 0 /c2
velocities perpendicular û· V̂ 0 = 0) =⇒ V = u + γ −1 V 0 . (15.90)
Note that if either u or V 0 is equal to c, the resultant expression has |V | = c as well. One
can’t boost the speed of light!
Let’s revisit briefly the example in section 15.3.4. For an observer, in the K frame, the
relative velocity of S and A is c − u, because even though we must boost the velocity
−c x̂ of the left-moving light wave by u x̂, the result is still −c x̂, according to our velocity
addition formula. Thus, the relative speed of A and S is c − u, which means that
d(β) d 1 d 1−β d
t∗A = = · = · = γ (1 − β) , (15.91)
c+u γ c+u γc 1 − β 2 c
since d(β) = γ −1 d. This result is exactly as found in section 15.3.4 by other means. A
corresponding analysis yields t∗B = γ (1 + β) d/c. again in agreement with the earlier result.
Here, it is crucial to account for the Lorentz contraction of the distance between the source
S and the observers A and B as measured in the K frame.
15.3.8 Four-velocity and four-acceleration
In nonrelativistic mechanics, the velocity V = dx

dt is locally tangent to a particle’s trajectory.
In relativistic mechanics, one defines the four-velocity,
dxα dxα

γ
uα ≡ =p = , (15.92)
ds 2
1 − β c dt γβ
which is locally tangent to the world line of a particle. Note that
gαβ uα uβ = 1 . (15.93)
The four-acceleration is defined as
duν d2xν
wν ≡ = . (15.94)
ds ds2
Note that u · w = 0, so the 4-velocity and 4-acceleration are orthogonal with respect to the
Minkowski metric.
15.4. THREE KINDS OF RELATIVISTIC ROCKETS 305
15.4 Three Kinds of Relativistic Rockets
15.4.1 Constant acceleration model
Consider a rocket which undergoes constant acceleration along x̂. Clearly the rocket has
no rest frame per se, because its velocity is changing. However, this poses no serious
obstacle to discussing its relativistic motion. We consider a frame K 0 in which the rocket
is instantaneously at rest. In such a frame, the rocket’s 4-acceleration is w0 α = (0, a/c2 ),
where we suppress the transverse coordinates y and z. In an inertial frame K, we have

d
α γ γ γ̇
w = = . (15.95)
ds γβ c γ β̇ + γ̇β
Transforming w0 α into the K frame, we have
γβa/c2

α γ γβ 0
w = = . (15.96)
γβ γ a/c2 γa/c2
Taking the upper component, we obtain the equation

!
βa d β a
γ̇ = =⇒ p = , (15.97)
c dt 1− β2 c
the solution of which, with β(0) = 0, is

s 2
at at
β(t) = √ , γ(t) = 1+ . (15.98)
c + a2 t2
2 c
The proper time for an observer moving with the rocket is thus
Zt
c dt1 c at
τ= p = sinh−1 .
c2 + a2 t21 a c
0
For large times t c/a, the proper time grows logarithmically in t, which is parametrically
slower. To find the position of the rocket, we integrate ẋ = cβ, and obtain, with x(0) = 0,
Zt
a ct dt c p 2
x(t) = p 1 1 = c + a2 t2 − c . (15.99)
c2 + a2 t21 a
0
It is interesting to consider the situation in the frame K 0 . We then have
β(τ ) = tanh(aτ /c) , γ(τ ) = cosh(aτ /c) . (15.100)

For an observer in the frame K 0 , the distance he has traveled is ∆x0 (τ ) = ∆x(τ )/γ(τ ), as
we found in eqn. 15.70. Now x(τ ) = (c2 /a) cosh(aτ /c) − 1 , hence

c2
∆x0 (τ ) = 1 − sech(aτ /c) . (15.101)
a
For τ c/a, we expand sech(aτ /c) ≈ 1 − 12 (aτ /c)2 and find x0 (τ ) = 21 aτ 2 , which clearly is
the nonrelativistic limit. For τ → ∞, however, we have ∆x0 (τ ) → c2 /a is finite! Thus, while
the entire Universe is falling behind the accelerating observer, it all piles up at a horizon a
distance c2 /a behind it, in the frame of the observer. The light from these receding objects
is increasingly red-shifted (see section 15.6 below), until it is no longer visible. Thus, as
John Baez describes it, the horizon is “a dark plane that appears to be swallowing the
entire Universe!” In the frame of the inertial observer, however, nothing strange appears to
be happening at all!
15.4.2 Constant force with decreasing mass
Suppose instead the rocket is subjected to a constant force F0 in its instantaneous rest
frame, and furthermore that the rocket’s mass satisfies m(τ ) = m0 (1 − ατ ), where τ is the
proper time for an observer moving with the rocket. Then from eqn. 15.97, we have
F0 d(γβ) d(γβ)
= = γ −1
m0 (1 − ατ ) dt dτ

1 dβ d 1 1+β
= = 2 ln , (15.102)
1 − β 2 dτ dτ 1−β
after using the chain rule, and with dτ /dt = γ −1 . Integrating, we find
1 − (1 − ατ )r

1+β 2F0
ln = ln 1 − ατ =⇒ β(τ ) = , (15.103)
1−β αm0 c 1 + (1 − ατ )r
with r = 2F0 /αm0 c. As τ → α−1 , the rocket loses all its mass, and it asymptotically
approaches the speed of light.
It is convenient to write " #

r 1
β(τ ) = tanh ln , (15.104)
2 1 − ατ
in which case
" #
dt r 1
γ= = cosh ln (15.105)
dτ 2 1 − ατ
" #
1 dx r 1
= sinh ln . (15.106)
c dτ 2 1 − ατ
15.4. THREE KINDS OF RELATIVISTIC ROCKETS 307
Integrating the first of these from τ = 0 to τ = α−1 , we find t∗ ≡ t τ = α−1 is

F0 2 −1
h i
2 F0
Z1  α − α if α >

 mc mc
1
t∗ = dσ σ −r/2 + σ r/2 = (15.107)
2α 
F0

0
∞ if α ≤

mc .
Since β(τ = α−1 ) = 1, this is the time in the K frame when the rocket reaches the speed of
light.
15.4.3 Constant ejecta velocity
Our third relativistic rocket model is a generalization of what is commonly known as the
rocket equation in classical physics. The model is one of a rocket which is continually
ejecting burnt fuel at a velocity −u in the instantaneous rest frame of the rocket. The
nonrelativistic rocket equation follows from overall momentum conservation:
dprocket + dpfuel = d(mv) + (v − u) (−dm) = 0 , (15.108)
since if dm < 0 is the differential change in rocket mass, the differential ejecta mass is −dm.
This immediately gives

m0
m dv + u dm = 0 =⇒ v = u ln , (15.109)
m
where the rocket is assumed to begin at rest, and where m0 is the initial mass of the rocket.
Note that as m → 0 the rocket’s speed increases without bound, which of course violates
special relativity.
In relativistic mechanics, as we shall see in section 15.5, the rocket’s momentum, as described
by an inertial observer, is p = γmv, and its energy is γmc2 . We now write two equations
for overall conservation of momentum and energy:
d(γmv) + γe ve dme = 0 (15.110)

d(γmc2 ) + γe (dme c2 ) = 0 , (15.111)
where ve is the velocity of the ejecta in the inertial frame, dme is the differential mass of
2 −1/2
the ejecta, and γe = 1 − vc2e . From the second of these equations, we have
γe dme = −d(γm) , (15.112)
which we can plug into the first equation to obtain
(v − ve ) d(γm) + γm dv = 0 . (15.113)
Before solving this, we remark that eqn. 15.112 implies that dme < |dm| – the differential
mass of the ejecta is less than the mass lost by the rocket! This is Einstein’s famous equation
E = mc2 at work – more on this later.
To proceed, we need to use the parallel velocity addition formula of eqn. 15.89 to find ve :
2
v−u u 1 − vc2
ve = =⇒ v − ve = . (15.114)
1 − uv
c2 1 − uv
c 2
We now define βu = u/c, in which case eqn, 15.113 becomes
βu (1 − β 2 ) d(γm) + (1 − ββu ) γm dβ = 0 . (15.115)
Using dγ = γ 3 β dβ, we observe a felicitous cancellation of terms, leaving

dm dβ
βu + =0. (15.116)
m 1 − β2

m
β = tanh βu ln 0 . (15.117)
m
Note that this agrees with the result of eqn. 15.104, if we take βu = F0 /αmc.
15.5 Relativistic Mechanics
Relativistic particle dynamics follows from an appropriately extended version of Hamilton’s

principle δS = 0. The action S must be a Lorentz scalar. The action for a free particle is
Zb Ztb r
v2
S x(t) = −mc ds = −mc2 dt 1 − 2 .

(15.118)
c
a ta
Thus, the free particle Lagrangian is

r 2 2
2 v2 2 1 2 1 2 v
L = −mc 1 − 2 = −mc + 2 mv + 8 mc + ... . (15.119)
c c2
Thus, L can be written as an expansion in powers of v 2 /c2 . Note that L(v = 0) = −mc2 .
We interpret this as −U0 , where U0 = mc2 is the rest energy of the particle. As a constant,
it has no consequence for the equations of motion. The next term in L is the familiar
nonrelativistic kinetic energy, 21 mv 2 . Higher order terms are smaller by increasing factors
of β 2 = v 2 /c2 .
We can add a potential U (x, t) to obtain

r
ẋ2
L(x, ẋ, t) = −mc2 1− − U (x, t) . (15.120)
c2
15.5. RELATIVISTIC MECHANICS 309
The momentum of the particle is
∂L
p= = γmẋ . (15.121)
∂ ẋ
The force is F = −∇U as usual, and Newton’s Second Law still reads ṗ = F . Note that

v v̇
ṗ = γm v̇ + 2 γ 2 v . (15.122)
c
Thus, the force F is not necessarily in the direction of the acceleration a = v̇. The
Hamiltonian, recall, is a function of coordinates and momenta, and is given by
p
H = p · ẋ − L = m2 c4 + p2 c2 + U (x, t) . (15.123)
Since ∂L/∂t = 0 for our case, H is conserved by the motion of the particle. There are two
limits of note:
p2
|p| mc (non-relativistic) : H = mc2 + + U + O(p4 /m4 c4 ) (15.124)
2m
|p| mc (ultra-relativistic) : H = c|p| + U + O(mc/p) . (15.125)
Expressed in terms of the coordinates and velocities, we have H = E, the total energy, with
E = γmc2 + U . (15.126)
In particle physics applications, one often defines the kinetic energy T as
T = E − U − mc2 = (γ − 1)mc2 . (15.127)
When electromagnetic fields are included,

r
2 ẋ2 q
L(x, ẋ, t) = −mc 1−− q φ + A · ẋ
c2 c
q dx µ
= −γmc2 − Aµ , (15.128)
c dt
where the electromagnetic 4-potential is Aµ = (φ , A). Recall Aµ = gµν Aν has the sign of
its spatial components reversed. One the has
∂L q
p= = γmẋ + A , (15.129)
∂ ẋ c
and the Hamiltonian is r q 2

H= m2 c4 + p − A + q φ . (15.130)
c
15.5.1 Relativistic harmonic oscillator
From E = γmc2 + U , we have

" 2 #
mc2

2 2
ẋ = c 1− . (15.131)
E − U (x)
1 2
Consider the one-dimensional harmonic oscillator potential U (x) = 2 kx . We define the
turning points as x = ±b, satisfying
E − mc2 = U (±b) = 12 kb2 . (15.132)
Now define the angle θ via x ≡ b cos θ, and further define the dimensionless parameter
= kb2 /4mc2 . Then, after some manipulations, one obtains
p
1 + sin2 θ
θ̇ = ω0 , (15.133)
1 + 2 sin2 θ
p
with ω0 = k/m as in the nonrelativistic case. Hence, the problem is reduced to quadra-
tures (a quaint way of saying ‘doing an an integral’):
Zθ
1 + 2 sin2 ϑ
t(θ) − t0 = ω0−1 dϑ p . (15.134)
1 + sin2 ϑ
θ0
While the result can be expressed in terms of elliptic integrals, such an expression is not
particularly illuminating. Here we will content ourselves with computing the period T ():
π
Z2
4 1 + 2 sin2 ϑ
T () = dϑ p (15.135)
ω0 1 + sin2 ϑ
0
π
Z2
4
= dϑ 1 + 32 sin2 ϑ − 58 2 sin4 ϑ + . . .
ω0
0
2π n 2
o
= · 1 + 34 − 15
64 + . . . . (15.136)
ω0
Thus, for the relativistic harmonic oscillator, the period does depend on the amplitude,
unlike the nonrelativistic case.
15.5.2 Energy-momentum 4-vector
Let’s focus on the case where U (x) = 0. This is in fact a realistic assumption for subatomic
particles, which propagate freely between collision events.
15.5. RELATIVISTIC MECHANICS 311
The differential proper time for a particle is given by
ds
dτ = = γ −1 dt , (15.137)
c
where xµ = (ct, x) are coordinates for the particle in an inertial frame. Thus,
dx E dx0
p = γmẋ = m , = mcγ = m , (15.138)
dτ c dτ
with x0 = ct. Thus, we can write the energy-momentum 4-vector as


E/c
dxµ  px 
pµ = m =
 py  .
 (15.139)
dτ
pz
Note that pν = mcuν , where uν is the 4-velocity of eqn. 15.92. The four-momentum satisfies
the relation
E2
pµ pµ = 2 − p2 = m2 c2 . (15.140)
c
The relativistic generalization of force is
dpµ
fµ =

= γF ·v/c , γF , (15.141)
dτ
where F = dp/dt as usual.
The energy-momentum four-vector transforms covariantly under a Lorentz transformation.

This means
ν
pµ = Lµν p0 . (15.142)
If frame K 0 moves with velocity u = cβ x̂ relative to frame K, then
E c−1 E 0 + β p0x p0x + βc−1 E 0

= p , px = p , py = p0y , pz = p0z . (15.143)
c 1 − β2 1 − β2
In general, from eqns. 15.50, 15.51, and 15.52, we have
E E0
=γ + γβp0k (15.144)
c c
E
pk = γβ + γp0k (15.145)
c
p⊥ = p0⊥ (15.146)
where pk = β̂·p and p⊥ = p − (β̂·p) β̂.

15.5.3 4-momentum for massless particles
For a massless particle, such as a photon, we have pµ pµ = 0, which means E 2 = p2 c2 . The

4-momentum may then be written pµ = |p| , p . We define the 4-wavevector k µ by the
relation pµ = h̄k µ , where h̄ = h/2π and h is Planck’s constant. We also write ω = ck, with
E = h̄ω.
15.6 Relativistic Doppler Effect
The 4-wavevector k µ = ω/c , k for electromagnetic radiation satisfies k µ kµ = 0. The

energy-momentum 4-vector is pµ = h̄k µ . The phase φ(xµ ) = −kµ xµ = k · x − ωt of a plane

wave is a Lorentz scalar. This means that the total number of wave crests (i.e. φ = 2πn)
emitted by a source will be the total number observed by a detector.
Suppose a moving source emits radiation of angular frequency ω 0 in its rest frame. Then
k 0 µ = Lµν (−β) k ν
 
γ −γ βx −γ βy −γ βz

ω/c
−γ βx 1 + (γ − 1) β̂x β̂x (γ − 1) β̂x β̂y (γ − 1) β̂x β̂z   kx 
 
=   .
−γ βy (γ − 1) β̂x β̂y 1 + (γ − 1) β̂y β̂y (γ − 1) β̂y β̂z   k y 
−γ βz (γ − 1) β̂x β̂z (γ − 1) β̂y β̂z 1 + (γ − 1) β̂z β̂z kz
(15.147)
This gives
ω0 ω ω
= γ − γ β · k = γ (1 − β cos θ) , (15.148)
c c c
where θ = cos−1 (β̂ · k̂) is the angle measured in K between β̂ and k̂. Solving for ω, we have
p
1 − β2
ω= ω , (15.149)
1 − β cos θ 0
where ω0 = ω 0 is the angular frequency in the rest frame of the moving source. Thus,
s
1+β
θ = 0 ⇒ source approaching ⇒ ω= ω (15.150)
1−β 0
p
θ = 12 π ⇒ source perpendicular ⇒ ω= 1 − β 2 ω0 (15.151)
s
1−β
θ=π ⇒ source receding ⇒ ω= ω . (15.152)
1+β 0
Recall the non-relativistic Doppler effect:
ω0
ω= . (15.153)
1 − (V /c) cos θ
15.6. RELATIVISTIC DOPPLER EFFECT 313
Figure 15.7: Alice’s big adventure.
We see that approaching sources have their frequencies shifted higher; this is called the
blue shift, since blue light is on the high frequency (short wavelength) end of the optical
spectrum. By the same token, receding sources are red-shifted to lower frequencies.
15.6.1 Romantic example
Alice and Bob have a “May-December” thang going on. Bob is May and Alice December,
if you get my drift. The social stigma is too much to bear! To rectify this, they decide
that Alice should take a ride in a space ship. Alice’s itinerary takes her along a sector of
a circle of radius R and angular span of Θ = 1 radian, as depicted in fig. 15.7. Define
O ≡ (r = 0), P ≡ (r = R, φ = − 12 Θ), and Q ≡ (r = R, φ = 21 Θ). Alice’s speed along the
first leg (straight from O to P) is va = 35 c. Her speed along the second leg (an arc from
P to Q) is vb = 12 4
13 c. The final leg (straight from Q to O) she travels at speed vc = 5 c.
Remember that the length of an circular arc of radius R and angular spread α (radians) is
` = αR.
(a) Alice and Bob synchronize watches at the moment of Alice’s departure. What is the
elapsed time on Bob’s watch when Alice returns? What is the elapsed time on Alice’s
watch? What must R be in order for them to erase their initial 30 year age difference?
Solution : In Bob’s frame, Alice’s trip takes a time
R RΘ R
∆t = + +
cβa cβb cβc
R 5 13 5 4R
= + + = . (15.154)
c 3 12 4 c
The elapsed time on Alice’s watch is

R RΘ R
∆t0 = + +
cγa βa cγb βb cγc βc
R 5 4 13 5 5 3
5R
= · + · + · = . (15.155)
c 3 5 12 13 4 5 2c
Thus, ∆T = ∆t − ∆t0 = 3R/2c and setting ∆T = 30 yr, we find R = 20 ly. So Bob will
have aged 80 years and Alice 50 years upon her return. (Maybe this isn’t such a good plan
after all.)
(b) As a signal of her undying love for Bob, Alice continually shines a beacon throughout
her trip. The beacon produces monochromatic light at wavelength λ0 = 6000 Å (frequency
f0 = c/λ0 = 5 × 1014 Hz). Every night, Bob peers into the sky (with a radiotelescope),
hopefully looking for Alice’s signal. What frequencies fa , fb , and fc does Bob see?
Solution : Using the relativistic Doppler formula, we have

s
1 − βa
fa = × f0 = 12 f0
1 + βa
q
5
fb = 1 − βb2 × f0 = 13 f0
s
1 + βc
fc = × f0 = 3f0 . (15.156)
1 − βc
(c) Show that the total number of wave crests counted by Bob is the same as the number
emitted by Alice, over the entire trip.
Solution : Consider first the O–P leg of Alice’s trip. The proper time elapsed on Alice’s
watch during this leg is ∆t0a = R/cγa βa , hence she emits Na0 = Rf0 /cγa βa wavefronts during
this leg. Similar considerations hold for the P–Q and Q–O legs, so Nb0 = RΘf0 /cγb βb and
Nc0 = Rf0 /cγc βc .
Although the duration of the O–P segment of Alice’s trip takes a time ∆ta = R/cβa in Bob’s
frame, he keeps receiving the signal at the Doppler-shifted frequency fa until the wavefront
emitted when Alice arrives at P makes its way back to Bob. That takes an extra time R/c,
hence the number of crests emitted for Alice’s O–P leg is
s
R R 1 − βa Rf0
Na = + × f0 = = Na0 , (15.157)
cβa c 1 + βa cγa βa
since the source is receding from the observer.
During the P–Q leg, we have θ = 12 π, and Alice’s velocity is orthogonal to the wavevector k,
which is directed radially inward. Bob’s first signal at frequency fb arrives a time R/c after
15.7. RELATIVISTIC KINEMATICS OF PARTICLE COLLISIONS 315
Alice passes P, and his last signal at this frequency arrives a time R/c after Alice passes Q.
Thus, the total time during which Bob receives the signal at the Doppler-shifted frequency
fb is ∆tb = RΘ/c, and
RΘ RΘf0
q
Nb = · 1 − βb2 × f0 = = Nb0 . (15.158)
cβb cγ βb b
Finally, during the Q–O home stretch, Bob first starts to receive the signal at the Doppler-
shifted frequency fc a time R/c after Alice passes Q, and he continues to receive the signal
until the moment Alice rushes into his open and very flabby old arms when she makes it
back to O. Thus, Bob receives the frequency fc signal for a duration ∆tc − R/c, where
∆tc = R/cβc . Thus,
s
R R 1 + βc Rf0
Na = − × f0 = = Nc0 , (15.159)
cβc c 1 − βc cγc βc
since the source is approaching.
Therefore, the number of wavelengths emitted by Alice will be precisely equal to the number
received by Bob – none of the waves gets lost.
15.7 Relativistic Kinematics of Particle Collisions
As should be expected, special relativity is essential toward the understanding of subatomic

particle collisions, where the particles themselves are moving at close to the speed of light. In
our analysis of the kinematics of collisions, we shall find it convenient to adopt the standard
convention on units, where we set c ≡ 1. Energies will typically be given in GeV, where
1 GeV = 109 eV = 1.602 × 10−10 J. Momenta will then be in units of GeV/c, and masses
in units of GeV/c2 . With c ≡ 1, it is then customary to quote masses in energy units. For
example, the mass of the proton in these units is mp = 938 MeV, and mπ− = 140 MeV.
For a particle of mass M , its 4-momentum satisfies Pµ P µ = M 2 (remember c = 1). Consider

now an observer with 4-velocity U µ . The energy of the particle, in the rest frame of the
observer is E = P µ Uµ . For example, if P µ = (M, 0, 0, 0) is its rest frame, and U µ = (γ , γβ),
then E = γM , as we have already seen.
Consider next the emission of a photon of 4-momentum P µ = (h̄ω/c, h̄k) from an object
with 4-velocity V µ , and detected in a frame with 4-velocity U µ . In the frame of the detector,
the photon energy is E = P µ Uµ , while in the frame of the emitter its energy is E 0 = P µ Vµ .
If U µ = (1, 0, 0, 0) and V µ = (γ , γβ), then E = h̄ω and E 0 = h̄ω 0 = γh̄(ω − β · k) =
γh̄ω(1 − β cos θ), where θ = cos−1 β̂ · k̂ . Thus, ω = γ −1 ω 0 /(1 − β cos θ). This recapitulates
our earlier derivation in eqn. 15.148.
Consider next the interaction of several particles. If in a given frame the 4-momenta of the
reactants are Piµ , where n labels the reactant ‘species’, and the 4-momenta of the products
are Qµj , then if the collision is elastic, we have that total 4-momentum is conserved, i.e.
N N̄
Piµ Qµj ,
X X
= (15.160)
i=1 j=1
where there are N reactants and N̄ products. For massive particles, we can write
Piµ = γi mi 1 , vi ) , Qµj = γ̄j m̄j 1 , v̄j ) , (15.161)
while for massless particles,
Piµ = h̄ki 1 , k̂

, ˆ .
Qµj = h̄k̄j 1 , k̄ (15.162)
15.7.1 Spontaneous particle decay into two products
Consider first the decay of a particle of mass M into two particles. We have P µ = Qµ1 + Qµ2 ,
hence in the rest frame of the (sole) reactant, which is also called the ‘center of mass’
(CM) frame since the total 3-momentum vanishes therein, we have M = E1 + E2 . Since
EiCM = γ CM mi , and γi ≥ 1, clearly we must have M > m1 + m2 , or else the decay cannot
possibly conserve energy. To analyze further, write P µ − Qµ1 = Qµ2 . Squaring, we obtain
M 2 + m21 − 2Pµ Qµ1 = m22 . (15.163)
The dot-product P · Q1 is a Lorentz scalar, and hence may be evaluated in any frame.
Let us first consider the CM frame, where P µ = M (1, 0, 0, 0), and Pµ Qµ1 = M E1CM , where
E1CM is the energy of n = 1 product in the rest frame of the reactant. Thus,
M 2 + m21 − m22 M 2 + m22 − m21

E1CM = , E2CM = , (15.164)
2M 2M
where the second result follows merely from switching the product labels. We may now
write Qµ1 = (E1CM , pCM ) and Qµ2 = (E2CM , −pCM ), with
(pCM )2 = (E1CM )2 − m21 = (E2CM )2 − m22

M − m21 − m22 2 m1 m2 2
2
= − . (15.165)
2M M
In the laboratory frame, we have P µ = γM (1 , V ) and Qµi = γi mi (1 , Vi ). Energy and

momentum conservation then provide four equations for the six unknowns V1 and V2 . Thus,
there is a two-parameter family of solutions, assuming we regard the reactant velocity V L as
fixed, corresponding to the freedom to choose p̂CM in the CM frame solution above. Clearly
the three vectors V , V1 , and V2 must lie in the same plane, and with V fixed, only one
additional parameter is required to fix this plane. The other free parameter may be taken
Figure 15.8: Spontaneous decay of a single reactant into two products.
to be the relative angle θ1 = cos−1 V̂ · V̂1 (see fig. 15.8). The angle θ2 as well as the speed

V2 are then completely determined. We can use eqn. 15.163 to relate θ1 and V1 :
M 2 + m21 − m22 = 2M m1 γγ1 1 − V V1 cos θ1 .

(15.166)
It is convenient to express both γ1 and V1 in terms of the energy E1 :

s
E1 m2
q
γ1 = , V1 = 1 − γ1−2 = 1 − 21 . (15.167)
m1 E1
This results in a quadratic equation for E1 , which may be expressed as

p
(1 − V 2 cos2 θ1 )E12 − 2 1 − V 2 E1CM E1 + (1 − V 2 )(E1CM )2 + m21 V 2 cos2 θ1 = 0 , (15.168)
the solutions of which are

√ q
1 − V 2 E1CM ± V cos θ1 (1 − V 2 )(E1CM )2 − (1 − V 2 cos2 θ1 )m21
E1 = . (15.169)
1 − V 2 cos2 θ1
The discriminant is positive provided
CM 2
E1 1 − V 2 cos2 θ1
> , (15.170)
m1 1−V2
which means
V −2 − 1
sin2 θ1 < ≡ sin2 θ1∗ , (15.171)
(V1CM )−2 − 1
where s 2
CM m1
V1 = 1− (15.172)
E1CM
is the speed of product 1 in the CM frame. Thus, for V < V1CM < 1, the scattering angle θ1
may take on any value, while for larger reactant speeds V1CM < V < 1 the quantity sin2 θ1
cannot exceed a critical value.
15.7.2 Miscellaneous examples of particle decays
Let us now consider some applications of the formulae in eqn. 15.164:
• Consider the decay π 0 → γγ, for which m1 = m2 = 0. We then have E1CM = E2CM =
1 CM
2 M . Thus, with M = mπ 0 = 135 MeV, we have E1 = E2CM = 67.5 MeV for the
photon energies in the CM frame.
• For the reaction K + −→ µ+ + νµ we have M = mK + = 494 MeV and m1 = mµ− =

106 MeV. The neutrino mass is m2 ≈ 0, hence E2CM = 236 MeV is the emitted neu-
trino’s energy in the CM frame.
• A Λ0 hyperon with a mass M = mΛ0 = 1116 MeV decays into a proton (m1 = mp =
938 MeV) and a pion m2 = mπ− = 140 MeV). The CM energy of the emitted proton
is E1CM = 943 MeV and that of the emitted pion is E2CM = 173 MeV.
15.7.3 Threshold particle production with a stationary target
Consider now a particle of mass M1 moving with velocity V1 = V1 x̂, incident upon a
stationary target particle of mass M2 , as indicated in fig. 15.9. Let the product masses be
m1 , m2 , . . . , mN 0 . The 4-momenta of the reactants and products are
P1µ = E1 , P1 P2µ = M2 1 , 0 Qµj = εj , pj .

, , (15.173)
Note that E12 − P12 = M12 and ε2j − p2j = m2j , with j ∈ {1, 2, . . . , N 0 }.
Conservation of momentum means that

N0
P1µ P2µ Qµj .

X
+ = (15.174)
j=1
In particular, taking the µ = 0 component, we have

N 0
X
E 1 + M2 = εj , (15.175)
j=1
which certainly entails

N 0
X
E1 ≥ mj − M2 (15.176)
j=1
since εj = γj mj ≥ mj . But can the equality ever be achieved? This would only be the case
if γj = 1 for all j, i.e. the final velocities are all zero. But this itself is quite impossible,
since the initial state momentum is P .
Figure 15.9: A two-particle initial state, with a stationary target in the LAB frame, and an
N 0 -particle final state.
To determine the threshold energy E1thr , we compare the length of the total momentum
vector in the LAB and CM frames:
(P1 + P2 )2 = M12 + M22 + 2E1 M2 (LAB) (15.177)

N0
!2
X
CM
= εj (CM) . (15.178)
j=1
Thus,
P 2
N0 CM
j=1 εj − M12 − M22
E1 = (15.179)
2M2
and we conclude
P 2
N0
j=1 mj − M12 − M22
THR
E1 ≥ E1 = . (15.180)
2M2
Note that in the CM frame it is possible for each εCM

j = mj .
PN 0
Finally, we must have E1THR ≥ j=1 mj − M2 . This then requires
N 0
X
M1 + M2 ≤ mj . (15.181)
j=1
15.7.4 Transformation between frames
Consider a particle with 4-velocity uµ in frame K and consider a Lorentz transformation

between this frame and a frame K 0 moving relative to K with velocity V . We may write
uµ = γ , γv cos θ , γv sin θ n̂⊥ u0µ = γ 0 , γ 0 v 0 cos θ0 , γ 0 v 0 sin θ0 n̂0⊥ .

, (15.182)
According to the general transformation rules of eqns. 15.50, 15.51, and 15.52, we may
write
γ = Γ γ 0 + Γ V γ 0 v 0 cos θ0 (15.183)
0 0 0 0
γv cos θ = Γ V γ + Γ γ v cos θ (15.184)
0 0 0
γv sin θ = γ v sin θ (15.185)
n̂⊥ = n̂0⊥ , (15.186)
where the x̂ axis is taken to be V̂ , and where Γ ≡ (1 − V 2 )−1/2 . Note that the last two of
these equations may be written as a single vector equation for the transverse components.
Dividing the eqn. 15.185 by eqn. 15.184, we obtain the result
sin θ0
tan θ = . (15.187)
V
Γ v0 + cos θ0
We can then use eqn. 15.183 to relate v 0 and cos θ0 :
−1
p Γ
γ0 1 + V v 0 cos θ0 .

= 1 − v 02 = (15.188)
γ
Squaring both sides, we obtain a quadratic equation whose roots are
−Γ 2 V cos θ0 ± γ 4 − Γ 2 γ 2 (1 − V 2 cos2 θ0 )
p
0
v = . (15.189)
γ 2 + Γ 2 V 2 cos2 θ0
CM frame mass and velocity
To find the velocity of the CM frame, simply write
N N N
!
Piµ
X X X
µ
Ptot = = γi mi , γi mi vi (15.190)
i=1 i=1 i=1
≡ Γ M (1 , V ) . (15.191)
Then
N
!2 N
!2
X X
2
M = γi mi − γi mi vi (15.192)
i=1 i=1
and
PN
i=1 γi mi vi
V = P N
. (15.193)
i=1 γi mi
Figure 15.10: Compton scattering of a photon and an electron.
15.7.5 Compton scattering
An extremely important example of relativistic scattering occurs when a photon scatters

off an electron: e− + γ −→ e− + γ (see fig. 15.10). Let us work in the rest frame of the
reactant electron. Then we have
Peµ = me (1, 0) , Peeµ = me (γ , γV ) (15.194)
for the initial and final 4-momenta of the electron. For the photon, we have
Pγµ = (ω , k) , Peγµ = (e
ω , k)
e , (15.195)
where we’ve set h̄ = 1 as well. Conservation of 4-momentum entails
Pγµ − Peγµ = Peeµ − Peµ . (15.196)
Thus,
ω−ω
e, k−k
e = m γ − 1 , γV .
e (15.197)
Squaring each side, we obtain
2
e 2 = 2ω ω

ω−ω e − k−k e (cos θ − 1)

= m2e (γ − 1)2 − γ 2 V 2
= 2m2e (1 − γ)
e − ω) .
= 2me ω (15.198)
Here we have used |k| = ω for photons, and also (γ − 1) me = ω − ω

e , from eqn. 15.197.
Restoring the units h̄ and c, we find the Compton formula

1 1 h̄
− = 1 − cos θ . (15.199)
ω
e ω me c2
This is often expressed in terms of the photon wavelengths, as

4πh̄
sin2 1

λ̃ − λ = 2θ , (15.200)
me c
showing that the wavelength of the scattered light increases with the scattering angle in the
rest frame of the target electron.
15.8 Covariant Electrodynamics
We begin with the following expression for the Lagrangian density of charged particles
coupled to an electromagnetic field, and then show that the Euler-Lagrange equations re-
capitulate Maxwell’s equations. The Lagrangian density is
1 1
L=− Fµν F µν − jµ Aµ . (15.201)
16π c
Here, Aµ = (φ , A) is the electromagnetic 4-potential , which combines the scalar field φ
and the vector field A into a single 4-vector. The quantity Fµν is the electromagnetic field
strength tensor and is given by
Fµν = ∂µ Aν − ∂ν Aµ . (15.202)
Note that as defined Fµν = −Fνµ is antisymmetric. Note that, if i = 1, 2, 3 is a spatial

index, then
1 ∂Ai ∂A0
F0i = − − = Ei (15.203)
c ∂t ∂xi
∂Ai ∂Aj
Fij = − = − ijk Bk . (15.204)
∂xj ∂xi
Here we have used Aµ = (A0 , A) and Aµ = (A0 , −A), as well as ∂µ = (c−1 ∂t , ∇).
IMPORTANT : Since the electric and magnetic fields E and B are not part of a 4-vector,
we do not use covariant / contravariant notation for their components. Thus, Ei is the ith
component of the vector E. We will not write E i with a raised index, but if we did, we’d
mean the same thing: E i = Ei . By contrast, for the spatial components of a four-vector
like Aµ , we have Ai = −Ai .
Explicitly, then, we have

   
0 Ex Ey Ez 0 −Ex −Ey −Ez
−Ex 0 −Bz By  E 0 −Bz By 
   
Fµν =  , F µν = x  ,
−Ey Bz −Bx  −Bx 

0 Ey Bz 0
−Ez −By Bx 0 Ez −By Bx 0
(15.205)
15.8. COVARIANT ELECTRODYNAMICS 323
where F µν = g µα g νβ Fαβ . Note that when comparing F µν and Fµν , the components with
one space and one time index differ by a minus sign. Thus,
1 E2 − B2
− Fµν F µν = , (15.206)
16π 8π
which is the electromagnetic Lagrangian density. The j · A term accounts for the interaction
between matter and electromagnetic degrees of freedom. We have
1 1
j Aµ = % φ − j · A , (15.207)
c µ c
where
µ c% µ φ
j = , A = , (15.208)
j A
where % is the charge density and j is the current density. Charge conservation requires
∂%
∂µ j µ = + ∇·j = 0 . (15.209)
∂t
We shall have more to say about this further on below.
Let us now derive the Euler-Lagrange equations for the action functional,
Z
−1 4 1 µν −1 µ
S = −c dx F F + c jµ A . (15.210)
16π µν
We first vary with respect to Aµ . Clearly
δFµν = ∂µ δAν − ∂ν δAµ . (15.211)
We then have
1 µν −1 ν 1 µν
δL = ∂ F − c j δAν − ∂µ F δAν . (15.212)
4π µ 4π
Ignoring the boundary term, we obtain Maxwell’s equations,
∂µ F µν = 4πc−1 j ν (15.213)
The ν = k component of these equations yields
∂0 F 0k + ∂i F jk = −∂0 Ek − jkl ∂j Bl = 4πc−1 j k , (15.214)
which is the k component of the Maxwell-Ampère law,

4π 1 ∂E
∇×B = j+ . (15.215)
c c ∂t
The ν = 0 component reads
4π 0
∂i F i0 = j ⇒ ∇·E = 4π% , (15.216)
c
which is Gauss’s law. The remaining two Maxwell equations come ‘for free’ from the very
definitions of E and B:
1 ∂A
E = −∇A0 − (15.217)
c ∂t
B =∇×A , (15.218)
which imply
1 ∂B
∇×E =− (15.219)
c ∂t
∇·B =0 . (15.220)
15.8.1 Lorentz force law
This has already been worked out in chapter 7. Here we reiterate our earlier derivation.
The 4-current may be written as
µ
X Z dXnµ (4)
j (x, t) = c qn dτ δ (x − X) . (15.221)
n
dτ
Thus, writing Xnµ = ct , Xn (t) , we have

X
j 0 (x, t) =

qn c δ x − Xn (t) (15.222)
n
X
j(x, t) = qn Ẋn (t) δ x − Xn (t) . (15.223)
n
The Lagrangian for the matter-field interaction term is then

Z
−1
d3x j 0 A0 − j · A

L = −c
X qn

=− qn φ(Xn , t) − A(Xn , t) · Ẋn , (15.224)
n
c
where φ = A0 . For each charge qn , this is equivalent to a particle with velocity-dependent

potential energy
q
U (x, t) = q φ(x, t) − A(r, t) · ẋ , (15.225)
c
where x = Xn .
1 2
Let’s work out the equations of motion. We assume a kinetic energy T = 2 mẋ for the
charge. We then have
d ∂L ∂L
= (15.226)
dt ∂ ẋ ∂x
with L = T − U , which gives

q dA q
m ẍ + = −q ∇φ + ∇(A · ẋ) , (15.227)
c dt c
or, in component notation,
q ∂Ai j q ∂Ai ∂φ q ∂Aj j
m ẍi + ẋ + = −q + ẋ , (15.228)
c ∂xj c ∂t ∂xi c ∂xi
which is to say
q ∂Ai q ∂Aj ∂Ai

i ∂φ
m ẍ = −q i − + − ẋj . (15.229)
∂x c ∂t c ∂xi ∂xj
It is convenient to express the cross product in terms of the completely antisymmetric tensor
of rank three, ijk :
∂Ak
Bi = ijk , (15.230)
∂xj
and using the result
ijk imn = δjm δkn − δjn δkm , (15.231)
we have ijk Bi = ∂ j Ak − ∂ k Aj , and
∂φ q ∂Ai q
m ẍi = −q − + ijk ẋj Bk , (15.232)
∂xi c ∂t c
or, in vector notation,
q ∂A q
m ẍ = −q ∇φ − + ẋ × (∇ × A)
c ∂t c
q
= q E + ẋ × B , (15.233)
c
which is, of course, the Lorentz force law.
15.8.2 Gauge invariance
The action S = c−1 d4x L admits a gauge invariance. Let Aµ → Aµ + ∂ µ Λ, where Λ(x, t)
R
is an arbitrary scalar function of spacetime coordinates. Clearly

Fµν → Fµν + ∂µ ∂ν Λ − ∂ν ∂µ Λ = Fµν , (15.234)
and hence the fields E and B remain invariant under the gauge transformation, even though
the 4-potential itself changes. What about the matter term? Clearly
−c−1 j µ Aµ → − c−1 j µ Aµ − c−1 j µ ∂µ Λ
= −c−1 j µ Aµ + c−1 Λ ∂µ j µ − ∂µ c−1 Λ j µ .

(15.235)
Once again we ignore the boundary term. We may now invoke charge conservation to write
∂µ j µ = 0, and we conclude that the action is invariant! Woo hoo! Note also the very deep
connection
gauge invariance ←→ charge conservation . (15.236)
Figure 15.11: Homer celebrates the manifest gauge invariance of classical electromagnetic
theory.
15.8.3 Transformations of fields
One last detail remains, and that is to exhibit explicitly the Lorentz transformation prop-
erties of the electromagnetic field. For the case of vectors like Aµ , we have
Aµ = Lµν A0 ν . (15.237)
The E and B fields, however, appear as elements in the field strength tensor F µν . Clearly
this must transform as a tensor:
F µν = Lµα Lνβ F 0 αβ = Lµα F 0 αβ Ltβ ν . (15.238)
We can write a general Lorentz transformation as a product of a rotation Lrot and a boost
Lboost . Let’s first see how rotations act on the field strength tensor. We take
!
11×1 01×3
L = Lrot = , (15.239)
03×1 R3×3
where Rt R = 1, i.e. R ∈ O(3) is an orthogonal matrix. We must compute

! !
−Ek0

µ 0 αβ t ν 1 0 0 1 0
L αF Lβ =
0 Rij Ej0 − jkm Bm 0 t
0 Rkl
!
0 −Ek0 Rkl
t
= . (15.240)
Rij Ej0 − jkm Rij Rlk Bm 0
Thus, we conclude
El = Rlk Ek0 (15.241)
0
iln Bn = jkm Rij Rlk Bm . (15.242)
Now for any 3 × 3 matrix R we have
jks Rij Rlk Rrs = det(R) ilr , (15.243)
and therefore
0
jkm Rij Rlk Bm = jkm Rij Rlk Rnm Rns Bs0
= det(R) iln Rns Bs0 , (15.244)
Therefore,
Ei = Rij Ej0 , Bi = det(R) · Rij Bj0 . (15.245)
For any orthogonal matrix, Rt R = 1 gives that det(R) = ±1. The extra factor of det(R)
in the transformation properties of B is due to the fact that the electric field transforms as
a vector , while the magnetic field transforms as a pseudovector . Under space inversion, for
example, where R = −1, the electric field is odd under this transformation (E → −E) while
the magnetic field is even (B → +B). Similar considerations hold in particle mechanics for
the linear momentum, p (a vector) and the angular momentum L = r × p (a pseudovector).
The analogy is not complete, however, because while both p and L are odd under the
operation of time-reversal, E is even while B is odd.
OK, so how about boosts? We can write the general boost, from eqn. 15.37, as

γ γ β̂
L= (15.246)
γ β̂ I + (γ − 1)Pβ
where Pβij = β̂i β̂j is the projector onto the direction of β. We now compute
!
t 0t
−E γβ t

µ 0 αβ t ν γ γβ 0 γ
L αF Lβ = . (15.247)
γβ I + (γ − 1)P E 0 − jkm Bm 0 γβ I + (γ − 1)P
Carrying out the matrix multiplications, we obtain
E = γ(E 0 − β × B 0 ) − (γ − 1)(β̂ · E 0 )β̂ (15.248)

B = γ(B 0 + β × E 0 ) − (γ − 1)(β̂ · B 0 )β̂ . (15.249)
Expressed in terms of the components Ek , E⊥ , Bk , and B⊥ , one has
Ek = Ek0 0 0

, E⊥ = γ E⊥ − β × B⊥ (15.250)
Bk = Bk0 0 0

, B⊥ = γ B⊥ + β × E⊥ . (15.251)
Recall that for any vector ξ, we write
ξk = β̂ · ξ (15.252)
ξ⊥ = ξ − (β̂ · ξ) β̂ , (15.253)
so that β̂ · ξ⊥ = 0.
15.8.4 Invariance versus covariance
We saw that the laws of electromagnetism were gauge invariant. That is, the solutions to
the field equations did not change under a gauge transformation Aµ → Aµ + ∂ µΛ. With
respect to Lorentz transformations, however, the theory is Lorentz covariant. This means
that Maxwell’s equations in different inertial frames take the exact same form, ∂µ F µν =
4πc−1 j ν , but that both the fields and the sources transform appropriately under a change
in reference frames. The sources are described by the current 4-vector j µ = (c% , j) and
transform as
c% = γc%0 + γβjk0 (15.254)
jk = γβc%0 + γjk0 (15.255)
0
j⊥ = j⊥ . (15.256)
The fields transform according to eqns. 15.250 and 15.251.
Consider, for example, a static point charge q located at the origin in the frame K 0 , which
moves with velocity u x̂ with respect to K. An observer in K 0 measures a charge density
%0 (x0 , t0 ) = q δ(x0 ). The electric and magnetic fields in the K 0 frame are then E 0 = q r̂ 0 /r0 2
and B 0 = 0. For an observer in the K frame, the coordinates transform as
ct = γct0 + γβx0 ct0 = γct − γβx (15.257)
0 0 0
x = γβct + γx x = −γβct + γx , (15.258)
as well as y = y 0 and z = z 0 . The observer in the K frame sees instead a charge at
xµ = (ct , ut , 0 , 0) and both a charge density as well as a current density:
%(x, t) = γ%(x0 , t0 ) = q δ(x − ut) δ(y) δ(z) (15.259)
j(x, t) = γβc %(x0 , t0 ) x̂ = u q δ(x − ut) δ(y) δ(z) x̂ . (15.260)
OK, so much for the sources. How about the fields? Expressed in terms of Cartesian
coordinates, the electric field in K 0 is given by
x0 x̂ + y 0 ŷ + z 0 ẑ
E 0 (x0 , t0 ) = q 3/2 . (15.261)
x0 2 + y 0 2 + z 0 2
From eqns. 15.250 and 15.251, we have Ex = Ex0 and Bx = Bx0 = 0. Furthermore, we have
Ey = γEy0 , Ez = γEz0 , By = −γβEz0 , and Bz = γβEy0 . Thus,
(x − ut)x̂ + y ŷ + z ẑ
E(x, t) = γq 3/2 (15.262)
γ 2 (x − ut)2 + y 2 + z 2
γu y ẑ − z ŷ
B(x, t) = q 3/2 . (15.263)
c γ 2 (x − ut)2 + y 2 + z 2
15.9. APPENDIX I : THE POLE, THE BARN, AND RASHOMAN 329
Let us define
R(t) = (x − ut) x̂ + y ŷ + z ẑ . (15.264)
We further define the angle θ ≡ cos−1 β̂ · R̂ . We may then write

qR 1 − β2
E(x, t) = 3
· 3/2
R 1 − β 2 sin2 θ
q β̂ × R 1 − β2
B(x, t) = · 3/2 . (15.265)
R3 1 − β 2 sin2 θ
The fields are therefore enhanced in the transverse directions: E⊥ /Ek = γ 3 .
15.9 Appendix I : The Pole, the Barn, and Rashoman
Akira Kurosawa’s 1950 cinematic masterpiece, Rashoman, describes a rape, murder, and
battle from four different and often contradictory points of view. It poses deep questions
regarding the nature of truth. Psychologists sometimes refer to problems of subjective
perception as the Rashoman effect. In literature, William Faulkner’s 1929 novel, The Sound
and the Fury, which describes the tormented incestuous life of a Mississippi family, also is
told from four points of view. Perhaps Faulkner would be a more apt comparison with
Einstein, since time plays an essential role in his novel. For example, Quentin’s watch,
given to him by his father, represents time and the sweep of life’s arc (“Quentin, I give you
the mausoleum of all hope and desire...”). By breaking the watch, Quentin symbolically
attempts to escape time and fate. One could draw an analogy to Einstein, inheriting a watch
from those who came before him, which he too broke – and refashioned. Did Faulkner know
of Einstein? But I digress.
Consider a relativistic runner carrying a pole of proper length `, as depicted in fig. 15.12.
He runs toward a barn of proper length ` at velocity u = cβ. Let the frame of the barn
be K and the frame of the runner be K 0 . Recall that the Lorentz transformations between
frames K and K 0 are given by
ct = γct0 + γx0 ct0 = γct − γβx (15.266)

0 0 0
x = γβct + γx x = −γβct + γx . (15.267)
We define the following points. Let A denote the left door of the barn and B the right
door. Furthermore, let P denote the left end of the pole and Q its right end. The spacetime
coordinates for these points in the two frames are clearly .
A = (ct , 0) P 0 = (ct0 , 0) (15.268)

0 0
B = (ct , `) Q = (ct , `) (15.269)
Figure 15.12: A relativistic runner carries a pole of proper length ` and runs into a barn of
proper length `.
We now compute A0 and B 0 in frame K 0 , as well as P and Q in frame K:
A0 = (γct , −γβct) B 0 = (γct − γβ` , −γβct + γ`) (15.270)

0 0 0 0 −1
≡ (ct , −βct ) ≡ (ct , −βct + γ `) . (15.271)
Similarly,
P = (γct0 , γβct0 ) Q = (γct0 + γβ` , γβct0 + γ`) (15.272)

−1
≡ (ct , βct) ≡ (ct , βct + γ `) . (15.273)
We now define four events, by the coincidences of A and B with P and Q:
• Event I : The right end of the pole enters the left door of the barn. This is described
by Q = A in frame K and by Q0 = A0 in frame K 0 .
• Event II : The right end of the pole exits the right door of the barn. This is described
by Q = B in frame K and by Q0 = B 0 in frame K 0 .
• Event III : The left end of the pole enters the left door of the barn. This is described
by P = A in frame K and by P 0 = A0 in frame K 0 .
• Event IV : The left end of the pole exits the right door of the barn. This is described
by P = B in frame K and by P 0 = B 0 in frame K 0 .
Mathematically, we have in frame K that
`
I: Q=A ⇒ tI = − (15.274)
γu
`
II : Q=B ⇒ tII = (γ − 1) (15.275)
γu
III : P =A ⇒ tIII = 0 (15.276)
`
IV : P =B ⇒ tIV = (15.277)
u
15.10. APPENDIX II : PHOTOGRAPHING A MOVING POLE 331
In frame K 0 , however
`
I: Q0 = A0 ⇒ t0I = − (15.278)
u
`
II : Q0 = B 0 ⇒ t0II = −(γ − 1) (15.279)
γu
III : P 0 = A0 ⇒ t0III = 0 (15.280)
`
IV : P 0 = B0 ⇒ t0IV = (15.281)
γu
Thus, to an observer in frame K, the order of events is I, III, II, and IV, because
tI < tIII < tII < tIV . (15.282)
For tIII < t < tII , he observes that the pole is entirely in the barn. Indeed, the right door can
start shut and the left door open, and sensors can automatically and, for the purposes of
argument, instantaneously trigger the closing of the left door immediately following event
III and the opening of the right door immediately prior to event II. So the pole can be
inside the barn with both doors shut!
But now for the Rashoman effect: according to the runner, the order of events is I, II, III,
and IV, because
t0I < t0II < t0III < t0IV . (15.283)
At no time does the runner observe the pole to be entirely within the barn. Indeed, for
t0II < t0 < t0III , both ends of the pole are sticking outside of the barn!
15.10 Appendix II : Photographing a Moving Pole
What is the length ` of a moving pole of proper length `0 as measured by an observer at

rest? The answer would appear to be γ −1 `0 , as we computed in eqn. 15.71. However, we
should be more precise when we we speak of ‘length’. The relation `(β) = γ −1 `0 tells us
the instantaneous end-to-end distance as measured in the observer’s rest frame K. But an
actual experiment might not measure this quantity.
For example, suppose a relativistic runner carrying a pole of proper length `0 runs past a
measuring rod which is at rest in the rest frame K of an observer. The observer takes a
photograph of the moving pole as it passes by. Suppose further that the angle between the
observer’s line of sight and the velocity u of the pole is α, as shown in fig. 15.13. What is
the apparent length `(α, u) of the pole as observed in the photograph? (I.e. the pole will
appear to cover a portion of the measuring rod which is of length `.)
The point here is that the shutter of the camera is very fast (otherwise the image will appear
blurry). In our analysis we will assume the shutter opens and closes instantaneously. Let’s
define two events:
Figure 15.13: An object of proper length ` and moving with velocity u, when photographed
˜
from an angle α, appears to have a length `.
• Event 1 : photon γ1 is emitted by the rear end of the pole.
• Event 2 : photon γ2 is emitted by the front end of the pole.
Both photons must arrive at the camera’s lens simultaneously. Since, as shown in the figure,
the path of photon #1 is longer by a distance ` cos α, where ` is the apparent length of the
pole, γ2 must be emitted a time ∆t = c−1 ` cos α after γ1 . Now if we Lorentz transform from
frame K to frame K 0 , we have
∆x0 = γ∆x − γβ∆t . (15.284)
But ∆x0 = `0 is the proper length of the pole, and ∆x = ` is the apparent length. With
c∆t = ` cos α, then, we have
γ −1 `0
`= . (15.285)
1 − β cos α
When α = 90◦ , we recover the familiar Lorentz-Fitzgerald contraction `(β) = γ −1 `0 . This
is because the photons γ1 and γ2 are then emitted simultaneously, and the photograph
measures the instantaneous end-to-end distance of the pole as measured in the observer’s
rest frame K. When cos α 6= 0, however, the two photons are not emitted simultaneously,
and the apparent length is given by eqn. 15.285.
Chapter 16
Dynamical Systems
16.1 Introduction
16.1.1 Phase space and phase curves
Dynamics is the study of motion through phase space. The phase space of a given dynamical
system is described as an N -dimensional manifold, M. A (differentiable) manifold M is a
topological space that is locally diffeomorphic to RN .1 Typically in this course M will RN
itself, but other common examples include the circle S 1 , the torus T 2 , the sphere S 2 , etc.
Let gt : M → M be a one-parameter family of transformations from M to itself, with

gt=0 = 1, the identity. We call gt the t-advance mapping. It satisfies the composition rule
gt gs = gt+s . (16.1)
Let us choose a point ϕ0 ∈ M. Then we write ϕ(t) = gt ϕ0 , which also is in M. The set

gt ϕ0 t ∈ R , ϕ0 ∈ M is called a phase curve. A graph of the motion ϕ(t) in the product
space R × M is called an integral curve.
16.1.2 Vector fields
The velocity vector V (ϕ) is given by the derivative

d
V (ϕ) = gt ϕ . (16.2)
dt t=0
The velocity V (ϕ) is an element of the tangent space to M at ϕ, abbreviated TMϕ .

If M is N -dimensional, then so is each TMϕ (for all p). However, M and TMϕ may
1
A diffeomorphism F : M → N is a differentiable map with a differentiable inverse. This is a special
type of homeomorphism, which is a continuous map with a continuous inverse.
333
334 CHAPTER 16. DYNAMICAL SYSTEMS
Figure 16.1: An example of a phase curve.
differ topologically. For example, if M = S1 , the circle, the tangent space at any point is
isomorphic to R.
For our purposes, we will take ϕ = (ϕ1 , . . . , ϕN ) to be an N -tuple, i.e. a point in RN . The
equation of motion is then
d
ϕ(t) = V ϕ(t) . (16.3)
dt
Note that any N th order ODE, of the general form
dNx dN −1x

dx
= F x, , . . . , N −1 , (16.4)
dtN dt dt
may be represented by the first order system ϕ̇ = V (ϕ). To see this, define ϕk =
dk−1x/dtk−1 , with k = 1, . . . , N . Thus, for j < N we have ϕ̇j = ϕj+1 , and ϕ̇N = f .
In other words,
ϕ̇ V (ϕ)
z  }| { 
z }| {

ϕ ϕ2
 .1 ..
.
 
d 
 .
 
= .

 . (16.5)

dt ϕ
  
ϕ 
N −1  N
ϕN F ϕ1 , . . . , ϕ N
16.1.3 Existence / uniqueness / extension theorems
Theorem : Given ϕ̇ = V (ϕ) and ϕ(0), if each V (ϕ) is a smooth vector field over some
open set D ∈ M, then for ϕ(0) ∈ D the initial value problem has a solution on some
finite time interval (−τ, +τ ) and the solution is unique. Furthermore, the solution has a
unique extension forward or backward in time, either indefinitely or until ϕ(t) reaches the
boundary of D.
Corollary : Different trajectories never intersect!

16.1.4 Linear differential equations
A homogeneous linear N th order ODE,

dNx dN −1x dx
N
+ cN −1 N −1
+ . . . + c1 + c0 = 0 (16.6)
dt dt dt
may be written in matrix form, as
M
  z }| {  
ϕ1 0 1 0 ··· 0 ϕ1
d  ϕ2 
  0 0 1 ··· 0  ϕ 
  2
 .  = .. .. .. ..  .  . (16.7)
 
dt  ..   . . . .   .. 
ϕN −c0 −c1 −c2 · · · −cN −1 ϕN
Thus,
ϕ̇ = M ϕ , (16.8)
and if the coefficients ck are time-independent, i.e. the ODE is autonomous, the solution is
obtained by exponentiating the constant matrix Q:
ϕ(t) = exp(M t) ϕ(0) ; (16.9)
the exponential of a matrix may be given meaning by its Taylor series expansion. If the
ODE is not autonomous, then M = M (t) is time-dependent, and the solution is given by
the path-ordered exponential,
( Zt )
0 0
ϕ(t) = P exp dt M (t ) ϕ(0) , (16.10)
0
As defined, the equation ϕ̇ = V (ϕ) is autonomous, since gt depends only on t and on no

other time variable. However, by extending the phase space from M to R × M, which is
of dimension (N + 1), one can describe arbitrary time-dependent ODEs.
16.1.5 Lyapunov functions
For a general dynamical system ϕ̇ = V (ϕ), a Lyapunov function L(ϕ) is a function which
satisfies
∇L(ϕ) · V (ϕ) ≤ 0 . (16.11)
There is no simple way to determine whether a Lyapunov function exists for a given dy-
namical system, or, if it does exist, what the Lyapunov function is. However, if a Lyapunov
function can be found,
then this severely limits the possible behavior of the system. This
is because L ϕ(t) must be a monotonic function of time:
d dϕ
L ϕ(t) = ∇L · = ∇L(ϕ) · V (ϕ) ≤ 0 . (16.12)
dt dt
Thus, the system evolves toward a local minimum of the Lyapunov function. In general
this means that oscillations are impossible in systems for which a Lyapunov function exists.
For example, the relaxational dynamics of the magnetization M of a system are sometimes
modeled by the equation
dM ∂F
= −Γ , (16.13)
dt ∂M
where F (M, T ) is the free energy of the system. In this model, assuming constant temper-
2
ature T , Ḟ = F 0 (M ) Ṁ = −Γ F 0 (M ) ≤ 0. So the free energy F (M ) itself is a Lyapunov

function, and it monotonically decreases during the evolution of the system. We shall meet
up with this example again in the next chapter when we discuss imperfect bifurcations.
16.2 N = 1 Systems
We now study phase flows in a one-dimensional phase space, governed by the equation
du
= f (u) . (16.14)
dt
Again, the equation u̇ = h(u, t) is first order, but not autonomous, and it corresponds to
the N = 2 system,
d u h(u, t)
= . (16.15)
dt t 1
The equation 20.39 is easily integrated:
Zu
du du0
= dt =⇒ t − t0 = . (16.16)
f (u) f (u0 )
u0
This gives t(u); we must then invert this relationship to obtain u(t).
Example : Suppose f (u) = a − bu, with a and b constant. Then

du
dt = = −b−1 d ln(a − bu) (16.17)
a − bu
whence
!
1 a − b u(0) a a
t = ln =⇒ u(t) = + u(0) − exp(−bt) . (16.18)
b a − b u(t) b b
Even if one cannot analytically obtain u(t), the behavior is very simple, and easily obtained
by graphical analysis. Sketch the function f (u). Then note that

f (u) > 0 u̇ > 0 ⇒ move to right

u̇ = f (u) =⇒ f (u) < 0 u̇ < 0 ⇒ move to left (16.19)

f (u) = 0 u̇ = 0 ⇒ fixed point

16.2. N = 1 SYSTEMS 337
Figure 16.2: Phase flow for an N = 1 system.
The behavior of N = 1 systems is particularly simple: u(t) flows to the first stable fixed
point encountered, where it then (after a logarithmically infinite time) stops. The motion
is monotonic – the velocity u̇ never changes sign. Thus, oscillations never occur for N = 1
phase flows.2
16.2.1 Classification of fixed points (N = 1)
A fixed point u∗ satisfies f (u∗ ) = 0. Generically, f 0 (u∗ ) 6= 0 at a fixed point.3 Suppose

f 0 (u∗ ) < 0. Then to the left of the fixed point, the function f (u < u∗ ) is positive, and the
flow is to the right, i.e. toward u∗ . To the right of the fixed point, the function f (u > u∗ ) is
negative, and the flow is to the left, i.e. again toward u∗ . Thus, when f 0 (u∗ ) < 0 the fixed
point is said to be stable, since the flow in the vicinity of u∗ is to u∗ . Conversely, when
f 0 (u∗ ) > 0, the flow is always away from u∗ , and the fixed point is then said to be unstable.
Indeed, if we linearize about the fixed point, and let ≡ u − u∗ , then
˙ = f 0 (u∗ ) + 21 f 00 (u∗ ) 2 + O(3 ) , (16.20)
and dropping all terms past the first on the RHS gives
h i
(t) = exp f 0 (u∗ ) t (0) . (16.21)
The deviation decreases exponentially for f 0 (u∗ ) < 0 and increases exponentially for f (u∗ ) >
0. Note that

1
t() = 0 ∗ ln , (16.22)
f (u ) (0)
so the approach to a stable fixed point takes a logarithmically infinite time. For the unstable
case, the deviation grows exponentially, until eventually the linearization itself fails.
2
When I say ‘never’ I mean ‘sometimes’ – see the section 16.3.
3
The system f (u∗ ) = 0 and f 0 (u∗ ) = 0 is overdetermined, with two equations for the single variable u∗ .
Figure 16.3: Flow diagram for the logistic equation.
16.2.2 Logistic equation
This model for population growth was first proposed by Verhulst in 1838. Let N denote
the population in question. The dynamics are modeled by the first order ODE,
dN N
= rN 1 − , (16.23)
dt K
where N , r, and K are all positive. For N K the growth rate is r, but as N increases a
quadratic nonlinearity kicks in and the rate vanishes for N = K and is negative for N > K.
The nonlinearity models the effects of competition between the organisms for food, shelter,
or other resources. Or maybe they crap all over each other and get sick. Whatever.
There are two fixed points, one at N ∗ = 0, which is unstable (f 0 (0) = r > 0). The other, at
N ∗ = K, is stable (f 0 (K) = −r). The equation is adimensionalized by defining ν = N/K
and s = rt, whence
ν̇ = ν(1 − ν) . (16.24)
Integrating,
dν ν ν0
= d ln = ds =⇒ ν(s) = . (16.25)
ν(1 − ν) 1−ν

ν0 + 1 − ν0 exp(−s)
As s → ∞, ν(s) = 1 − ν0−1 − 1 e−s + O(e−2s ), and the relaxation to equilibrium (ν ∗ = 1)

is exponential, as usual.
Another application of this model is to a simple autocatalytic reaction, such as
A+X *
) 2X , (16.26)
i.e. X catalyses the reaction A −→ X. Assuming a fixed concentration of A, we have
ẋ = κ+ a x − κ− x2 , (16.27)
where x is the concentration of X, and κ± are the forward and backward reaction rates.
16.2. N = 1 SYSTEMS 339
α
Figure 16.4: f (u) = A u − u∗ , for α > 1 and α < 1.

16.2.3 Singular f (u)

α
Suppose that in the vicinity of a fixed point we have f (u) = A u − u∗ , with A > 0. We

now analyze both sides of the fixed point.
u < u∗ : Let = u∗ − u. Then
1−α 1−α
˙ = −A α =⇒ = 0 − At , (16.28)
1−α 1−α
hence
h i 1
1−α
(t) = 1 + (α − 1)A α−1
0 t . (16.29)
This, for α < 1 the fixed point = 0 is reached in a finite time: (tc ) = 0, with
1−α
0
tc = . (16.30)
(1 − α) A
For α > 1, we have limt→∞ (t) = 0, but (t) > 0 ∀ t < ∞.
The fixed point u = u∗ is now half-stable – the flow from the left is toward u∗ but from the
right is away from u∗ . Let’s analyze the flow on either side of u∗ .
u > u∗ : Let = u − u∗ . Then ˙ = A α , and

h i 1
1−α
(t) = 1 + (1 − α)A 0α−1 t . (16.31)
For α < 1, (t) escapes to = ∞ only after an infinite time. For α > 1, the escape to
infinity takes a finite time: (tc ) = ∞, with
1−α
0
tc = . (16.32)
(α − 1) A
In both cases, higher order terms in the (nonanalytic) expansion of f (u) about u = u∗ will
eventually come into play.
Figure 16.5: Solution to ˙ = ∓A α .
16.2.4 Recommended exercises
It is constructive to sketch the phase flows for the following examples:

v̇ = −g u̇ = A sin(u)
mv̇ = −mg − γv u̇ = A (u − a)(u − b)(u − c)
2
mv̇ = −mg − cv sgn(v) u̇ = au2 − bu3 .
In each case, identify all the fixed points and assess their stability. Assume all constants A,
a, b, c, γ, etc. are positive.
16.2.5 Non-autonomous ODEs
Non-autonomous ODEs of the form u̇ = h(u, t) are in general impossible to solve by quadra-
tures. One can always go to the computer, but it is worth noting that in the separable case,
h(u, t) = f (u) g(t), one can obtain the solution
Zu Zt
du du0
= g(t) dt =⇒ = dt0 g(t0 ) , (16.33)
f (u) f (u0 )
u0 0
which implicitly gives u(t). Note that u̇ may now change sign, and u(t) may even oscillate.
For an explicit example, consider the equation
u̇ = A (u + 1) sin(βt) , (16.34)
the solution of which is

A
u(t) = −1 + u0 + 1 exp 1 − cos(βt) . (16.35)
β
In general, the non-autonomous case defies analytic solution. Many have been studied, such
as the Riccati equation,
du
= P (t)u2 + Q(t)u + R(t) . (16.36)
dt
16.3. FLOWS ON THE CIRCLE 341
Riccati equations have the special and remarkable property that one can generate all solu-
tions (i.e. with arbitrary boundary condition u(0) = u0 ) from any given solution (i.e. with
any boundary condition).
16.3 Flows on the Circle
We had remarked that oscillations are impossible for the equation u̇ = f (u) because the
flow is to the first stable fixed point encountered. If there are no stable fixed points, the
flow is unbounded. However, suppose phase space itself is bounded, e.g. a circle S1 rather
than the real line R. Thus,
θ̇ = f (θ) , (16.37)
with f (θ + 2π) = f (θ). Now if there are no fixed points, θ(t) endlessly winds around the
circle, and in this sense we can have oscillations.
16.3.1 Nonuniform oscillator
A particularly common example is that of the nonuniform oscillator,
θ̇ = ω − sin θ , (16.38)
which has applications to electronics, biology, classical mechanics, and condensed matter
physics. Note that the general equation θ̇ = ω − A sin θ may be rescaled to the above form.
A simple application is to the dynamics of a driven, overdamped pendulum. The equation
of motion is
I θ̈ + b θ̇ + Iω02 sin θ = N , (16.39)
where I is the moment of inertia, b is the damping parameter, N is the external torque
(presumed constant), and ω0 is the frequency of small oscillations when b = N = 0. When
b is large, the inertial term I θ̈ may be neglected, and after rescaling we arrive at eqn. 16.38.
The book by Strogatz provides a biological example of the nonuniform oscillator: fireflies.
An individual firefly will on its own flash at some frequency f . This can be modeled by the
equation φ̇ = β, where β = 2πf is the angular frequency. A flash occurs when φ = 2πn for
n ∈ ZZ. When subjected to a periodic stimulus, fireflies will attempt to synchronize their
flash to the flash of the stimulus. Suppose the stimulus is periodic with angular frequency
Ω. The firefly synchronization is then modeled by the equation
φ̇ = β − A sin(φ − Ωt) . (16.40)
Here, A is a measure of the firefly’s ability to modify its natural frequency in response to
the stimulus. Note that when 0 < φ − Ωt < π , i.e. when the firefly is leading the stimulus,
the dynamics tell the firefly to slow down. Conversely, when −π < φ − Ωt < 0, the firefly
Figure 16.6: Flow for the nonuniform oscillator θ̇ = ω − sin θ for three characteristic values
of ω.
is lagging the stimulus, the the dynamics tell it to speed up. Now focus on the difference
θ ≡ φ − Ωt. We have
φ̇ = β − Ω − A sin φ , (16.41)
which is the nonuniform oscillator. We can adimensionalize by defining
β−Ω
s ≡ At , ω≡ , (16.42)
A
dθ
yielding ds = ω − sin θ.
Fixed points occur only for ω < 1, for sin θ = ω. To integrate, set z = exp(iθ), in which
case
ż = − 21 (z 2 − 2iωz − 1)
= − 12 (z − z− )(z − z+ ) , (16.43)
√
where ν = 1 − ω 2 and z± = iω ± ν. Note that for ω 2 < 1, ν is real, and z(t → ∓∞) = z± .
This equation can easily be integrated, yielding
(z(0) − z− ) z+ − (z(0) − z+ ) z− exp(νt)
z(t) = , (16.44)
(z(0) − z− ) − (z(0) − z+ ) exp(νt)
For ω 2 > 1, the motion is periodic, with period
Z2π
dθ 2π
T = =√ . (16.45)
|ω| − sin θ ω2 − 1
0
The situation is depicted in Fig. 16.6.
16.4 Appendix I : Evolution of Phase Space Volumes
Recall the general form of a dynamical system, ϕ̇ = V (ϕ). Usually we are interested in
finding integral curves ϕ(t). However, consider for the moment a collection of points in
16.5. APPENDIX II : LYAPUNOV CHARACTERISTIC EXPONENTS 343
phase space comprising a region R. As the dynamical system evolves, this region will also
evolve, so that R = R(t). We now ask: how does the volume of R(t),
Z

vol R(t) = dµ , (16.46)
R(t)
where dµ = dϕ1 dϕ2 · · · dϕN is the phase space measure, change with time. We have,
explicitly,
Z

vol R(t + dt) = dµ
R(t+dt)
Z
∂ϕi (t + dt)
= dµ
∂ϕj (t)

R(t)
Z n o
= dµ 1 + ∇·V dt + O (dt)2 , (16.47)
R(t)
since
∂ϕi (t + dt) ∂Vi
dt + O (dt)2 ,

= δij + (16.48)
∂ϕj (t) ∂ϕj ϕ(t)

and, using ln det M = Tr ln M ,
det(1 + A) = 1 + TrA + O(2 ) . (16.49)
Thus,
Z
d
vol R(t) = dµ ∇·V (16.50)
dt
R(t)
Z
= dΣ n̂ · V , (16.51)
∂R(t)
where in the last line we have used Stokes’ theorem to convert the volume integral over R
to a surface integral over its boundary ∂R.
16.5 Appendix II : Lyapunov Characteristic Exponents
Suppose ϕ(t) is an integral curve – i.e. a solution of ϕ̇ = V (ϕ). We now ask: how do
nearby trajectories behave? Do they always remain close to ϕ(t) for all t? To answer this,
e ≡ ϕ(t) + η(t), in which case
we write ϕ(t)
d
ηi (t) = Mij (t) ηj (t) + O η 2 ,

(16.52)
dt
where
∂Vi
Mij (t) = . (16.53)
∂ϕj ϕ(t)
The solution, valid to first order in δϕ, is
ηi (t) = Qij (t, t0 ) ηj (t0 ) , (16.54)
where the matrix Q(t, t0 ) is given by the path ordered exponential,

( Zt )
Q(t, t0 ) = P exp dt0 M (t0 ) (16.55)
t0
∆t ∆t ∆t
≡ lim 1+ M (tN −1 ) · · · 1 + M (t1 ) 1 + M (t0 ) , (16.56)
N →∞ N N N
with ∆t = t − t0 and tj = t0 + (j/N )∆t. P is the path ordering operator , which places
earlier times to the right:

0 0
A(t) B(t ) if t > t

PA(t) B(t0 ) = (16.57)

 0 0
B(t ) A(t) if t < t .
The distinction is important if A(t), B(t0 ) 6= 0. Note that Q satisfies the composition

property,
Q(t, t0 ) = Q(t, t1 ) Q(t1 , t0 ) (16.58)
for any t1 ∈ [t0 , t]. When M is time-independent, as in the case of a fixed point where
V (ϕ∗ ) = 0, thepath ordered exponential reduces to the ordinary exponential, and Q(t, t0 ) =
exp M (t − t0 ) .
Generally it is impossible to analytically compute path-ordered exponentials. However, the

following example may be instructive. Suppose

M1 if t/T ∈ 2j, 2j + 1

M (t) = (16.59)

M2 if t/T ∈ 2j + 1, 2j + 2 ,

for all integer j. M (t) is a ‘matrix-valued square wave’, with period 2T . Then, integrating
over one period, from t = 0 to t = 2T , we have
( Z2T )
A ≡ exp dt M (t) = e(M1 +M2 ) T (16.60)
0
( Z2T )
AP ≡ P exp dt M (t) = eM2 T eM1 T . (16.61)
0
16.5. APPENDIX II : LYAPUNOV CHARACTERISTIC EXPONENTS 345
In general, A 6= AP , so the path ordering has a nontrivial effect4 .
The Lyapunov exponents are defined in the following manner. Let ê be an N -dimensional
unit vector. Define
!
1 η(t)
Λ(ϕ0 , ê) ≡ lim lim ln , (16.62)
t→∞ b→0 t − t0 η(t0 )
η(t0 )=b ê

where k · k denotes the Euclidean norm of a vector, and where ϕ0 = ϕ t0 . A theorem due
to Oseledec guarantees that there are N such values Λi (ϕ0 ), depending on the choice of ê,
for a given ϕ0 . Specifically, the theorem guarantees that the matrix
1/(t−t0 )
Q̂ ≡ Qt Q (16.63)
converges in the limit t → ∞ for almost all ϕ0 . The eigenvalues Λi correspond to the differ-
ent eigenspaces of R. Oseledec’s theorem (also called the ‘multiplicative ergodic theorem’)
guarantees that the eigenspaces of Q either grow (Λi > 1) or shrink (Λi < 1) exponen-
th
tially fast. That
is, the norm any vector lying in the i eigenspace of Q will behave as
exp Λi (t − t0 ) , for t → ∞.
Note that while Q̂ = Q̂t is symmetric by construction, Q is simply a general real-valued

N × N matrix. The left and right eigenvectors of a matrix M ∈ GL(N, R) will in general
be different. The set of eigenvalues λα is, however, common to both sets of eigenvectors.
Let {ψα } be the right eigenvectors and {χ∗α } the left eigenvectors, such that
Mij ψα,j = λα ψα,i (16.64)

χ∗α,i Mij = λα χ∗α,j . (16.65)
We can always choose the left and right eigenvectors to be orthonormal, viz.
χα ψβ = χ∗α,i ψβ,j = δαβ .

(16.66)
−1
Indeed, we can define the matrix Siα = ψα,i , in which case Sαj = χ∗α,j , and
S −1 M S = diag λ1 , . . . , λN .

(16.67)
The matrix M can always be decomposed into its eigenvectors, as

X
Mij = λα ψα,i χ∗α,j . (16.68)
α
If we expand u in terms of the right eigenvectors,

X
η(t) = Cβ (t) ψβ (t) , (16.69)
β
4
ˆ ˜
If M1 , M2 = 0 then A = AP .
then upon taking the inner product with χα , we find that Cα obeys

Ċα + χα ψ̇β Cβ = λα Cα . (16.70)
If ψ̇β = 0, e.g. if M is time-independent, then Cα (t) = Cα (0) eλα t , and
Cα (0)
X zX }| {
ηi (t) = ηj (0) χ∗α,j eλα t ψα,i . (16.71)
α j
Thus, the component of η(t) along ψα increases exponentially with time if Re(λα ) > 0, and
decreases exponentially if Re(λα ) < 0.
Chapter 17
Bifurcations
17.1 Types of Bifurcations
17.1.1 Saddle-node bifurcation
We remarked above how f 0 (u) is in general nonzero when f (u) itself vanishes, since two
equations in a single unknown is an overdetermined set. However, consider the function
F (x, α), where α is a control parameter. If we demand F (x, α) = 0 and ∂x F (x, α) = 0,
we have two equations in two unknowns, and in general there will be a zero-dimensional
solution set consisting of points (xc , αc ). The situation is depicted in Fig. 17.1.
Let’s expand F (x, α) in the vicinity of such a point (xc , αc ):

1 ∂ 2F

∂F ∂F
F (x, α) = F (xc , αc ) + (x − xc ) + (α − αc ) + (x − xc )2
∂x ∂α 2 ∂x2
(xc ,αc ) (xc ,αc ) (xc ,αc )
2 2

∂F 1 ∂F
(α − αc )2 + . . .

+ (x − xc ) (α − αc ) + (17.1)
∂x ∂α 2 ∂α2
(xc ,αc ) (xc ,αc )
2
= A (α − αc ) + B (x − xc ) + . . . , (17.2)
where
p we keep terms of lowest order in√ the deviations δu and δr. If we now rescale u ≡
B/A (x − xc ), r ≡ α − αc , and τ = AB t, we have, neglecting the higher order terms,
we obtain the ‘normal form’ of the saddle-node bifurcation,
du
= r + u2 . (17.3)
dτ
The evolution of the

√ flow is depicted in Fig. 17.2.√ For r < 0 there are two fixed points –
one stable (u∗ = − −r) and one unstable (u = + −r). At r = 0 these two nodes coalesce
and annihilate each other. (The point u∗ = 0 is half-stable precisely at r = 0.) For r > 0
there are no longer any fixed points in the vicinity of u = 0. In Fig. 17.3 we show the flow
in the extended (r, u) plane. The unstable and stable nodes annihilate at r = 0.
347
348 CHAPTER 17. BIFURCATIONS
Figure 17.1: Evolution of F (x, α) as a function of the control parameter α.
17.1.2 Transcritical bifurcation
Another situation which arises frequently is the transcritical bifurcation. Consider the
equation ẋ = f (x) in the vicinity of a fixed point x∗ .
dx
= f 0 (x∗ ) (x − x∗ ) + 12 f 00 (x∗ )(x − x∗ )2 + . . . . (17.4)
dt
We rescale u ≡ β (x−x∗ ) with β = − 21 f 00 (x∗ ) and define r ≡ f 0 (x∗ ) as the control parameter,
to obtain, to order u2 ,
du
= ru − u2 . (17.5)
dt
Note that the sign of the u2 term can be reversed relative to the others by sending u → −u
and r → −r.
What happens in the transcritical bifurcation is an exchange of stability of the fixed points
Figure 17.2: The saddle-node bifurcation u̇ = r + u2 .

17.1. TYPES OF BIFURCATIONS 349
Figure 17.3: Flow diagram in (r, u) space for the saddle-node bifurcation u̇ = r + u2 .
at u∗ = 0 and u∗ = r as r passes through zero. This is depicted graphically in figs. 17.4

and 17.5.
Consider a crude model of a laser threshold. Let n be the number of photons in the laser
cavity, and N the number of excited atoms in the cavity. The dynamics of the laser are
approximated by the equations
ṅ = GN n − kn (17.6)
N = N0 − αn . (17.7)
Here G is the gain coefficient and k the photon decay rate. N0 is the pump strength, and α
is a numerical factor. The first equation tells us that the number of photons in the cavity
grows with a rate GN − k; gain is proportional to the number of excited atoms, and the
loss rate is a constant cavity-dependent quantity (typically through the ends, which are
Figure 17.4: The transcritical bifurcation u̇ = ru − u2 .

Figure 17.5: Extended phase space (r, u) flow diagram for the transcritical bifurcation.
semi-transparent). The second equation says that the number of excited atoms is equal to
the pump strength minus a term proportional to the number of photons (since the presence
of a photon means an excited atom has decayed). Putting them together,
ṅ = (GN0 − k) n − αGn2 , (17.8)
which exhibits a transcritical bifurcation at pump strength N0 = G/k. For N0 < G/k the
system acts as a lamp; for N0 > G/k the system acts as a laser.
17.1.3 Pitchfork bifurcation
The pitchfork bifurcation is commonly encountered in systems in which there is an overall

parity symmetry (u → −u). There are two classes of pitchfork - supercritical and subcritical.
The normal form of the supercritical bifurcation is
u̇ = ru − u3 , (17.9)
√
which has fixed points at u∗ = 0 and u∗ = ± r. Thus, the situation is as depicted in fig.
17.6 (top panel). For r < 0 there is a single stable fixed point at u∗ = 0. For r > 0, u∗ = 0
√
is unstable, and flanked by two stable fixed points at u∗ = ± r.
If we send u → −u, r → −r, and t → −t, we obtain the subcritical pitchfork , depicted in
fig. 17.6 (bottom panel). The fixed point structure in both cases is shown in Fig. 17.7.
Figure 17.6: Top: supercritical pitchfork bifurcation u̇ = ru − u3 . Bottom: subcritical

pitchfork bifurcation u̇ = ru + u3 .
17.1.4 Imperfect bifurcation
The imperfect bifurcation occurs when a symmetry-breaking term is added to the pitchfork.
The normal form contains two control parameters:
u̇ = h + ru − u3 . (17.10)
Here, the constant h breaks the parity symmetry if u → −u. This equation arises from
a crude model of magnetization dynamics. Let M be the magnetization of a sample, and
F (M ) the free energy. Assuming M is small, we can expand F (M ) as
F (M ) = −HM + 12 aM 2 + 14 bM 4 + . . . , (17.11)
where H is the external magnetic field, and a and b are temperature-dependent constants.
This is called the Landau expansion of the free energy. We assume b > 0 in order that the
minimum of F (M ) not lie at infinity. The dynamics of M (t) are modeled by
dM ∂F
= −Γ , (17.12)
dt ∂M
with Γ > 0. Thus, the magnetization evolves toward a local minimum in the free energy.
Note that the free energy is a decreasing function of time:
∂F 2

dF ∂F dM
= = −Γ . (17.13)
dt ∂M dt ∂M
Figure 17.7: Fixed points and their stabilities for the supercritical (left) and subcritical
(right) pitchfork bifurcations. Solid lines are for stable fixed points; dashed lines are unstable
fixed points.
By rescaling M ≡ αu with α = (b Γ)−1/2 and defining r ≡ −(Γ/b)1/2 and h ≡ (Γ3 b)1/2 H,

we obtain the normal form
∂f
u̇ = h + ru − u3 = − (17.14)
∂u
f (u) = − 12 ru2 + 14 u4 − hu . (17.15)
Here, f (u) is a scaled version of the free energy.
Fixed points satisfy the equation
u3 − ru − h = 0 , (17.16)
and correspond to extrema in f (u). By the fundamental theorem of algebra, this cubic
polynomial may be uniquely factorized over the complex plane. Since the coefficients are
real, the complex conjugate ū satisfies the same equation as u, hence there are two possi-
bilities for the roots: either (i) all three roots are real, or (ii) one root is real and the other
two are a complex conjugate pair. Clearly for r < 0 we are in situation (ii), since u3 − ru
is then monotonically increasing for u ∈ R, and therefore takes the value h precisely once
for u real. For r > 0, there is a region h ∈ − hc (r), hc (r) over which there are three real
roots. To find hc (r), we demand f 00 (u) = 0 as well as f 0 (u) = 0, which says that two roots
2
have merged, forming an inflection point. One easily finds hc (r) = 3√ 3
r3/2 .
Examples of the function f (u) for r > 0 are shown in Fig. 17.8 for three different values
of h. For |h| < hc (r) there are three extrema satisfying f 0 (u∗ ) = 0: u∗1 < u∗2 < 0 < u∗3 ,
assuming (without loss of generality) that h > 0. Clearly u∗1 is a local minimum, u∗2 a local
maximum, and u∗3 the global minimum of the function f (u). The ‘phase diagram’ for this
system, plotted in the (r, h) control parameter space, is shown in Fig. 17.9.
Figure 17.8: Scaled free energy f (u) = − 12 ru2 + 1/4u4 − hu. Green curve: h = hc (r) =
√ √
2
√ r3/2 . Brown curve: h = √2 r3/2 . Blue curve: h < √2 r3/2 .
3 3 3 3 3 3
In Fig. 17.10 we plot the fixed points u∗ (r) for fixed h. A saddle-node bifurcation occurs
3
at r = rc (h) = 22/3 |h|2/3 . For h = 0 this reduces to the supercritical pitchfork; for finite h
the pitchfork is deformed and even changed topologically. Finally, in Fig. 17.11 we show
the behavior of u∗ (h) for fixed r. When r < 0 the curve retraces itself as h is ramped up
and down, but for r > 0 the system exhibits the phenomenon of hysteresis, i.e. there is an
irreversible aspect to the behavior. Fig, 17.11 shows a hysteresis loop when r > 0.
Figure 17.9: Phase diagram for the imperfect bifurcation u̇ = h + ru − u3 .

Figure 17.10: u∗ (r) at fixed h for the imperfect bifurcation. This is in a sense a deformed
supercritical pitchfork.
17.2 Examples
17.2.1 Magnetization dynamics
A magnetic system has a free energy
F (M ) = 1
2 a M 2 − 13 c M 3 + 14 b M 4 ,
Figure 17.11: u∗ (h) at fixed r for the imperfect bifurcation. For r < 0 the behavior is
completely reversible. For r > 0, a regime of irreversibility sets in between −hc (r) and
+hc (r).
17.2. EXAMPLES 355
Figure 17.12: Fixed point diagram for ṁ = r m + m2 − m3 . Solid blue portions are stable
fixed points; dashed red portions are unstable fixed points. A saddle-node bifurcation sets
in at r = − 41 . and a transcritical bifurcation at r = 0.
with b and c both positive. The evolution of the magnetization is given by

dM ∂F
= −Γ ,
dt ∂M
where Γ is a positive constant. Note that F (M ) is a Lyapunov function for the magnetiza-
tion dynamics.
(a) Show that M and t can be rescaled to dimensionless variables m and τ , respectively,
such that
dm ∂f
=− , (17.17)
dτ ∂m
where
f (m, r) = − 12 r m2 − 13 m3 + 14 m4 . (17.18)
Provide expressions for r, m, and τ .
Solution : Write M = α m and t = β τ , and obtain

dm
= −Γ aβ m + Γ αβ c m2 − Γ α2 β b m3 (17.19)
dτ
= r m + m2 − m3 , (17.20)
which gives Γ aβ = −r, Γ αβ c = 1, and Γ α2 β b = 1. Thus,

c b ab
α= , β= , r=− . (17.21)
b Γ c2 c2
Figure 17.13: Sketch of the potential f (m) = − 12 r m2 − 13 m3 + 14 m4 for r = − 12 (red),

r = −1/4 (magenta), r = − 29 (yellow), r = − 16 (green), and r = + 100
1
(blue).
(b) Sketch the set of fixed points m∗ as a function of the control parameter r. Label stable
and unstable branches. Identify and classify all bifurcations.
Solution : The fixed points are solutions of
∂f
= (m2 − m − r) m = 0 . (17.22)
∂m
Clearly m = 0 always is a solution. There are two other roots,
√
m± = 12 1 ± 1 + 4r . (17.23)
For r < − 14 , these roots are complex. At r = − 14 there is a saddle-node bifurcation, and
two new roots appear on either side of m = 21 . At r = 0, there is a transcritical bifurcation.
See fig. 17.12.
(c) Sketch the free energy f (m, r) versus m at (at least) k + 1 values of r, where k is the
number of bifurcations you found in (b). Choose representative values of r to the left and
right of each bifurcation.
Solution : Sketches of f (m, r) are shown in fig. 17.13. At the saddle-node bifurcation,
r = − 14 , m = 12 , and one finds f = 192
1
> 0. Setting f (m) = 0 and f 0 (m) = 0 gives r = − 29 .
Thus, for r < − 29 the global minimum of f (m, r) occurs at m = 0. For − 41 < r < − 29 , there
is another local minimum at m = m+ , and a local maximum at m = m− . For − 29 < r < 0,
the global minimum lies at m = m+ , and m = 0 is a local minimum. Finally, at r = 0 there
17.2. EXAMPLES 357
Figure 17.14: Sketch of phase flow for ṁ = r(t) m + m2 − m3 as r is slowly ramped up and
down, starting from an initial value r(0) < − 14 , and with two initial conditions on m. Blue
arrows: m(0) > 0; red arrows: m(0) < 0.
is a transcritical bifurcation, and for r > 0 there is a local minimum at m = m− , a global

minimum at m = m+ , and a local maximum at m = 0.
(d) Suppose r is slowly increased starting from a large negative value (to the left of all
bifurcations) to a large positive value (to the right of all bifurcations). If m initially is
positive, sketch the hysteresis curve and identify any regions of irreversible behavior. Next,
assume that m initially is negative, and do the same. Explain your results. You may assume
that m flows quickly toward toe first stable fixed point encountered.
Solution : The phase flows are sketched in fig. 17.14. Note that hysteresis occurs for
m(0) > 0, and there is no hysteresis for m(0) < 0. This is because the initial flow brings
m close to, but still on the original side of, the stable fixed point m = 0. Incidentally, the
flow to m = m+ for r > 0 may proceed slowly at first, since m = 0 is a fixed point. If mhas
approached very close to m = 0 while r < 0, it may take a long while for m to ‘roll off’
m = 0 once r crosses the transcritical bifurcation at r = 0.
17.2.2 Population dynamics
Consider the dynamics of a harvested population,

N
Ṅ = rN 1 − − H(N ) , (17.24)
K
where r, K > 0, and where H(N ) is the harvesting rate.

Figure 17.15: Phase flow for the constantly harvested population. If the harvesting rate H0
is too large, the population always shrinks. The region N < 0 is unphysical.
(a) Suppose H(N ) = H0 is a constant. Sketch the phase flow, and identify and classify all
fixed points.
Solution : We examing Ṅ = f (N ) with

r N2
f (N ) = + r N − H0 . (17.25)
K
Setting f 0 (N ) = 0 yields N = 12 K. f (N ) is a downward-opening parabola whose maximum

value is f 12 K = 14 r K − H0 . Thus, if H0 > 14 r K, the harvesting rate is too large and the

population always shrinks. A saddle-node bifurcation occurs at this value of H0 , and for
larger harvesting rates, there are fixed points at
r
4H0
N± = 21 K ± 12 K 1 − , (17.26)
rK
with N− unstable and N+ stable. See fig. 17.15.
(b) One defect of the constant harvesting rate model is that N = 0 is not a fixed point. To
remedy this, consider the following model for H(N ):
B N2
H(N ) = , (17.27)
N 2 + A2
where A and B are (positive) constants. Show that one can rescale (N, t) to (n, τ ), such
that
dn n n2
=γn 1− − 2 , (17.28)
dτ c n +1
17.2. EXAMPLES 359
Figure 17.16: Plot 2

n
of h(n) = n/(n +1) (thick black curve). Straight lines show the function
y(n) = γ 1 − c for different values of c and γ. The red line is tangent to the inflection
point of h(n). See the analysis in the text.
where γ and c are positive constants. Provide expressions for n, τ , γ, and c.
Solution : Examining the denominator of H(N ), we must take N = An. Dividing both
sides of Ṅ = f (N ) by B, we obtain
A dN rA A n2
= n 1− n − 2 ,
B dt B K n +1
from which we glean τ = Bt/A, γ = rA/B, and c = K/A.
(c) Show that for c sufficiently small that there is a unique asymptotic (τ → ∞) value for
the (scaled) population n, for any given value of γ. Thus, there are no bifurcations as a
function of the control parameter γ for c fixed and c < c∗ .
(d) Show that for c > c∗ , there are two bifurcations as a function of γ, and that for γ1∗ < γ <
γ2∗ the asymptotic solution is bistable, i.e. there are two stable values for n(τ → ∞). Sketch
the solution set ‘phase diagram’ in the (c, γ) plane. Hint: Sketch the functions γ(1 − n/c)
and n/(n2 + 1). The n 6= 0 fixed points are given by the intersections of these two curves.
Determine the boundary of the bistable region in the (c, γ) plane parametrically in terms
of n. Find c∗ and γ1∗ (c) = γ2∗ (c).
Figure 17.17: Phase diagram for the equation ṅ = γ(1 − n/c) n − n2 /(n2 + 1), labeling n 6= 0
fixed points. (The point n = 0 is always unstable.)
Solution (c) and (d) : We examine

( )
dn n n
= g(n) = γ 1− − 2 n. (17.29)
dτ c n +1
There is an unstable fixed point at n = 0, where g 0 (0) = γ > 0. The other fixed points
occur when the term in the curvy brackets vanishes. In fig. 17.16 we plot the function
h(n) ≡ n/(n2 + 1) versus n. We seek the intersection of this function with a two-parameter
family of straight lines, given by y(n) = γ (1 − n/c). The n-intercept is c and the y-intercept
is γ. Provided c is large enough, there are two bifurcations as a function of γ, which we call
γ± (c). These are shown as the blue and green lines in figure 17.16.
Both bifurcations are of the saddle-node type. We determine the curves γ± (c) by requiring
that h(n) is tangent to y(n), which gives two equations:S
n n
h(n) = 2 =γ 1− = y(n) (17.30)
n +1 c
1 − n2 γ
h0 (n) = 2 2
= − = y 0 (n) . (17.31)
(n + 1) c
Together, these give γ(c) parametrically, i.e. as γ(n) and c(n):
2n3 2n3
γ(n) = , c(n) = . (17.32)
(n2 + 1)2 (n2 − 1)
Since h(n) is maximized for n = 1, where h(1) = 12 , there is no bifurcation occurring at

values n < 1. If we plot γ(n) versus c(n) over the allowed range of n, we obtain the phase
17.3. APPENDIX I : THE BLETCH 361
diagram in fig. 17.17. The cusp occurs at (c∗ , γ ∗ ), and is determined by the requirement
that the two bifurcations coincide. This supplies a third condition, namely that h0 (n) = 0,
where
2n (n2 − 3)
h00 (n) = . (17.33)
(n2 + 1)3
√ √ √
Hence n = 3, which gives c∗ = 3 3 and γ ∗ = 3 8 3 . For c < c∗ , there are no bifurcations
at any value of γ.
17.3 Appendix I : The Bletch
Problem: The bletch is a disgusting animal native to the Forest of Jkroo on the planet
Barney. The bletch population obeys the equation
dN
= aN 2 − bN 3 , (17.34)
dt
where N is the number of bletches, and a and b are constants. (Bletches reproduce asexually,
but only when another bletch is watching. However, when there are three bletches around,
they beat the @!!*$&* out of each other.)
(a) Sketch the phase flow for N . (Strange as the bletch is, you can still rule out N < 0.)
Identify and classify all fixed points.
(b) The bletch population is now harvested (they make nice shoes). To model this, we
add an extra term to the dynamics:
dN
= −hN + aN 2 − bN 3 , (17.35)
dt
where h is the harvesting rate. Show that the phase flow now depends crucially on h,
in that there are two qualitatively different flows, depending on whether h < hc (a, b)
or h > hc (a, b). Find the critical harvesting rate hc (a, b) and sketch the phase flows
for the two different regimes.
(c) In equilibrium, the rate at which bletches are harvested is R = hN ∗ , where N ∗ is

the equilibrium bletch population. Suppose we start with h = 0, in which case N ∗
is given by the value of N at the stable fixed point you found in part (a). Now let
h be increased very slowly from zero. As h is increased, the equilibrium population
changes. Sketch R versus h. What value of h achieves the biggest bletch harvest?
What is the corresponding value of Rmax ?
Figure 17.18: Phase flow for the scaled bletch population, ṅ = n2 − n3 .
Solution:
(a) Setting the RHS of eqn. 17.34 to zero suggests the rescaling
a b
N= n , t= τ . (17.36)
b a2
This results in
dn
= n2 − n3 . (17.37)
dτ
The point n = 0 is a (nonlinearly) repulsive fixed point, and n = 1, corresponding to
N = a/b, is attractive. The flow is shown in fig. 17.18.
By the way, the dynamics can be integrated, using the method of partial fractions, to
yield
1 1 n 1 − n0
− + ln · =τ . (17.38)
n0 n n0 1 − n
(b) Upon rescaling, the harvested bletch dynamics obeys the equation
dn
= −νn + n2 − n3 , (17.39)
dτ
where ν = bh/a2 is the dimensionless harvesting rate. Setting the RHS to zero yields
n(n2 − n + ν) = 0, with solutions n∗ = 0 and
√
n∗± = 12 ± 1/4 − ν . (17.40)
Thus, for ν > 1/4 the only fixed point (for real n) is at n∗ = 0 (stable) – the bletch
population is then overharvested . For ν > 1/4, there are three solutions: a stable fixed
√
point at n∗ = 0, an unstable fixed point at n∗ = 12 − 1/4 − ν, and a stable fixed
√
point at n∗ = 21 + 1/4 − ν. The critical harvesting rate is νc = 1/4, which means
hc = a2 /4b.
Figure 17.19: Phase flow for the harvested bletch population, ṅ = −νn + n2 − n3 .
a3
(c) The scaled bletch harvest is given by r = ν n∗+ (ν). Note R = h N+∗ = b2
r. The
optimal harvest occurs when ν n∗ is a maximum, which means we set
√

d 1
ν + ν 1/4 − ν = 0 =⇒ νopt = 29 . (17.41)
dν 2
17.4. APPENDIX II : LANDAU THEORY OF PHASE TRANSITIONS 363
Figure 17.20: Scaled bletch harvest r versus scaled harvesting rate ν. Optimal harvesting
occurs for νopt = 29 . The critical harvesting rate is νc = 14 , at which point the harvest
discontinuously drops to zero.
Thus, n∗+ (νopt ) = 23 and ropt = 274

, meaning R = 4a3 /27 b2 . Note that at ν = νc = 1/4
that n∗+ (νc ) = 21 , hence r(νc ) = 18 , which is smaller than (νopt ) = 23 . The harvest r(ν)
discontinuously drops to zero at ν = νc , since for ν > νc the flow is to the only stable
fixed point at n∗ = 0.
17.4 Appendix II : Landau Theory of Phase Transitions
Landau’s theory of phase transitions is based on an expansion of the free energy of a

thermodynamic system in terms of an order parameter , which is nonzero in an ordered
phase and zero in a disordered phase. For example, the magnetization M of a ferromagnet
in zero external field but at finite temperature typically vanishes for temperatures T > Tc ,
where Tc is the critical temperature, also called the Curie temperature in a ferromagnet.
A low order expansion in powers of the order parameter is appropriate sufficiently close to
the phase transition, i.e. at temperatures such that the order parameter, if nonzero, is still
small.
The simplest example is the quartic free energy,
f (m, h = 0, t) = f0 + 12 am2 + 1/4bm4 , (17.42)
where f0 = f0 (t), a = a(t), and b = b(t). Here, t is a dimensionless measure of the

temperature. If for example the local exchange energy in the ferromagnet is J, then we
might define t = kB T /J. Let us assume b > 0, which is necessary if the free energy is to be
bounded from below1 . The equation of state ,

∂f
= 0 = am + bm3 , (17.43)
∂m
p
has three solutions in the complex m plane: (i) m = 0, and (ii/iii) ± −a/b . The latter
two solutions lie along the (physical) real axis if a < 0. We assume that there exists a
unique temperature tc where a(tc ) = 0. Minimizing f , we find
a2
t < tc : f (t) = f0 − (17.44)
4b
t > tc : f (t) = f0 . (17.45)
The free energy is continuous at tc since a(tc ) = 0. The specific heat, however, is discontin-
uous across the transition, with
2
tc a0 (tc )

∂ 2
2
+
−
a
c tc − c tc = −tc 2 =− . (17.46)
∂t t=tc 4b 2b(tc )
17.4.1 Cubic terms in Landau theory : first order transitions
Next, consider a free energy with a cubic term,
f = f0 + 12 am2 − 13 ym3 + 1/4bm4 , (17.47)
with b > 0 for stability. Without loss of generality, we may assume y > 0 (else send
m → −m). Note that we no longer have m → −m, or ZZ2 , symmetry. The cubic term
favors positive m. What is the phase diagram in the (a, y) plane?
Extremizing the free energy with respect to m, we obtain

∂f
− 0 = am − ym2 + bm3 . (17.48)
∂m
This cubic equation factorizes into a linear and quadratic piece, and hence may be solved
simply. The three solutions are m = 0 and
r
y y 2 a
m = m± ≡ ± − . (17.49)
2b 2b b
We now see that for y 2 < 4ab there is only one real solution, at m = 0, while for y 2 > 4ab
there are three real solutions. Which solution has lowest free energy? To find out, we
compare the energy f (0) with f (m+ )2 . Thus, we set
2
f (m) = f (0) =⇒ 1
2 am − 13 ym3 + 1/4bm4 = 0 , (17.50)
1
It is always the case that f is bounded from below, on physical grounds. Were b negative, we’d have to
consider higher order terms in the Landau expansion.
2
We needn’t waste our time considering the m = m− solution, since the cubic term prefers positive m.
Indeed, it is obvious that f (m− ) > f (0).
Figure 17.21: Behavior of the quartic free energy f (m) = 12 am2 − 13 ym3 + 1/4bm4 . A:
y 2 < 4ab ; B: 4ab < y 2 < 29 ab ; C and D: y 2 > 92 ab. The thick black line denotes a line of
first order transitions, where the order parameter is discontinuous across the transition.
and we now have two quadratic equations to solve simultaneously:
0 = a − ym + bm2 (17.51)
0 = 12 a − 13 ym + 1/4bm2 = 0 . (17.52)
Eliminating the quadratic term gives m = 3a/y. Finally, substituting m = m+ gives us a

relation between a, b, and y:
y 2 = 92 ab . (17.53)
Thus, we have the following:
y2
a> : 1 real root m = 0
4b
y 2 2y 2
>a> : 3 real roots; minimum at m = 0
4b 9b r
2y 2 y y 2 a
>a : 3 real roots; minimum at m = + −
9b 2b 2b b
The solution m = 0 lies at a local minimum of the free energy for a > 0 and at a local
2 2
maximum for a < 0. Over the range y4b > a > 2y 9b , then, there is a global minimum at
Figure 17.22: Behavior of the sextic free energy f (m) = 12 am2 + 1/4bm4 + 16 cm6 . A: a > 0
√
and b > 0 ; B: a < 0 and b > 0 ; C: a < 0 and b < 0 ; D: a > 0 and b < − √43 ac ; E: a > 0
√ √ √
and − √43 ac < b < −2 ac ; F: a > 0 and −2 ac < b < 0. The thick dashed line is a line
of second order transitions, which meets the thick solid line of first order transitions at the
tricritical point, (a, b) = (0, 0).
m = 0, a local minimum at m = m+ , and a local maximum at m = m− , with m+ > m− > 0.

2
For 2y
9b > a > 0, there is a local minimum at a = 0, a global minimum at m = m+ , and
a local maximum at m = m− , again with m+ > m− > 0. For a < 0, there is a local
maximum at m = 0, a local minimum at m = m− , and a global minimum at m = m+ , with
m+ > 0 > m− . See fig. 17.21.
17.4.2 Sixth order Landau theory : tricritical point
Finally, consider a model with ZZ2 symmetry, with the Landau free energy
f = f0 + 12 am2 + 1/4bm4 + 16 cm6 , (17.54)
with c > 0 for stability. We seek the phase diagram in the (a, b) plane. Extremizing f with
respect to m, we obtain
∂f
= 0 = m (a + bm2 + cm4 ) , (17.55)
∂m
which is a quintic with five solutions over the complex m plane. One solution is obviously
m = 0. The other four are
v s
u
u b b 2 a
m=± − ± − . (17.56)
t
2c 2c c
For each ± symbol in the above equation, there are two options, hence four roots in all.
If a > 0 and b > 0, then four of the roots are imaginary and there is a unique minimum at
m = 0.
For a < 0, there are only three solutions to f 0 (m) = 0 for real m, since the − choice for
the ± sign under the radical leads to imaginary roots. One of the solutions is m = 0. The
other two are s r
b b 2 a
m=± − + − . (17.57)
2c 2c c
√
The most interesting situation is a > 0 and b < 0. If a > 0 and b < −2 ac, all five roots
are real. There must be three minima, separated by two local maxima. Clearly if m∗ is a
solution, then so is −m∗ . Thus, the only question is whether the outer minima are of lower
energy than the minimum at m = 0. We assess this by demanding f (m∗ ) = f (0), where
m∗ is the position of the largest root (i.e. the rightmost minimum). This gives a second
quadratic equation,
0 = 21 a + 1/4bm2 + 16 m4 , (17.58)
which together with equation 17.55 gives
√
b = − √43 ac . (17.59)
Thus, we have the following, for fixed a > 0:

√
b > −2 ac : 1 real root m = 0
√ √
−2 ac > b > − √43 ac : 5 real roots; minimum at m = 0
s r
√ b b 2 a
− √43 ac > b : 5 real roots; minima at m = ± − + −
2c 2c c
The point (a, b) = (0, 0), which lies at the confluence of a first order line and a second order
line, is known as a tricritical point.
17.4.3 Hysteresis for the sextic potential
Consider the sextic potential of eqn. 17.54 with b < 0 and c > 0, and endowed with the
dynamics
∂f
γ ṁ = − . (17.60)
∂m
Figure 17.23: Free energy f (u) = − 12 ru2 − 1/4u4 + 16 u6 at different values of the control
parameter r.
By a suitable rescaling of the magnetization m and the time t, we can write this equation
as
u̇ = −f 0 (u) = ru + u3 − u5 , (17.61)
where the free energy as a function of the scaled order parameter is
f (u) = − 12 ru2 − 1/4u4 + 61 u6 . (17.62)
The fixed points satisfy f 0 (u) = (u4 − u2 − r) u = 0. The roots of the quadratic equation
√
are at u2 = 12 ± r + 1/4 . Thus, recapitulating our earlier results,
r < −1/4 : u∗ = 0
∗ ∗
q √ ∗
q √
−1/4 < r < 0 : u = 0 , u = ± 2 + 1/4 + r , u = ± 12 − 1/4 + r
1
∗ ∗
q √
r>0 : u = 0 , u = ± 12 + 1/4 + r
The situation is depicted in fig. 17.23.
The fixed point structure from these equations is depicted in fig. 17.24. When the control
parameter r is large and negative, the flow is toward the sole fixed point at u∗ =0, which
is stable. At r = −1/4, two simultaneous saddle-node bifurcations take place at u∗ = ± √12 ;
the stable branch is the outer branch in both cases. At r = 0 there is a subcritical pitchfork
bifurcation, and the fixed point at u∗ = 0 becomes unstable.
Suppose one starts off with r − 41 with some value u > 0. The flow u̇ = −f 0 (u) then
rapidly results in u → 0+ . This is the ‘high temperature phase’ in which there is no
Figure 17.24: Fixed points f 0 (u∗ ) = 0 with stable fixed points in solid blue and unstable
fixed points in red dashed curves. The flow u̇ = −f 0 (u) is also shown, with f (u) given in
eqn. 17.62.
magnetization. Now let r increase slowly. The scaled magnetization u(t) = u∗ r(t) will

remain pinned at the fixed point u∗ = 0+ . By 0+ , we mean that u is a positive infinitesimal.

As r passes −1/4, two new stable values of u∗ appear, but our system remains at u = 0+ ,
since u∗ = 0 is a stable fixed point. But after the subcritical pitchfork, u∗ = 0 becomes
1/2
unstable. The magnetization u(t) then rapidly flows to u∗ (r) = 21 + (1/4 + r)1/2 , and
follows this curve for all r > 0.
Now suppose we start decreasing r (increasing temperature). The magnetization follows

1/2
the stable fixed point u∗ (r) = 21 + (1/4 + r)1/2 past r = 0, and all the way down to
r = −1/4, at which point this fixed point is annihilated at a saddle-node bifurcation. The
flow then rapidly takes u → u∗ = 0+ , where it remains as r continues to be lowered further.
The hysteresis loop for this system is shown in fig. 17.25.
Figure 17.25: Hysteresis loop (in magenta arrows) under slow changes of the control pa-
rameter r for the dynamics u̇ = −f 0 (u), where f (u) is the sextic potential of eqn. 17.62,
assuming u > 0. An equivalent loop exists upon reflection u → −u about the r axis.
Chapter 18
Two-Dimensional Phase Flows
We’ve seen how, for one-dimensional dynamical systems u̇ = f (u), the possibilities in terms
of the behavior of the system are in fact quite limited. Starting from an arbitrary initial
condition u(0), the phase flow is monotonically toward the first stable fixed point encoun-
tered. (That point may lie at infinity.) No oscillations are possible1 . For N = 2 phase
flows, a richer set of possibilities arises, as we shall now see.
18.1 Harmonic Oscillator and Pendulum
18.1.1 Simple harmonic oscillator
A one-dimensional harmonic oscillator obeys the equation of motion,
d2 x
m = −kx , (18.1)
dt2
where m is the mass and k the force constant (of a spring). If we define v = ẋ, this may be
written as the N = 2 system,

d x 0 1 x v
= = , (18.2)
dt v −Ω 2 0 v −Ω 2 x
p
where Ω = k/m has the dimensions of frequency (inverse time). The solution is well
known:
v0
x(t) = x0 cos(Ωt) + sin(Ωt) (18.3)
Ω
v(t) = v0 cos(Ωt) − Ω x0 sin(Ωt) . (18.4)
1
If phase space itself is multiply connected, e.g. a circle, then the system can oscillate by moving around
the circle.
371
372 CHAPTER 18. TWO-DIMENSIONAL PHASE FLOWS
Figure 18.1: Phase curves for the harmonic oscillator.
The phase curves are ellipses:
Ω x2 (t) + Ω −1 v 2 (t) = C , (18.5)
where the constant C = Ω x20 + Ω −1 v02 . A sketch of the phase curves and of the phase flow
is shown in Fig. 18.1. Note that the x and v axes have different dimensions. Note also that
the origin is a fixed point, however, unlike the N = 1 systems studied in the first lecture,
here the phase flow can avoid the fixed points, and oscillations can occur.
Incidentally, eqn. 18.2 is linear, and may be solved by the following method. Write the
equation as ϕ̇ = M ϕ, with

x 0 1
ϕ= and M= (18.6)
ẋ −Ω 2 0
The formal solution to ϕ̇ = M ϕ is
ϕ(t) = eM t ϕ(0) . (18.7)
What do we mean by the exponential of a matrix? We mean its Taylor series expansion:
eM t = 11 + M t + 1
2! M 2 t2 + 1
3! M 3 t3 + . . . . (18.8)
Note that

0 1 0 1
M2 =
−Ω 2 0 −Ω 2 0
−Ω 2

0
= = −Ω 2 11 , (18.9)
0 −Ω 2
hence
M 2k = (−Ω 2 )k 11 , M 2k+1 = (−Ω 2 )k M . (18.10)
18.1. HARMONIC OSCILLATOR AND PENDULUM 373
Thus,
∞ ∞
Mt
X 1 X 1
e = (−Ω 2 t2 )k · 11 + (−Ω 2 t2 )k · M t
(2k)! (2k + 1)!
k=0 k=0
= cos(Ωt) · 11 + Ω −1 sin(Ωt) · M
Ω −1 sin(Ωt)

cos(Ωt)
= . (18.11)
−Ω sin(Ωt) cos(Ωt)
Plugging this into eqn. 18.7, we obtain the desired solution.
For the damped harmonic oscillator, we have

2 0 1
ẍ + 2β ẋ + Ω x = 0 =⇒ M= 2 . (18.12)
−Ω −2β
The phase curves then spiral inward to the fixed point at (0, 0).
18.1.2 Pendulum
Next, consider the simple pendulum, composed of a mass point m affixed to a massless rigid
rod of length `.
m`2 θ̈ = −mg` sin θ . (18.13)
This is equivalent to

d θ ω
= , (18.14)
dt ω −Ω 2 sin θ
p
where ω = θ̇ is the angular velocity, and where Ω = g/` is the natural frequency of small
oscillations.
The phase curves for the pendulum are shown in Fig. 18.2. The small oscillations of the
pendulum are essentially the same as those of a harmonic oscillator. Indeed, within the
small angle approximation, sin θ ≈ θ, and the pendulum equations of motion are exactly
those of the harmonic oscillator. These oscillations are called librations. They involve
a back-and-forth motion in real space, and the phase space motion is contractable to a
point, in the topological sense. However, if the initial angular velocity is large enough, a
qualitatively different kind of motion is observed, whose phase curves are rotations. In this
case, the pendulum bob keeps swinging around in the same direction, because, as we’ll see
in a later lecture, the total energy is sufficiently large. The phase curve which separates
these two topologically distinct motions is called a separatrix .
Figure 18.2: Phase curves for the simple pendulum. The separatrix divides phase space
into regions of vibration and libration.
18.2 General N = 2 Systems
The general form to be studied is

!
d x Vx (x, y)
= . (18.15)
dt y Vy (x, y)
Special cases include autonomous second order ODEs, viz.

d x v
ẍ = f (x, ẋ) =⇒ = , (18.16)
dt v f (x, v)
of the type which occur in one-dimensional mechanics.
18.2.1 The damped driven pendulum
Another example is that of the damped and driven harmonic oscillator,

d2φ dφ
2
+γ + sin φ = j . (18.17)
ds ds
This is equivalent to a model of a resistively and capacitively shunted Josephson junction,
depicted in fig. 18.3. If φ is the superconducting phase difference across the junction, the
18.2. GENERAL N = 2 SYSTEMS 375
Figure 18.3: . The resistively and capacitively shunted Josephson junction. The Josephson
junction is the X element at the bottom of the figure.
current through the junction is given by IJ = Ic sin φ, where Ic is the critical current.
The current carried by the resistor is IR = V /R from Ohm’s law, and the current from
the capacitor is IC = Q̇. Finally, the Josephson relation relates the voltage V across the
junction to the superconducting phase difference φ: V = (h̄/2e) φ̇. Summing up the parallel
currents, we have that the total current I is given by
h̄C h̄
I= φ̈ + φ̇ + Ic sin φ , (18.18)
2e 2eR
which, again, is equivalent to a damped, driven pendulum.
This system also has a mechanical analog. Define the ‘potential’

U (φ) = −Ic cos φ − Iφ . (18.19)
The equation of motion is then
h̄C h̄ ∂U
φ̈ + φ̇ = − . (18.20)
2e 2eR ∂φ
Thus, the combination h̄C/2e plays the role of the inertial term (mass, or moment of inertia),
while the combination h̄/2eR plays the role of a damping coefficient. The potential U (φ)
is known as the tilted washboard potential , for obvious reasons. (Though many of you have
perhaps never seen a washboard.)
The model is adimensionalized by defining the Josephson plasma frequency ωp and the RC
time constant τ : r
2eIc
ωp ≡ , τ ≡ RC . (18.21)
h̄C
The dimensionless combination ωp τ then enters the adimensionalized equation as the sole
control parameter:
I d2φ 1 dφ
= 2+ + sin φ , (18.22)
Ic ds ωp τ ds
where s = ωp t. In the Josephson junction literature, the quantity β ≡ 2eIc R2 C/h̄ = (ωp τ )2 ,
known as the McCumber-Stewart parameter, is a dimensionless measure of the damping
(large β means small damping). In terms of eqn. 18.17, we have γ = (ωp τ )−1 and j = I/Ic .
We can write the second order ODE of eqn. 18.17 as two coupled first order ODEs:

d φ ω
= , (18.23)
dt ω j − sin φ − γ ω
where ω = φ̇. Phase space is a cylinder, S1 × R1 .
The quantity ωp τ typically ranges from 10−3 to 103 in Josephson junction applications.
If ωp τ is small, then the system is heavily damped, and the inertial term d2φ/ds2 can be
neglected. One then obtains the N = 1 system
dφ
γ = j − sin φ . (18.24)
ds
If |j| < 1, thenpφ(s) evolves to the first stable fixed point encountered, where φ∗ = sin−1 (j)
and cos φ∗ = 1 − j 2 . Since φ(s) → φ∗ is asymptotically a constant, the voltage drop V
must then vanish, as a consequence of the Josephson relation V = (h̄/2e) φ̇. This, there is
current flowing with no voltage drop!
If |j| > 1, the RHS never vanishes, in which case φ(s) is monotonic. We then can integrate
the differential equation
h̄ dφ
dt = · . (18.25)
2eR I − Ic sin φ
Asymptotically the motion is periodic, with the period T obtained by integrating over the
interval φ ∈ [0, 2π]. One finds
h̄ 2π
T = ·p . (18.26)
2eR I 2 − Ic2
The time-averaged voltage drop is then
h̄ h̄ 2π p
hV i = hφ̇i = · = R I 2 − Ic2 . (18.27)
2e 2e T
This is the physics of the current-biased resistively and capacitively shunted Josephson junc-
tion in the strong damping limit. It is ‘current-biased’ because we are specifying the current
I. Note that Ohm’s law is recovered at large values of I.
For general ωp τ , we can still say quite a bit. At a fixed point, both components of the
vector field V (φ, ω) must vanish. This requires ω = 0 and j = sin φ. Therefore, there are
two fixed points for |j| < 1, one a saddle point and the other a stable spiral. For |j| > 1
there are no fixed points, and asymptotically the function φ(t) tends to a periodic limit
cycle φLC (t). The flow is sketched for two representative values of j in Fig. 18.4.
Figure 18.4: Phase flows for the equation φ̈ + γ −1 φ̇ + sin φ = j. Left panel: 0 < j < 1;
note the separatrix (in black), which flows into the stable and unstable fixed points. Right
panel: j > 1. The red curve overlying the thick black dot-dash curve is a limit cycle.
18.2.2 Classification of N = 2 fixed points
Suppose we have solved the fixed point equations Vx (x∗ , y ∗ ) = 0 and Vy (x∗ , y ∗ ) = 0. Let us
now expand about the fixed point, writing

∂Vx ∂Vx
ẋ = (x − x∗ ) + (y − y ∗ ) + . . . (18.28)
∂x ∂y
(x∗ ,y ∗ ) (x∗ ,y ∗ )

∂Vy
∂Vy

ẏ = (x − x∗ ) + (y − y ∗ ) + . . . . (18.29)
∂x ∂y
(x∗ ,y ∗ ) (x∗ ,y ∗ )
We define
u1 = x − x∗ , u2 = y − y ∗ , (18.30)
which, to linear order, satisfy
M
 z }| {  


u1 a b u1
d       + O(u2 ) .
= (18.31)
dt
u2 c d u2
Figure 18.5: Complete classification of fixed points for the N = 2 system.
The formal solution to u̇ = M u is
u(t) = exp(M t) u(0) , (18.32)

P∞ 1
where exp(M t) = n=0 n! (M t)n is the exponential of the matrix M t.
The behavior of the system is determined by the eigenvalues of M , which are roots of the
characteristic equation P (λ) = 0, where
P (λ) = det(λ1 − M )
= λ2 − T λ + D , (18.33)
with T = a + d = Tr(M ) and D = ad − bc = det(M ). The two eigenvalues are therefore

p
λ± = 12 T ± T 2 − 4D . (18.34)
To see why the eigenvalues control the behavior, let us expand u(0) in terms of the eigen-
vectors of M . Since M is not necessarily symmetric, we should emphasize that we expand
u(0) in terms of the right eigenvectors of M , which satisfy
M ψa = λ a ψa , (18.35)
where the label a runs over the symbols + and −, as in (18.34). We write
X
u(t) = Ca (t) ψa . (18.36)
a
Since (we assume) the eigenvectors are linearly independent, the equation u̇ = M u becomes
Ċa = λa Ca , (18.37)
Figure 18.6: Fixed point zoo for N = 2 systems. Not shown: unstable versions of node,
spiral, and star (reverse direction of arrows to turn stable into unstable).
with solution
Ca (t) = eλa t Ca (0) . (18.38)
Thus, the coefficients of the eigenvectors ψa will grow in magnitude if |λa | > 1, and will
shrink if |λa | < 1.
18.2.3 The fixed point zoo
• Saddles : When D < 0, both eigenvalues are real; one is positive and one is negative,
i.e. λ+ > 0 and λ− < 0. The right eigenvector ψ− is thus the stable direction while
ψ+ is the unstable direction.
• Nodes : When 0 < D < 14 T 2 , both eigenvalues are real and of the same sign. Thus,
both right eigenvectors correspond to stable or to unstable directions, depending on
whether T < 0 (stable; λ− < λ+ < 0) or T < 0 (unstable; λ+ > λ− > 0). If λ± are
distinct, one can distinguish fast and slow eigendirections, based on the magnitude of
the eigenvalues.
• Spirals : When D > 14 T 2 , the discriminant T 2 − 4D is negative, and the eigenvalues

come in a complex conjugate pair: λ− = λ∗+ . The real parts are given by Re(λ± ) = 12 T ,
so the motion is stable (i.e. collapsing to the fixed point) if T < 0 and unstable (i.e.
diverging from the fixed point) if T > 0. The motion is easily shown to correspond
Figure 18.7: Phase portrait for an N = 2 flow including saddles (A,C), unstable spiral (B),
and limit cycle (D).
to a spiral. One can check that the spiral rotates counterclockwise for a > d and
clockwise for a < d.
√
• Degenerate Cases : When T = 0 we have λ± = ± −D. For D < 0 we have a
√
saddle, but for D > 0 both eigenvalues are imaginary: λ± = ±i D. The orbits do
not collapse to a point, nor do they diverge to infinity, in the t → ∞ limit, as they do
in the case of the stable and unstable spiral. The fixed point is called a center , and
it is surrounded by closed trajectories.
When D = 14 T 2 , the discriminant vanishes and the eigenvalues are degenerate. If the
rank of M is two, the fixed point is a stable (T < 0) or unstable (T > 0) star . If M
is degenerate and of rank one, the fixed point is a degenerate node.
When D = 0, one of the eigenvalues vanishes. This indicates a fixed line in phase
space, since any point on that line will not move. The fixed line can be stable or
unstable, depending on whether the remaining eigenvalue is negative (stable, T < 0),
or positive (unstable, T > 0).
Putting it all together, an example of a phase portrait is shown in Fig. 18.7. Note the
presence of an isolated, closed trajectory, which is called a limit cycle. Many self-sustained
physical oscillations, i.e. oscillations with no external forcing, exhibit limit cycle behavior.
Limit cycles, like fixed points, can be stable or unstable, or partially stable. Limit cycles
are inherently nonlinear. While the linear equation ϕ̇ = M ϕ can have periodic solutions
if M has purely imaginary eigenvalues, these periodic trajectories are not isolated , because
λ ϕ(t) is also a solution. The amplitude of these linear oscillations is fixed by the initial
conditions, whereas for limit cycles, the amplitude is inherent from the dynamics itself, and
the initial conditions are irrelevant (for a stable limit cycle).
In fig. 18.8 we show simple examples of stable, unstable, and half-stable limit cycles. As
we shall see when we study nonlinear oscillations, the Van der Pol oscillator,
ẍ + µ(x2 − 1) ẋ + x = 0 , (18.39)
Figure 18.8: Stable, unstable, and half-stable limit cycles.
with µ > 0 has a stable limit cycle. The physics is easy to apprehend. The coefficient of the
ẋ term in the equation of motion is positive for |x| > 1 and negative for |x| < 1. Interpreting
this as a coefficient of friction, we see that the friction is positive, i.e. dissipating energy,
when |x| > 1 but negative, i.e. accumulating energy, for |x| < 1. Thus, any small motion
with |x| < 1 is amplified due to the negative friction, and would increase without bound
were it not for the fact that the friction term reverses its sign and becomes dissipative for
|x| > 1. The limit cycle for µ 1 is shown in fig. 18.9.
18.2.4 Fixed points for N = 3 systems
For an N = 2 system, there are five generic types of fixed points. They are classified
according to the eigenvalues of the linearized dynamics at the fixed point. For a real 2 × 2
matrix, the eigenvalues must be real or else must be a complex conjugate pair. The five
types of fixed points are then
λ 1 > 0 , λ2 > 0 : (1) unstable node

λ 1 > 0 , λ2 < 0 : (2) saddle point
λ 1 < 0 , λ2 < 0 : (3) stable node
Re(λ1 ) > 0 , λ2 = λ∗1 : (4) unstable spiral
Re(λ1 ) < 0 , λ2 = λ∗1 : (5) stable spiral
How many possible generic fixed points are there for an N = 3 system?
For a general real 3×3 matrix M , the characteristic polynomial P (λ) = det(λ−M ) satisfies
P (λ∗ ) = P (λ). Thus, if λ is a root then so is λ∗ . This means that the eigenvalues are either
real or else come in complex conjugate pairs. There are then eight generic possibilities for
Figure 18.9: Limit cycle of the Van der Pol oscillator for µ 1 Credit: Wikipedia.
the three eigenvalues:
λ 1 > 0 , λ2 > 0 , λ3 > 0 : (1) unstable node

λ 1 > 0 , λ2 > 0 , λ3 < 0 : (2) (+ + −) saddle
λ 1 > 0 , λ2 < 0 , λ3 < 0 : (3) (+ − −) saddle
λ 1 < 0 , λ2 < 0 , λ3 < 0 : (4) stable note
Re(λ1 ) > 0 , λ2 = λ∗1 , λ3 > 0 : (5) unstable spiral-node
Re(λ1 ) > 0 , λ2 = λ∗1 , λ3 < 0 : (6) (+ + −) unstable spiral-saddle
Re(λ1 ) < 0 , λ2 = λ∗1 , λ3 > 0 : (7) (− − +) unstable spiral-saddle
Re(λ1 ) < 0 , λ2 = λ∗1 , λ3 < 0 : (8) stable spiral-node
18.3 Andronov-Hopf Bifurcation
A bifurcation between a spiral and a limit cycle is known as an Andronov-Hopf bifurcation.

As a simple example, consider the N = 2 system,
ẋ = ax − by − C(x2 + y 2 ) x (18.40)
2 2
ẏ = bx + ay − C(x + y ) y , (18.41)
where a, b, and C are real. Clearly the origin is a fixed point, at which one finds the
eigenvalues λ = a ± ib. Thus, the fixed point is a stable spiral if a < 0 and an unstable
spiral if a > 0.
18.3. ANDRONOV-HOPF BIFURCATION 383
Figure 18.10: Hopf bifurcation: for C > 0 the bifurcation is supercritical, between stable
spiral and stable limit cycle. For C < 0 the bifurcation is subcritical, between unstable
spiral and unstable limit cycle. The bifurcation occurs at a = 0 in both cases.
Written in terms of the complex variable z = x + iy, these two equations collapse to the
single equation
ż = (a + ib) z − C |z|2 z . (18.42)
The dynamics are also simple in polar coordinates r = |z|, θ = arg(z):
ṙ = ar − Cr3 (18.43)
θ̇ = b . (18.44)
The phase diagram, pfor fixed b > 0, is depicted in Fig. 18.10. For positive a/C, there is a
limit cycle at r = a/C. In both cases, the limit cycle disappears as a crosses the value
a∗ = 0 and is replaced by a stable (a < 0, C > 0) or unstable (a > 0, C < 0) spiral.
This example also underscores the following interesting point. Adding a small nonlinear
term C has no fundamental effect on the fixed point behavior so long as a 6= 0, when the
fixed point is a stable or unstable spiral. In general, fixed points which are attractors (stable
spirals or nodes), repellers (unstable spirals or nodes), or saddles are robust with respect
to the addition of a small nonlinearity. But the fixed point behavior in the marginal cases
– centers, stars, degenerate nodes, and fixed lines – is strongly affected by the presence of
even a small nonlinearity. In this example, the FP is a center when a = 0. But as the
(r, θ) dynamics shows, a small nonlinearity will destroy the center and turn the FP into an
attractor (C > 0) or a repeller (C < 0).
18.4 Population Biology : Lotka-Volterra Models
Consider two species with populations N1 and N2 , respectively. We model the evolution of
these populations by the coupled ODEs
dN1
= aN1 + bN1 N2 + cN12 (18.45)
dt
dN2
= dN2 + eN1 N2 + f N22 , (18.46)
dt
where {a, b, c, d, e, f } are constants. We can eliminate some constants by rescaling. This
results in the following:
ẋ = σx + σ 0 xy + Ax2 (18.47)
00 2
ẏ = By + σ xy + Cy , (18.48)
where σ, σ 0 , and σ 00 can each take on one of two possible values ±1. The remaining
coefficients may also be of either sign. The values and especially the signs of the various
coefficients have a physical (or biological) significance. For example, if σ 0 > 0 it means that
x grows due to the presence of y. The effect of y on x may be of the same sign (σ 0 σ 00 = 1)
or of opposite sign (σ 0 σ 00 = −1).
18.4.1 Rabbits and foxes
As an example, consider the model
ẋ = x − xy , ẏ = −βy + xy . (18.49)
The quantity x might represent the (scaled) population of rabbits and y the population of
foxes in an ecosystem. There are two fixed points – at (0, 0) and at (β, 1). Linearizing the
dynamics about these fixed points, one finds that (0, 0) is a saddle while (β, 1) is a center.
Let’s do this explicitly.
The first step is to find the fixed points (x∗ , y ∗ ). To do this, we set ẋ = 0 and ẏ = 0. From
ẋ = x(1 − y) = 0 we have that x = 0 or y = 1. Suppose x = 0. The second equation,
ẏ = (x − β)y then requires y = 0. So P1 = (0, 0) is a fixed point. The other possibility is
that y = 1, which then requires x = β. So P2 = (β, 1) is the second fixed point. Those are
the only possibilities.
We now compute the linearized dynamics at these fixed points. The linearized dynamics
are given by ϕ̇ = M ϕ, with
  
∂ ẋ/∂x ∂ ẋ/∂y 1−y −x
M = =  . (18.50)
∂ ẏ/∂x ∂ ẏ/∂y y x−β
18.4. POPULATION BIOLOGY : LOTKA-VOLTERRA MODELS 385
Figure 18.11: Phase flow for the rabbits vs. foxes Lotka-Volterra model of eq. 18.49.
Evaluating M at P1 and P2 , we find

   
1 0 0 −β
M1 =   , M2 =   . (18.51)
0 −β 1 0
The eigenvalues are easily found:

p
P1 : λ+ = 1 P2 : λ+ = i β (18.52)
p
λ− = −β λ− = −i β . (18.53)
Thus P1 is a saddle point and P2 is a center.
18.4.2 Rabbits and sheep
In the rabbits and foxes model of eq. 18.49, the rabbits are the food for the foxes. This
means σ 0 = −1 but σ 00 = +1 – the fox population is enhanced by the presence of rabbits,
but the rabbit population is diminished by the presence of foxes. Consider now a model in
which the two species (rabbits and sheep, say) compete for food:
ẋ = x − xy − Ax2 , ẏ = By − xy − Cy 2 , (18.54)
with A, B, and C all positive. Note that when either population x or y vanishes, the
remaining population is governed by the logistic equation, i.e. it will flow to a nonzero fixed
point.
Figure 18.12: Two possible phase flows for the rabbits vs. sheep model of eq. 18.54.
The matrix of derivatives, which is to be evaluated at each fixed point in order to assess its
stability, is
   
∂ ẋ/∂x ∂ ẋ/∂y 1 − y − 2Ax −x
M = =  . (18.55)
∂ ẏ/∂x ∂ ẏ/∂y −y B − x − 2Cy
At each fixed point, we must evaluate D = det(M ) and T = Tr (M ) and apply the classifi-
cation scheme of Fig. 18.5.
P1 = (0, 0) : This is the trivial

state with
no rabbits (x = 0) and no sheep (y = 0). The
1 0
linearized dynamics gives M1 = , which corresponds to an unstable node.
0 B
P −1 , 0) : Here we have rabbits but no sheep. The linearized dynamics gives M =

2 = (A −1 2
1 −A
. For AB > 1 this is a saddle; for AB < 1 it is a stable node.
0 B − A−1
(0, C −1 B) : Here
P3 = we have sheep but no rabbits. The linearized dynamics gives
1−C B −1 0
M3 = . For B > C this is a stable node; for B < C it is a saddle.
−C −1 B −B
P4 = (x4 , y4 ) : There is one remaining fixed point – a nontrivial one where both x and y
are nonzero. To find it, we set ẋ = ẏ = 0, and divide out by x and y respectively, to get
Ax + y = 1 , x + Cy = B , (18.56)
the solution of which is

!
B−C 1 − AB
P4 = , . (18.57)
1 − AC 1 − AC
18.5. INDEX THEORY 387

Figure 18.13: Two singularities with index +1. The direction field Vb = V /V is shown in
both cases.
The linearized dynamics then gives

 
AC − AB C −B
1
M4 =   , (18.58)
1 − AC
AB − 1 ABC − C
giving
AC + ABC − AB − C (B − C)(AB − 1)
T = , D= . (18.59)
1 − AC 1 − AC
The classification of this fixed point can vary with parameters. For example, if A = 1,
B = 12 , and C = 14 , find T = − 12 and D = − 16 , corresponding to a saddle point. In this
case, it is the fate of one population to die out at the expense of the other. Note that
AB < 1 and B > C, so P2 and P3 are both stable nodes. Which species lives and which
dies depends on initial conditions. If on the other hand A = 2, B = 1, and C = 2, find
T = − 43 and D = 13 , and P4 is a stable node, while P2 and P3 are saddles. The situation is
depicted in Fig. 18.12.
18.5 Index Theory
Consider a smooth two-dimensional vector field V (ϕ). The angle that the vector V makes
with respect to the ϕ
b1 and ϕ
b2 axes is a scalar field,

−1 V2 (ϕ)
Θ(ϕ) = tan . (18.60)
V1 (ϕ)
Figure 18.14: Two singularities with index −1.
So long as V has finite length, the angle Θ is well-defined. In particular, we expect that we
can integrate ∇Θ over a closed curve C in phase space to get
I
dϕ · ∇Θ = 0 . (18.61)
C
However, this can fail if V (ϕ) vanishes (or diverges) at one or more points in the interior
of C. In general, if we define
I
1
WC (V ) = dϕ · ∇Θ , (18.62)
2π
C
then WC (V ) ∈ ZZ is an integer valued function of C, which is the change in Θ around the

curve C. This must be an integer, because Θ is well-defined only up to multiples of 2π.
Note that differential changes of Θ are in general well-defined.
Thus, if V (ϕ) is finite, meaning neither infinite nor infinitesimal, i.e. V neither diverges
nor vanishes anywhere in int(C), then WC (V ) = 0. Assuming that V never diverges, any
singularities in Θ must arise from points where V = 0, which in general occurs at isolated
points, since it entails two equations in the two variables (ϕ1 , ϕ2 ).
The index of a two-dimensional vector field V (ϕ) at a point ϕ is the integer-valued winding
Figure 18.15: Left panel: a singularity with index +2. Right panel: two singularities each
with index +1. Note that the long distance behavior of V is the same in both cases.
of V about that point:

I
1
ind (V ) = lim dϕ · ∇Θ (18.63)
ϕ0 a→0 2π
Ca (ϕ0 )
V1 ∇V2 − V2 ∇V1
I
1
= lim dϕ · , (18.64)
a→0 2π V12 + V22
Ca (ϕ0 )
where Ca (ϕ0 ) is a circle of radius a surrounding the point ϕ0 . The index of a closed curve
C is given by the sum of the indices at all the singularities enclosed by the curve:2
X
WC (V ) = ind (V ) . (18.65)
ϕi
ϕi ∈ int(C)
As an example, consider the vector fields plotted in fig. 18.13. We have:
V = (x , y) =⇒ Θ=θ (18.66)
1
V = (y , −x) =⇒ Θ=θ+ 2π . (18.67)
The index is the same, +1, in both cases, even though the first corresponds to an unstable
node and the second to a center. Any N = 2 fixed point with det M > 0 has index +1.
2
Technically, we should weight the index at each enclosed singularity by the signed number of times the
curve C encloses that singularity. For simplicity and clarity, we assume that the curve C is homeomorphic
to the circle S1 .
Figure 18.16: A vector field with index −2.
Fig. 18.14 shows two vector fields, each with index −1:
V = (x , −y) =⇒ Θ = −θ (18.68)
1
V = (y , x) =⇒ Θ = −θ + 2π . (18.69)
In both cases, the fixed point is a saddle.
As an example of the content of eqn. 18.65, consider the vector fields in eqn. 18.15. The
left panel shows the vector field V = (x2 − y 2 , 2xy), which has a single fixed point, at the
origin (0 , 0), of index +2. The right panel shows the vector field V = (1+x2 −y 2 , x+2xy),
which has fixed points (x∗ , y ∗ ) at (0 , 1) and (0 , −1). The linearized dynamics is given by
the matrix   
∂ ẋ ∂ ẋ 
∂x ∂y 2x −2y
M = =  . (18.70)
  
∂ ẏ ∂ ẏ
∂x ∂y
1 + 2y 2x
Thus,
0 −2 0 2
M(0,1) = , M(0,−1) = . (18.71)
2 0 −2 0
At each of these fixed points, we have T = 0 and D = 4, corresponding to a center, with
index +1. If we consider a square-ish curve Caround the periphery of each figure, the vector
field is almost the same along such a curve for both the left and right panels, and the
winding number is WC (V ) = +2.
Finally, consider the vector field shown in fig. 18.16, with V = (x2 − y 2 , −2xy). Clearly
Θ = −2θ, and the index of the singularity at (0 , 0) is −2.
To recapitulate some properties of the index / winding number:
• The index ind ϕ (V ) of an N = 2 vector field V at a point ϕ0 is the winding number

0
of V about that point.
• The winding number WC (V ) of a curve C is the sum of the indices of the singularities
enclosed by that curve.
• Smooth deformations of C do not change its winding number. One must instead
“stretch” C over a fixed point singularity in order to change WC (V ).
• Uniformly rotating each vector in the vector field by an angle β has the effect of
sending Θ → Θ + β; this leaves all indices and winding numbers invariant.
• Nodes and spirals, whether stable or unstable, have index +1 (ss do the special cases
of centers, stars, and degenerate nodes). Saddle points have index −1.
• Clearly any closed orbit must lie on a curve C of index +1.
18.5.1 Gauss-Bonnet Theorem
There is a deep result in mathematics, the Gauss-Bonnet theorem, which connects the local
geometry of a two-dimensional manifold to its global topological structure. The content of
the theorem is as follows:
Z X
dA K = 2π χ(M) = 2π ind (V ) , (18.72)
i ϕi
M
where M is a 2-manifold (a topological space locally homeomorphic to R R 2 ), κ is the local

Gaussian curvature of M, which is given by K = (R1 R2 )−1 , where R1,2 are the principal
radii of curvature at a given point, and dA is the differential area element. The quantity
χ(M) is called the Euler characteristic of M and is given by χ(M) = 2 − 2g, where g is
the genus of M, which is the number of holes (or handles) of M. Furthermore, V (ϕ) is
any smooth vector field on M, and ϕi are the singularity points of that vector field, which
are fixed points of the dynamics ϕ̇ = V (ϕ).
To apprehend the content of the Gauss-Bonnet theorem, it is helpful to consider an example.

Let M = S2 be the unit 2-sphere, as depicted in fig. 18.17. At any point on the unit 2-
sphere, the radii of curvature are degenerate and both equal to R = 1, hence K = 1. If we
integrate the Gaussian curvature over the sphere, we thus get 4π = 2π χ S2 , which says
χ(S2 ) = 2−2g = 2, which agrees with g = 0 for the sphere. Furthermore, the Gauss-Bonnet
theorem says that any smooth vector field on S2 must have a singularity or singularities,
with the total index summed over the singularities equal to +2. The vector field sketched
in the left panel of fig. 18.17 has two index +1 singularities, which could be taken at the
north and south poles, but which could be anywhere. Another possibility, depicted in the
right panel of fig. 18.17, is that there is a one singularity with index +2.
Figure 18.17: Two smooth vector fields on the sphere S2 , which has genus g = 0. Left
panel: two index +1 singularities. Right panel: one index +2 singularity.
In fig. 18.18 we show examples of manifolds with genii g = 1 and g = 2. The case g = 1 is
the familiar 2-torus, which is topologically equivalent to a product of circles: T2 ' S1 × S1 ,
and is thus coordinatized by two angles θ1 and θ2 . A smooth vector field pointing in the
direction of
increasing θ1 never vanishes, and thus has no singularities, consistent with g = 1
2
χ
and T = 0. Topologically, one can define a torus as the quotient space R R / ZZ2 , or as a
2
square with opposite sides identified. This is what mathematicians call a ‘flat torus’ – one
with curvature K = 0 everywhere. Of course, such a torus cannot be embedded in three-
Figure 18.18: Smooth vector fields on the torus T2 (g = 1), and on a 2-manifold M of genus
g = 2.
18.6. POINCARÉ-BENDIXSON THEOREM 393
dimensional Euclidean space; a two-dimensional figure embedded in a three-dimensional

Euclidean space inherits a metric due to the embedding, and for a physical torus, like the
surface of a bagel, the Gaussian curvature is only zero on average.
The g = 2 surface M shown in the right panel of fig. 18.18 has Euler characteristic
χ(M) = −2, which means that any smooth vector field on M must have singularities with
indices totalling −2. One possibility, depicted in the figure, is to have two saddle points
with index −1; one of these singularities is shown in the figure (the other would be on the
opposite side).
18.6 Poincaré-Bendixson Theorem
Although N = 2 systems are much richer than N = 1 systems, they are still ultimately
rather impoverished in terms of their long-time behavior. If an orbit does not flow off
to infinity or asymptotically approach a stable fixed point (node or spiral or nongeneric
example), the only remaining possibility is limit cycle behavior. This is the content of the
Poincaré-Bendixson theorem, which states:
• IF Ω is a compact (i.e. closed and bounded) subset of phase space,
• AND ϕ̇ = V (ϕ) is continuously differentiable on Ω,
• AND Ω contains no fixed points (i.e. V (ϕ) never vanishes in Ω),
• AND a phase curve ϕ(t) is always confined to Ω,
THEN ϕ(t) is either closed or approaches a closed trajectory in the limit t → ∞.
Thus, under the conditions of the theorem, Ω must contain a closed orbit.
One way to prove that ϕ(t) is confined to Ω is to establish that V · n̂ ≤ 0 everywhere on

the boundary ∂Ω, which means that the phase flow is always directed inward (or tangent)
along the boundary. Let’s analyze an example from the book by Strogatz. Consider the
system
ṙ = r(1 − r2 ) + λ r cos θ (18.73)

θ̇ = 1 , (18.74)
with 0 < λ < 1. Then define

√ √
a≡ 1−λ , b≡ 1+λ (18.75)
and n o
Ω ≡ (r, θ) a < r < b . (18.76)
On the boundaries of Ω, we have

r=a ⇒ ṙ = λ a 1 + cos θ (18.77)

r=b ⇒ ṙ = −λ b 1 − cos θ . (18.78)
We see that the radial component of the flow is inward along both r = a and r = b. Thus,
any trajectory which starts inside Ω can never escape. The Poincaré-Bendixson theorem
tells us that the trajectory will approach a stable limit cycle in the limit t → ∞.
It is only with N ≥ 3 systems that the interesting possibility of chaotic behavior emerges.
18.7 Appendix : Example Problem
Consider the two-dimensional phase flow,
ẋ = 21 x + xy − 2x3 (18.79)
ẏ = 52 y + xy − y 2 . (18.80)
(a) Find and classify all fixed points.
Solution : We have
1
+ y − 2x2

ẋ = x 2 (18.81)
5

ẏ = y 2 +x−y . (18.82)
The matrix of first derivatives is

 
∂ ẋ ∂ ẋ 1
+ y − 6x2

∂x ∂y 2 x
M =  =  . (18.83)
 
∂ ẏ ∂ ẏ 5
∂x ∂y
y 2 + x − 2y
There are six fixed points.
(x, y) = (0, 0) : The derivative matrix is

1
2 0
M= 5 . (18.84)
0 2
The determinant is D = 54 and the trace is T = 3. Since D < 41 T 2 and T > 0, this is an
unstable node. (Duh! One can read off both eigenvalues are real and positive.) Eigenvalues:
λ1 = 12 , λ2 = 52 .
(x, y) = (0, 52 ) : The derivative matrix is

3 0
M= 5 , (18.85)
2 − 52
18.7. APPENDIX : EXAMPLE PROBLEM 395
Figure 18.19: Sketch of phase flow for ẋ = 12 x + xy − 2x3 , ẏ = 52 y + xy − y 2 . Fixed point

classifications are in the text.
for which D = − 15 1
2 and T = 2 . The determinant is negative, so this is a saddle. Eigenvalues:
5
λ1 = − 2 , λ2 = 3.
(x, y) = (− 12 , 0) : The derivative matrix is
−1 − 21

M= , (18.86)
0 2
for which D = −2 and T = +1. The determinant is negative, so this is a saddle. Eigenvalues:
λ1 = −1, λ2 = 2.
(x, y) = ( 12 , 0) : The derivative matrix is
−1 12

M= , (18.87)
0 3
for which D = −3 and T = +2. The determinant is negative, so this is a saddle. Eigenvalues:
λ1 = −1, λ2 = 3.
(x, y) = ( 32 , 4) : This is one root obtained by setting y = x + 52 and the solving 1

2 + y − 2x2 =
3 + x − 2x2 = 0, giving x = −1 and x = + 32 . The derivative matrix is
−9 32

M= , (18.88)
4 −4
1
for which D = 30 and T = −13. Since D < 4 T 2 and T < 0, this corresponds to a stable
node. Eigenvalues: λ1 = −10, λ2 = −3.
(x, y) = (−1, 32 ) : This is the second root obtained by setting y = x + 52 and the solving
1 2 2 3
2 + y − 2x = 3 + x − 2x = 0, giving x = −1 and x = + 2 . The derivative matrix is

−4 −1
M= 3 , (18.89)
2 − 32
for which D = 15 11
2 and T = − 2 . Since D <
1
4 T 2 and T < 0, this corresponds to a stable
node. Eigenvalues: λ1 = −3, λ2 = − 25 .
(b) Sketch the phase flow.
Solution : The flow is sketched in fig. 18.19. Thanks to Evan Bierman for providing the
Mathematica code.
Chapter 19
Nonlinear Oscillators
19.1 Weakly Perturbed Linear Oscillators
Consider a nonlinear oscillator described by the equation of motion
ẍ + Ω02 x = h(x) . (19.1)
Here, is a dimensionless parameter, assumed to be small, and h(x) is a nonlinear function

of x. In general, we might consider equations of the form
ẍ + Ω02 x = h(x, ẋ) , (19.2)
such as the van der Pol oscillator,
ẍ + µ(x2 − 1)ẋ + Ω02 x = 0 . (19.3)
First, we will focus on nondissipative systems, i.e. where we may write mẍ = −∂x V , with
V (x) some potential.
As an example, consider the simple pendulum, which obeys
θ̈ + Ω02 sin θ = 0 , (19.4)
where Ω02 = g/`, with ` the length of the pendulum. We may rewrite his equation as
θ̈ + Ω02 θ = Ω02 (θ − sin θ)

= 1
6 Ω02 θ3 − 1
120 Ω02 θ5 + . . . (19.5)
The RHS above is a nonlinear function of θ. We can define this to be h(θ), and take = 1.
397
398 CHAPTER 19. NONLINEAR OSCILLATORS
19.1.1 Naı̈ve Perturbation theory and its failure
Let’s assume though that is small, and write a formal power series expansion of the solution
x(t) to equation 19.1 as
x = x0 + x1 + 2 x2 + . . . . (19.6)
We now plug this into 19.1. We need to use Taylor’s theorem,
h(x0 + η) = h(x0 ) + h0 (x0 ) η + 12 h00 (x0 ) η 2 + . . . (19.7)
with
η = x1 + 2 x2 + . . . . (19.8)
Working out the resulting expansion in powers of is tedious. One finds
n o
h(x) = h(x0 ) + h0 (x0 ) x1 + 2 h0 (x0 ) x2 + 21 h00 (x0 ) x21 + . . . . (19.9)
Equating terms of the same order in , we obtain a hierarchical set of equations,
ẍ0 + Ω02 x0 = 0 (19.10)

ẍ1 + Ω02 x1 = h(x0 ) (19.11)
ẍ2 + Ω02 x2 = h0 (x0 ) x1 (19.12)
ẍ3 + Ω02 x3 = h0 (x0 ) x2 + 12 h00 (x0 ) x21 (19.13)
et cetera, where prime denotes differentiation with respect to argument. The first of these
is easily solved: x0 (t) = A cos(Ω0 t + ϕ), where A and ϕ are constants. This solution then
is plugged in at the next order, to obtain an inhomogeneous equation for x1 (t). Solve for
x1 (t) and insert into the following equation for x2 (t), etc. It looks straightforward enough.
The problem is that resonant forcing terms generally appear in the RHS of each equation
of the hierarchy past the first. Define θ ≡ Ω0 t + ϕ. Then x0 (θ) is an even periodic function

of θ with period 2π, hence so is h(x0 ). We may then expand h x0 (θ) in a Fourier series:
∞
X
h A cos θ = hn (A) cos(nθ) . (19.14)
n=0
The n = 1 term leads to resonant forcing. Thus, the solution for x1 (t) is
∞
X hn (A) h1 (A)
x1 (t) = 2
cos(nΩ0 t + nϕ) + t sin(Ω0 t + ϕ) , (19.15)
n=0
1−n 2Ω0
(n6=1)
which increases linearly with time. As an example, consider a cubic nonlinearity with
h(x) = r x3 , where r is a constant. Then using
3
cos(3θ) = 4 cos(θ) + 14 cos(3θ) , (19.16)
3 1
we have h1 = 4 rA3 , h3 = 4 rA3 .
19.1. WEAKLY PERTURBED LINEAR OSCILLATORS 399
19.1.2 Poincaré-Lindstedt method
The problem here is that the nonlinear oscillator has a different frequency than its linear
counterpart. Indeed, if we assume the frequency Ω is a function of , with
Ω() = Ω0 + Ω1 + 2 Ω2 + . . . , (19.17)
then subtracting the unperturbed solution from the perturbed one and expanding in yields
cos(Ωt) − cos(Ω0 t) = − sin(Ω0 t) (Ω − Ω0 ) t − 12 cos(Ω0 t) (Ω − Ω0 )2 t2 + . . .
n o
= − sin(Ω0 t) Ω1 t − 2 sin(Ω0 t) Ω2 t + 21 cos(Ω0 t) Ω12 t2 + O(3 ) .
(19.18)
What perturbation theory can do for us is to provide a good solution up to a given time,
provided that is sufficiently small . It will not give us a solution that is close to the true
answer for all time. We see above that in order to do that, and to recover the shifted
frequency Ω(), we would have to resum perturbation theory to all orders, which is a
daunting task.
The Poincaré-Lindstedt method obviates this difficulty by assuming Ω = Ω() from the
outset. Define a dimensionless time s ≡ Ωt and write 19.1 as
d2x
Ω2 + Ω02 x = h(x) , (19.19)
ds2
where
x = x0 + x1 + 2 x2 + . . . (19.20)
2 2
Ω = a0 + a1 + a2 + . . . . (19.21)
We now plug the above expansions into 19.19:
d2x0 d2x1 2

2 2 d x2
a0 + a1 + a2 + . . . + 2 + + ...
ds2 ds ds2
+ Ω02 x0 + x1 + 2 x2 + . . .

n o
= h(x0 ) + 2 h0 (x0 ) x1 + 3 h0 (x0 ) x2 + 21 h00 (x0 ) x21 + . . . (19.22)
Now let’s write down equalities at each order in :
d2x0
a0 + Ω02 x0 = 0 (19.23)
ds2
d2x1 2 d2x0
a0 + Ω x
0 1 = h(x0 ) − a1 (19.24)
ds2 ds2
d2x2 2 0 d2x0 d2x1

a0 + Ω x
0 2 = h (x0 ) x1 − a2 − a1 , (19.25)
ds2 ds2 ds2
et cetera.
The first equation of the hierarchy is immediately solved by
a0 = Ω02 , x0 (s) = A cos(s + ϕ) . (19.26)
At O(), then, we have
d2x1
+ x1 = Ω0−2 h A cos(s + ϕ) + Ω0−2 a1 A cos(s + ϕ) .

2
(19.27)
ds
The LHS of the above equation has a natural frequency of unity (in terms of the dimension-
less time s). We expect h(x0 ) to contain resonant forcing terms, per 19.14. However, we
now have the freedom to adjust the undetermined coefficient a1 to cancel any such resonant
term. Clearly we must choose
h1 (A)
aa = − . (19.28)
A
The solution for x1 (s) is then
∞
1 X hn (A)
x1 (s) = 2 cos(ns + nϕ) , (19.29)
Ω0 n=0 1 − n2
(n6=1)
which is periodic and hence does not increase in magnitude without bound, as does 19.15.
The perturbed frequency is then obtained from
h1 (A) h1 (A)
Ω 2 = Ω02 − + O(2 ) =⇒ Ω() = Ω0 − + O(2 ) . (19.30)
A 2AΩ0
Note that Ω depends on the amplitude of the oscillations.
As an example, consider an oscillator with a quartic nonlinearity in the potential, i.e.

h(x) = r x3 . Then
h A cos θ = 43 rA3 cos θ + 14 rA3 cos(3θ) .

(19.31)
We then obtain, setting = 1 at the end of the calculation,
3 rA2
Ω = Ω0 − + ... (19.32)
8 Ω0
where the remainder is higher order in the amplitude A. In the case of the pendulum,
θ̈ + Ω02 θ = 16 Ω02 θ3 + O θ5 ,

(19.33)
1
and with r = 6 Ω02 and θ0 (t) = θ0 sin(Ωt), we find
2π 2π n o
T (θ0 ) = = · 1+ 1
16 θ02 + . . . . (19.34)
Ω Ω0
19.2. MULTIPLE TIME SCALE METHOD 401
One can check that this is correct to lowest nontrivial order in the amplitude, using the
exact result for the period,
4
K sin2 12 θ0 ,

T (θ0 ) = (19.35)
Ω0
where K(x) is the complete elliptic integral.
The procedure can be continued to the next order, where the free parameter a2 is used to
eliminate resonant forcing terms on the RHS.
A good exercise to test one’s command of the method is to work out the lowest order
nontrivial corrections to the frequency of an oscillator with a quadratic nonlinearity, such
as h(x) = rx2 . One finds that there are no resonant forcing terms at first order in , hence
one must proceed to second order to find the first nontrivial corrections to the frequency.
19.2 Multiple Time Scale Method
Another method of eliminating secular terms (i.e. driving terms which oscillate at the
resonant frequency of the unperturbed oscillator), and one which has applicability beyond
periodic motion alone, is that of multiple time scale analysis. Consider the equation
ẍ + x = h(x, ẋ) , (19.36)
where is presumed small, and h(x, ẋ) is a nonlinear function of position and/or velocity.
We define a hierarchy of time scales: tn ≡ n t. There is a normal time scale t0 , slow time
scale t1 , a ‘superslow’ time scale t1 , etc. Thus,
d ∂ ∂ ∂
= + + 2 + ...
dt ∂t0 ∂t1 ∂t2
∞
X ∂
= n . (19.37)
∂tn
n=0
Next, we expand
∞
X
x(t) = n xn (t0 , t1 , . . .) . (19.38)
n=0
Thus, we have
∞
X ∞
2 X ∞ ∞ ∞ ∞
X !
n ∂ k
X
k
X
k
X
n ∂ k
xk + xk = h xk , xk .
∂tn ∂tn
n=0 k=0 k=0 k=0 n=0 k=0
We now evaluate this order by order in :

2
0 ∂
O( ) : + 1 x0 = 0 (19.39)
∂t20
2
∂ 2x0

1 ∂ ∂x0
O( ) : + 1 x1 = −2 + h x0 , (19.40)
∂t20 ∂t0 ∂t1 ∂t0
2 2 2 ∂ 2x0

∂ ∂ x1 ∂ x0
O(2 ) : + 1 x2 = −2 − 2 −
∂t20 ∂t0 ∂t1 ∂t0 ∂t2 ∂t21

∂h ∂h ∂x1 ∂x0
+ x + + , (19.41)
∂x 1 ∂ ẋ ∂t0 ∂t1

{x0 ,ẋ0 } {x0 ,ẋ0 }
et cetera. The expansion gets more and more tedious with increasing order in .
Let’s carry this procedure out to first order in . To order 0 ,

x0 = A cos t0 + φ , (19.42)
where A and φ are arbitrary (at this point) functions of {t1 , t2 , . . .}. Now we solve the next
equation in the hierarchy, for x1 . Let θ ≡ t0 + φ. Then ∂t0 = ∂θ and we have
2
∂ ∂A ∂φ
2
+ 1 x1 = 2 sin θ + 2 A cos θ + h A cos θ, −A sin θ . (19.43)
∂θ ∂t1 ∂t1
Since the arguments of h are periodic under θ → θ + 2π, we may expand h in a Fourier
series:
∞ ∞
X X
h(θ) ≡ h A cos θ, −A sin θ = αk (A) sin(kθ) + βk (A) cos(kθ) . (19.44)
k=1 k=0
The inverse of this relation is

Z2π
dθ
αk (A) = h(θ) sin(kθ) (k > 0) (19.45)
π
0
Z2π
dθ
β0 (A) = h(θ) (19.46)
2π
0
Z2π
dθ
βk (A) = h(θ) cos(kθ) (k > 0) . (19.47)
π
0
We now demand that the secular terms on the RHS – those terms proportional to cos θ and
sin θ – must vanish. This means
∂A
2 + α1 (A) = 0 (19.48)
∂t1
∂φ
2A + β1 (A) = 0 . (19.49)
∂t1
These two first order equations require two initial conditions, which is sensible since our
initial equation ẍ + x = h(x, ẋ) is second order in time.
With the secular terms eliminated, we may solve for x1 :

∞
X αk (A) βk (A)
x1 = sin(kθ) + cos(kθ) + C0 cos(θ + η0 ) . (19.50)
1 − k2 1 − k2
k6=1
Note: (i) the k = 1 terms are excluded from the sum, and (ii) an arbitrary solution to the
homogeneous equation, i.e. eqn. 19.43 with the right hand side set to zero, is included. The
constants C0 and η0 are arbitrary functions of t1 , t2 , etc. .
The equations for A and φ are both first order in t1 . They will therefore involve two
constants of integration – call them A0 and φ0 . At second order, these constants are taken
as dependent upon the superslow time scale t2 . The method itself may break down at this
order. (See if you can find out why.)
Let’s apply this to the nonlinear oscillator ẍ+sin x = 0, also known as the simple pendulum.
We’ll expand the sine function to include only the lowest order nonlinear term, and consider
ẍ + x = 1
6 x3 . (19.51)
We’ll assume is small and take = 1 at the end of the calculation. This will work provided
the amplitude of the oscillation is itself small. To zeroth order, we have x0 = A cos(t + φ),
as always. At first order, we must solve
2
∂ ∂A ∂φ
2
+ 1 x1 = 2 sin θ + 2 A cos θ + 16 A2 cos3 θ (19.52)
∂θ ∂t1 ∂t1
∂A ∂φ
=2 sin θ + 2 A cos θ + 241
A3 cos(3θ) + 18 A3 cos θ .
∂t1 ∂t1
We eliminate the secular terms by demanding
∂A ∂φ
=0 , 1
= − 16 A2 , (19.53)
∂t1 ∂t1
1
hence A = A0 and φ = − 16 A20 t1 + φ0 , and
1 2

x(t) = A0 cos t − 16 A0 t + φ0
1 3 3 2

− 192 A0 cos 3t − 16 A0 t + 3φ0 + . . . , (19.54)
which reproduces the result obtained from the Poincaré-Lindstedt method.
19.2.1 Duffing oscillator
Consider the equation

ẍ + 2µẋ + x + x3 = 0 . (19.55)
This describes a damped nonlinear oscillator. Here we assume both the damping coefficient
µ̃ ≡ µ as well as the nonlinearity both depend linearly on the small parameter . We may
write this equation in our standard form ẍ + x = h(x, ẋ), with h(x, ẋ) = −2µẋ − x3 .
For > 0, which we henceforth assume, it is easy to see that the only fixed point is
(x, ẋ) = (0, 0). The linearized flow in the vicinity of the fixed point is given by

d x 0 1 x
= + O(x3 ) . (19.56)
dt ẋ −1 −2µ ẋ
The determinant is D = 1 and the trace is T = −2µ. Thus, provided µ < 1, the fixed
point is a stable spiral; for µ > 1 the fixed point becomes a stable node.
We employ the multiple time scale method to order . We’ll save some indices and write
τ ≡ t0 = t and T ≡ t1 = t. We have x0 = A cos(τ + φ) to zeroth order, as usual. The
nonlinearity is expanded in a Fourier series in θ = τ + φ:
h x0 , ∂τ x0 = 2µA sin θ − A3 cos3 θ

= 2µA sin θ − 34 A3 cos θ − 14 A3 cos 3θ . (19.57)
Thus, α1 (A) = 2µA and β1 (A) = − 34 A3 . We now solve the first order equations,
∂A
= − 21 α1 (A) = −µA =⇒ A(T ) = A0 e−µT (19.58)
∂T
as well as
∂φ β (A) 3A20
=− 1 = 38 A20 e−2µT 1 − e−2µT .

=⇒ φ(T ) = φ0 + (19.59)
∂T 2A 16µ
After elimination of the secular terms, we may read off

1 3

x1 (τ, T ) = 32 A (T ) cos 3τ + 3φ(T ) . (19.60)
Finally, we have
3A20
x(t) = A0 e−µt cos t + 1 − e−2µt + φ0

16µ
9A20
1
A30 e−3µt cos 3t + 1 − e−2µt + 3φ0 .

+ 32 (19.61)
16µ
19.2.2 Forced Duffing oscillator
The forced, damped linear oscillator,
ẍ + 2µẋ + x = f0 cos Ωt (19.62)

has the solution

x(t) = xh (t) + C(Ω) cos Ωt + δ(Ω) , (19.63)
where
xh (t) = A+ eλ+ t + A− eλ− t , (19.64)
p
where λ± = −µ ± µ2 − 1 are the roots of λ2 + 2µλ + 1 = 0. The ‘susceptibility’ C and
phase shift δ are given by

1 −1 2µΩ
C(Ω) = p , δ(Ω) = tan . (19.65)
(Ω 2 − 1)2 + 4µ2 Ω 2 1 − Ω2
The homogeneous solution, xh (t), is a transient and decays exponentially with time, since
Re(λ± ) < 0. The asymptotic behavior is a phase-shifted oscillation at the driving frequency
Ω.
Now let’s add a nonlinearity. We study the equation
ẍ + x = h(x, ẋ) + f0 cos(t + νt) . (19.66)
Note that amplitude of the driving term, f0 cos(Ωt), is assumed to be small, i.e. propor-
tional to , and the driving frequency Ω = 1 + ν is assumed to be close to resonance.
(The resonance frequency of the unperturbed oscillator is ωres = 1.) Were the driving fre-
quency far from resonance, it could be dealt with in the same manner as the non-secular
terms encountered thus far. The situation when Ω is close to resonance deserves our special
attention.
At order 0 , we still have x0 = A cos(τ + φ), with τ = t and T = t. At order 1 , we must
solve
2
∂ 0 0

+ 1 x 1 = 2A sin θ + 2Aφ cos θ + h A cos θ , −A sin θ + f0 cos(θ − ψ)
∂θ2
X
= αk sin(kθ) + βk cos(kθ) + 2A0 + α1 + f0 sin ψ sin θ
k6=1

+ 2Aψ 0 + 2Aν + β1 + f0 cos ψ cos θ , (19.67)
where ψ ≡ φ(T ) − νT , and where the prime denotes differentiation with respect to T . We
must therefore solve
dA
= − 12 α1 (A) − 12 f0 sin ψ (19.68)
dT
dψ β (A) f
= −ν − 1 − 0 cos ψ . (19.69)
dT 2A 2A
If we assume that {A, ψ} approaches a fixed point of these dynamics, then at the fixed point
these equations provide a relation between the amplitude A, the ‘detuning’ parameter ν,
and the drive f0 :
h i2 h i2
α1 (A) + 2νA + β1 (A) = f02 . (19.70)
Figure 19.1: Amplitude A versus detuning ν for the forced Duffing oscillator for three values
of the drive f0 . The critical drive is f0,c = 316
5/4 µ
3/2 . For f > f , there is hysteresis as a
0 0,c
function of the detuning.
Thus far our approach has been completely general. We now restrict our attention to the
Duffing equation, for which
α1 (A) = 2µA , β1 (A) = − 43 A3 , (19.71)
which yields the cubic equation
A6 − 16
3 νA
4
+ 64
9 (µ
2
+ ν 2 )A2 − 16 2
9 f0 =0. (19.72)
Analyzing the cubic is a good exercise. Setting y = A2 , we define
G(y) ≡ y 3 − 16
3 ν y2 + 64
9 (µ
2
+ ν 2) y , (19.73)
and we seek a solution to G(y) = 16 2

Setting G0 (y) = 0, we find roots at
9 f0 .
p
y± = 16
9 ν ± 8
9 ν 2 − 3µ2 . (19.74)
Thus, ν y± ≥ 0, and thus, since y = A2 must be positive, there is a unique solution to

√
G(y) = 16 2

9 f0 for ν ∈ − ∞, 3 µ , for all f0 .
√
For ν > 3 µ, the function G(y) has a local maximum at y = y− and a local minimum at
y = y+ . There are then three solutions for y(ν) for f0 ∈ f0− , f0+ , where f0± = 43 G(y∓ ).
p
If we define κ ≡ ν/µ, then

q p
± 8 3/2
f0 = 9 µ κ3 + 9κ ± κ2 − 3 . (19.75)
16
The phase diagram is shown in Fig. 19.2. The minimum value for f0 is f0,c = 35/4
µ3/2 ,
√
which occurs at κ = 3.
Thus far we have assumed that the (A, ψ) dynamics evolves to a fixed point. We should
check to make sure that this fixed point is in fact stable. To do so, we evaluate the linearized
dynamics at the fixed point. Writing A = A∗ + δA and ψ = ψ ∗ + δψ, we have

d δA δA
=M , (19.76)
dT δψ δψ
Figure 19.2: Phase diagram for the forced Duffing oscillator.
with
 
∂ Ȧ ∂ Ȧ
− 21 f0 cos ψ
 
∂A ∂ψ −µ
M = =
   
∂ ψ̇ ∂ ψ̇ 3 f0 f0
∂A ∂ψ 4A + 2A2
cos ψ 2A sin ψ
νA − 38 A3
 
−µ
=  . (19.77)
9 ν
8A − A −µ
One then has T = −2µ and
D = µ2 + ν − 83 A2 ν − 98 A2 .

(19.78)
Setting D = 14 T 2 = µ2 sets the boundary between stable spiral and stable node. Setting
D = 0 sets the boundary between stable node and saddle. The fixed point structure is as
shown in Fig. 19.3.
19.2.3 Van der Pol oscillator
Let’s apply this method to another problem, that of the van der Pol oscillator,
ẍ + (x2 − 1) ẋ + x = 0 , (19.79)
Figure 19.3: Amplitude versus detuning for the forced Duffing oscillator for ten equally
spaced values of f0 between µ3/2 and 10 µ3/2 . The critical value is f0,c = 4.0525 µ3/2 . The
red and blue curves are boundaries for the fixed point classification.
with > 0. The nonlinear term acts as a frictional drag for x > 1, and as a ‘negative
friction’ (i.e. increasing the amplitude) for x < 1. Note that the linearized equation at the
fixed point (x = 0, ẋ = 0) corresponds to an unstable spiral for < 2.
For the van der Pol oscillator, we have h(x, ẋ) = (1 − x2 ) ẋ, and plugging in the zeroth order
solution x0 = A cos(t + φ) gives

∂x0
= 1 − A2 cos2 θ − A sin θ

h x0 ,
∂t0
= − A + 41 A3 sin θ + 14 A3 sin(3θ) ,

(19.80)
with θ ≡ t + φ. Thus, α1 = −A + 14 A3 and β1 = 0, which gives φ = φ0 and

∂A
2 = A − 14 A3 . (19.81)
∂t1
The equation for A is easily integrated:

8 dA 2 1 1 A
dt1 = − = − − dA = d ln
A (A2 − 4) A A−2 A+2 A2 − 4
2
=⇒ A(t1 ) = q . (19.82)
4

1− 1− A20
exp(−t1 )
Thus,
2 cos(t + φ0 )
x0 (t) = q . (19.83)
1 − 1 − A42 exp(−t)

0
This behavior describes the approach to the limit cycle 2 cos(t + φ0 ). With the elimination
of the secular terms, we have
1

1 3 4 sin 3t + 3φ0
x1 (t) = − 32 A sin(3θ) = − 3/2 . (19.84)
4

1 − 1 − A2 exp(−t)
0
19.2.4 Forced van der Pol oscillator
Consider now a weakly dissipative, weakly forced van der Pol oscillator, governed by the
equation
ẍ + (x2 − 1) ẋ + x = f0 cos(t + νt) , (19.85)
where the forcing frequency is Ω = 1 + ν, which is close to the natural frequency ω0 = 1.
We apply the multiple time scale method, with h(x, ẋ) = (1 − x2 ) ẋ. As usual, the lowest

order solution is x0 = A(T ) cos τ + φ(T ) , where τ = t and T = t. Again, we define
θ ≡ τ + φ(T ) and ψ(T ) ≡ φ(T ) − νT . From
h(A cos θ, −A sin θ) = 14 A3 − A sin θ + 14 A3 sin(3θ) ,

(19.86)
we arrive at
2
∂ 2 x0

∂ ∂x0
+ 1 x1 = −2 + h x0 ,
∂θ2 ∂τ ∂T ∂τ
− A + 2A0 + f0 sin ψ sin θ
1 3

= 4A
+ 2A ψ 0 + 2νA + f0 cos ψ cos θ + 14 A3 sin(3θ) .

(19.87)
We eliminate the secular terms, proportional to sin θ and cos θ, by demanding

dA
= 12 A − 18 A3 − 12 f0 sin ψ (19.88)
dT
dψ f0
= −ν − cos ψ . (19.89)
dT 2A
Stationary solutions have A0 = ψ 0 = 0, hence cos ψ = −2ν A/f0 , and hence

2
f02 = 4ν 2A2 + 1 + 14 A2 A2
1 6
= 16 A − 12 A4 + (1 + 4ν 2 )A2 . (19.90)
For this solution, we have

x0 = A∗ cos(τ + νT + ψ ∗ ) , (19.91)
Figure 19.4: Amplitude versus detuning for the forced van der Pol oscillator. Fixed point
classifications are abbreviated SN (stable node), SS (stable spiral), UN (unstable node), US
(unstable spiral), and SP (saddle point).
and the oscillator’s frequency is the forcing frequency Ω. This oscillator is thus entrained.
To proceed further, let y = A2 , and consider the cubic equation

1 3
F (y) = 16 y − 12 y 2 + (1 + 4ν 2 ) y − f02 = 0 . (19.92)
Setting F 0 (y) = 0, we find the roots of F 0 (y) lie at y± = 43 (2 ± u), where u = (1 − 12 ν 2 )1/2 .
1
Thus, the roots are complex for ν 2 > 12 , in which case F (y) is monotonically increasing,
and there is a unique solution to F (y) = 0. Since F (0) = −f02 < 0, that solution satisfies
1
y > 0. For ν 2 < 12 , there are two local extrema at y = y± . When Fmin = F (y+ ) <
0 < F (y− ) = Fmax , the cubic F (y) has three real, positive roots. This is equivalent to the
condition
8 3
− 27 u + 89 u2 < 32 2 8 3 8 2
27 − f0 < 27 u + 9 u . (19.93)
We can say even more by exploring the behavior of (19.4,19.5) in the vicinity of the fixed
points. Writing A = A∗ + δA and ψ = ψ ∗ + δψ, we have
3 ∗2
νA∗
  1  
dT δA 2 1 − 4A δA
 =   . (19.94)
1 ∗2
−ν/A∗ 1

dT δψ 2 1 − 4 A δψ
The eigenvalues of the linearized dynamics at the fixed point are given by λ± = 12 T ±
√
T 2 − 4D , where T and D are the trace and determinant of the linearized equation. Recall
Figure 19.5: Phase diagram for the weakly forced van der Pol oscillator in the (ν 2 , f02 ) plane.
Inset shows detail. Abbreviations for fixed point classifications are as in Fig. 19.4.
now the classification scheme for fixed points of two-dimensional phase flows, discussed in
section 18.2.2. To recapitulate, when D < 0, we have λ− < 0 < λ+ and the fixed point is
a saddle. For 0 < 4D < T 2 , both eigenvalues have the same sign, so the fixed point is a
node. For 4D > T 2 , the eigenvalues form a complex conjugate pair, and the fixed point is
a spiral. A node/spiral fixed point is stable if T < 0 and unstable if T > 0. For our forced
van der Pol oscillator, we have
T = 1 − 12 A∗ 2 (19.95)
∗2 ∗4
1 3
+ ν2 .

D= 4 1−A + 16 A (19.96)
From these results we can obtain q the plot of Fig. 19.4, where amplitude is shown versus
detuning. Note that for f0 < 32

27 there is a region ν− , ν+ of hysteretic behavior in which
varying the detuning parameter ν is not a reversible process. The phase diagram in the
(ν 2 , f02 ) is shown in Fig. 19.5.
Finally, we can make the following statement about the global dynamics (i.e. not simply in
the vicinity of a fixed point). For large A, we have
dA dψ
= − 18 A3 + . . . , = −ν + . . . . (19.97)
dT dT
This flow is inward, hence if the flow is not to a stable fixed point, it must be attracted
Figure 19.6: Forced van der Pol system with = 0.1, ν = 0.4 for three values of f0 . The
limit entrained solution becomes unstable at f0 = 1.334.
to a limit cycle. The limit cycle necessarily involves several frequencies. This result – the
generation of new frequencies by nonlinearities – is called heterodyning.
We can see heterodyning in action in the van der Pol system. In Fig. 19.5, the blue line
which separates stable and unstable spiral solutions is given by f02 = 8ν 2 + 12 . For example,
if we take ν = 0.40 then the boundary lies at f0 = 1.334. For f0 < 1.334, we expect
heterodyning, as the entrained solution is unstable. For f > 1.334 the solution is entrained
19.3. RELAXATION OSCILLATIONS 413
and oscillates at a fixed frequency. This behavior is exhibited in Fig. 19.6.
19.3 Relaxation Oscillations
We saw how to use multiple time scale analysis to identify the limit cycle of the van der
Pol oscillator when is small. Consider now the opposite limit, where the coefficient of the
damping term is very large. We generalize the van der Pol equation to
ẍ + µ Φ(x) ẋ + x = 0 , (19.98)
and suppose µ 1. Define now the variable
Zx
ẋ
y≡ + dx0 Φ(x0 )
µ
0
ẋ
= + F (x) , (19.99)
µ
where F 0 (x) = Φ(x). (y is sometimes called the Liènard variable, and (x, y) the Liènard
plane.) Then the original second order equation may be written as two coupled first order
equations:
x
ẏ = − (19.100)
µ

ẋ = µ y − F (x) . (19.101)
Figure 19.7: Relaxation oscillations in the so-called Liènard plane (x, y). The system rapidly
flows to a point on the curve y = F (x), and then crawls slowly along this curve. The slow
motion takes x from −b to −a, after which the system executes a rapid jump to x = +b,
then a slow retreat to x = +a, followed by a rapid drop to x = −b.
Figure 19.8: A sketch of the limit cycle for the relaxation oscillation studied in this section.
Since µ 1, the first of these equations is slow and the second one fast. The dynamics
rapidly achieves y ≈ F (x), and then slowly evolves along the curve y = F (x), until it is
forced to make a large, fast excursion.
A concrete example is useful. Consider F (x) of the form sketched in Fig. 19.7. This is what
one finds for the van der Pol oscillator, where Φ(x) = x2 − 1 and F (x) = 31 x3 − x. The limit
cycle behavior xLC (t) is sketched in Fig. 19.8. We assume Φ(x) = Φ(−x) for simplicity.
When µ 1 we can determine approximately the period of the limit cycle. Assuming
y = F (x) throughout the slow portion of the cycle, we have
x dx
ẏ = F 0 (x) ẋ = − =⇒ dt = −µ Φ(x) . (19.102)
µ x
The period is then
Zb
Φ(x)
T ' 2µ dx , (19.103)
x
a
where F 0 (±a) = Φ(±a) = 0 and F (±b) = F (∓a). For the van der Pol oscillator, with
Φ(x) = x2 − 1, we have a = 1, b = 2, and T ' (3 − 2 ln 2) µ.
The limit cycle itself is easily obtained in the large µ limit. Assuming F (x) = −F (−x), and
Figure 19.9: Limit cycle for large µ relaxation oscillations, shown in the phase plane (x, ẋ).
with F 0 (±a) = 0 and F (±b) = F (∓a) as above, we have that
x
x ∈ [ −b, −a ] (ẋ > 0) : ẋ ≈ − (19.104)
µ F 0 (x)

x ∈ [ −a, b ] (ẋ > 0) : ẋ ≈ µ F (b) − F (x) (19.105)
x
x ∈ [ a, b ] (ẋ < 0) : ẋ ≈ − (19.106)
µ F 0 (x)

x ∈ [ −b, a ] (ẋ < 0) : ẋ ≈ µ F (a) − F (x) . (19.107)
A sketch of the limit cycle is given in fig. 19.9, showing the slow and fast portions.
19.3.1 Example problem

ẍ + µ |x| − 1 ẋ + x = 0 . (19.108)
Sketch the trajectory in the Liènard plane, and find the approximate period of the limit
cycle for µ 1.

Figure 19.10: Relaxation oscillations for ẍ + µ |x| − 1 ẋ + x = 0 plotted in the Liénard
plane. The solid black curve is y = F (x) = 12 x2 sgn(x) − x. The variable y is defined to be
y = µ−1 ẋ + F (x). Along slow portions of the limit cycle, y ' F (x).
Solution : We define

1 2
+ 2 x − x if x > 0

F 0 (x) = |x| − 1 ⇒ F (x) = (19.109)

 1 2
− 2 x − x if x < 0 .
We therefore have
x
ẋ = µ y − F (x) , ẏ = − , (19.110)
µ
with y ≡ µ−1 ẋ + F (x).
Setting F 0 (x) = 0 we find x = ±a, where a = 1 and F (±a) = ∓ 12 . We also find F (±b) =
√
F (∓a), where b = 1 + 2. Thus, the limit cycle is as follows: (i) fast motion from x = −a
to x = +b, (ii) slow relaxation from x = +b to x = +a, (iii) fast motion from x = +a to
x = −b, and (iv) slow relaxation from x = −b to x = −a. The period is approximately the
time it takes for the slow portions of the cycle. Along these portions, we have y ' F (x),
and hence ẏ ' F 0 (x) ẋ. But ẏ = −x/µ, so
x F 0 (x)
F 0 (x) ẋ ' − ⇒ dt = −µ dx , (19.111)
µ x
Figure 19.11: Liénard plots for systems with one (left) and two (right) relaxation oscillations.
which we integrate to obtain

√
Za 1+
Z 2
F 0 (x)

1
T ' −2µ dx = 2µ dx 1 − (19.112)
x x
b 1
h√ √ i
= 2µ 2 − ln 1 + 2 ' 1.066 µ . (19.113)
19.3.2 Multiple limit cycles
For the equation

ẍ + µ F 0 (x) ẋ + x = 0 , (19.114)
it is illustrative to consider what sort of F (x) would yield more than one limit cycle. Such
an example is shown in fig. 19.11.
In polar coordinates, it is very easy to construct such examples. Consider, for example, the
system
ṙ = sin(πr) + cos θ (19.115)
θ̇ = b r , (19.116)
with || < 1. First consider the case = 0. Clearly the radial flow is outward for sin(πr) > 0
and inward for sin(πr) < 0. Thus, we have stable limit cycles at r = 2n + 1 and unstable
limit cycles at r = 2n, for all n ∈ ZZ. With 0 < || < 1, we have
1
sin−1 , 2n + 1 − 1
sin−1

ṙ > 0 for r ∈ 2n + π π (19.117)
1
sin−1 , 2n + 2 − 1
sin−1

ṙ < 0 for r ∈ 2n + 1 + π π (19.118)
The Poincaré-Bendixson theorem then guarantees the existence of stable and unstable limit
cycles. We can put bounds on the radial extent of these limit cycles.
1
sin−1 , 2n + 1 + 1
sin−1

stable limit cycle : r ∈ 2n + 1 − π π (19.119)
1
sin−1 , 2n + 1
sin−1

unstable limit cycle : r ∈ 2n − π π (19.120)
Note that an unstable limit cycle is a repeller, which is to say that it is stable (an attractor)
if we run the dynamics backwards, sending t → −t.
19.3.3 Example problem
Consider the nonlinear oscillator,
ẍ + µ Φ(x) ẋ + x = 0 ,
with µ 1. For each case in fig. 19.12, sketch the flow in the Liènard plane, starting with
a few different initial conditions. For which case(s) do relaxation oscillations occur?
Solution : Recall the general theory of relaxation oscillations. We define

Zx
ẋ ẋ
y≡ + dx0 Φ(x0 ) = + F (x) ,
µ µ
0
Figure 19.12: Three instances of Φ(x).

19.4. PARAMETRIC OSCILLATOR 419
Figure 19.13: Phase flows in the Liénard plane for the three examples in fig. 19.12.
in which case the second order ODE for the oscillator may be written as two coupled first
order ODEs:
x
ẏ = − , ẋ = µ y − F (x) .
µ
Since µ 1, the first of these equations is slow and the second one fast. The dynamics
rapidly achieves y ≈ F (x), and then slowly evolves along the curve y = F (x), until it is
forced to make a large, fast excursion.
To explore the dynamics in the Liènard plane, we plot F (x) versus x, which means we must
integrate Φ(x). This is done for each of the three cases in fig. 19.12.
Note that a fixed point corresponds to x = 0 and ẋ = 0. In the Liènard plane, this means
x = 0 and y = F (0). Linearizing by setting x = δx and y = F (0) + δy, we have1
µ δy − µF 0 (0) δx −µF 0 (0) µ

d δx δx
= = .
dt δy −µ−1 δx −µ−1 0 δy
The linearized map has trace T = −µ F 0 (0) and determinant D = 1. Since µ 1 we have
0 < D < 41 T 2 , which means the fixed point is either a stable node, for F 0 (0) > 0, or an
unstable node, for F 0 (0) < 0. In cases (a) and (b) the fixed point is a stable node, while in
case (c) it is unstable. The flow in case (a) always collapses to the stable node. In case (b)
the flow either is unbounded or else it collapses to the stable node. In case (c), all initial
conditions eventually flow to a unique limit cycle exhibiting relaxation oscillations.
19.4 Parametric Oscillator

ẍ + ω02 (t) x = 0 , (19.121)
1
We could, of course, linearize about the fixed point in (x, ẋ) space and obtain the same results.
Figure 19.14: Phase diagram for the parametric oscillator in the (θ, ) plane. Thick black
lines correspond to T = ±2. Blue regions: |T | < 2. Red regions: T > 2. Magenta regions:
T < −2.
where the oscillation frequency is a function of time. Equivalently,

M (t) ϕ(t)
z }| { z}|{
d x 0 1 x
= . (19.122)
dt ẋ −ω02 (t) 0 ẋ
The formal solution is the path-ordered exponential,
( Zt )
ϕ(t) = P exp dt0 M (t0 ) ϕ(0) . (19.123)
0
Let’s consider an example in which


(1 + ) ω0
 if 2nτ ≤ t ≤ (2n + 1)τ
ω(t) = (19.124)

(1 − ) ω0 if (2n + 1)τ ≤ t ≤ (2n + 2)τ .

Define ϕn ≡ ϕ(2nτ ). Then
ϕn+1 = exp(M− τ ) exp(M+ τ ) ϕn , (19.125)

19.4. PARAMETRIC OSCILLATOR 421
where
0 1
M± = 2 , (19.126)
−ω± 0
with ω± ≡ (1 ± ) ω0 . Note that M±2 = −ω± 2 · 1 is a multiple of the identity. Evaluating the
Taylor series for the exponential, one finds

!
−1
cos ω± τ ω± sin ω± τ
exp(M± t) = , (19.127)
−ω± sin ω± τ cos ω± τ
from which we derive

a b
Q≡ = exp(M− τ ) exp(M+ τ ) (19.128)
c d
! !
−1 −1
cos ω− τ ω− sin ω− τ cos ω+ τ ω+ sin ω+ τ
=
−ω− sin ω− τ cos ω− τ −ω+ sin ω+ τ cos ω+ τ
with
ω+
a = cos ω− τ cos ω+ τ − sin ω− τ sin ω+ τ (19.129)
ω−
1 1
b= cos ω− τ sin ω+ τ + sin ω− τ cos ω+ τ (19.130)
ω+ ω−
c = −ω+ cos ω− τ sin ω+ τ − ω− sin ω− τ cos ω+ τ (19.131)
ω−
d = cos ω− τ cos ω+ τ − sin ω− τ sin ω+ τ . (19.132)
ω+
Note that det exp(M± τ ) = 1, hence detQ = 1. Also note that
P (λ) = det Q − λ · 1 = λ2 − T λ + ∆ ,

(19.133)
where
T = a + d = Tr Q (19.134)
∆ = ad − bc = detQ . (19.135)
The eigenvalues of Q are p

λ± = 12 T ± 1
2 T 2 − 4∆ . (19.136)
In our case, ∆ = 1. There are two cases to consider:
|T | < 2 : λ+ = λ∗− = eiδ , δ = cos−1 12 T (19.137)

|T | > 2 : λ+ = λ−1
− = ±e µ
, µ= cosh−1 12 |T | . (19.138)
When |T | < 2, ϕ remains bounded; when |T | > 2, |ϕ| increases exponentially with time.
Note that phase space volumes are preserved by the dynamics.
To investigate more fully, let θ ≡ ω0 τ . The period of the ω0 oscillations is δt = 2τ , i.e.

ωpump = π/τ is the frequency at which the system is ‘pumped’. We compute the trace of
Q and find
1 cos(2θ) − 2 cos(2θ)
2T = 1 − 2
. (19.139)
We are interested in the boundaries in the (θ, ) plane where |T | = 2. Setting T = +2, we
write θ = nπ + δ, which means ω0 ≈ nωpump . Expanding for small δ and , we obtain the
relation 1/2
2 4 2
δ
δ = θ ⇒ = . (19.140)
nπ
Setting T = −2, we write θ = (n + 12 )π + δ, i.e. ω0 ≈ (n + 12 ) ωpump . This gives
δ 2 = 2 ⇒ = ±δ . (19.141)
The full phase diagram in the (θ, ) plane is shown in Fig. 19.14. A physical example is
pumping a swing. By extending your legs periodically, youp effectively change the length
`(t) of the pendulum, resulting in a time-dependent ω0 (t) = g/`(t).
19.5 Appendix I : Multiple Time Scale Analysis to O(2 )
Problem : A particle of mass m moves in one dimension subject to the potential
x3
U (x) = 12 m ω02 x2 + 13 m ω02 , (19.142)
a
where is a dimensionless parameter.
(a) Find the equation of motion for x. Show that by rescaling x and t you can write this
equation in dimensionless form as
d2 u
+ u = −u2 . (19.143)
ds2
Solution : The equation of motion is
mẍ = −U 0 (x) (19.144)

x2
= −mω02 x − mω02 . (19.145)
a
We now define s ≡ ω0 t and u ≡ x/a, yielding
d2 u
+ u = −u2 . (19.146)
ds2
19.5. APPENDIX I : MULTIPLE TIME SCALE ANALYSIS TO O(2 ) 423
(b) You are now asked to perform an O 2 multiple time scale analysis of this problem,

writing
τ =s , T = s , Ξ = 2 s ,
and
u = u0 + u1 + 2 u2 + . . . .
This results in a hierarchy of coupled equations for the functions {un }. Derive the first
three equations in the hierarchy.
Solution : We have
d ∂ ∂ ∂
= + + 2 + ... . (19.147)
ds ∂τ ∂T ∂Ξ
Therefore
2
∂τ + ∂T + 2 ∂Ξ + . . . u0 + u1 + 2 u2 + . . . + u0 + u1 + 2 u2 + . . .

(19.148)
2
= − u0 + u1 + 2 u2 + . . . .

(19.149)
Expanding and then collecting terms order by order in , we derive the hierarchy. The first
three levels are
∂τ2 u0 + u0 = 0 (19.150)
∂τ2 u1 + u1 = −2 ∂τ ∂T u0 − u20 (19.151)
∂τ2 u2 + u2 = −2 ∂τ ∂Ξ u0 − ∂T ∂T u0 − 2 ∂τ ∂T u1 − 2 u0 u1 . (19.152)
(c) Show that there is no frequency shift to first order in .
Solution : At the lowest (first) level of the hierarchy, the solution is

u0 = A(T, Ξ) cos τ + φ(T, Ξ) .
At the second level, then,
∂ 2 u1 ∂A ∂φ
2
+ u1 = 2 sin(τ + φ) + 2A cos(τ + φ) − A2 cos2(τ + φ) .
∂τ ∂T ∂T
We eliminate the resonant forcing terms on the RHS by demanding
∂A ∂φ
=0 and =0.
∂T ∂T
Thus, we must have A = A(Ξ) and φ = φ(Ξ). To O(), then, φ is a constant, which means
there is no frequency shift at this level of the hierarchy.
(d) Find u0 (s) and u1 (s).

Solution :The equation for u1 is that of a non-resonantly forced harmonic oscillator. The
solution is easily found to be
u1 = − 12 A2 + 16 A2 cos(2τ + 2φ) .
We now insert this into the RHS of the third equation in the hierarchy:
∂ 2 u2 ∂ 2 u0
+ u 2 = −2 − 2 u0 u1
∂τ 2 ∂τ ∂Ξ
∂A ∂φ n o
=2 sin(τ + φ) + 2A cos(τ + φ) − 2A cos(τ + φ) − 12 A2 + 16 A2 cos(2τ + 2φ)
∂Ξ ∂Ξ
∂A ∂φ
=2 sin(τ + φ) + 2A + 56 A3 cos(τ + φ) − 16 A3 cos(3τ + 3φ) .
∂Ξ ∂Ξ
Setting the coefficients of the resonant terms on the RHS to zero yields
∂A
=0 ⇒ A = A0
∂Ξ
∂φ
2A + 56 A3 = 0 ⇒ 5
φ = − 12 A20 Ξ .
∂Ξ
Therefore,
u0 (s) u1 (s)
z }| { z }| {
5 2 2
A0 s + 16 A20 cos 2s − 56 2 A20 s − 12 A20 +O 2

u(s) = A0 cos s − 12
19.6 Appendix II : MSA and Poincaré-Lindstedt Methods
19.6.1 Problem using multiple time scale analysis
2 −3
Consider the central force law F (r) = −k rβ .
2
(a) Show that a stable circular orbit exists at radius r0 = (`2 /µk)1/β .
Solution : For a circular orbit, the effective radial force must vanish:
`2 `2 k
Feff (r) = 3
+ F (r) = 3
− 3−β 2 = 0 . (19.153)
µr µr r
2
Solving for r = r0 , we have r0 = (`2 /µk)1/β . The second derivative of Ueff (r) at this point
is
00 0 3`2 k β 2 `2
Ueff (r0 ) = −Feff (r0 ) = 4 + (β 2 − 3) 4−β 2 = , (19.154)
µr0 r µr04
0
which is manifestly positive. Thus, the circular orbit at r = r0 is stable.

19.6. APPENDIX II : MSA AND POINCARÉ-LINDSTEDT METHODS 425
(b) Show that the geometric equation for the shape of the orbit may be written
d2 s
+ s = K(s) (19.155)
dφ2
where s = 1/r, and

1−β 2
s
K(s) = s0 , (19.156)
s0
with s0 = 1/r0 .
Solution : We have previously derived (e.g. in the notes) the equation
d2 s µ
+ s = − 2 2 F (s−1 ) . (19.157)
dφ2 ` s
From the given F (r), we then have
d2 s µk 2
2
+ s = 2 s1−β ≡ K(s) , (19.158)
dφ `
2
where s0 ≡ (µk/`2 )1/β = 1/r0 , and where
1−β 2
s
K(s) = s0 . (19.159)
s0
(c) Writing s ≡ (1 + u) s0 , show that u satisfies
1 d2 u
+ u = a1 u2 + a2 u3 + . . . . (19.160)
β 2 dφ2
Find a1 and a2 .
Solution : Writing s ≡ s0 (1 + u), we have
d2 u 2
2
+ 1 + u = (1 + u)1−β
dφ
= 1 + (1 − β 2 ) u + 12 (−β 2 )(1 − β 2 ) u2
+ 16 (−1 − β 2 )(−β 2 )(1 − β 2 ) u3 + . . . . (19.161)
Thus,
1 d2 u
+ u = a1 u2 + a2 u3 + . . . , (19.162)
β 2 dφ2
where
a1 = − 12 (1 − β 2 ) , a2 = 16 (1 − β 4 ) . (19.163)
(d) Now let us associate a power of ε with each power of the deviation u and write
1 d2 u
+ u = ε a1 u2 + ε2 a2 u3 + . . . , (19.164)
β 2 dφ2
Solve this equation using the method of multiple scale analysis (MSA). You will have to go
to second order in the multiple scale expansion, writing
X ≡ βφ , Y ≡ ε βφ , Z ≡ ε2 βφ (19.165)
and hence
1 d ∂ ∂ ∂
= +ε + ε2 + ... . (19.166)
β dφ ∂X ∂Y ∂Z
Further writing
u = u0 + ε u1 + ε2 u2 + . . . , (19.167)
derive the equations for the multiple scale analysis, up to second order in ε.
Solution : We now associate one power of ε with each additional power of u beyond order
u1 . In this way, a uniform expansion in terms of ε will turn out to be an expansion in
powers of the amplitude of the oscillations. We’ll see how this works below. We then have
1 d2 u
+ u = a1 ε u2 + a2 ε2 u3 + . . . , (19.168)
β 2 dφ2
with ε = 1. We now perform a multiple scale analysis, writing
X ≡ βφ , Y ≡ ε βφ , Z ≡ ε2 βφ . (19.169)
This entails
1 d ∂ ∂ ∂
= +ε + ε2 + ... . (19.170)
β dφ ∂X ∂Y ∂Z
We also expand u in powers of ε, as
u = u0 + ε u1 + ε2 u2 + . . . . (19.171)
Thus, we obtain
2
∂X + ε ∂Y + ε2 ∂Z + . . . (u0 + εu1 + ε2 u2 + . . . ) + (u0 + εu1 + ε2 u2 + . . . )
= ε a1 (u0 + εu1 + ε2 u2 + . . . )2 + ε2 a2 (u0 + εu1 + ε2 u2 + . . . )3 + . . . . (19.172)
We now extract a hierarchy of equations, order by order in powers of ε.
We find, out to order ε2 ,

∂ 2 u0
O(ε0 ) : + u0 = 0 (19.173)
∂X 2
∂ 2 u1 ∂ 2 u0
O(ε1 ) : + u 1 = −2 + a1 u20 (19.174)
∂X 2 ∂Y ∂X
∂ 2 u2 ∂ 2 u0 ∂ 2 u0 ∂ 2 u1
O(ε2 ) : + u 2 = −2 − − 2 + 2a1 u0 u1 + a2 u30 . (19.175)
∂X 2 ∂Z ∂X ∂Y 2 ∂Z ∂X
(e) Show that there is no shift of the angular period ∆φ = 2π/β if one works only to
leading order in ε.
Solution : The O(ε0 ) equation in the hierarchy is solved by writing
u0 = A cos(X + ψ) , (19.176)
where
A = A(Y, Z) , ψ = ψ(Y, Z) . (19.177)
We define θ ≡ X + ψ(Y, Z), so we may write u0 = A cos θ. At the next order, we obtain
∂ 2 u1 ∂A ∂ψ
+ u1 = 2 sin θ + 2A cos θ + a1 A2 cos θ
∂θ2 ∂Y ∂Y
∂A ∂ψ
=2 sin θ + 2A cos θ + 21 a1 A2 + 12 a1 A2 cos 2θ . (19.178)
∂Y ∂Y
In order that there be no resonantly forcing terms on the RHS of eqn. 20.39, we demand
∂A ∂ψ
=0 , =0 ⇒ A = A(Z) , ψ = ψ(Z) . (19.179)
∂Y ∂Y
The solution for u1 is then
u1 (θ) = 12 a1 A2 − 16 a1 A2 cos 2θ . (19.180)
Were we to stop at this order, we could ignore Z = ε2 βφ entirely, since it is of order ε2 ,

and the solution would be
u(φ) = A0 cos(βφ + ψ0 ) + 12 εa1 A20 − 16 εa1 A20 cos(2βφ + 2ψ0 ) . (19.181)
The angular period is still ∆φ = 2π/β, and, starting from a small amplitude solution at
order ε0 we find that to order ε we must add a constant shift proportional to A20 , as well as
a second harmonic term, also proportional to A20 .
(f) Carrying out the MSA to second order in ε, show that the shift of the angular period
vanishes only if β 2 = 1 or β 2 = 4.
Solution : Carrying out the MSA to the next order, O(ε2 ), we obtain
∂ 2 u2 ∂A ∂ψ 2
1
− 16 a1 A2 cos 2θ + a2 A3 cos3 θ

+ u2 = 2 sin θ + 2A cos θ + 2a1 A cos θ 2 a1 A
∂θ2 ∂Z ∂Z
∂A ∂ψ 5 2
+ 34 a2 A3 cos θ + − 61 a21 + 14 a2 A3 cos 3θ .

=2 sin θ + 2A cos θ + 6 a1
∂Z ∂Z
(19.182)
Now in order to make the resonant forcing terms on the RHS vanish, we must choose
∂A
=0 (19.183)
∂Z
as well as
∂ψ 5 2
+ 38 a2 A2

=− 12 a1 (19.184)
∂Z
1
= − 24 (β 2 − 4)(β 2 − 1) . (19.185)
The solutions to these equations are trivial:

2
A(Z) = A0 , ψ(Z) = ψ0 − 1
24 (β − 1)(β 2 − 4)A20 Z . (19.186)
With the resonant forcing terms eliminated, we may write

∂ 2 u2
+ u2 = − 16 a21 + 14 a2 A3 cos 3θ ,

2
(19.187)
∂θ
with solution
2
u2 = 1
96 (2a1 − 3a2 ) A3 cos 3θ
1
β 2 (β 2 − 1) A20 cos 3X + 3ψ(Z) .

= 96 (19.188)
The full solution to second order in this analysis is then

u(φ) = A0 cos(β 0 φ + ψ0 ) + 12 εa1 A20 − 16 εa1 A20 cos(2β 0 φ + 2ψ0 )
+ 1 2 2
96 ε (2a1 − 3a2 ) A30 cos(3β 0 φ + 3ψ0 ) . (19.189)
with n o
β0 = β · 1 − 1
24 ε2 (β 2 − 1)(β 2 − 4)A20 . (19.190)
The angular period shifts:
2π 2π n o
∆φ = = · 1+ 1
24 ε2 (β 2 − 1)(β 2 − 4)A20 + O(ε3 ) . (19.191)
β0 β
Note that there is no shift in the period, for any amplitude, if β 2 = 1 (i.e. Kepler potential)
or β 2 = 4 (i.e. harmonic oscillator).
19.6.2 Solution using Poincaré-Lindstedt method
Recall that geometric equation for the shape of the (relative coordinate) orbit for the two
body central force problem is
d2 s
+ s = K(s) (19.192)
dφ2
1−β 2
s
K(s) = s0 (19.193)
s0
2
where s = 1/r, s0 = (l2 /µk)1/β is the inverse radius of the stable circular orbit, and
2
f (r) = −krβ −3 is the central force. Expanding about the stable circular orbit, one has
d2 y
+ β 2 y = 12 K 00 (s0 ) y 2 + 16 K 000 (s0 ) y 3 + . . . , (19.194)
dφ2
where s = s0 (1 + y), with
β 2
0 2 s0
K (s) = (1 − β ) (19.195)
s
1+β 2
00 2 2 s0
K (s) = −β (1 − β ) (19.196)
s
2+β 2
s0
K 000 (s) = β 2 (1 − β 2 ) (1 + β 2 ) . (19.197)
s
Thus,
d2 y
+ β 2 y = a1 y 2 + 2 a2 y 3 , (19.198)
dφ2
with = 1 and
a1 = − 21 β 2 (1 − β 2 ) (19.199)
a2 = + 16 β 2 (1 − β 2 ) (1 + β 2 ) . (19.200)
Note that we assign one factor of for each order of nonlinearity beyond order y 1 . Note
also that while y here corresponds to u in eqn. 19.162, the constants a1,2 here are a factor
of β 2 larger than those defined in eqn. 19.163.
We now apply the Poincaré-Lindstedt method, by defining θ = Ω φ, with

Ω2 = Ω20 + Ω21 + 2 Ω22 + . . . (19.201)
and
y(θ) = y0 (θ) + y1 (θ) + 2 y2 (θ) + . . . . (19.202)
We therefore have
d d
=Ω (19.203)
dφ dθ
and
Ω20 + Ω21 +2 Ω22 + . . . y000 + y100 + 2 y200 + . . . + β 2 y0 + y1 + 2 y2 + . . .

(19.204)
2 3
y0 + y1 + 2 y2 + . . . + 2 a2 y0 + y1 + 2 y2 + . . . .

= a1
(19.205)
We now extract equations at successive orders of . The first three in the hierarchy are
Ω20 y000 + β 2 y0 = 0 (19.206)
Ω21 y000 + Ω20 y100 + β 2 y1 = a1 y02 (19.207)
Ω22 y000 + Ω21 y100 + Ω20 y200 + β 2 y2 = 2 a1 y0 y1 + a2 y03 , (19.208)

where prime denotes differentiation with respect to θ.
To order 0 , the solution is Ω20 = β 2 and
y0 (θ) = A cos(θ + δ) , (19.209)
where A and δ are constants.
At order 1 , we have
β 2 y100 + y1 = −Ω21 y000 + a1 y02

= Ω21 A cos(θ + δ) + a1 A2 cos2 (θ + δ)

= Ω21 A cos(θ + δ) + 12 a1 A2 + 12 a1 A2 cos(2θ + 2δ) . (19.210)
The secular forcing terms on the RHS are eliminated by the choice Ω21 = 0. The solution is
then
a1 A2 n 1
o
y1 (θ) = 1 − 3 cos(2θ + 2δ) . (19.211)
2 β2
At order 2 , then, we have
β 2 y200 + y2 = −Ω22 y000 − Ω21 y100 + 2 a1 y1 y1 + a2 y03

a2 A3 n o
= Ω22 A cos(θ + δ) + 1 2 1 − 13 cos(2θ + 2δ) cos(θ + δ) + a2 A3 cos2 (θ + δ)
β
2 3 a21 A3 1

2 5 a1 A 3 3 3
= Ω2 + + 4 a2 A A cos(θ + δ) + − + 4 a2 A cos(3θ + 3δ) .
6β 2 6β 2
(19.212)
The resonant forcing terms on the RHS are eliminated by the choice

Ω22 = − 65 β −2 a21 + 34 a2 A3
h i
1 2
= − 24 β (1 − β 2 ) 5 (1 − β 2 ) + 3 (1 + β 2 )
1 2
= − 12 β (1 − β 2 ) (4 − β 2 ) . (19.213)
Thus, the frequency shift to this order vanishes whenever β 2 = 0, β 2 = 1, or β 2 = 4. Recall

2
the force law is F (r) = −C rβ −3 , so we see that there is no shift – hence no precession –
for inverse cube, inverse square, or linear forces.
19.7 Appendix III : Modified van der Pol Oscillator
Consider the nonlinear oscillator
ẍ + (x4 − 1) ẋ + x = 0 . (19.214)
19.7. APPENDIX III : MODIFIED VAN DER POL OSCILLATOR 431
Figure 19.15: Sketch of phase flow and nullclines for the oscillator ẍ + (x4 − 1) ẋ + x = 0.
Red nullclines: v̇ = 0; blue nullcline: ẋ = 0.
Analyze this using the same approach we apply to the van der Pol oscillator.
(a) Sketch the vector field ϕ̇ for this problem. It may prove convenient to first identify the
nullclines, which are the curves along which ẋ = 0 or v̇ = 0 (with v = ẋ). Argue that a
limit cycle exists.
Solution : There is a single fixed point, at the origin (0, 0), for which the linearized
dynamics obeys
d x 0 1 x
= + O(x4 v) . (19.215)
dt v −1 v
One finds T = and D = 1 for the trace and determinant, respectively. The origin is an
unstable spiral for 0 < < 2 and an unstable node for > 2.
The nullclines are sketched in Fig. 19.15. One has

1 x
ẋ = 0 ↔ v = 0 , v̇ = 0 ↔ v = . (19.216)
1 − x4
The flow at large distances from the origin winds once around the origin and spirals in. The
flow close to the origin spirals out ( < 2) or flows radially out ( > 2). Ultimately the flow
must collapse to a limit cycle, as can be seen in the accompanying figures.
(b) In the limit 0 < ε 1, use multiple time scale analysis to obtain a solution which
reveals the approach to the limit cycle.
Solution : We seek to solve the equation
ẍ + x = h(x, ẋ) , (19.217)
with
h(x, ẋ) = (1 − x4 ) ẋ . (19.218)
Employing the multiple time scale analysis to lowest nontrivial order, we write τ ≡ t, T ≡ t,
x = x0 + x1 + . . . (19.219)
and identify terms order by order in . At O(0 ), this yields
∂ 2 x0
+ x0 = 0 x0 = A cos(τ + φ) , (19.220)
∂τ 2
where A = A(T ) and φ = φ(T ). At O(1 ), we have
∂ 2 x1 ∂ 2 x0

∂x0
+ x1 = −2 + h x0 ,
∂τ 2 ∂T ∂τ ∂τ
∂A ∂φ
=2 sin θ + 2A cos θ + h A cos θ, −A sin θ (19.221)
∂T ∂T
with θ = τ + φ(T ) as usual. We also have
h(A cos θ, −A sin θ = A5 sin θ cos θ − A sin θ

= 18 A5 − A sin θ + 163
A5 sin 3θ + 1
A5 sin 5θ .

16 (19.222)
To eliminate the resonant terms in eqn. 19.221, we must choose
∂A 1 5 ∂φ
= 12 A − 16 A , =0. (19.223)
∂T ∂T
The A equation is similar to the logistic equation. Clearly A = 0 is an unstable fixed point,
and A = 81/4 ≈ 1.681793 is a stable fixed point. Thus, the amplitude of the oscillations
will asymptotically approach A∗ = 81/4 . (Recall the asymptotic amplitude in the van der
Pol case was A∗ = 2.)
To integrate the A equation, substitute y = √1 A2 , and obtain

8
dy y2 1
dT = = 1
d ln ⇒ y 2 (T ) = . (19.224)
y(1 − y 2 ) 2 1 − y2 1+ (y0−2 − 1) exp(−2T )
We then have
!1/4
1/4 1/2 8
A(T ) = 8 y (T ) = −4 . (19.225)
1 + (8A0 − 1) exp(−2T )
(c) In the limit 1, find the period of relaxation oscillations, using Liénard plane
analysis. Sketch the orbit of the relaxation oscillation in the Liénard plane.
Solution : Our nonlinear oscillator may be written in the form
dF (x)
ẍ + +x=0 , (19.226)
dt
with
F (x) = 15 x5 − x . (19.227)
Note Ḟ = (x4 − 1) ẋ. Now we define the Liénard variable
ẋ
y≡ + F (x) , (19.228)

and in terms of (x, y) we have
h i x
ẋ = y − F (x) , ẏ = − . (19.229)

As we have seen in the notes, for large the motion in the (x, y) plane is easily analyzed.
x(t) must move quickly over to the curve y = F (x), at which point the motion slows down
and slowly creeps along this curve until it can no longer do so, at which point another big
fast jump occurs. The jumps take place between the local extrema of F (x), which occur
for F 0 (a) = a4 − 1 = 0, i.e. at a = ±1, and points on the curve with the same values of
F (a). Thus, we solve F (−1) = 54 = 15 b5 − b and find the desired root at b∗ ≈ 1.650629. The
period of the relaxation oscillations, for large , is
Zb
F 0 (x) h ib
T ≈ 2 dx = · 12 x4 − 2 ln x ≈ 2.20935 . (19.230)
x a
a
(d) Numerically integrate the equation (19.214) starting from several different initial con-
ditions.
Solution : The accompanying Mathematica plots show x(t) and v(t) for this system for
two representative values of .
Figure 19.16: Vector field and phase curves for the oscillator ẍ + (x4 − 1) ẋ + x = 0, with
= 1 and starting from (x0 , v0 ) = (1, 1).
Figure 19.17: Solution to the oscillator equation ẍ + (x4 − 1) ẋ + x = 0 with = 1 and

initial conditions (x0 , v0 ) = (1, 3). x(t) is shown in red and v(t) in blue. Note that x(t)
resembles a relaxation oscillation for this moderate value of .
Figure 19.18: Vector field and phase curves for the oscillator ẍ + (x4 − 1) ẋ + x = 0, with
= 0.25 and starting from (x0 , v0 ) = (1, 1). As → 0, the limit cycle is a circle of radius
A∗ = 81/4 ≈ 1.682.
Figure 19.19: Solution to the oscillator equation ẍ + (x4 − 1) ẋ + x = 0 with = 0.25 and
initial conditions (x0 , v0 ) = (1, 3). x(t) is shown in red and v(t) in blue. As → 0, the
amplitude of the oscillations tends to A∗ = 81/4 ≈ 1.682.
Chapter 20
Hamiltonian Mechanics
20.1 The Hamiltonian
Recall that L = L(q, q̇, t), and

∂L
pσ = . (20.1)
∂ q̇σ
The Hamiltonian, H(q, p) is obtained by a Legendre transformation,
n
X
H(q, p) = pσ q̇σ − L . (20.2)
σ=1
Note that
n
X ∂L ∂L ∂L
dH = pσ dq̇σ + q̇σ dpσ − dqσ − dq̇ − dt
∂qσ ∂ q̇σ σ ∂t
σ=1
n
X ∂L ∂L
= q̇σ dpσ − dqσ − dt . (20.3)
∂qσ ∂t
σ=1
Thus, we obtain Hamilton’s equations of motion,

∂H ∂H ∂L
= q̇σ , =− = −ṗσ (20.4)
∂pσ ∂qσ ∂qσ
and
dH ∂H ∂L
= =− . (20.5)
dt ∂t ∂t
Some remarks:
• As an example, consider a particle moving in three dimensions, described by spherical

polar coordinates (r, θ, φ). Then
L = 21 m ṙ2 + r2 θ̇2 + r2 sin2 θ φ̇2 − U (r, θ, φ) .

(20.6)
437
438 CHAPTER 20. HAMILTONIAN MECHANICS
We have
∂L ∂L ∂L
pr = = mṙ , pθ = = mr2 θ̇ , pφ = = mr2 sin2 θ φ̇ , (20.7)
∂ ṙ ∂ θ̇ ∂ φ̇
and thus
H = pr ṙ + pθ θ̇ + pφ φ̇ − L
p2r p2θ p2φ

= + + + U (r, θ, φ) . (20.8)
2m 2mr2 2mr2 sin2 θ
∂H dH
Note that H is time-independent, hence ∂t = dt = 0, and therefore H is a constant
of the motion.
∂L
• In order to obtain H(q, p) we must invert the relation pσ = ∂ q̇σ = pσ (q, q̇) to obtain
q̇σ (q, p). This is possible if the Hessian,
∂pα ∂ 2L
= (20.9)
∂ q̇β ∂ q̇α ∂ q̇β
is nonsingular. This is the content of the ‘inverse function theorem’ of multivariable

calculus.
• Define the rank 2n vector, ξ, by its components,
(
qi if 1 ≤ i ≤ n
ξi = (20.10)
pi−n if n ≤ i ≤ 2n .
Then we may write Hamilton’s equations compactly as
∂H
ξ˙i = Jij , (20.11)
∂ξj
where
!
0n×n 1n×n
J= (20.12)
−1n×n 0n×n
is a rank 2n matrix. Note that J t = −J, i.e. J is antisymmetric, and that J 2 =

−12n×2n . We shall utilize this ‘symplectic structure’ to Hamilton’s equations shortly.
20.2. MODIFIED HAMILTON’S PRINCIPLE 439
20.2 Modified Hamilton’s Principle
We have that
Ztb Ztb

0 = δ dt L = δ dt pσ q̇σ − H (20.13)
ta ta
Ztb
∂H ∂H
= dt pσ δ q̇σ + q̇σ δpσ − δq − δp
∂qσ σ ∂pσ σ
ta
Ztb ( )
∂H ∂H tb
= dt − ṗσ + δqσ + q̇σ − δpσ + pσ δqσ ,
∂qσ ∂pσ ta
ta
assuming δqσ (ta ) = δqσ (tb ) = 0. Setting the coefficients of δqσ and δpσ to zero, we recover
Hamilton’s equations.
20.3 Phase Flow is Incompressible
A flow for which ∇ · v = 0 is incompressible – we shall see why in a moment. Let’s check
that the divergence of the phase space velocity does indeed vanish:
n
X ∂ q̇σ ∂ ṗσ
∇ · ξ̇ = +
∂qσ ∂pσ
σ=1
2n
X ∂ ξ˙i X ∂ 2H
= = Jij =0. (20.14)
∂ξi ∂ξi ∂ξj
i=1 i,j
Now let ρ(ξ, t) be a distribution on phase space. Continuity implies
∂ρ
+ ∇ · (ρ ξ̇) = 0 . (20.15)
∂t
Invoking ∇ · ξ̇ = 0, we have that
Dρ ∂ρ
= + ξ̇ · ∇ρ = 0 , (20.16)
Dt ∂t
where Dρ/Dt is sometimes

called the convective derivative – it is the total derivative of the
function ρ ξ(t), t , evaluated at a point ξ(t) in phase space which moves according to the
dynamics. This says that the density in the “comoving frame” is locally constant.
20.4 Poincaré Recurrence Theorem
Let gτ be the ‘τ -advance mapping’ which evolves points in phase space according to Hamil-
ton’s equations
∂H ∂H
q̇i = + , ṗi = − (20.17)
∂pi ∂qi
for a time interval ∆t = τ . Consider a region Ω in phase space. Define gτn Ω to be the
nth image of Ω under the mapping gτ . Clearly gτ is invertible; the inverse is obtained by
integrating the equations of motion backward in time. We denote the inverse of gτ by gτ−1 .
By Liouville’s theorem, gτ is volume preserving when acting on regions in phase space, since
the evolution of any given point is Hamiltonian. This follows from the continuity equation
for the phase space density,
∂%
+ ∇ · (u%) = 0 (20.18)
∂t
where u = {q̇, ṗ} is the velocity vector in phase space, and Hamilton’s equations, which
say that the phase flow is incompressible, i.e. ∇ · u = 0:
n
X ∂ q̇i ∂ ṗi
∇·u = +
∂qi ∂pi
i=1
n
( )
X ∂ ∂H ∂ ∂H
= + − =0. (20.19)
∂qi ∂pi ∂pi ∂qi
i=1
Thus, we have that the convective derivative vanishes, viz.

D% ∂%
≡ + u · ∇% = 0 , (20.20)
Dt ∂t
which guarantees that the density remains constant in a frame moving with the flow.
The proof of the recurrence theorem is simple. Assume that gτ is invertible and volume-
preserving, as is the case for Hamiltonian flow. Further assume that phase space volume
is finite. Since the energy is preserved in the case of time-independent Hamiltonians, we
simply ask that the volume of phase space at fixed total energy E be finite, i.e.
Z

dµ δ E − H(q, p) < ∞ , (20.21)
where dµ = dq dp is the phase space uniform integration measure.
Theorem: In any finite neighborhood Ω of phase space there exists a point ϕ0 which will
return to Ω after n applications of gτ , where n is finite.
Proof: Assume the theorem fails; we will show this assumption results in a contradiction.
Consider the set Υ formed from the union of all sets gτm Ω for all m:
∞
[
Υ= gτm Ω (20.22)
m=0
20.5. KAC RING MODEL 441
We assume that the set {gτm Ω | m ∈ Z , m ≥ 0} is disjoint. The volume of a union of disjoint
sets is the sum of the individual volumes. Thus,
∞
X
vol(Υ) = vol(gτm Ω)
m=0
∞
X
= vol(Ω) · 1=∞, (20.23)
m=1
since vol(gτm Ω) = vol(Ω) from volume preservation. But clearly Υ is a subset of the entire
phase space, hence we have a contradiction, because by assumption phase space is of finite
volume.
Thus, the assumption that the set {gτm Ω | m ∈ Z , m ≥ 0} is disjoint fails. This means that
there exists some pair of integers k and l, with k 6= l, such that gτk Ω ∩ gτl Ω 6= ∅. Without
loss of generality we may assume k > l. Apply the inverse gτ−1 to this relation l times to get
gτk−l Ω∩Ω 6= ∅. Now choose any point ϕ ∈ gτn Ω∩Ω, where n = k −l, and define ϕ0 = gτ−n ϕ.
Then by construction both ϕ0 and gτn ϕ0 lie within Ω and the theorem is proven.
Each of the two central assumptions – invertibility and volume preservation – is crucial.
Without either of them, the proof fails. Consider, for example, a volume-preserving map
which is not invertible. An example might be a mapping f : R → R which takes any real
number to its fractional part. Thus, f (π) = 0.14159265 . . .. Let us restrict our attention
to intervals of width less than unity. Clearly f is then volume preserving. The action of f
on the interval [2, 3) is to map it to the interval [0, 1). But [0, 1) remains fixed under the
action of f , so no point within the interval [2, 3) will ever return under repeated iterations
of f . Thus, f does not exhibit Poincaré recurrence.
Consider next the case of the damped harmonic oscillator. In this case, phase space volumes
contract. For a one-dimensional oscillator obeying ẍ+2β ẋ+Ω02 x = 0 one has ∇·u = −2β <
0 (β > 0 for damping). Thus the convective derivative obeys Dt % = −(∇·u)% = +2β% which
says that the density increases exponentially in the comoving frame, as %(t) = e2βt %(0).
Thus, phase space volumes collapse, and are not preserved by the dynamics. In this case, it
is possible for the set Υ to be of finite volume, even if it is the union of an infinite number of
sets gτn Ω, because the volumes of these component sets themselves decrease exponentially,
as vol(gτn Ω) = e−2nβτ vol(Ω). A damped pendulum, released from rest at some small angle
θ0 , will not return arbitrarily close to these initial conditions.
20.5 Kac Ring Model
The implications of the Poincaré recurrence theorem are surprising – even shocking. If one
takes a bottle of perfume in a sealed, evacuated room and opens it, the perfume molecules
will diffuse throughout the room. The recurrence theorem guarantees that after some finite
time T all the molecules will go back inside the bottle (and arbitrarily close to their initial
Figure 20.1: Left: A configuration of the Kac ring with N = 16 sites and F = 4 flippers. The
flippers, which live on the links, are represented by blue dots. Right: The ring system after
one time step. Evolution proceeds by clockwise rotation. Spins passing through flippers are
flipped.
velocities as well). The hitch is that this could take a very long time, e.g. much much longer
than the age of the Universe.
On less absurd time scales, we know that most systems come to thermodynamic equilibrium.
But how can a system both exhibit equilibration and Poincaré recurrence? The two concepts
seem utterly incompatible!
A beautifully simple model due to Kac shows how a recurrent system can exhibit the
phenomenon of equilibration. Consider a ring with N sites. On each site, place a ‘spin’
which can be in one of two states: up or down. Along the N links of the system, F of
them contain ‘flippers’. The configuration of the flippers is set at the outset and never
changes. The dynamics of the system are as follows: during each time step, every spin
moves clockwise a distance of one lattice spacing. Spins which pass through flippers reverse
their orientation: up becomes down, and down becomes up.
The ‘phase space’ for this system consists of 2N discrete configurations. Since each configu-
ration maps onto a unique image under the evolution of the system, phase space ‘volume’ is
preserved. The evolution is invertible; the inverse is obtained simply by rotating the spins
counterclockwise. Figure ?? depicts an example configuration for the system, and its first
iteration under the dynamics.
Suppose the flippers were not fixed, but moved about randomly. In this case, we could focus
on a single spin and determine its configuration probabilistically. Let pn be the probability
that a given spin is in the up configuration at time n. The probability that it is up at time
(n + 1) is then
pn+1 = (1 − x) pn + x (1 − pn ) , (20.24)
20.5. KAC RING MODEL 443
Figure 20.2: Two simulations of the Kac ring model, each with N = 1000 sites and with
F = 100 flippers (top panel) and F = 24 flippers (bottom panel). The red line shows the
magnetization as a function of time, starting from an initial configuration in which 90% of
the spins are up. The blue line shows the prediction of the Stosszahlansatz , which yields
an exponentially decaying magnetization with time constant τ .
where x = F/N is the fraction of flippers in the system. In words: a spin will be up at
time (n + 1) if it was up at time n and did not pass through a flipper, or if it was down at
time n and did pas through a flipper. If the flipper locations are randomized at each time
step, then the probability of flipping is simply x = F/N . Equation 20.24 can be solved
immediately:
pn = 12 + (1 − 2x)n (p0 − 21 ) , (20.25)
which decays exponentially to the equilibrium value of peq = 12 with time scale τ =
−1/ ln |1 − 2x|. If we define the magnetization m ≡ (N↑ − N↓ )/N , then m = 2p − 1,
so mn = (1 − 2x)n m0 . The equilibrium magnetization is meq = 0. Note that for 21 < x < 1
that the magnetization reverses sign each time step, as well as decreasing exponentially in
magnitude.
The assumption that leads to equation 20.24 is called the Stosszahlansatz . The resulting
dynamics are irreversible: the magnetization inexorably decays to zero. However, the Kac
ring model is purely deterministic, and the Stosszahlansatz can at best be an approximation
to the true dynamics. Clearly the Stosszahlansatz fails to account for correlations such as
the following: if spin i is flipped at time n, then spin i+1 will have been flipped at time n−1.
Indeed, since the dynamics of the Kac ring model are invertible and volume preserving, it
must exhibit Poincaré recurrence.
The model is trivial to simulate. The results of such a simulation are shown in figure 20.2
for a ring of N = 1000 sites, with F = 100 and F = 24 flippers. Note how the magnetization
decays and fluctuates about the equilibrium value eq = 0, but that after N iterations m
Figure 20.3: Simulations of the Kac ring model. Top: N = 1000 sites with F = 900 flippers.
The flipper density x = F/N is greater than 21 , so the magnetization reverses sign every
time step. Only 100 iterations are shown, and the blue curve depicts the absolute value of
the magnetization within the Stosszahlansatz . Bottom: N = 25, 000 sites with F = 1000
flippers. Note that the fluctuations about the ‘equilibrium’ magnetization m = 0 are much
smaller than in the N = 1000 site simulations.
recovers its initial value: mN = m0 . The recurrence time for this system is simply N if F is
even, and 2N if F is odd, since every spin will then have flipped an even number of times.
In figure 20.3 we plot two other simulations. The top panel shows what happens when
x > 12 , so that the magnetization wants to reverse its sign with every iteration. The bottom
panel shows a simulation for a larger ring, with N = 25000 sites. Note that the fluctuations
in m about equilibrium are smaller than in the cases with N = 1000 sites. Why?
20.6 Poisson Brackets
The time evolution of any function F (q, p) over phase space is given by
n
d ∂F X ∂F ∂F
F q(t), p(t), t = + q̇ + ṗ
dt ∂t ∂qσ σ ∂pσ σ
σ=1
∂F
≡ + F, H , (20.26)
∂t
20.7. CANONICAL TRANSFORMATIONS 445
where the Poisson bracket {· , ·} is given by

n
X ∂A ∂B ∂A ∂B
A, B ≡ − (20.27)
∂qσ ∂pσ ∂pσ ∂qσ
σ=1
2n
X ∂A ∂B
= Jij . (20.28)
∂ξi ∂ξj
i,j=1
Properties of the Poisson bracket:
• Antisymmetry:
f, g = − g, f . (20.29)
• Bilinearity: if λ is a constant, and f , g, and h are functions on phase space, then

f + λ g, h = f, h + λ{g, h . (20.30)
Linearity in the second argument follows from this and the antisymmetry condition.
• Associativity:
f g, h = f g, h + g f, h . (20.31)
• Jacobi identity:
f, {g, h} + g, {h, f } + h, {f, g} = 0 . (20.32)
Some other useful properties:
∂A dA
◦ If {A, H} = 0 and ∂t = 0, then
= 0, i.e. A(q, p) is a constant of the motion.
dt

◦ If {A, H} = 0 and {B, H} = 0, then {A, B}, H = 0. If in addition A and B have
no explicit time dependence, we conclude that {A, B} is a constant of the motion.
◦ It is easily established that
{qα , qβ } = 0 , {pα , pβ } = 0 , {qα , pβ } = δαβ . (20.33)
20.7 Canonical Transformations
20.7.1 Point transformations in Lagrangian mechanics
In Lagrangian mechanics, we are free to redefine our generalized coordinates, viz.
Qσ = Qσ (q1 , . . . , qn , t) . (20.34)
This is called a “point transformation.” The transformation is invertible if

∂Qα
det 6= 0 . (20.35)
∂qβ
The transformed Lagrangian, L̃, written as a function of the new coordinates Q and veloc-
ities Q̇, is

L̃ Q, Q̇, t) = L q(Q, t), q̇(Q, Q̇, t) . (20.36)
Finally, Hamilton’s principle,

Ztb
δ dt L̃(Q, Q̇, t) = 0 (20.37)
t1
with δQσ (ta ) = δQσ (tb ) = 0, still holds, and the form of the Euler-Lagrange equations
remains unchanged:

∂ L̃ d ∂ L̃
− =0. (20.38)
∂Qσ dt ∂ Q̇σ
The invariance of the equations of motion under a point transformation may be verified
explicitly. We first evaluate

d ∂ L̃ d ∂L ∂ q̇α d ∂L ∂qα
= = , (20.39)
dt ∂ Q̇σ dt ∂ q̇α ∂ Q̇σ dt ∂ q̇α ∂Qσ
where the relation

∂ q̇α ∂qα
= (20.40)
∂ Q̇σ ∂Qσ
follows from
∂qα ∂qα
q̇α = Q̇ + . (20.41)
∂Qσ σ ∂t
Now we compute
∂ L̃ ∂L ∂qα ∂L ∂ q̇α
= +
∂Qσ ∂qα ∂Qσ ∂ q̇α ∂Qσ
∂ 2 qα ∂ 2 qα

∂L ∂qα ∂L
= + Q̇ 0 +
∂qα ∂Qσ ∂ q̇α ∂Qσ ∂Qσ0 σ ∂Qσ ∂t

d ∂L ∂qα ∂L d ∂qα
= +
dt ∂ q̇σ ∂Qσ ∂ q̇α dt ∂Qσ

d ∂L ∂qα d ∂ L̃
= = , (20.42)
dt ∂ q̇σ ∂Qσ dt ∂ Q̇σ
where the last equality is what we obtained earlier in eqn. 20.39.

20.7.2 Canonical transformations in Hamiltonian mechanics
In Hamiltonian mechanics, we will deal with a much broader class of transformations – ones
which mix all the q 0 s and p0 s. The general form for a canonical transformation (CT) is

qσ = qσ Q1 , . . . , Qn ; P1 , . . . , Pn ; t (20.43)

pσ = pσ Q1 , . . . , Qn ; P1 , . . . , Pn ; t , (20.44)
with σ ∈ {1, . . . , n}. We may also write

ξi = ξi Ξ1 , . . . , Ξ2n ; t , (20.45)
with i ∈ {1, . . . , 2n}. The transformed Hamiltonian is H̃(Q, P, t).
What sorts of transformations are allowed? Well, if Hamilton’s equations are to remain
invariant, then
∂ H̃ ∂ H̃
Q̇σ = , Ṗσ = − , (20.46)
∂Pσ ∂Qσ
which gives
∂ Q̇σ ∂ Ṗσ ∂ Ξ̇i
+ =0= . (20.47)
∂Qσ ∂Pσ ∂Ξi
I.e. the flow remains incompressible in the new (Q, P ) variables. We will also require that
phase space volumes are preserved by the transformation, i.e.

∂Ξi ∂(Q, P )
det =
= 1 . (20.48)
∂ξj ∂(q, p)
Additional conditions will be discussed below.
20.7.3 Hamiltonian evolution
Hamiltonian evolution itself defines a canonical transformation. Let ξi = ξi (t) and ξi0 =
ξi (t + dt). Then from the dynamics ξ˙i = Jij ∂ξ
∂H
j
, we have
∂H
dt + O dt2 .

ξi (t + dt) = ξi (t) + Jij (20.49)
∂ξj
Thus,
∂ξi0

∂ ∂H 2

= ξ + Jik dt + O dt
∂ξj ∂ξj i ∂ξk
∂ 2H
dt + O dt2 .

= δij + Jik (20.50)
∂ξj ∂ξk
Now, using the result
det 1 + M = 1 + Tr M + O(2 ) ,

(20.51)
we have
0
∂ξi ∂ 2H 2

∂ξj = 1 + Jjk ∂ξj ∂ξk dt + O dt (20.52)

= 1 + O dt2 .

(20.53)
20.7.4 Symplectic structure
We have that
∂H
ξ˙i = Jij . (20.54)
∂ξj
Suppose we make a time-independent canonical transformation to new phase space coordi-

nates, Ξa = Ξa (ξ). We then have
∂Ξa ˙ ∂Ξa ∂H
Ξ̇a = ξj = Jjk . (20.55)
∂ξj ∂ξj ∂ξk
But if the transformation is canonical, then the equations of motion are preserved, and we
also have
∂ H̃ ∂ξk ∂H
Ξ̇a = Jab = Jab . (20.56)
∂Ξb ∂Ξb ∂ξk
Equating these two expressions, we have
∂H −1 ∂H
Maj Jjk = Jab Mkb , (20.57)
∂ξk ∂ξk
where
∂Ξa
Maj ≡ (20.58)
∂ξj
is the Jacobian of the transformation. Since the equality must hold for all ξ, we conclude
−1
MJ = J Mt =⇒ M JM t = J . (20.59)
A matrix M satisfying M M t = 1 is of course an orthogonal matrix. A matrix M satisfying

M JM t = J is called symplectic. We write M ∈ Sp(2n), i.e. M is an element of the group
of symplectic matrices 1 of rank 2n.
The symplectic property of M guarantees that the Poisson brackets are preserved under a
1
Note that the rank of a symplectic matrix is always even. Note also M JM t = J implies M t JM = J.
canonical transformation:
∂A ∂B
A, B ξ
= Jij
∂ξi ∂ξj
∂A ∂Ξa ∂B ∂Ξb
= Jij
∂Ξa ∂ξi ∂Ξb ∂ξj
t
∂A ∂B
= Mai Jij Mjb
∂Ξa ∂Ξb
∂A ∂B
= Jab
∂Ξa ∂Ξb

= A, B Ξ . (20.60)
20.7.5 Generating functions for canonical transformations
For a transformation to be canonical, we require
Ztb n o Ztb n o
δ dt pσ q̇σ − H(q, p, t) = 0 = δ dt Pσ Q̇σ − H̃(Q, P, t) . (20.61)
ta ta
This is satisfied provided

n o dF
pσ q̇σ − H(q, p, t) = λ Pσ Q̇σ − H̃(Q, P, t) + , (20.62)
dt
where λ is a constant. For canonical transformations, λ = 1.2 Thus,
∂F ∂F
H̃(Q, P, t) = H(q, p, t) + Pσ Q̇σ − pσ q̇σ + q̇σ + Q̇σ
∂qσ ∂Qσ
∂F ∂F ∂F
+ ṗσ + Ṗσ + . (20.63)
∂pσ ∂Pσ ∂t
Thus, we require
∂F ∂F ∂F ∂F
= pσ , = −Pσ , =0 , =0. (20.64)
∂qσ ∂Qσ ∂pσ ∂Pσ
The transformed Hamiltonian is
∂F
H̃(Q, P, t) = H(q, p, t) + . (20.65)
∂t
2
Solutions of eqn. 20.62 with λ 6= 1 are known as extended canonical transformations. We can always
rescale coordinates and/or momenta to achieve λ = 1.
There are four possibilities, corresponding to the freedom to make Legendre transformations
with respect to each of the arguments of F (q, Q) :
; pσ = + ∂F ∂F1


F1 (q, Q, t) 1
∂qσ , Pσ = − ∂Q σ
(type I)





; pσ = + ∂F ∂F2

F2 (q, P, t) − Pσ Qσ
 2
, Qσ = + ∂P (type II)

 ∂qσ σ
F (q, Q, t) =

F3 (p, Q, t) + pσ qσ ; qσ = − ∂F 3 ∂F3
, Pσ = − ∂Q (type III)



 ∂pσ σ





F4 (p, P, t) + pσ qσ − Pσ Qσ ; qσ = − ∂F 4
, Qσ = + ∂P∂F4
(type IV)

∂pσ σ
In each case (γ = 1, 2, 3, 4), we have

∂Fγ
H̃(Q, P, t) = H(q, p, t) + . (20.66)
∂t
Let’s work out some examples:
• Consider the type-II transformation generated by

F2 (q, P ) = Aσ (q) Pσ , (20.67)
where Aσ (q) is an arbitrary function of the {qσ }. We then have

∂F2 ∂F2 ∂Aα
Qσ = = Aσ (q) , pσ = = P . (20.68)
∂Pσ ∂qσ ∂qσ α
Thus,
∂qα
Qσ = Aσ (q) , Pσ = p . (20.69)
∂Qσ α
This is a general point transformation of the kind discussed in eqn. 20.34. For a general
−1
linear point transformation, Qα = Mαβ qβ , we have Pα = pβ Mβα , i.e. Q = M q,
P = p M −1 . If Mαβ = δαβ , this is the identity transformation. F2 = q1 P3 + q3 P1
interchanges labels 1 and 3, etc.
• Consider the type-I transformation generated by
F1 (q, Q) = Aσ (q) Qσ . (20.70)
We then have
∂F1 ∂Aα
pσ = = Q (20.71)
∂qσ ∂qσ α
∂F1
Pσ = − = −Aσ (q) . (20.72)
∂Qσ
Note that Aσ (q) = qσ generates the transformation

q −P
−→ . (20.73)
p +Q
• A mixed transformation is also permitted. For example,
F (q, Q) = q1 Q1 + (q3 − Q2 ) P2 + (q2 − Q3 ) P3 (20.74)
is of type-I with respect to index σ = 1 and type-II with respect to indices σ = 2, 3.

The transformation effected is
Q1 = p1 Q2 = q3 Q3 = q2 (20.75)
P1 = −q1 P2 = p 3 P3 = p2 . (20.76)
• Consider the harmonic oscillator,
p2
H(q, p) = + 1 kq 2 . (20.77)
2m 2
If we could find a time-independent canonical transformation such that
r
p 2 f (P )
p = 2mf (P ) cos Q , q= sin Q , (20.78)
k
where f (P ) is some function of P , then we’d have H̃(Q, P ) = f (P ), which is cyclic in
Q. To find this transformation, we take the ratio of p and q to obtain
√
p = mk q ctn Q , (20.79)
which suggests the type-I transformation

√
F1 (q, Q) = 1
2 mk q 2 ctn Q . (20.80)
This leads to
√
∂F1 √ ∂F1 mk q 2
p= = mk q ctn Q , P =− = . (20.81)
∂q ∂Q 2 sin2 Q
Thus, √ r
2P k
q= √
4
sin Q =⇒ f (P ) = P = ωP , (20.82)
mk m
p
where ω = k/m is the oscillation frequency. We therefore have
H̃(Q, P ) = ωP , (20.83)
whence P = E/ω. The equations of motion are
∂ H̃ ∂ H̃
Ṗ = − =0 , Q̇ = =ω , (20.84)
∂Q ∂P
which yields r
2E
Q(t) = ωt + ϕ0 , q(t) = 2
sin ωt + ϕ0 . (20.85)
mω
20.8 Hamilton-Jacobi Theory
We’ve stressed the great freedom involved in making canonical transformations. Coordi-
nates and momenta, for example, may be interchanged – the distinction between them is
purely a matter of convention! We now ask: is there any specially preferred canonical trans-
formation? In this regard, one obvious goal is to make the Hamiltonian H̃(Q, P, t) and the
corresponding equations of motion as simple as possible.
Recall the general form of the canonical transformation:
∂F
H̃(Q, P ) = H(q, p) + , (20.86)
∂t
with
∂F ∂F
= pσ =0 (20.87)
∂qσ ∂pσ
∂F ∂F
= −Pσ =0. (20.88)
∂Qσ ∂Pσ
We now demand that this transformation result in the simplest Hamiltonian possible, that
is, H̃(Q, P, t) = 0. This requires we find a function F such that
∂F ∂F
= −H , = pσ . (20.89)
∂t ∂qσ
The remaining functional dependence may be taken to be either on Q (type I) or on P

(type II). As it turns out, the generating function F we seek is in fact the action, S, which
is the integral of L with respect to time, expressed as a function of its endpoint values.
20.8.1 The action as a function of coordinates and time
We have seen how the action S[η(τ )] is a functional of the path η(τ ) and a function of the
endpoint values {qa , ta } and {qb , tb }. Let us define the action function S(q, t) as
Zt
S(q, t) = dτ L η, η̇, τ ) , (20.90)
ta
where η(τ ) starts at (qa , ta ) and ends at (q, t). We also require that η(τ ) satisfy the Euler-
Lagrange equations,

∂L d ∂L
− =0 (20.91)
∂ησ dτ ∂ η̇σ
20.8. HAMILTON-JACOBI THEORY 453
Let us now consider a new path, η̃(τ ), also starting at (qa , ta ), but ending at (q + dq, t + dt),
and also satisfying the equations of motion. The differential of S is

dS = S η̃(τ ) − S η(τ )
t+dt
Z Zt
= ˙ τ ) − dτ L η, η̇, τ )
dτ L(η̃, η̃, (20.92)
ta ta
Zt ( )
∂L h i ∂L h i
η̃˙ σ (τ ) − η̇σ (τ ) ˙

= dτ η̃σ (τ ) − ησ (τ ) + + L η̃(t), η̃(t), t dt
∂ησ ∂ η̇σ
ta
Zt ( )h
∂L d ∂L i
= dτ − η̃σ (τ ) − ησ (τ )
∂ησ dτ ∂ η̇σ
ta
h
∂L i
˙

+ η̃σ (t) − ησ (t) + L η̃(t), η̃(t), t dt
∂ η̇σ t

= 0 + πσ (t) δησ (t) + L η(t), η̇(t), t dt + O(δq · dt) , (20.93)

∂L
πσ = , (20.94)
∂ η̇σ
and
δησ (τ ) ≡ η̃σ (τ ) − ησ (τ ) . (20.95)
Note that the differential dqσ is given by
dqσ = η̃σ (t + dt) − ησ (t) (20.96)

= η̃σ (t + dt) − η̃σ (t) + η̃σ (t) − ησ (t)
= η̃˙ σ (t) dt + δησ (t)
= q̇σ (t) dt + δησ (t) + O(δq · dt) . (20.97)
Thus, with πσ (t) ≡ pσ , we have

dS = pσ dqσ + L − pσ q̇σ dt
= pσ dqσ − H dt . (20.98)
We therefore obtain
∂S ∂S dS
= pσ , = −H , =L. (20.99)
∂qσ ∂t dt
What about the lower limit at ta ? Clearly there are n + 1 constants associated with this
limit: q1 (ta ), . . . , qn (ta ); ta . Thus, we may write
S = S(q1 , . . . , qn ; Λ1 , . . . , Λn , t) + Λn+1 , (20.100)

Figure 20.4: A one-parameter family of paths q(s; ).
where our n + 1 constants are {Λ1 , . . . , Λn+1 }. If we regard S as a mixed generator, which
is type-I in some variables and type-II in others, then each Λσ for 1 ≤ σ ≤ n may be chosen
to be either Qσ or Pσ . We will define
(
∂S +Qσ if Λσ = Pσ
Γσ = = (20.101)
∂Λσ −Pσ if Λσ = Qσ
For each σ, the two possibilities Λσ = Qσ or Λσ = Pσ are of course rendered equivalent by

a canonical transformation (Qσ , Pσ ) → (Pσ , −Qσ ).
20.8.2 The Hamilton-Jacobi equation
Since the action S(q, Λ, t) has been shown to generate a canonical transformation for which
H̃(Q, P ) = 0. This requirement may be written as
∂S ∂S ∂S
H q1 , . . . , q n , ,..., ,t + =0. (20.102)
∂q1 ∂qn ∂t
This is the Hamilton-Jacobi equation (HJE). It is a first order partial differential equation
in n + 1 variables, and in general is nonlinear (since kinetic energy is generally a quadratic
function of momenta). Since H̃(Q, P, t) = 0, the equations of motion are trivial, and
Qσ (t) = const. , Pσ (t) = const. (20.103)
Once the HJE is solved, one must invert the relations Γσ = ∂S(q, Λ, t)/∂Λσ to obtain
q(Q, P, t). This is possible only if
∂ 2S

det 6= 0 , (20.104)
∂qα ∂Λβ
which is known as the Hessian condition.
It is worth noting that the HJE may have several solutions. For example, consider the case
of the free particle, with H(q, p) = p2 /2m. The HJE is
1 ∂S 2 ∂S

+ =0. (20.105)
2m ∂q ∂t
One solution of the HJE is
m (q − Λ)2
S(q, Λ, t) = . (20.106)
2t
For this we find
∂S m Γ
Γ = = − (q − Λ) ⇒ q(t) = Λ − t . (20.107)
∂Λ t m
Here Λ = q(0) is the initial value of q, and Γ = −p is minus the momentum.
Another equally valid solution to the HJE is

√
S(q, Λ, t) = q 2mΛ − Λ t . (20.108)
This yields r r
∂S 2m Λ
Γ = =q − t ⇒ q(t) = (t + Γ ) . (20.109)
∂Λ Λ 2m
For
p this solution, Λ is the energy and Γ may be related to the initial value of q(t) =
Γ Λ/2m.
20.8.3 Time-independent Hamiltonians
When H has no explicit time dependence, we may reduce the order of the HJE by one,
writing
S(q, Λ, t) = W (q, Λ) + T (Λ, t) . (20.110)
The HJE becomes
∂W ∂T
H q, =− . (20.111)
∂q ∂t
Note that the LHS of the above equation is independent of t, and the RHS is independent
of q. Therefore, each side must only depend on the constants Λ, which is to say that each
side must be a constant, which, without loss of generality, we take to be Λ1 . Therefore
S(q, Λ, t) = W (q, Λ) − Λ1 t . (20.112)

The function W (q, Λ) is called Hamilton’s characteristic function. The HJE now takes the
form
∂W ∂W
H q1 , . . . , q n , ,..., = Λ1 . (20.113)
∂q1 ∂qn
Note that adding an arbitrary constant C to S generates the same equation, and simply
shifts the last constant Λn+1 → Λn+1 + C. This is equivalent to replacing t by t − t0 with
t0 = C/Λ1 , i.e. it just redefines the zero of the time variable.
20.8.4 Example: one-dimensional motion
As an example of the method, consider the one-dimensional system,
p2
H(q, p) = + U (q) . (20.114)
2m
The HJE is
1 ∂S 2

+ U (q) = Λ . (20.115)
2m ∂q
which may be recast as
∂S
q
= 2m Λ − U (q) , (20.116)
∂q
with solution
√ Zq p
S(q, Λ, t) = 2m dq 0 Λ − U (q 0 ) − Λ t . (20.117)
We now have
∂S
q
p= = 2m Λ − U (q) , (20.118)
∂q
as well as
q(t)
dq 0
r Z
∂S m
Γ = = p −t . (20.119)
∂Λ 2 Λ − U (q 0 )
Thus, the motion q(t) is given by quadrature:
Zq(t)
dq 0
r
m
Γ +t= p , (20.120)
2 Λ − U (q 0 )
where Λ and Γ are constants. The lower limit on the integral is arbitrary and merely shifts
t by another constant. Note that Λ is the total energy.
20.8.5 Separation of variables
It is convenient to first work an example before discussing the general theory. Consider the
following Hamiltonian, written in spherical polar coordinates:
potential U (r,θ,φ)
}| z {
p2θ p2φ

1 2 B(θ) C(φ)
H= p + + + A(r) + 2 + 2 2 . (20.121)
2m r r2 r2 sin2 θ r r sin θ
We seek a solution with the characteristic function
W (r, θ, φ) = Wr (r) + Wθ (θ) + Wφ (φ) . (20.122)

The HJE is then
1 ∂Wr 2 ∂Wθ 2 ∂Wφ 2

1 1
+ +
2m ∂r 2mr2 ∂θ 2mr2 sin2 θ ∂φ
B(θ) C(φ)
+ A(r) + 2
+ 2 2 = Λ1 = E . (20.123)
r r sin θ
Multiply through by r2 sin2 θ to obtain

( )
1 ∂Wφ 2
2
1 ∂W θ
+ C(φ) = − sin2 θ + B(θ)
2m ∂φ 2m ∂θ
( 2 )
1 ∂W r
− r2 sin2 θ + A(r) − Λ1 . (20.124)
2m ∂r
The LHS is independent of (r, θ), and the RHS is independent of φ. Therefore, we may set
1 ∂Wφ 2

+ C(φ) = Λ2 . (20.125)
2m ∂φ
Proceeding, we replace the LHS in eqn. 20.124 with Λ2 , arriving at

( )
1 ∂Wθ 2 1 ∂Wr 2

Λ2 2
+ B(θ) + = −r + A(r) − Λ1 . (20.126)
2m ∂θ sin2 θ 2m ∂r
The LHS of this equation is independent of r, and the RHS is independent of θ. Therefore,
1 ∂Wθ 2

Λ2
+ B(θ) + = Λ3 . (20.127)
2m ∂θ sin2 θ
We’re left with

1 ∂Wr 2

Λ3
+ A(r) + 2 = Λ1 . (20.128)
2m ∂r r
The full solution is therefore
√ Zr r
Λ3
S(q, Λ, t) = 2m dr0 Λ1 − A(r0 ) − 0 2 (20.129)
r
√ Zθ r
Λ2
+ 2m dθ0 Λ3 − B(θ0 ) −
sin2 θ0
√ Zφ q
+ 2m dφ0 Λ2 − C(φ0 ) − Λ1 t . (20.130)
We then have
dr0
Z r(t)
pm
∂S 2
Γ1 = = q −t (20.131)
∂Λ1
Λ1 − A(r0 ) − Λ3 r0 −2
Z θ(t) pm 0 Z φ(t) p m 0
∂S 2 dθ dφ
Γ2 = =− q + q 2 (20.132)
∂Λ2
sin2 θ0 Λ3 − B(θ0 ) − Λ2 csc2 θ0 Λ2 − C(φ0 )
Z r(t) pm 0 Z θ(t) pm 0
∂S 2 dr 2 dθ
Γ3 = =− q + q . (20.133)
∂Λ3
r0 2 Λ1 − A(r0 ) − Λ3 r0 −2 Λ3 − B(θ0 ) − Λ2 csc2 θ0
The game plan here is as follows. The first of the above trio of equations is inverted to yield
r(t) in terms of t and constants. This solution is then invoked in the last equation (the upper
limit on the first integral on the RHS) in order to obtain an implicit equation for θ(t), which
is invoked in the second equation to yield an implicit equation for φ(t). The net result is
the motion of the system in terms of time t and the six constants (Λ1 , Λ2 , Λ3 , Γ1 , Γ2 , Γ3 ). A
seventh constant, associated with an overall shift of the zero of t, arises due to the arbitrary
lower limits of the integrals.
In general, the separation of variables method begins with3

n
X
W (q, Λ) = Wσ (qσ , Λ) . (20.134)
σ=1
Each Wσ (qσ , Λ) may be regarded as a function of the single variable qσ , and is obtained by
satisfying an ODE of the form4

dWσ
Hσ q σ , = Λσ . (20.135)
dqσ
We then have
∂Wσ ∂W
pσ = , Γσ = + δσ,1 t . (20.136)
∂qσ ∂Λσ
Note that while each Wσ depends on only a single qσ , it may depend on several of the Λσ .
20.8.6 Example #2 : point charge plus electric field
Consider a potential of the form

k
U (r) = − Fz , (20.137)
r
3
Here we assume complete separability. A given system may only be partially separable.
4
` Hσ (qσ , pσ ) ´may also depend on several of the Λα . See e.g. eqn. 20.128, which is of the form
Hr r, ∂r Wr , Λ3 = Λ1 .
which corresponds to a charge in the presence of an external point charge plus an external
electric field. This problem is amenable to separation in parabolic coordinates, (ξ, η, ϕ):
p p
x = ξη cos ϕ , y = ξη sin ϕ , z = 12 (ξ − η) . (20.138)
Note that
p p
ρ≡ x2 + y 2 = ξη (20.139)
p
r = ρ2 + z 2 = 12 (ξ + η) . (20.140)
T = 12 m ρ̇2 + ρ2 ϕ̇2 + ż 2

˙2
η̇ 2

ξ
= 18 m (ξ + η) + + 12 m ξη ϕ̇2 , (20.141)
ξ η
and hence the Lagrangian is

!
ξ˙2 η̇ 2 2k
L= 1
8 m (ξ + η) + + 12 m ξη ϕ̇2 − + 1 F (ξ − η) . (20.142)
ξ η ξ+η 2
Thus, the conjugate momenta are
∂L ξ˙
pξ = = 14 m (ξ + η) (20.143)
∂ ξ˙ ξ
∂L η̇
pη = = 14 m (ξ + η) (20.144)
∂ η̇ η
∂L
pϕ = = m ξη ϕ̇ , (20.145)
∂ ϕ̇
and the Hamiltonian is
H = pξ ξ˙ + pη η̇ + pϕ ϕ̇ (20.146)
!
2 ξ p2ξ + η p2η p2ϕ 2k
= + + − 1 F (ξ − η) . (20.147)
m ξ+η 2mξη ξ + η 2
Notice that ∂H/∂t = 0, which means dH/dt = 0, i.e. H = E ≡ Λ1 is a constant of the

motion. Also, ϕ is cyclic in H, so its conjugate momentum pϕ is a constant of the motion.
We write
S(q, Λ) = W (q, Λ) − Et (20.148)

= Wξ (ξ, Λ) + Wη (η, Λ) + Wϕ (ϕ, Λ) − Et . (20.149)
with E = Λ1 . Clearly we may take
Wϕ (ϕ, Λ) = Pϕ ϕ , (20.150)
where Pϕ = Λ2 . Multiplying the Hamilton-Jacobi equation by 12 m (ξ + η) then gives

2
Pϕ2

dWξ
ξ + + mk − 14 F ξ 2 − 12 mEξ
dξ 4ξ
dWη 2 Pϕ2 1 2 1

= −η − − 4 F η + 2 mEη ≡ Υ , (20.151)
dη 4η
where Υ = Λ3 is the third constant: Λ = (E, Pϕ , Υ ). Thus,

q Z ξ s
z }| { Υ − mk 1 Pϕ2
S ξ, η, ϕ; E, Pϕ , Υ = dξ 0 12 mE +

+ mF ξ 0−
ξ0 4
| {z } 4ξ 0 2
Λ
Z η s
Υ Pϕ2
+ dη 0 12 mE − 0 − 14 mF η 0 − 0 2
η 4η
+ Pϕ ϕ − Et . (20.152)
20.8.7 Example #3 : Charged Particle in a Magnetic Field
The Hamiltonian is
1 e 2
H=p− A . (20.153)
2m c
We choose the gauge A = Bxŷ, and we write
S(x, y, P1 , P2 ) = Wx (x, P1 , P2 ) + Wy (y, P1 , P2 ) − P1 t . (20.154)
Note that here we will consider S to be a function of {qσ } and {Pσ }.
The Hamilton-Jacobi equation is then
∂Wx 2 eBx 2

∂Wy
+ − = 2mP1 . (20.155)
∂x ∂y c
We solve by writing
2 2
dWx eBx
Wy = P2 y ⇒ + P2 − = 2mP1 . (20.156)
dx c
This equation suggests the substitution
cP2 c p
x= + 2mP1 sin θ . (20.157)
eB eB
in which case
∂x c p
= 2mP1 cos θ (20.158)
∂θ eB
and
∂Wx ∂Wx ∂θ eB 1 ∂Wx
= · = √ . (20.159)
∂x ∂θ ∂x c 2mP1 cos θ ∂θ
Substitution this into eqn. 20.156, we have
∂Wx 2mcP1
= cos2 θ , (20.160)
∂θ eB
with solution
mcP1 mcP1
Wx = θ+ sin(2θ) . (20.161)
eB 2eB
We then have
∂Wx ∂Wx ∂x p
px = = = 2mP1 cos θ (20.162)
∂x ∂θ ∂θ
and
∂Wy
py = = P2 . (20.163)
∂y
The type-II generator we seek is then
mcP1 mcP1
S(q, P, t) = θ+ sin(2θ) + P2 y − P1 t , (20.164)
eB 2eB
where
eB −1 cP2
θ= √ sin x− . (20.165)
c 2mP1 eB
Note that, from eqn. 20.157, we may write
c mc 1 c p
dx = dP2 + √ sin θ dP1 + 2mP1 cos θ dθ , (20.166)
eB eB 2mP1 eB
from which we derive

∂θ tan θ ∂θ 1
=− , = −√ . (20.167)
∂P1 2P1 ∂P2 2mP1 cos θ
These results are useful in the calculation of Q1 and Q2 :
∂S
Q1 =
∂P1
mc mcP1 ∂θ mc mcP1 ∂θ
= θ+ + sin(2θ) + cos(2θ) −t
eB eB ∂P1 2eB eB ∂P1
mc
= θ−t (20.168)
eB
and
∂S
Q2 =
∂P2
mcP1 ∂θ
=y+ 1 + cos(2θ)
eB ∂P2
c p
=y− 2mP1 cos θ . (20.169)
eB
Now since H̃(P, Q) = 0, we have that Q̇σ = 0, which means that each Qσ is a constant. We
therefore have the following solution:
x(t) = x0 + A sin(ωc t + δ) (20.170)

y(t) = y0 + A cos(ωc t + δ) , (20.171)
where ωc = eB/mc is the ‘cyclotron frequency’, and
cP2 c p
x0 = , y0 = Q2 , δ ≡ ωc Q1 , A= 2mP1 . (20.172)
eB eB
20.9 Action-Angle Variables
20.9.1 Circular Phase Orbits: Librations and Rotations
In a completely integrable system, the Hamilton-Jacobi equation may be solved by separa-

tion of variables. Each momentum pσ is a function of only its corresponding coordinate qσ
plus constants – no other coordinates enter:
∂Wσ
pσ = = pσ (qσ , Λ) . (20.173)
∂qσ
The motion satisfies

Hσ (qσ , pσ ) = Λσ . (20.174)
The level sets of Hσ are curves Cσ . In general, these curves each depend on all of the
constants Λ, so we write Cσ = Cσ (Λ). The curves Cσ are the projections of the full motion
onto the (qσ , pσ ) plane. In general we will assume the motion, and hence the curves Cσ ,
is bounded . In this case, two types of projected motion are possible: librations and rota-
tions. Librations are periodic oscillations about an equilibrium position. Rotations involve
the advancement of an angular variable by 2π during a cycle. This is most conveniently
illustrated in the case of the simple pendulum, for which
p2φ
+ 12 Iω 2 1 − cos φ .

H(pφ , φ) = (20.175)
2I
20.9. ACTION-ANGLE VARIABLES 463
Figure 20.5: Phase curves for the simple pendulum, showing librations (in blue), rotations
(in green), and the separatrix (in red). This phase flow is most correctly viewed as taking
place on a cylinder, obtained from the above sketch by identifying the lines φ = π and
φ = −π.
• When E < I ω 2 , the momentum pφ vanishes at φ = ± cos−1 (2E/Iω 2 ). The system

executes librations between these extreme values of the angle φ.
• When E > I ω 2 , the kinetic energy is always positive, and the angle advances mono-
tonically, executing rotations.
In a completely integrable system, each Cσ is either a libration or a rotation5 . Both librations

and rotations are closed curves. Thus, each Cσ is in general homotopic to (= “can be
continuously distorted to yield”) a circle, S 1 . For n freedoms, the motion is therefore
confined to an n-torus, T n :
n times
z }| {
Tn = S1 × S1 × · · · × S1 . (20.176)
These are called invariant tori (or invariant manifolds). There are many such tori, as there
are many Cσ curves in each of the n two-dimensional submanifolds.
Invariant tori never intersect! This is ruled out by the uniqueness of the solution to the
dynamical system, expressed as a set of coupled ordinary differential equations.
Note also that phase space is of dimension 2n, while the invariant tori are of dimension n.
Phase space is ‘covered’ by the invariant tori, but it is in general difficult to conceive of how
this happens. Perhaps the most accessible analogy is the n = 1 case, where the ‘1-tori’ are
just circles. Two-dimensional phase space is covered noninteracting circular orbits. (The
orbits are topologically equivalent to circles, although geometrically they may be distorted.)
It is challenging to think about the n = 2 case, where a four-dimensional phase space is
filled by nonintersecting 2-tori.
5
Cσ may correspond to a separatrix, but this is a nongeneric state of affairs.
20.9.2 Action-Angle Variables
For a completely integrable system, one can transform canonically from (q, p) to new co-
ordinates (φ, J) which specify a particular n-torus T n as well as the location on the torus,
which is specified by n angle variables. The {Jσ } are ‘momentum’ variables which specify
the torus itself; they are constants of the motion since the tori are invariant. They are
called action variables. Since J˙σ = 0, we must have
∂H
J˙σ = − =0 =⇒ H = H(J) . (20.177)
∂φσ
The {φσ } are the angle variables.
The coordinate φσ describes the projected motion along Cσ , and is normalized by

I
dφσ = 2π (once around Cσ ) . (20.178)
Cσ
The dynamics of the angle variables are given by
∂H
φ̇σ = ≡ νσ (J) . (20.179)
∂Jσ
Thus,
φσ (t) = φσ (0) + νσ (J) t . (20.180)

The νσ (J) are frequencies describing the rate at which the Cσ are traversed; Tσ (J) =
2π/νσ (J) is the period.
20.9.3 Canonical Transformation to Action-Angle Variables
The {Jσ } determine the {Cσ }; each qσ determines a point on Cσ . This suggests a type-II
transformation, with generator F2 (q, J):
∂F2 ∂F2
pσ = , φσ = . (20.181)
∂qσ ∂Jσ
Note that6
∂ 2F2
I I I I
∂F2 ∂
2π = dφσ = d = dqσ = pσ dqσ , (20.182)
∂Jσ ∂Jσ ∂qσ ∂Jσ
Cσ Cσ Cσ Cσ
` ∂F2 ´ 2
6
In general, we should write d ∂J σ
= ∂J∂σ F∂q
2
α
dqα with a sum over α. However, in eqn. 20.182 all
coordinates and momenta other than qσ and pσ are held fixed. Thus, α = σ is the only term in the sum
which contributes.
which suggests the definition I

1
Jσ = pσ dqσ . (20.183)
2π
Cσ
I.e. Jσ is (2π)−1 times the area enclosed by Cσ .
If, separating variables, X

W (q, Λ) = Wσ (qσ , Λ) (20.184)
σ
is Hamilton’s characteristic function for the transformation (q, p) → (Q, P ), then

I
1 ∂Wσ
Jσ = dqσ = Jσ (Λ) (20.185)
2π ∂qσ
Cσ
is a function only of the {Λα } and not the {Γα }. We then invert this relation to obtain
Λ(J), to finally obtain
X
F2 (q, J) = W q, Λ(J) = Wσ qσ , Λ(J) . (20.186)
σ
Thus, the recipe for canonically transforming to action-angle variable is as follows:
P
(1) Separate and solve the Hamilton-Jacobi equation for W (q, Λ) = σ Wσ (qσ , Λ).
(2) Find the orbits Cσ – the level sets of satisfying Hσ (qσ , pσ ) = Λσ .

1
H ∂Wσ
(3) Invert the relation Jσ (Λ) = 2π ∂qσ dqσ to obtain Λ(J).
Cσ
(4) F2 (q, J) = σ Wσ qσ , Λ(J) is the desired type-II generator7 .

P
20.9.4 Example : Harmonic Oscillator
The Hamiltonian is
p2
H= + 1 mω02 q 2 , (20.187)
2m 2
hence the Hamilton-Jacobi equation is
dW 2

+ m2 ω02 q 2 = 2mΛ . (20.188)
dq
Thus,
dW
q
p= = ± 2mΛ − m2 ω02 q 2 . (20.189)
dq
7
Note that F2 (q, J) is time-independent. I.e. we are not transforming to H̃ = 0, but rather to H̃ = H̃(J).
We now define 1/2

√

2Λ
q≡ sin θ ⇒ p= 2mΛ cos θ , (20.190)
mω02
in which case
I Z2π
1 1 2Λ Λ
J= p dq = · · dθ cos2 θ = . (20.191)
2π 2π ω0 ω0
0
Solving the HJE, we write

dW ∂q dW
= · = 2J cos2 θ . (20.192)
dθ ∂θ dq
Integrating,
W = Jθ + 12 J sin 2θ , (20.193)
up to an irrelevant constant. We then have

∂W 1
∂θ
φ= = θ + 2 sin 2θ + J 1 + cos 2θ . (20.194)
∂J q ∂J q
p
To find (∂θ/∂J)q , we differentiate q = 2J/mω0 sin θ:
r
sin θ 2J ∂θ 1
dq = √ dJ + cos θ dθ ⇒ =− tan θ . (20.195)
2mω0 J mω0 ∂J q 2J
Plugging this result into eqn. 20.194, we obtain φ = θ. Thus, the full transformation is
2J 1/2
p
q= sin φ , p = 2mω0 J cos φ . (20.196)
mω0
The Hamiltonian is
H = ω0 J , (20.197)
hence φ̇ = ∂H
∂J = ω0 and J˙ = − ∂H
∂φ = 0, with solution φ(t) = φ(0) + ω0 t and J(t) = J(0).
20.9.5 Example : Particle in a Box
Consider a particle in an open box of dimensions Lx × Ly moving under the influence of

gravity. The bottom of the box lies at z = 0. The Hamiltonian is
p2x p2y p2
H= + + z + mgz . (20.198)
2m 2m 2m
Step one is to solve the Hamilton-Jacobi equation via separation of variables. The Hamilton-
Jacobi equation is written
1 ∂Wx 2 1 ∂Wy 2 1 ∂Wz 2

+ + + mgz = E ≡ Λz . (20.199)
2m ∂x 2m ∂y 2m ∂z
Figure 20.6: The librations Cz and Cx . Not shown is Cy , which is of the same shape as Cx .
We can solve for Wx,y by inspection:

p p
Wx (x) = 2mΛx x , Wy (y) = 2mΛy y . (20.200)
We then have8
q
Wz0 (z) = − 2m Λz − Λx − Λy − mgz

(20.201)
√
2 2 3/2
Wz (z) = √ Λz − Λx − Λy − mgz . (20.202)
3 mg
p
Step two is to find the Cσ . Clearly px,y = 2mΛx,y . For fixed px , the x motion proceeds
from x = 0 to x = Lx and back, with corresponding motion for y. For x, we have
q
pz (z) = Wz0 (z) =

2m Λz − Λx − Λy − mgz , (20.203)
and thus Cz is a truncated parabola, with zmax = (Λz − Λx − Λy )/mg.
Step three is to compute J(Λ) and invert to obtain Λ(J). We have
I ZLx p
1 1 Lx p
Jx = px dx = dx 2mΛx = 2mΛx (20.204)
2π π π
Cx 0
I ZLy
1 1 p Ly p
Jy = py dy = dy 2mΛy = 2mΛy (20.205)
2π π π
Cy 0
8
Our choice of signs in taking the square roots for Wx0 , Wy0 , and Wz0 is discussed below.
and
I zZmax
1 1
q
Jz = pz dz = dx 2m Λz − Λx − Λy − mgz
2π π
Cz 0
√
2 2 3/2
= √ Λz − Λx − Λy . (20.206)
3π m g
We now invert to obtain
π2 2 π2
Λx = J , Λy = J2 (20.207)
2mL2x x 2mL2y y
√
3π m g 2/3 2/3 π2 π2

2
Λz = √ Jz + 2
Jx + J2 .
2 y
(20.208)
2 2 2mL x 2mL y
2m2/3 g 1/3 z 3/2

πx πy 2/3
F2 x, y, z, Jx , Jy , Jz = J + J + π Jz − . (20.209)
Lx x Ly y (3π)2/3
We now find
∂F2 πx ∂F2 πy
φx = = , φy = = (20.210)
∂Jx Lx ∂Jy Ly
and s
2m2/3 g 1/3 z
r
∂F2 z
φz = =π 1− 2/3
=π 1− , (20.211)
∂Jz (3πJz ) zmax
where
(3πJz )2/3
zmax (Jz ) = . (20.212)
2m2/3 g 1/3
The momenta are
∂F2 πJx ∂F2 πJy
px = = , py = = (20.213)
∂x Lx ∂y Ly
and
√ !1/2
√ 3π m g 2/3 2/3

∂F2
pz = = − 2m √ Jz − mgz . (20.214)
∂z 2 2
We note that the angle variables φx,y,z seem to be restricted to the range [0, π], which
seems to be at odds with eqn. 20.182. Similarly, the momenta px,y,z all seem to be positive,
whereas we know the momenta reverse sign when the particle bounces off a wall. The origin
of the apparent discrepancy is that when we solved for the functions Wx,y,z , we had to take
a square root in
√ each case, and we chose a particular branch of the square root. So rather
than Wx (x) = 2mΛx x, we should have taken
(√
2mΛx x if px > 0
Wx (x) = √ (20.215)
2mΛx (2Lx − x) if px < 0 .
√
The relation Jx = (Lx /π) 2mΛx is unchanged, hence
(
(πx/Lx ) Jx if px > 0
Wx (x) = (20.216)
2πJx − (πx/Lx ) Jx if px < 0 .
and
(
πx/Lx if px > 0
φx = (20.217)
π(2Lx − x)/Lx if px < 0 .
Now the angle variable φx advances by 2π during the cycle Cx . Similar considerations apply
to the y and z sectors.
20.9.6 Kepler Problem in Action-Angle Variables
This is discussed in detail in standard texts, such as Goldstein. The potential is V (r) =
−k/r, and the problem is separable. We write9
W (r, θ, φ) = Wr (r) + Wθ (θ) + Wϕ (ϕ) , (20.218)
hence
1 ∂Wr 2 ∂Wθ 2 ∂Wϕ 2

1 1
+ + + V (r) = E ≡ Λr . (20.219)
2m ∂r 2mr2 ∂θ 2mr2 sin2 θ ∂ϕ
Separating, we have
1 dWϕ 2
I
dWϕ p
= Λϕ ⇒ Jϕ = dϕ = 2π 2mΛϕ . (20.220)
2m dϕ dϕ
Cϕ
Next we deal with the θ coordinate:
1 dWθ 2

Λϕ
= Λθ − ⇒
2m dθ sin2 θ
p Zθ0 q

Jθ = 4 2mΛθ dθ 1 − Λϕ /Λθ csc2 θ
0
√ p p
= 2π 2m Λθ − Λϕ , (20.221)
9
We denote the azimuthal angle by ϕ to distinguish it from the AA variable φ.
where θ0 = sin−1 (Λϕ /Λθ ). Finally, we have10
1 dWr 2

k Λθ
=E+ − 2 ⇒
2m dr r r
I s
k Λθ
Jr = dr 2m E + − 2
r r
Cr
s
2m
= −(Jθ + Jϕ ) + πk , (20.222)
|E|
where we’ve assumed E < 0, i.e. bound motion.
Thus, we find
2π 2 mk 2
H=E=− 2 . (20.223)
Jr + Jθ + Jϕ
Note that the frequencies are completely degenerate:
!1/2
∂H 4π 2 mk 2 π 2 mk 2
ν ≡ νr,θ,ϕ = = 3 = . (20.224)
∂Jr,θ,ϕ Jr + Jθ + Jϕ 2|E|3
This threefold degeneracy may be removed by a transformation to new AA variables,

n o n o
(φr , Jr ), (φθ , Jθ ), (φϕ , Jϕ ) −→ (φ1 , J1 ), (φ2 , J2 ), (φ3 , J3 ) , (20.225)
using the type-II generator
F2 (φr , φθ , φϕ ; J1 , J2 , J3 ) = (φϕ − φθ ) J1 + (φθ − φr ) J2 + φr J3 , (20.226)
which results in
∂F2 ∂F2
φ1 = = φϕ − φθ Jr = = J3 − J2 (20.227)
∂J1 ∂φr
∂F2 ∂F2
φ2 = = φθ − φr Jθ = = J2 − J1 (20.228)
∂J2 ∂φθ
∂F2 ∂F2
φ3 = = φr Jϕ = = J1 . (20.229)
∂J3 ∂φϕ
The new Hamiltonian is

2π 2 mk 2
H(J1 , J2 , J3 ) = − , (20.230)
J32
whence ν1 = ν2 = 0 and ν3 = ν.
10
The details of performing the integral around Cr are discussed in e.g. Goldstein.
20.9.7 Charged Particle in a Magnetic Field
For the case of the charged particle in a magnetic field, studied above in section 20.8.7, we
found
cP2 c p
x= + 2mP1 sin θ (20.231)
eB eB
and p
px = 2mP1 cos θ , py = P2 . (20.232)
The action variable J is then
I Z2π
2mcP1 mcP1
J= px dx = dθ cos2 θ = . (20.233)
eB eB
0
We then have
W = Jθ + 12 J sin(2θ) + P y , (20.234)
where P ≡ P2 . Thus,
∂W
φ=
∂J
∂θ
= θ + 12 sin(2θ) + J 1 + cos(2θ)

∂J
1 2 tan θ
= θ + 2 sin(2θ) + 2J cos θ · −
2J
=θ . (20.235)
The other canonical pair is (Q, P ), where

r
∂W 2cJ
Q= =y− cos φ . (20.236)
∂P eB
Therefore, we have
r r
cP 2cJ 2cJ
x= + sin φ , y =Q+ cos φ (20.237)
eB eB eB
and r
2eBJ
px = cos φ , py = P . (20.238)
c
The Hamiltonian is
p2x eBx 2

1
H= + p −
2m 2m y c
eBJ eBJ
= cos2 φ + sin2 φ
mc mc
= ωc J , (20.239)
where ωc = eB/mc. The equations of motion are
∂H ∂H
φ̇ = = ωc , J˙ = − =0 (20.240)
∂J ∂φ
and
∂H ∂H
Q̇ = =0 , Ṗ = − =0. (20.241)
∂P ∂Q
Thus, Q, P , and J are constants, and φ(t) = φ0 + ωc t.
20.9.8 Motion on Invariant Tori
The angle variables evolve as
φσ (t) = νσ (J) t + φσ (0) . (20.242)
Thus, they wind around the invariant torus, specified by {Jσ } at constant rates. In general,
while each φσ executed periodic motion around a circle, the motion of the system as a whole
is not periodic, since the frequencies νσ (J) are not, in general, commensurate. In order for
the motion to be periodic, there must exist a set of integers, {lσ }, such that
n
X
lσ νσ (J) = 0 . (20.243)
σ=1
This means that the ratio of any two frequencies νσ /να must be a rational number. On a
given torus, there are several possible orbits, depending on initial conditions φ(0). However,
since the frequencies are determined by the action variables, which specify the tori, on a
given torus either all orbits are periodic, or none are.
In terms of the original coordinates q, there are two possibilities:
∞ ∞
(σ)
X X
qσ (t) = ··· Al1 l2 ···ln eil1 φ1 (t) · · · eiln φn (t)
l1 =−∞ ln =−∞
X
≡ Aσl eil·φ(t) (libration) (20.244)
l
or
X
qσ (t) = qσ◦ φσ (t) + Blσ eil·φ(t) (rotation) . (20.245)
l
For rotations, the variable qσ (t) increased by ∆qσ = 2π qσ◦ .

20.10. CANONICAL PERTURBATION THEORY 473
20.10 Canonical Perturbation Theory
20.10.1 Canonical Transformations and Perturbation Theory
Suppose we have a Hamiltonian
H(ξ, t) = H0 (ξ, t) + H1 (ξ, t) , (20.246)
where is a small dimensionless parameter. Let’s implement a type-II transformation,

generated by S(q, P, t):11
∂
H̃(Q, P, t) = H(q, p, t) + S(q, P, t) . (20.247)
∂t
Let’s expand everything in powers of :
qσ = Qσ + q1,σ + 2 q2,σ + . . . (20.248)

pσ = Pσ + p1,σ + 2 p2,σ + . . . (20.249)
H̃ = H̃0 + H̃1 + 2 H̃2 + . . . (20.250)
S = q|σ{zPσ} + S1 + 2 S2 + . . . . (20.251)
identity
transformation
Then
∂S ∂S1 ∂S2
Qσ = = qσ + + 2 + ... (20.252)
∂Pσ ∂Pσ ∂Pσ

∂S1 ∂S2 2
= Qσ + q1,σ + + q2,σ + + ...
∂Pσ ∂Pσ
and
∂S ∂S1 ∂S2
pσ = = Pσ + + 2 + ... (20.253)
∂qσ ∂qσ ∂qσ
= Pσ + p1,σ + 2 p2,σ + . . . . (20.254)
We therefore conclude, order by order in ,
∂Sk ∂Sk
qk,σ = − , pk,σ = + . (20.255)
∂Pσ ∂qσ
11
Here, S(q, P, t) is not meant to signify Hamilton’s principal function.
Now let’s expand the Hamiltonian:
∂S
H̃(Q, P, t) = H0 (q, p, t) + H1 (q, p, t) + (20.256)
∂t
∂H0 ∂H0
= H0 (Q, P, t) + (qσ − Qσ ) + (p − Pσ )
∂Qσ ∂Pσ σ
∂
+ H1 (Q, P, t) + S (Q, P, t) + O(2 )
∂t 1
!
∂H0 ∂S1 ∂H0 ∂S1 ∂S1
= H0 (Q, P, t) + − + + + H1 + O(2 )
∂Qσ ∂Pσ ∂Pσ ∂Qσ ∂t

∂S1
+ O(2 ) .

= H0 (Q, P, t) + H1 + S1 , H0 + (20.257)
∂t
In the above expression, we evaluate Hk (q, p, t) and Sk (q, P, t) at q = Q and p = P and

expand in the differences q − Q and p − P . Thus, we have derived the relation
H̃(Q, P, t) = H̃0 (Q, P, t) + H̃1 (Q, P, t) + . . . (20.258)
with
H̃0 (Q, P, t) = H0 (Q, P, t) (20.259)

∂S1
H̃1 (Q, P, t) = H1 + S1 , H0 + . (20.260)
∂t
The problem, though, is this: we have one equation, eqn, 20.260, for the two unknowns H̃1
and S1 . Thus, the problem is underdetermined. Of course, we could choose H̃1 = 0, which
basically recapitulates standard Hamilton-Jacobi theory. But we might just as well demand
that H̃1 satisfy some other requirement, such as that H̃0 + H̃1 being integrable.
Incidentally, this treatment is paralleled by one in quantum mechanics, where a unitary

transformation may be implemented to eliminate a perturbation to lowest order in a small
parameter. Consider the Schrödinger equation,
∂ψ
ih̄ = (H0 + H1 ) ψ , (20.261)
∂t
and define χ by
ψ ≡ eiS/h̄ χ , (20.262)
with
S = S1 + 2 S2 + . . . . (20.263)
As before, the transformation U ≡ exp(iS/h̄) collapses to the identity in the → 0 limit.
Now let’s write the Schrödinger equation for χ. Expanding in powers of , one finds
∂χ

1 ∂S1
ih̄ = H0 χ + H1 + S ,H + χ + . . . ≡ H̃ χ , (20.264)
∂t ih̄ 1 0 ∂t
where [A, B] = AB − BA is the commutator. Note the classical-quantum correspondence,
1
{A, B} ←→ [A, B] . (20.265)
ih̄
Again, what should we choose for S1 ? Usually the choice is made to make the O() term
in H̃ vanish. But this is not the only possible simplifying choice.
20.10.2 Canonical Perturbation Theory for n = 1 Systems
Henceforth we shall assume H(ξ, t) = H(ξ) is time-independent, and we write the perturbed
Hamiltonian as
H(ξ) = H0 (ξ) + H1 (ξ) . (20.266)
Let (φ0 , J0 ) be the action-angle variables for H0 . Then

H̃0 (φ0 , J0 ) = H0 q(φ0 , J0 ), p(φ0 , J0 ) = H̃0 (J0 ) . (20.267)
We define

H̃1 (φ0 , J0 ) = H1 q(φ0 , J0 ), p(φ0 , J0 ) . (20.268)
We assume that H̃ = H̃0 + H̃1 is integrable12 , so it, too, possesses action-angle vari-
ables, which we denote by (φ, J)13 . Thus, there must be a canonical transformation taking
(φ0 , J0 ) → (φ, J), with

H̃ φ0 (φ, J), J0 (φ, J) ≡ K(J) = E(J) . (20.269)
We solve via a type-II canonical transformation:
S(φ0 , J) = φ0 J + S1 (φ0 , J) + 2 S2 (φ0 , J) + . . . , (20.270)
where φ0 J is the identity transformation. Then
∂S ∂S1 ∂S2
J0 = =J + + 2 + ... (20.271)
∂φ0 ∂φ0 ∂φ0
∂S ∂S1 ∂S2
φ= = φ0 + + 2 + ... , (20.272)
∂J ∂J ∂J
and
E(J) = E0 (J) + E1 (J) + 2 E2 (J) + . . . (20.273)

= H̃0 (φ0 , J0 ) + H̃1 (φ0 , J0 ) . (20.274)
12
This is always true, in fact, for n = 1.
13
We assume the motion is bounded, so action-angle variables may be used.
We now expand H̃(φ0 , J0 ) in powers of J0 − J:

H̃(φ0 , J0 ) = H̃0 (φ0 , J0 ) + H̃1 (φ0 , J0 ) (20.275)
∂ H̃0 ∂ 2H̃
= H̃0 (J) + (J0 − J) + 01
(J − J)2 + . . .
∂J ∂J 2 0
2
∂ H̃1
+ H̃1 (φ0 , J0 ) + (J0 − J) + . . .
∂J

∂ H̃0 ∂S1
= H̃0 (J) + H̃1 (φ0 , J0 ) + (20.276)
∂J ∂φ0
!
∂ H̃0 ∂S2 1 ∂ 2H̃0 ∂S1 2 ∂ H̃1 ∂S1

+ + + 2 + . . . .
∂J ∂φ0 2 ∂J 2 ∂φ0 ∂J ∂φ0
Equating terms, then,
E0 (J) = H̃0 (J) (20.277)
∂ H̃0 ∂S1
E1 (J) = H̃1 (φ0 , J) + (20.278)
∂J ∂φ0
∂ H̃0 ∂S2 1 ∂ 2H̃0 ∂S1 2 ∂ H̃1 ∂S1

E2 (J) = + + . (20.279)
∂J ∂φ0 2 ∂J 2 ∂φ0 ∂J ∂φ0
How, one might ask, can we be sure that the LHS of each equation in the above hierarchy
depends only on J when each RHS seems to depend on φ0 as well? The answer is that we
use the freedom to choose each Sk to make this so. We demand each RHS be independent
of φ0 , which means it must be equal to its average, h RHS(φ0 ) i, where
Z2π

dφ0
f φ0 = f φ0 . (20.280)
2π
0
The average is performed at fixed J and not at fixed J0 . In this regard, we note that holding
J constant and increasing φ0 by 2π also returns us to the same starting point. Therefore,
J is a periodic function of φ0 . We must then be able to write
∞
X
Sk (φ0 , J) = Sk (J; m) eimφ0 (20.281)
m=−∞
for each k > 0, in which case

∂Sk 1
= Sk (2π) − Sk (0) = 0 . (20.282)
∂φ0 2π
Let’s see how this averaging works to the first two orders of the hierarchy. Since H̃0 (J) is
independent of φ0 and since ∂S1 /∂φ0 is periodic, we have
this vanishes!
z }| {

∂ H̃0 ∂S1
E1 (J) = H̃1 (φ0 , J) + (20.283)
∂J ∂φ0
and hence S1 must satisfy

∂S1 H̃1 − H̃1
= , (20.284)
∂φ0 ν0 (J)
where ν0 (J) = ∂ H̃0 /∂J. Clearly the RHS of eqn. 20.284 has zero average, and must be a
periodic function of φ0 . The solution is S1 = S1 (φ0 , J) + g(J), where g(J) is an arbitrary
function of J. However, g(J) affects only the difference φ − φ0 , changing it by a constant
value g 0 (J). So there is no harm in taking g(J) = 0.
Next, let’s go to second order in . We have

this vanishes!
2 z }| {

∂ H̃1 ∂S1 1∂ν0 ∂S1 ∂S2
E2 (J) = +2 + ν0 (J) . (20.285)
∂J ∂φ0 ∂J ∂φ1 ∂φ0
The equation for S2 is then

(
∂S2 1 ∂ H̃1
∂ H̃1 ∂ H̃1
∂ H̃1
= 2 H̃0 − H̃0 − H̃1 + H̃1
∂φ0 ν0 (J) ∂J ∂J ∂J ∂J
)
1 ∂ ln ν0
2
2
H̃1 − 2 H̃1 + 2 H̃1 − H̃12

+ . (20.286)
2 ∂J
The expansion for the energy E(J) is then

(
2

∂ H̃1
∂ H̃1
E(J) = H̃0 (J) + H̃1 + H̃1 − H̃1
ν0 (J) ∂J ∂J
)
1 ∂ ln ν0
2
2
+ H̃1 − H̃1 + O(3 ) . (20.287)
2 ∂J
Note that we don’t need S to find E(J)! The perturbed frequencies are
∂E
ν(J) = . (20.288)
∂J
Sometimes the frequencies are all that is desired. However, we can of course obtain the full
motion of the system via the succession of canonical transformations,
(φ, J) −→ (φ0 , J0 ) −→ (q, p) . (20.289)
20.10.3 Example : Nonlinear Oscillator
Consider the nonlinear oscillator with Hamiltonian

H0
z }| {
p2
H(q, p) = + 1 mν 2 q 2 + 14 αq 4 . (20.290)
2m 2 0
Figure 20.7: Action-angle variables for the harmonic oscillator.
The action-angle variables for the harmonic oscillator Hamiltonian H0 are
p2
φ0 = tan−1 mvq/p) , J0 = + 1 mν q 2 , (20.291)
2mν0 2 0
and the relation between (φ0 , J0 ) and (q, p) is further depicted in fig. 20.7. Note H0 = ν0 J0 .
For the full Hamiltonian, we have
r 4
1 2J0
H̃(φ0 , J0 ) = ν0 J0 + 4 α sin φ0
mν0
α
= ν0 J0 + 2 2 J02 sin4 φ0 . (20.292)
m ν0
We may now evaluate
Z2π
αJ 2 dφ0 3αJ 2
sin4 φ0 =

E1 (J) = H̃1 = 2 2 . (20.293)
m ν0 2π 8m2 ν02
0
The frequency, to order , is

3 αJ
ν(J) = ν0 + . (20.294)
4m2 ν02
Now to lowest order in , we may replace J by J0 = 21 mν0 A2 , where A is the amplitude of
the q motion. Thus,
3α
ν(A) = ν0 + . (20.295)
8mν0
This result agrees with that obtained via heavier lifting, using the Poincaré-Lindstedt
method.
Next, let’s evaluate the canonical transformation (φ0 , J0 ) → (φ, J). We have
∂S1 αJ 2
ν0 = 2 2 38 − sin4 φ0 ⇒
∂φ0 m ν0
αJ 2
3 + 2 sin2 φ0 sin φ0 cos φ0 + O(2 ) .

S(φ0 , J) = φ0 J + 2 3 (20.296)
8m ν0
Thus,
∂S αJ
3 + 2 sin2 φ0 sin φ0 cos φ0 + O(2 )

φ= = φ0 + 2 3 (20.297)
∂J 4m ν0
∂S αJ 2
+ O(2 ) .

J0 = =J+ 2 3 4 cos 2φ0 − cos 4φ0 (20.298)
∂φ0 8m ν0
Again, to lowest order, we may replace J by J0 in the above, whence
αJ02
4 cos 2φ0 − cos 4φ0 + O(2 )

J = J0 − 2 3 (20.299)
8m ν0
αJ0 2
sin 2φ0 + O(2 ) .

φ = φ0 + 3 3 + 2 sin φ0 (20.300)
8m2 ν0
To obtain (q, p) in terms of (φ, J) is not analytically tractable – the relations cannot be
analytically inverted.
20.10.4 n > 1 Systems : Degeneracies and Resonances
Generalizing the procedure we derived for n = 1, we obtain
∂S ∂S1 ∂S2
J0α = α = J α + α + 2 + ... (20.301)
∂φ0 ∂φ0 ∂φα0
∂S ∂S1 ∂S2
φα = α
= φα0 + α
+ 2 + ... (20.302)
∂J ∂J ∂J α
and
E0 (J ) = H̃0 (J ) (20.303)
∂S1
E1 (J ) = H̃0 (φ0 , J ) + ν0α (J ) (20.304)
∂φα0
∂ H̃0 ∂S2 1 ∂ν0α ∂S1 ∂S1 ∂S1
E2 (J ) = α + α + ν0α . (20.305)
∂Jα ∂φ0 β
2 ∂J ∂φ0 ∂φ0 β ∂φα0
We now implement the averaging procedure, with
Z2π 1 Z2π n
1 n dφ0 dφ0
f φ10 , . . . , φn0 , J 1 , . . . , J n .

f (J , . . . , J ) = ··· (20.306)
2π 2π
0 0
The equation for S1 is
∂S1 X0
ν0α Vl eil·φ ,

α = H̃1 − H̃1 ≡ − (20.307)
∂φ0
l
where l = {l1 , l2 , . . . , ln }, with each lσ an integer, and with l 6= 0. The solution is

X 0 Vl
S1 (φ0 , J ) = i eil·φ . (20.308)
l · ν0
l
where l · ν0 = lα ν0α . When two or more of the frequencies να (J) are commensurate, there
exists a set of integers l such that the denominator of D(l) vanishes. But even when the
0
frequencies are not rationally related, one can approximate the ratios ν0α /ν0α by rational
numbers, and for large enough l the denominator can become arbitrarily small.
Periodic time-dependent perturbations present a similar problem. Consider the system

H(φ, J , t) = H0 (J ) + V (φ, J , t) , (20.309)
where V (t + T ) = V (t). This means we may write
X
V (φ, J , t) = Vk (φ, J ) e−ikΩt (20.310)
k
XX
= V̂k,l (J ) eil·φ e−ikΩt . (20.311)
k l
by Fourier transforming from both time and angle variables; here Ω = 2π/T . Note that
∗ =V
V (φ, J , t) is real if Vk,l −k,−l . The equations of motion are
∂H X
J˙α = − α = −i lα V̂k,l (J ) eil·φ e−ikΩt (20.312)
∂φ
k,l
∂H X ∂ V̂k,l (J )
φ̇α = + α
= ν0α (J ) + eil·φ e−ikΩt . (20.313)
∂J ∂J α
k,l
We now expand in :
φα = φα0 + φα1 + 2 φα2 + . . . (20.314)
α
J = J0α + J1α + 2
J2α + ... . (20.315)
To order 0 , J α = J0α and φα0 = ν0α t + β0α . To order 1 ,
X
J˙1α = −i lα V̂k,l (J0 ) ei(l·ν0 −kΩ)t ei·β0 (20.316)
k,l
and
∂ν0α β X ∂ V̂k,l (J ) i(l·ν0 −kΩ)t il·β0
φ̇α1 = J + e e , (20.317)
∂J β 1 ∂J α
k,l
where derivatives are evaluated at J = J0 . The solution is:

X lα V̂k,l (J0 )
J1α = ei(l·ν0 −kΩ)t eil·β0 (20.318)
kΩ − l · ν0
k,l
( )
α l β V̂ (J ) ∂ V̂ (J)
∂ν k,l 0 k,l 1
φα1 = 0
+ ei(l·ν0 −kΩ)t eil·β0 . (20.319)
∂J β (kΩ − l · ν0 )2 ∂J α kΩ − l · ν0
When the resonance condition,

kΩ = l · ν0 (J0 ) , (20.320)
holds, the denominators vanish, and the perturbation theory breaks down.
20.10.5 Particle-Wave Interaction
Consider a particle of charge e moving in the presence of a constant magnetic field B = B ẑ

and a space- and time-varying electric field E(x, t), described by the Hamiltonian
1 2
H= p − ec A + eV0 cos(k⊥ x + kz z − ωt) , (20.321)
2m
where is a dimensionless expansion parameter. Working in the gauge A = Bxŷ, from our
earlier discussions in section 20.8.7, we may write
r
p2z

k⊥ P 2J
H = ωc J + + eV0 cos kz z + + k⊥ sin φ − ωt . (20.322)
2m mωc mωc
Here, r r
P 2J 2J
x= + sin φ , y =Q+ cos φ , (20.323)
mωc mωc mωc
with ωc = eB/mc, the cyclotron frequency. We now make a mixed canonical transformation,
generated by
k⊥ P
F = φJ 0 + kz z + − ωt K 0 − P Q0 , (20.324)
mωc
where the new sets of conjugate variables are (φ0 , J 0 ) , (Q0 , P 0 ) , (ψ 0 , K 0 ) . We then have

∂F ∂F
φ0 = =φ J= = J0 (20.325)
∂J 0 ∂φ
∂F k⊥ K 0 ∂F
Q=− =− + Q0 P0 = − =P (20.326)
∂P mωc ∂Q0
∂F k⊥ P ∂F
ψ0 = 0
= kz z + − ωt pz = = kz K 0 . (20.327)
∂K mωc ∂z
The transformed Hamiltonian is
∂F
H0 = H +
∂t r
kz2 0 2 2J 0

0 0 0 0
= ωc J + K − ωK + eV0 cos ψ + k⊥ sin φ . (20.328)
2m mωc
We will now drop primes and simply write H = H0 + H1 , with

kz2 2
H0 = ω c J + K − ωK (20.329)
2m r

2J
H1 = eV0 cos ψ + k⊥ sin φ . (20.330)
mωc
When = 0, the frequencies associated with the φ and ψ motion are
∂H0 ∂H0 k2 K
ωφ0 = = ωc , ωψ0 = = z − ω = kz vz − ω , (20.331)
∂φ ∂ψ m
where vz = pz /m is the z-component of the particle’s velocity. Now let us solve eqn. 20.307:
∂S1 ∂S1
ωφ0 + ωψ0 = h H1 i − H1 . (20.332)
∂φ ∂ψ
This yields
r
kz2 K

∂S1 ∂S1 2J
ωc + −ω = −eA0 cos ψ + k⊥ sin φ
∂φ m ∂ψ mωc
∞ r !
X 2J
= −eA0 Jn k⊥ cos(ψ + nφ) , (20.333)
n=−∞
mωc
where we have used the result

∞
X
iz sin θ
e = Jn (z) einθ . (20.334)
n=−∞
The solution for S1 is

s !
X eV0 2J¯
S1 = J k sin(ψ + nφ) . (20.335)
n
ω − nωc − kz2 K̄/m n ⊥ mωc
We then have new action variables J¯ and K̄, where

∂S1
J = J¯ + + O(2 ) (20.336)
∂φ
∂S1
K = K̄ + + O(2 ) . (20.337)
∂ψ
Defining the dimensionless variable
r
2J
λ ≡ k⊥ , (20.338)
mωc
we obtain the result
mωc2 mωc2
X nJn (λ) cos(ψ + nφ)
2 2
2 λ̄ = 2 λ − kz2 K
+ O(2 ) , (20.339)
2eV0 k⊥ 2eV0 k⊥ n
ω
−n− ωc mωc
¯
p
where λ̄ = k⊥ 2J/mω 14
c.
14
Note that the argument of Jn in eqn. 20.339 is λ and not λ̄. This arises because we are computing the
new action J¯ in terms of the old variables (φ, J) and (ψ, K).
Figure 20.8: Plot of λ versus ψ for φ = 0 (Poincaré section) for ω = 30.11 ωc Top panels are
nonresonant invariant curves calculated to first order. Bottom panels are exact numerical
dynamics, with x symbols marking the initial conditions. Left panels: weak amplitude
(no trapping). Right panels: stronger amplitude (shows trapping). From Lichtenberg and
Lieberman (1983).
We see that resonances occur whenever

ω k2 K
− z =n, (20.340)
ωc mωc
for any integer n. Let us consider the case kz = 0, in which the resonance condition is
ω = nωc . We then have
λ̄2 λ2 X n Jn (λ) cos(ψ + nφ)
= − ω , (20.341)
2α 2α n ωc − n
where
E0 ck⊥
α= · (20.342)
B ωc
is a dimensionless measure of the strength of the perturbation, with E0 ≡ k⊥ V0 . In Fig.
20.8 we plot the level sets for the RHS of the above equation λ(ψ) for φ = 0, for two different
values of the dimensionless amplitude α, for ω/ωc = 30.11 (i.e. off resonance). Thus, when
the amplitude is small, the level sets are far from a primary resonance, and the analytical and
numerical results are very similar (left panels). When the amplitude is larger, resonances
may occur which are not found in the lowest order perturbation treatment. However, as
is apparent from the plots, the gross features of the phase diagram are reproduced by
perturbation theory. What is missing is the existence of ‘chaotic islands’ which initially
emerge in the vicinity of the trapping regions.
20.11 Adiabatic Invariants
Adiabatic perturbations are slow, smooth, time-dependent perturbations to a dynamical

system. A classic example: a pendulum with a slowly varying length l(t). Suppose λ(t)
is the adiabatic parameter. We write H = H q, p; λ(t) . All explicit time-dependence to
H comes through λ(t). Typically, a dimensionless parameter may be associated with the
perturbation:
1 d ln λ
= , (20.343)
ω0 dt
where ω0 is the natural frequency of the system when λ is constant. We require 1 for
adiabaticity. In adiabatic processes, the action variables are conserved to a high degree of
accuracy. These are the adiabatic invariants. For example, for the harmonix oscillator, the
action is J = E/ν. While E and ν may vary considerably during the adiabatic process,
their ratio is very nearly fixed. As a consequence, assuming small oscillations,
2J
E = νJ = 12 mgl θ02 ⇒ θ0 (l) ≈ √ 3/2 , (20.344)
m gl
so θ0 (`) ∝ l−3/4 .
Suppose that for fixed λ the Hamiltonian is transformed to action-angle variables via the
generator S(q, J; λ). The transformed Hamiltonian is
∂S dλ
H̃(φ, J, t) = H(φ, J; λ) + , (20.345)
∂λ dt
where
H(φ, J; λ) = H q(φ, J; λ), p(φ, J; λ); λ) . (20.346)
We assume n = 1 here. Hamilton’s equations are now
∂ H̃ ∂ 2S dλ
φ̇ = + = ν(J; λ) + (20.347)
∂J ∂λ ∂J dt
∂ H̃ ∂ 2S dλ
J˙ = − =− . (20.348)
∂φ ∂λ ∂φ dt
The second of these may be Fourier decomposed as
X ∂Sm (J; λ) imφ
J˙ = −iλ̇ m e , (20.349)
m
∂λ
20.11. ADIABATIC INVARIANTS 485
Figure 20.9: A mechanical mirror.
hence
Z∞
X ∂Sm (J; λ) dλ imφ
∆J = J(t = +∞) − J(t = −∞) = −i m dt e . (20.350)
m
∂λ dt
−∞
Since λ̇ is small, we have φ(t) = ν t + β, to lowest order. We must therefore evaluate

integrals such as
Z∞
∂Sm (J; λ) dλ imνt
Im = dt e . (20.351)
∂λ dt
−∞
The term in curly brackets is a smooth, slowly varying function of t. Call it f (t). We
presume f (t) can be analytically continued off the real t axis, and that its closest singularity
in the complex t plane lies at t = ±iτ , in which case I behaves as exp(−|m|ντ ). Consider,
for example, the Lorentzian,
Z∞
C
f (t) = ⇒ dt f (t) eimνt = πτ e−|m|ντ , (20.352)
1 + (t/τ )2
−∞
which is exponentially small in the time scale τ . Because of this, only m = ±1 need be
considered. What this tells us is that the change ∆J may be made arbitrarily small by a
sufficiently slowly varying λ(t).
20.11.1 Example: mechanical mirror
Consider a two-dimensional version of a mechanical mirror, depicted in fig. 20.9. A particle

bounces between two curves, y = ±D(x), where |D0 (x)| << 1. The bounce time is τb⊥ =
2D/vy . We assume τ L/vx , where vx,y are the components of the particle’s velocity, and
L is the total length of the system. There are, therefore, many bounces, which means the
particle gets to sample the curvature in D(x).
The adiabatic invariant is the action,
ZD Z−D
1 1 2
J= dy m vy + dy m (−vy ) = mvy D(x) . (20.353)
2π 2π π
−D D
Thus,
π2J 2
E = 12 m vx2 + vy2 ) = 12 mvx2 + , (20.354)
8mD2 (x)
or 2
2E πJ
vx2 = − . (20.355)
m 2mD(x)
The particle is reflected in the throat of the device at horizontal coordinate x∗ , where
πJ
D(x∗ ) = √ . (20.356)
8mE
20.11.2 Example: magnetic mirror
Consider a particle of charge e moving in the presence of a uniform magnetic field B = B ẑ.
Recall the basic physics: velocity in the parallel direction vz is conserved, while in the plane
perpendicular to B the particle executes circular ‘cyclotron orbits’, satisfying
mv⊥2
e mcv⊥
= v⊥ B ⇒ ρ= , (20.357)
ρ c eB
where ρ is the radial coordinate in the plane perpendicular to B. The period of the orbits
is T = 2πρ.v⊥ = 2πmc/eB, hence their frequency is the cyclotron frequency ωc = eB/mc.
Now assume that the magnetic field is spatially dependent. Note that a spatially varying
B-field cannot be unidirectional:
∂Bz
∇ · B = ∇⊥ · B⊥ + =0. (20.358)
∂z
The non-collinear nature of B results in the drift of the cyclotron orbits. Nevertheless, if
the field B felt by the particle varies slowly on the time scale T = 2π/ωc , then the system
possesses an adiabatic invariant:
I I
1 1
mv + ec A · d`

J= p · d` = (20.359)
2π 2π
C C
I I
m e
= v · d` + B · n̂ dΣ . (20.360)
2π 2πc
C int(C)
The last two terms are of opposite sign, and one has
m ρeBz e
J =− · · 2πρ + · Bz · πρ2 (20.361)
2π mc 2πc
eBz ρ2 e m2 v⊥
2c
=− =− · ΦB (C) = − , (20.362)
2c 2πc 2eBz
where ΦB (C) is the magnetic flux enclosed by C.

20.11. ADIABATIC INVARIANTS 487
Figure 20.10: B field lines in a magnetic bottle.
The energy is
2
E = 21 mv⊥ + 12 mvz2 , (20.363)
hence we have
r
2
vz = E − MB . (20.364)
m
where
e e2
M ≡− J= ΦB (C) (20.365)
mc 2πmc2
is the magnetic moment. Note that vz vanishes when B = Bmax = E/M . When this limit
is reached, the particle turns around. This is a magnetic mirror . A pair of magnetic mirrors
may be used to confine charged particles in a magnetic bottle, depicted in fig. 20.10.
Let vk,0 , v⊥,0 , and Bk,0 be the longitudinal particle velocity, transverse particle velocity,
and longitudinal component of the magnetic field, respectively, at the point of injection.
Our two conservation laws (J and E) guarantee
vk2 (z) + v⊥
2 2
(z) = vk,0 2
+ v⊥,0 (20.366)
2
v⊥ (z)2 v⊥,0
= . (20.367)
Bk (z) Bk,0
This leads to reflection at a longitudinal coordinate z ∗ , where
v
u
u 2
vk,0
∗
Bk (z ) = Bk,0 t1 + 2 . (20.368)
v⊥,0
The physics is quite similar to that of the mechanical mirror.

20.11.3 Resonances
When n > 1, we have

X ∂Sm (J; λ) im·φ
J˙α = −iλ̇ mα e (20.369)
m
∂λ
Z∞
X
α ∂Sm (J; λ) dλ im·νt im·β
∆J = −i m dt e e . (20.370)
m
∂λ dt
−∞
Therefore, when m · ν(J) = 0 we have a resonance, and the integral grows linearly with
time – a violation of the adiabatic invariance of J α .
20.12 Appendix : Example Problem in Canonical Perturba-

tion Theory
Consider the Hamiltonian

p2 q3
H= + 12 m ω02 q 2 + 13 m ω02 ,
2m a
where is a small dimensionless parameter.
(a) Show that the oscillation frequency satisfies ν(J) = ω0 + O(2 ). That is, show that the
first order (in ) frequency shift vanishes.
Solution: It is good to recall the basic formulae

r
2J0 p
q= sin φ0 , p = 2m ω0 J0 cos φ0 (20.371)
mω0
as well as the results
∂S ∂S1 ∂S2
J0 = =J + + 2 + ... (20.372)
∂φ0 ∂φ0 ∂φ0
∂S ∂S1 ∂S2
φ= = φ0 + + 2 + ... , (20.373)
∂J ∂J ∂J
and
E0 (J) = H̃0 (J) (20.374)
∂ H̃0 ∂S1
E1 (J) = H̃1 (φ0 , J) + (20.375)
∂J ∂φ0
∂ H̃0 ∂S2 1 ∂ 2H̃0 ∂S1 2 ∂ H̃1 ∂S1

E2 (J) = + + . (20.376)
∂J ∂φ0 2 ∂J 2 ∂φ0 ∂J ∂φ0
20.12. APPENDIX : EXAMPLE PROBLEM IN CANONICAL PERTURBATION THEORY489
Expressed in action-angle variables,
H̃0 (φ0 , J) = ω0 J (20.377)

r
2 2ω0 3/2
H̃1 (φ0 , J) = J sin3 φ0 . (20.378)
3 ma2
∂ H̃0
Thus, ν0 = ∂J = ω0 .
Averaging the equation for E1 (J) yields

r

2 2ω0 3/2
3
E1 (J) = H̃1 (φ0 , J) = J sin φ0 = 0 . (20.379)
3 ma2
(b) Compute the frequency shift ν(J) to second order in .
Solution : From the equation for E1 , we also obtain
∂S1 1

= H̃1 − H̃1 . (20.380)
∂φ0 ν0
Inserting this into the equation for E2 (J) and averaging then yields

1 ∂ H̃1
1 ∂ H̃1
E2 (J) = H̃1 − H̃1 =− H̃1 (20.381)
ν0 ∂J ν0 ∂J
2
4ν0 J
6
=− sin φ0 (20.382)
3ma2
In computing the average of sin6 φ0 , it is good to recall the binomial theorem, or the Fi-
bonacci tree. The sixth order coefficents are easily found to be {1, 6, 15, 20, 15, 6, 1}, whence
1 6
sin6 φ0 = 6
eiφ0 − e−iφ0 (20.383)
(2i)
1

= 64 − 2 sin 6φ0 + 12 sin 4φ0 − 30 sin 2φ0 + 20 . (20.384)
Thus,
sin6 φ0 = 5

16 , (20.385)
whence
5 2 J2
E(J) = ω0 J − 12 (20.386)
ma2
and
∂E J
ν(J) = = ω0 − 56 2 . (20.387)
∂J ma2
(c) Find q(t) to order . Your result should be finite for all times.
Solution : From the equation for E1 (J), we have

s
∂S1 2 2J 3
=− sin3 φ0 . (20.388)
∂φ0 3 mω0 a2
s
2 2J 3
cos φ0 − 13 cos3 φ0

S1 (φ0 , J) = 2
(20.389)
3 mω0 a
J 3/2 1

=√ cos φ0 − 9 cos 3φ0 . (20.390)
2mω0 a2
Thus, with
S(φ0 , J) = φ0 J + S1 (φ0 , J) + . . . , (20.391)
we have
∂S 3 J 1/2 1

φ= = φ0 + √ cos φ0 − 9 cos 3φ 0 (20.392)
∂J 2 2mω0 a2
∂S J 3/2 1

J0 = =J−√ sin φ0 − 3 sin 3φ 0 . (20.393)
∂φ0 2mω0 a2
Inverting, we may write φ0 and J0 in terms of φ and J:
3 J 1/2 1

φ0 = φ + √ 9 cos 3φ − cos φ (20.394)
2 2mω0 a2
J 3/2 1

J0 = J + √ 3 sin 3φ − sin φ . (20.395)
2mω0 a2
Thus,
r
2J0
q(t) = sin φ0 (20.396)
mω0
r
2J δJ
= sin φ · 1 + + ... sin φ + δφ cos φ + . . . (20.397)
mω0 2J
r
2J J
1 + 13 cos 2φ + O 2 ,

= sin φ − (20.398)
mω0 mω0 a
with
φ(t) = φ(0) + ν(J) t . (20.399)
Chapter 21
Strange Attractors and Chaos : A

Preview
An attractor of a dynamical system ϕ̇ = V (ϕ) is the set of ϕ values that the system evolves
to after a sufficiently long time. For N = 1 the only possible attractors are stable fixed
points. For N = 2, we have stable nodes and spirals, but also stable limit cycles. For N > 3
the situation is qualitatively different, and a fundamentally new type of set, the strange
attractor, emerges.
A strange attractor is basically a bounded set on which nearby orbits diverge exponentially
(i.e. there exists at least one positive Lyapunov exponent). To envision such a set, consider
a flat rectangle, like a piece of chewing gum. Now fold the rectangle over, stretch it, and
squash it so that it maintains its original volume. Keep doing this. Two points which
started out nearby to each other will eventually, after a sufficiently large number of folds
and stretches, grow far apart. Formally, a strange attractor is a fractal , and may have
noninteger Hausdorff dimension. (We won’t discuss fractals and Hausdorff dimension here.)
21.1 The Lorenz Model
The canonical example of an N = 3 strange attractor is found in the Lorenz model. E.

N. Lorenz, in a seminal paper from the early 1960’s, reduced the essential physics of the
coupled partial differential equations describing Rayleigh-Benard convection (a fluid slab of
finite thickness, heated from below – in Lorenz’s case a model of the atmosphere warmed
by the ocean) to a set of twelve coupled nonlinear ordinary differential equations. Lorenz’s
intuition was that his weather model should exhibit recognizable patterns over time. What
he found instead was that in some cases, changing his initial conditions by a part in a
thousand rapidly led to totally different behavior. This sensitive dependence on initial
conditions is a hallmark of chaotic systems.
491
492 CHAPTER 21. STRANGE ATTRACTORS AND CHAOS : A PREVIEW
Figure 21.1: Evolution of the Lorenz equations for σ = 10, b = 83 , and r = 20, with initial
conditions (x, y, z) = (1, 1, 1), projected onto the (x, z) plane. The system is attracted by a
stable spiral.
The essential physics (or mathematics?) of Lorenz’s N = 12 system is elicited by the

reduced N = 3 system,
Ẋ = −σX + σY (21.1)
Ẏ = rX − Y − XZ (21.2)
Ż = XY − bZ , (21.3)
where σ, r, and b are all real and positive. Here t is the familiar time variable (appropriately
scaled), and (X, Y, Z) represent linear combinations of physical fields, such as global wind
current and poleward temperature gradient. These equations possess a symmetry under
(X, Y, Z) → (−X, −Y, Z), but what is most important is the presence of nonlinearities in
the second and third equations.
The Lorenz system is dissipative because phase space volumes contract:

∂ Ẋ ∂ Ẏ ∂ Ż
∇·V = + + = −(σ + b + 1) . (21.4)
∂X ∂Y ∂Z
Thus, volumes contract under the flow. Another property is the following. Let
F (X, Y, Z) = 12 X 2 + 12 Y 2 + 12 (Z − r − σ)2 . (21.5)
Then
Ḟ = X Ẋ + Y Ẏ + (Z − r − σ)Ż
2
= −σX 2 − Y 2 − b Z − 21 r − 12 σ + 14 b(r + σ)2 . (21.6)
Thus, Ḟ < 0 outside an ellipsoid, which means that all solutions must remain bounded in
phase space for all times.
21.1. THE LORENZ MODEL 493
Figure 21.2: Evolution of the Lorenz equations for σ = 10, b = 83 , and r = 28, with initial
conditions (X0 , Y0 , Z0 ) = (1, 1, 1), showing the ‘strange attractor’.
21.1.1 Fixed point analysis
Setting ẋ = ẏ = ż = 0, we have three possible solutions. One solution, which is always

present, is x∗ = y ∗ = z ∗ = 0. If we linearize about this solution, we obtain
    
δX −σ σ 0 δX
d   
δY = r −1 0   δY  . (21.7)
dt
δZ 0 0 −b δZ
The eigenvalues of the linearized dynamics are found to be

p
λ1,2 = − 12 (1 + σ) ± 12 (1 + σ)2 + 4σ(r − 1) (21.8)
λ3 = −b ,
and thus if 0 < r < 1 all three eigenvalues are negative, and the fixed point is a stable
node. If, however, r > 1, then λ3 > 0 and the fixed point is attractive in two directions but
repulsive in a third, corresponding to a three-dimensional version of a saddle point.
For r > 1, a new pair of solutions emerges, with
X ∗ = Y ∗ = ± b(r − 1) , Z∗ = r − 1 .
p
(21.9)
Linearizing about either one of these fixed points, we find

    
δX −σ σ 0 δX
d   
δY = 1 −1 −X ∗   δY  . (21.10)
dt
δZ X ∗ X ∗ −b δZ
Figure 21.3: X(t) for the Lorenz equations with σ = 10, b = 83 , r = 28, and initial conditions
(X0 , Y0 , Z0 ) = (−2.7, −3.9, 15.8), and initial conditions (X0 , Y0 , Z0 ) = (−2.7001, −3.9, 15.8).
The characteristic polynomial of the linearized map is
P (λ) = λ3 + (b + σ + 1) λ2 + b(σ + r) λ + 2b(r − 1) . (21.11)
Since b, σ, and r are all positive, P 0 (λ) > 0 for all λ ≥ 0. Since P (0) = 2b(r − 1) > 0,
we may conclude that there is always at least one eigenvalue λ1 which is real and negative.
The remaining two eigenvalues are either both real and negative, or else they occur as a
complex conjugate pair: λ2,3 = α ± iβ. The fixed point is stable provided α < 0. The
stability boundary lies at α = 0. Thus, we set
h i h i
P (iβ) = 2b(r − 1) − (b + σ + 1)β 2 + i b(σ + r) − β 2 β = 0 , (21.12)
which results in two equations. Solving these two equations for r(σ, b), we find
σ(σ + b + 3)
rc = . (21.13)
σ−b−1

The fixed point is stable for r ∈ 1, rc . These fixed points correspond to steady convection.
The approach to this fixed point is shown in Fig. 21.1.
The Lorenz system has commonly been studied with σ = 10 and b = 38 . This means that the
volume collapse is very rapid, since ∇· V = − 41
3 ≈ −13.67, leading to a volume contraction
−41/3 −6
' 1.16 × 10 per unit time. For these parameters, one also has rc = 470
of e 19 ≈ 24.74.
The capture by the strange attractor is shown in Fig. 21.2.
In addition to the new pair of fixed points, a strange attractor appears for r > rs ' 24.06.
In the narrow interval r ∈ [24.06, 24.74] there are then three stable attractors, two of which
correspond to steady convection and the third to chaos. Over this interval, there is also
hysteresis. I.e. starting with a convective state for r < 24.06, the system remains in the
convective state until r = 24.74, when the convective fixed point becomes unstable. The
system is then driven to the strange attractor, corresponding to chaotic dynamics. Reversing
21.1. THE LORENZ MODEL 495
Figure 21.4: Lorenz attractor for b = 83 , σ = 10, and r = 28. Maxima of Z are depicted by
stars.
the direction of r, the system remains chaotic until r = 24.06, when the strange attractor
loses its own stability.
Another simple N = 3 system which possesses a strange attractor is the Rössler system,
Ẋ = −Y − Z (21.14)
Ẏ = Z + aY (21.15)
Ż = b + Z(X − µ) , (21.16)
typically studied as a function of µ for a = b = 15 . In Fig. 21.6, we present results from

work by Crutchfield et al. (1980). The transition from simple limit cycle to strange attractor
proceeds via a sequence of period-doubling bifurcations, as shown in the figure. A convenient
diagnostic for examining this period-doubling route to chaos is the power spectral density,
or PSD, defined for a function F (t) as
Z∞ 2
dω 2
F (t) e−iωt = F̂ (ω) .

ΦF (ω) = (21.17)

2π
−∞
As one sees in Fig. 21.6, as µ is increased past each critical value, the PSD exhibits a series
of frequency halvings (i.e. period doublings). All harmonics of the lowest frequency peak
are present. In the chaotic region, where µ > µ∞ ≈ 4.20, the PSD also includes a noisy
broadband background.
Figure 21.5: Plot of relation between successive maxima Zn along the strange attractor for
the Lorenz system.
Figure 21.6: Period doubling bifurcations of the Rössler attractor, projected onto the (x, y)
plane, for eight values of µ, and corresponding power spectral density for Z(t). (a) µ = 2.6;
(b) µ = 3.5; (c) µ = 4.1; (d) µ = 4.18; (e) µ = 4.21; (f) µ = 4.23; (g) µ = 4.30; (h) µ = 4.60.
Chapter 22
Physics 110A-B Midterm and

Final Examinations
The following pages contain problems and solutions from midterm and final examps in
Physics 110A-B.
497
498 CHAPTER 22. PHYSICS 110A-B MIDTERM AND FINAL EXAMINATIONS
22.1 F05 Physics 110A Midterm #1
[1] A particle of mass m moves in the one-dimensional potential

x2 −x/a
U (x) = U0 e . (22.1)
a2
(a) Sketch U (x). Identify the location(s) of any local minima and/or maxima, and be sure
that your sketch shows the proper behavior as x → ±∞.
(b) Sketch a representative set of phase curves. Identify and classify any and all fixed points.
Find the energy of each and every separatrix.
2
(c) Sketch all the phase curves for motions with total energy E = 5 U0 . Do the same for
E = U0 . (Recall that e = 2.71828 . . . .)
(d) Derive and expression for the period T of the motion when |x| a.
Solution:
(a) Clearly U (x) diverges to +∞ for x → −∞, and U (x) → 0 for x → +∞. Setting
U 0 (x) = 0, we obtain the equation
x2

0 U0
U (x) = 2 2x − e−x/a = 0 , (22.2)
a a
with (finite x) solutions at x = 0 and x = 2a. Clearly x = 0 is a local minimum and x = 2a
a local maximum. Note U (0) = 0 and U (2a) = 4 e−2 U0 ≈ 0.541 U0 .
Figure 22.1: The potential U (x). Distances are here measured in units of a, and the
potential in units of U0 .
(b) Local minima of a potential U (x) give rise to centers in the (x, v) plane, while local
maxima give rise to saddles. In Fig. 22.2 we sketch the phase curves. There is a center at
22.1. F05 PHYSICS 110A MIDTERM #1 499
Figure 22.2: Phase curves for the potential U (x). The red curves show phase curves for
E = 52 U0 (interior, disconnected red curves, |v| < 1) and E = U0 (outlying red curve). The
separatrix is the dark blue curve which forms a saddle at (x, v) = (2, 0), and corresponds
to an energy E = 4 e−2 U0 .
(0, 0) and a saddle at (2a, 0). There is one separatrix, at energy E = U (2a) = 4 e−2 U0 ≈
0.541 U0 .
(c) Even without a calculator, it is easy to verify that 4 e−2 > 25 . One simple way is to
multiply both sides by 25 e2 to obtain 10 > e2 , which is true since e2 < (2.71828 . . .)2 < 10.
Thus, the energy E = 52 U0 lies below the local maximum value of U (2a), which means that
there are two phase curves with E = 25 U0 .
It is also quite obvious that the second energy value given, E = U0 , lies above U (2a), which
means that there is a single phase curve for this energy. One finds bound motions only for
x < 2 and 0 ≤ E < U (2a). The phase curves corresponding to total energy E = 52 U0 and
E = U0 are shown in Fig. 22.2.
(d) Expanding U (x) in a Taylor series about x = 0, we have
U0 2 x3 x4

U (x) = 2 x − + 2 + ... . (22.3)
a a 2a
The leading order term is sufficient for |x| a. The potential energy is then equivalent to
that of a spring, with spring constant k = 2U0 /a2 . The period is
r s
m ma2
T = 2π = 2π . (22.4)
k 2U0
[2] A forced, damped oscillator obeys the equation
ẍ + 2β ẋ + ω02 x = f0 cos(ω0 t) . (22.5)
You may assume the oscillator is underdamped.
(a) Write down the most general solution of this differential equation.
(b) Your solution should involve two constants. Derive two equations relating these con-
stants to the initial position x(0) and the initial velocity ẋ(0). You do not have to solve
these equations.
(c) Suppose ω0 = 5.0 s−1 , β = 4.0 s−1 , and f0 = 8 cm s−2 . Suppose further you are told that
x(0) = 0 and x(T ) = 0, where T = π6 s. Derive an expression for the initial velocity ẋ(0).
Solution: (a) The general solution with forcing f (t) = f0 cos(Ωt) is

x(t) = xh (t) + A(Ω) f0 cos Ωt − δ(Ω) , (22.6)
with i−1/2
h
−1 2βΩ
A(Ω) = (ω02 2 2
− Ω ) + 4β Ω 2 2
, δ(Ω) = tan 2 (22.7)
ω0 − Ω 2
and
xh (t) = C e−βt cos(νt) + D e−βt sin(νt) , (22.8)
p
with ν = ω02 − β 2 .
In our case, Ω = ω0 , in which case A = (2βω0 )−1 and δ = 1

2 π. Thus, the most general
solution is
f0
x(t) = C e−βt cos(νt) + D e−βt sin(νt) + sin(ω0 t) . (22.9)
2βω0
(b) We determine the constants C and D by the boundary conditions on x(0) and ẋ(0):
f0
x(0) = C , ẋ(0) = −βC + νD + . (22.10)
2β
Thus,
β 1 f0
C = x(0) , D= x(0) + ẋ(0) − . (22.11)
ν ν 2βν
(c) From x(0) = 0 we obtain C = 0. The constant D is then determined by the condition
at time t = T = 16 π.
Note that ν = ω02 − β 2 = 3.0 s−1 . Thus, with T = 16 π, we have νT = 12 π, and

p
f0
x(T ) = D e−βT + sin(ω0 T ) . (22.12)
2βω0
This determines D:
f0 βT
D=− e sin(ω0 T ) . (22.13)
2βω0
We now can write
f0
ẋ(0) = νD + (22.14)
2β

f0 ν βT
= 1− e sin(ω0 T ) (22.15)
2β ω0

= 1− 3
10 e2π/3 cm/s . (22.16)
Numerically, the value is ẋ(0) ≈ 0.145 cm/s .

[1] Two blocks connected by a spring of spring constant k are free to slide frictionlessly
along a horizontal surface, as shown in Fig. 22.3. The unstretched length of the spring is a.
Figure 22.3: Two masses connected by a spring sliding horizontally along a frictionless
surface.
(a) Identify a set of generalized coordinates and write the Lagrangian.

[15 points]
Solution : As generalized coordinates I choose X and u, where X is the position of the

right edge of the block of mass M , and X + u + a is the position of the left edge of the
block of mass m, where a is the unstretched length of the spring. Thus, the extension of
the spring is u. The Lagrangian is then
L = 12 M Ẋ 2 + 12 m(Ẋ + u̇)2 − 12 ku2

= 12 (M + m)Ẋ 2 + 12 mu̇2 + mẊ u̇ − 12 ku2 . (22.17)
(b) Find the equations of motion.

[15 points]
Solution : The canonical momenta are

∂L ∂L
pX ≡ = (M + m)Ẋ + mu̇ , pu ≡ = m(Ẋ + u̇) . (22.18)
∂ Ẋ ∂ u̇
The corresponding equations of motion are then
∂L
ṗX = FX = ⇒ (M + m)Ẍ + mü = 0 (22.19)
∂X
∂L
ṗu = Fu = ⇒ m(Ẍ + ü) = −ku . (22.20)
∂u
(c) Find all conserved quantities.

[10 points]
Solution : There are two conserved quantities. One is pX itself, as is evident from the
fact that L is cyclic in X. This is the conserved ‘charge’ Λ associated with the continuous
symmetry X → X + ζ. i.e. Λ = pX . The other conserved quantity is the Hamiltonian H,
since L is cyclic in t. Furthermore, because the kinetic energy is homogeneous of degree

two in the generalized velocities, we have that H = E, with
E = T + U = 21 (M + m)Ẋ 2 + 12 mu̇2 + mẊ u̇ + 12 ku2 . (22.21)
It is possible to eliminate Ẋ, using the conservation of Λ:

Λ − mu̇
Ẋ = . (22.22)
M +m
This allows us to write
Λ2 M m u̇2
E= + + 1 ku2 . (22.23)
2(M + m) 2(M + m) 2
(d) Find a complete solution to the equations of motion. As there are two degrees of
freedom, your solution should involve 4 constants of integration. You need not match initial
conditions, and you need not choose the quantities in part (c) to be among the constants.
[10 points]
Solution : Using conservation of Λ, we may write Ẍ in terms of ẍ, in which case

Mm
ü = −ku ⇒ u(t) = A cos(Ωt) + B sin(Ωt) , (22.24)
M +m
where r
(M + m)k
Ω= . (22.25)
Mm
For the X motion, we integrate eqn. 22.22 above, obtaining
Λt m
X(t) = X0 + − A cos(Ωt) − A + B sin(Ωt) . (22.26)
M +m M +m
There are thus four constants: X0 , Λ, A, and B. Note that conservation of energy says
Λ2
E= + 1 k(A2 + B 2 ) . (22.27)
2(M + m) 2
Alternate solution : We could choose X as the position of the left block and x as the
position of the right block. In this case,
L = 12 M Ẋ 2 + 12 mẋ2 − 12 k(x − X − b)2 . (22.28)
Here, b includes the unstretched length a of the spring, but may also include the size of
the blocks if, say, X and x are measured relative to the blocks’ midpoints. The canonical
momenta are
∂L ∂L
pX = = M Ẋ , px = = mẋ . (22.29)
∂ Ẋ ∂ ẋ

∂L
ṗX = FX = ⇒ M Ẍ = k(x − X − b) (22.30)
∂X
∂L
ṗx = Fx = ⇒ mẍ = −k(x − X − b) . (22.31)
∂x
The one-parameter family which leaves L invariant is X → X + ζ and x → x + ζ, i.e.
simultaneous and identical displacement of both of the generalized coordinates. Then
Λ = M Ẋ + mẋ , (22.32)
which is simply the x-component of the total momentum. Again, the energy is conserved:
E = 12 M Ẋ 2 + 21 mẋ2 + 12 k (x − X − b)2 . (22.33)
We can combine the equations of motion to yield
d2
Mm 2
x − X − b = −k (M + m) x − X − b , (22.34)
dt
which yields
x(t) − X(t) = b + A cos(Ωt) + B sin(Ωt) , (22.35)
From the conservation of Λ, we have
M X(t) + m x(t) = Λt + C , (22.36)
were C is another constant. Thus, we have the motion of the system in terms of four
constants: A, B, Λ, and C:
Λt + C
X(t) = − Mm
+m b + A cos(Ωt) + B sin(Ωt) + M + m (22.37)
M
Λt + C
x(t) = M +m b + A cos(Ωt) + B sin(Ωt) + . (22.38)
M +m
[2] A uniformly dense ladder of mass m and length 2` leans against a block of mass M ,
as shown in Fig. 22.4. Choose as generalized coordinates the horizontal position X of the
right end of the block, the angle θ the ladder makes with respect to the floor, and the
coordinates (x, y) of the ladder’s center-of-mass. These four generalized coordinates are not
all independent, but instead are related by a certain set of constraints.
Recall that the kinetic energy of the ladder can be written as a sum TCM + Trot , where
TCM = 21 m(ẋ2 + ẏ 2 ) is the kinetic energy of the center-of-mass motion, and Trot = 21 I θ̇2 ,
where I is the moment of inertial. For a uniformly dense ladder of length 2`, I = 31 m`2 .
Figure 22.4: A ladder of length 2` leaning against a massive block. All surfaces are fric-
tionless..
(a) Write down the Lagrangian for this system in terms of the coordinates X, θ, x, y, and
their time derivatives.
[10 points]
Solution : We have L = T − U , hence

L = 12 M Ẋ 2 + 21 m(ẋ2 + ẏ 2 ) + 12 I θ̇2 − mgy . (22.39)
(b) Write down all the equations of constraint.

[10 points]
Solution : There are two constraints, corresponding to contact between the ladder and the
block, and contact between the ladder and the horizontal surface:
G1 (X, θ, x, y) = x − ` cos θ − X = 0 (22.40)
G2 (X, θ, x, y) = y − ` sin θ = 0 . (22.41)
(c) Write down all the equations of motion.

[10 points]
Solution : Two Lagrange multipliers, λ1 and λ2 , are introduced to effect the constraints.
We have for each generalized coordinate qσ ,
k
d ∂L ∂L X ∂Gj
− = λj ≡ Qσ , (22.42)
j=1
where there are k = 2 constraints. We therefore have
M Ẍ = −λ1 (22.43)
mẍ = +λ1 (22.44)
mÿ = −mg + λ2 (22.45)
I θ̈ = ` sin θ λ1 − ` cos θ λ2 . (22.46)
These four equations of motion are supplemented by the two constraint equations, yielding
six equations in the six unknowns {X, θ, x, y, λ1 , λ2 }.
(d) Find all conserved quantities.

[10 points]
Solution : The Lagrangian and all the constraints are invariant under the transformation
X →X +ζ , x→x+ζ , y→y , θ→θ . (22.47)
The associated conserved ‘charge’ is

∂L ∂ q̃σ
Λ= = M Ẋ + mẋ . (22.48)
∂ q̇σ ∂ζ ζ=0
Using the first constraint to eliminate x in terms of X and θ, we may write this as
Λ = (M + m)Ẋ − m` sin θ θ̇ . (22.49)
The second conserved quantity is the total energy E. This follows because the Lagrangian
and all the constraints are independent of t, and because the kinetic energy is homogeneous
of degree two in the generalized velocities. Thus,
E = 12 M Ẋ 2 + 12 m(ẋ2 + ẏ 2 ) + 12 I θ̇2 + mgy (22.50)

Λ2
= + 12 I + m`2 − Mm +m m`2
sin2
θ θ̇2 + mg` sin θ , (22.51)
2(M + m)
where the second line is obtained by using the constraint equations to eliminate x and y in
terms of X and θ.
(e) What is the condition that the ladder detaches from the block? You do not have to solve
for the angle of detachment! Express the detachment condition in terms of any quantities
you find convenient.
[10 points]
Solution : The condition for detachment from the block is simply λ1 = 0, i.e. the normal
force vanishes.
Further analysis : It is instructive to work this out in detail (though this level of analysis
was not required for the exam). If we eliminate x and y in terms of X and θ, we find
x = X + ` cos θ y = ` sin θ (22.52)

ẋ = Ẋ − ` sin θ θ̇ ẏ = ` cos θ θ̇ (22.53)
2 2
ẍ = Ẍ − ` sin θ θ̈ − ` cos θ θ̇ ÿ = ` cos θ θ̈ − ` sin θ θ̇ . (22.54)
We can now write
λ1 = mẍ = mẌ − m` sin θ θ̈ − m` cos θ θ̇2 = −M Ẍ , (22.55)
which gives
(M + m)Ẍ = m` sin θ θ̈ + cos θ θ̇2 ,

(22.56)
and hence
Mm
` sin θ θ̈ + cos θ θ̇2 .

Qx = λ1 = − (22.57)
m+m
We also have
Qy = λ2 = mg + mÿ
= mg + m` cos θ θ̈ − sin θ θ̇2 .

(22.58)
We now need an equation relating θ̈ and θ̇. This comes from the last of the equations of
motion:
I θ̈ = ` sin θ λ1 − ` cos θλ2
m 2
= − MM+m ` sin2 θ θ̈ + sin θ cos θ θ̇2 − mg` cos θ − m`2 cos2 θ θ̈ − sin θ cos θ θ̇2

= −mg` cos θ − m`2 1 − m
M +m sin2 θ θ̈ + m
M +m m`2 sin θ cos θ θ̇2 . (22.59)
Collecting terms proportional to θ̈, we obtain
I + m`2 − Mm 2 m 2 2

+m sin θ θ̈ = M +m m` sin θ cos θ θ̇ − mg` cos θ . (22.60)
We are now ready to demand Qx = λ1 = 0, which entails

cos θ 2
θ̈ = − θ̇ . (22.61)
sin θ
Substituting this into eqn. 22.60, we obtain
I + m`2 θ̇2 = mg` sin θ .

(22.62)
Finally, we substitute this into eqn. 22.51 to obtain an equation for the detachment angle,
θ∗
Λ2 m`2

m
E− = 3− · sin θ · 12 mg` sin θ∗ .
2 ∗
(22.63)
2(M + m) M + m I + m`2
Figure 22.5: Plot of θ∗ versus θ0 for the ladder-block problem (eqn. 22.64). Allowed
solutions, shown in blue, have α ≥ 1, and thus θ∗ ≤ θ0 . Unphysical solutions, with α < 1,
are shown in magenta. The line θ∗ = θ0 is shown in red.
If our initial conditions are that the system starts from rest1 with an angle of inclination
θ0 , then the detachment condition becomes

m`2
sin θ0 = 23 sin θ∗ − 12 Mm
+m I+m`2
sin3 θ∗
= 3
2 sin θ∗ − 12 α−1 sin3 θ∗ , (22.64)
where
M I
α≡ 1+ 1+ . (22.65)
m m`2
Note that α ≥ 1, and that when M/m = ∞2 , we recover θ∗ = sin−1 23 sin θ0 . For finite

α, the ladder detaches at a larger value of θ∗ . A sketch of θ∗ versus θ0 is provided in Fig.

22.5. Note that, provided α ≥ 1, detachment always occurs for some unique value θ∗ for
each θ0 .
1
‘Rest’ means that the initial velocities are Ẋ = 0 and θ̇ = 0, and hence Λ = 0 as well.
2
I must satisfy I ≤ m`2 .
22.3. F05 PHYSICS 110A FINAL EXAM 509
22.3 F05 Physics 110A Final Exam
[1] Two blocks and three springs are configured as in Fig. 22.6. All motion is horizontal.
When the blocks are at rest, all springs are unstretched.
Figure 22.6: A system of masses and springs.
(a) Choose as generalized coordinates the displacement of each block from its equilibrium
position, and write the Lagrangian.
[5 points]

[5 points]
(c) Suppose
m1 = 2m , m2 = m , k1 = 4k , k2 = k , k3 = 2k ,
Find the frequencies of small oscillations.

[5 points]
(d) Find the normal modes of oscillation.

[5 points]
(e) At time t = 0, mass #1 is displaced by a distance b relative to its equilibrium position.

I.e. x1 (0) = b. The other initial conditions are x2 (0) = 0, ẋ1 (0) = 0, and ẋ2 (0) = 0.
Find t∗ , the next time at which x2 vanishes.
[5 points]
Solution
L = 12 m1 x21 + 12 m2 x22 − 12 k1 x21 − 12 k2 (x2 − x1 )2 − 12 k3 x22

(b) The T and V matrices are

! !
∂2T m1 0 ∂2U k1 + k2 −k2
Tij = = , Vij = =
∂ ẋi ∂ ẋj 0 m2 ∂xi ∂xj −k2 k2 + k3
(c) We have m1 = 2m, m2 = m, k1 = 4k, k2 = k, and k3 = 2k. Let us write ω 2 ≡ λ ω02 ,

p
where ω0 ≡ k/m. Then

2 2λ − 5 1
ω T−V=k .
1 λ−3
The determinant is
det (ω 2 T − V) = (2λ2 − 11λ + 14) k 2

= (2λ − 7) (λ − 2) k 2 .
There are two roots: λ− = 2 and λ+ = 72 , corresponding to the eigenfrequencies

r r
2k 7k
ω− = , ω+ =
m 2m
~ (a) = 0. Plugging in λ = 2 we have

(d) The normal modes are determined from (ωa2 T − V) ψ
~
for the normal mode ψ (−)
(−)
−1 1 ψ1 ~ (−) 1
=0 ⇒ ψ = C−
1 −1 ψ2(−) 1
Plugging in λ = 7 ~ (+)
we have for the normal mode ψ
2
(+)
2 1 ψ1 ~ (+) = C 1
=0 ⇒ ψ
1 12 ψ2(+) + −2
(a) (b)
The standard normalization ψi Tij ψj = δab gives
1 1
C− = √ , C2 = √ . (22.66)
3m 6m
(e) The general solution is

!
x1 1 1 1 1
=A cos(ω− t) + B cos(ω+ t) + C sin(ω− t) + D sin(ω+ t) .
x2 1 −2 1 −2
The initial conditions x1 (0) = b, x2 (0) = ẋ1 (0) = ẋ2 (0) = 0 yield
A = 23 b , B = 13 b , C=0 , D=0.
Thus,

x1 (t) = 13 b · 2 cos(ω− t) + cos(ω+ t)

x2 (t) = 23 b · cos(ω− t) − cos(ω+ t) .
Setting x2 (t∗ ) = 0, we find
2π
cos(ω− t∗ ) = cos(ω+ t∗ ) ⇒ π − ω− t = ω+ t − π ⇒ t∗ =
ω− + ω+
[2] Two point particles of masses m1 and m2 interact via the central potential
r2

U (r) = U0 ln 2 ,
r + b2
where b is a constant with dimensions of length.
(a) For what values of the relative angular momentum ` does a circular orbit exist? Find
the radius r0 of the circular orbit. Is it stable or unstable?
[7 points]
(c) For the case where a circular orbit exists, sketch the phase curves for the radial motion
in the (r, ṙ) half-plane. Identify the energy ranges for bound and unbound orbits.
[5 points]
(c) Suppose the orbit is nearly circular, with r = r0 +η, where |η| r0 . Find the equation
for the shape η(φ) of the perturbation.
[8 points]
(d) What is the angle ∆φ through which periapsis changes each cycle? For which value(s)
of ` does the perturbed orbit not precess?
[5 points]
Solution
(a) The effective potential is
`2
Ueff (r) = + U (r)
2µr2
`2 r2

= + U0 ln 2 .
2µr2 r + b2
where µ = m1 m2 /(m1 + m1 ) is the reduced mass. For a circular orbit, we must have
0 (r) = 0, or
Ueff
l2 0 2rU0 b2
= U (r) = .
µr3 r2 (r2 + b2 )
The solution is
b2 `2
r02 =
2µb2 U0 − `2
Since r02 > 0, the condition on ` is
p
` < `c ≡ 2µb2 U0
For large r, we have

`2

1
Ueff (r) = − U0 b2 · + O(r−4 ) .
2µ r2
Thus, for ` < `c the effective potential is negative for sufficiently large values of r. Thus,
over the range ` < `c , we must have Ueff,min < 0, which must be a global minimum, since
Ueff (0+ ) = ∞ and Ueff (∞) = 0. Therefore, the circular orbit is stable whenever it exists.
(b) Let ` = `c . The effective potential is then
Ueff (r) = U0 f (r/b) ,
where the dimensionless effective potential is
2
f (s) = − ln(1 + s−2 ) .
s2
The phase curves are plotted in Fig. 22.7.
(c) The energy is
E = 12 µṙ2 + Ueff (r)

2
`2 dr
= 4
+ Ueff (r) ,
2µr dφ
s−2 − ln(1 + s−2 ), with

Figure 22.7: Phase curves for the scaled effective potential f (s) = p
= √12 . Here, = `/`c . The dimensionless time variable is τ = t · U0 /mb2 .
where we’ve used ṙ = φ̇ r0 along with ` = µr2 φ̇. Writing r = r0 + η and differentiating E
with respect to φ, we find
µr04 00
η 00 = −β 2 η , β2 = U (r ) .
`2 eff 0
For our potential, we have
!
`2 `2
β2 = 2 − 2 =2 1− 2
µb U0 `c
The solution is
η(φ) = A cos(βφ + δ) (22.67)
where A and δ are constants.
(d) The change of periapsis per cycle is
∆φ = 2π β −1 − 1

If β > 1 then ∆φ < 0 and periapsis advances each cycle (i.e.it comes sooner with
p every
cycle). If β < 1 then ∆φ > 0 and periapsis recedes. For β = 1, which means ` = µb2 U0 ,
there is no precession and ∆φ = 0.
[3] A particle of charge e moves in three dimensions in the presence of a uniform magnetic
field B = B0 ẑ and a uniform electric field E = E0 x̂. The potential energy is
e
U (r, ṙ) = −e E0 x − B0 x ẏ ,
c
where we have chosen the gauge A = B0 x ŷ.
(a) Find the canonical momenta px , py , and pz .

[7 points]
(b) Identify all conserved quantities.
[8 points]

(c) Find a complete, general solution for the motion of the system x(t), y(t), x(t) .
[10 points]
Solution

e
L = 21 m(ẋ2 + ẏ 2 + ż 2 ) + B x ẏ + e E0 x .
c 0
∂L
px = = mẋ
∂ ẋ
∂L e
py = = mẏ + B0 x
∂ ẏ c
∂L
px = = mż
∂ ż
∂L
(b) There are three conserved quantities. First is the momentum py , since Fy = ∂y = 0.
∂L
Second is the momentum pz , since Fy = ∂z = 0. The third conserved quantity is the
Hamiltonian, since ∂L
∂t = 0. We have
H = px ẋ + py ẏ + pz ż − L
⇒ H = 12 m(ẋ2 + ẏ 2 + ż 2 ) − e E0 x
(c) The equations of motion are

e
ẍ − ωc ẏ = E
m 0
ÿ + ωc ẋ = 0
z̈ = 0 .
The second equation can be integrated once to yield ẏ = ωc (x0 − x), where x0 is a constant.
Substituting this into the first equation gives
e
ẍ + ωc2 x = ωc2 x0 + E .
m 0
This is the equation of a constantly forced harmonic oscillator. We can therefore write the
general solution as
eE0
x(t) = x0 + + A cos ω c t + δ
mωc2
eE0
y(t) = y0 − t − A sin ωc t + δ
mωc
z(t) = z0 + ż0 t

Note that there are six constants, A, δ, x0 , y0 , z0 , ż0 , are are required for the general
solution of three coupled second order ODEs.
[4] An N = 1 dynamical system obeys the equation
du
= ru + 2bu2 − u3 ,
dt
where r is a control parameter, and where b > 0 is a constant.
(a) Find and classify all bifurcations for this system.

[7 points]
(b) Sketch the fixed points u∗ versus r.

[6 points]
Now let b = 3. At time t = 0, the initial value of u is u(0) = 1. The control parameter
r is then increased very slowly from r = −20 to r = +20, and then decreased very
slowly back down to r = −20.
(c) What is the value of u when r = −5 on the increasing part of the cycle?
[3 points]
(d) What is the value of u when r = +16 on the increasing part of the cycle?
[3 points]
(e) What is the value of u when r = +16 on the decreasing part of the cycle?
[3 points]
(f) What is the value of u when r = −5 on the decreasing part of the cycle?
[3 points]
Solution
(a) Setting u̇ = 0 we obtain

(u2 − 2bu − r) u = 0 .
The roots are p
u = 0 , u = b ± b2 + r .
√
The roots at u = u± = b ± b2 + r are only present when r > −b2 . At r = −b2 there
is a saddle-node bifurcation. The fixed point u = u− crosses the fixed point at u = 0 at
r = 0, at which the two fixed points exchange stability. This corresponds to a transcritical
bifurcation. In Fig. 22.8 we plot u̇/b3 versus u/b for several representative values of r/b2 .
Note that, defining ũ = u/b, r̃ = r/b2 , and t̃ = b2 t that our N = 1 system may be written
dũ
= r̃ + 2ũ − ũ2 ũ ,

dt̃
which shows that it is only the dimensionless combination r̃ = r/b2 which enters into the
location and classification of the bifurcations.
(b) A sketch of the fixed points u∗ versus r is shown in Fig. 22.9. Note the two bifurcations
at r = −b2 (saddle-node) and r = 0 (transcritical).
(c) For r = −20 < −b2 = −9, the initial condition u(0) = 1 flows directly toward the stable
fixed point at u = 0. Since the approach to the FP is asymptotic, u remains slightly positive
even after a long time. When r = −5, the FP at u = 0 is still stable. Answer: u = 0.
(d) As soon as r becomes positive, the FP at u∗ =√0 becomes unstable, and u flows to the
upper branch u+ . When r = 16, we have u = 3 + 32 + 16 = 8. Answer: u = 8.
(e) Coming back down from larger r, the upper FP branch remains stable, thus, u = 8 at
r = 16 on the way down as well. Answer: u = 8.
(f) Now when r first becomes negative on the way down, the upper branch u+ remains
stable. Indeed it remains stable all the way down to r = −b2 , the location of the saddle-
node bifurcation, at which point the solution u = u+ simply vanishes and the flow is toward
u = 0 again. Thus, for√r = −5 on the way down, the system remains on the upper branch,
in which case u = 3 + 32 − 5 = 5. Answer: u = 5.
Figure 22.8: Plot of dimensionless ‘velocity’ u̇/b3 versus dimensionless ‘coordinate’ u/b for
several values of the dimensionless control parameter r̃ = r/b2 .
[1] A particle of mass m moves in the one-dimensional potential
U0 2 2
U (x) = 4
x − a2 . (22.68)
a
(a) Sketch U (x). Identify the location(s) of any local minima and/or maxima, and be sure
that your sketch shows the proper behavior as x → ±∞.
[15 points]
Solution : Clearly the minima lie at x = ±a and there is a local maximum at x = 0.
(b) Sketch a representative set of phase curves. Be sure to sketch any separatrices which
exist, and identify their energies. Also sketch all the phase curves for motions with total
energy E = 21 U0 . Do the same for E = 2 U0 .
[15 points]
Figure 22.9: Fixed points and their stability versus control parameter for the N = 1 system
u̇ = ru + 2bu2 − u3 . Solid lines indicate stable fixed points; dashed lines indicate unstable
fixed points. There is a saddle-node bifurcation at r = −b2 and a transcritical bifurcation
at r = 0. The hysteresis loop in the upper half plane u > 0 is shown. For u < 0 variations
of the control parameter r are reversible and there is no hysteresis.
Solution : See Fig. 22.10 for the phase curves. Clearly U (±a) = 0 is the minimum of the
potential, and U (0) = U0 is the local maximum and the energy of the separatrix. Thus,
E = 21 U0 cuts through the potential in both wells, and the phase curves at this energy form
two disjoint sets. For E < U0 there are four turning points, at
s r s r
E E
x1,< = −a 1 + , x1,> = −a 1 −
U0 U0
and s s
r r
E E
x2,< = a 1 − , x2,> = a 1 +
U0 U0
For E = 2 U0 , the energy is above that of the separatrix, and there are only two turning
points, x1,< and x2,> . The phase curve is then connected.
Figure 22.10: Sketch of the double well potential U (x) = (U0 /a4 )(x2 − a2 )2 , here with
distances in units of a, and associated phase curves. The separatrix is the phase curve
which runs through the origin. Shown in red is the phase curve for U = 21 U0 , consisting
of two deformed ellipses. For U = 2 U0 , the phase curve is connected, lying outside the
separatrix.

x
(c) The phase space dynamics are written as ϕ̇ = V (ϕ), where ϕ = . Find the upper
ẋ
and lower components of the vector field V .
[10 points]
Solution :
d x ẋ ẋ
= = . (22.69)
dt ẋ 1
−m U 0 (x) − 4U
a2
0
x (x2 − a2 )
(d) Derive and expression for the period T of the motion when the system exhibits small
oscillations about a potential minimum.
[10 points]
Solution : Set x = ±a + η and Taylor expand:

4U0 2
U (±a + η) = η + O(η 3 ) . (22.70)
a2
Equating this with 12 k η 2 , we have the effective spring constant k = 8U0 /a2 , and the small
oscillation frequency r r
k 8 U0
ω0 = = . (22.71)
m ma2
The period is 2π/ω0 .
[2] An R-L-C circuit is shown in fig. 22.11. The resistive element is a light bulb. The
inductance is L = 400 µH; the capacitance is C = 1 µF; the resistance is R = 32 Ω. The
voltage V (t) oscillates sinusoidally, with V (t) = V0 cos(ωt), where V0 = 4 V. In this problem,
you may neglect all transients; we are interested in the late time, steady state operation of
this circuit. Recall the relevant MKS units:
1Ω = 1V · s/C , 1F = 1C/V , 1 H = 1 V · s2 / C .
Figure 22.11: An R-L-C circuit in which the resistive element is a light bulb.
(a) Is this circuit underdamped or overdamped?

[10 points]
Solution : We have
R
ω0 = (LC)−1/2 = 5 × 104 s−1 , β= = 4 × 104 s−1 .
2L
Thus, ω02 > β 2 and the circuit is underdamped .
(b) Suppose the bulb will only emit light when the average power dissipated by the bulb is
greater than a threshold Pth = 29 W . For fixed V0 = 4 V, find the frequency range for ω over
which the bulb emits light. Recall that the instantaneous power dissipated by a resistor is
PR (t) = I 2 (t)R. (Average this over a cycle to get the average power dissipated.)
[20 points]
Solution : The charge on the capacitor plate obeys the ODE

Q
L Q̈ + R Q̇ + = V (t) .
C
The solution is
V0
Q(t) = Qhom (t) + A(ω) cos ωt − δ(ω) ,
L
with i−1/2
h
−1 2βω
A(ω) = (ω02 − ω 2 )2 + 4β 2 ω 2 , δ(ω) = tan .
ω02 − ω 2
Thus, ignoring the transients, the power dissipated by the bulb is
PR (t) = Q̇2 (t) R

V02 R
= ω 2 A2 (ω) sin2 ωt − δ(ω) .

L 2
Averaging over a period, we have h sin2 (ωt − δ) i = 12 , so
V02 R V02 4β 2 ω 2
h PR i = ω 2 A2 (ω) = · .
2L2 2 R (ω02 − ω 2 )2 + 4β 2 ω 2
1
Now V02 /2R = 4 W. So Pth = αV02 /2R, with α = 89 . We then set hPR i = Pth , whence
(1 − α) · 4β 2 ω 2 = α (ω02 − ω 2 )2 .
The solutions are

s
√ √
r
1−α 1−α 2
ω=± β+ β + ω02 = 3 3 ± 2 × 1000 s−1 .
α α
(c) Compare the expressions for the instantaneous power dissipated by the voltage source,
PV (t), and the power dissipated by the resistor PR (t) = I 2 (t)R. If PV (t) 6= PR (t), where
does the power extra power go or come from? What can you say about the averages of PV
and PR (t) over a cycle? Explain your answer.
[20 points]
Solution : The instantaneous power dissipated by the voltage source is

V0
PV (t) = V (t) I(t) = −ω A sin(ωt − δ) cos(ωt)
L
V0
= ωA sin δ − sin(2ωt − δ) .
2L
As we have seen, the power dissipated by the bulb is
V02 R
PR (t) = ω 2 A2 sin2(ωt − δ) .
L2
These two quantities are not identical, but they do have identical time averages over one
cycle:
V2
h PV (t) i = h PR (t) i = 0 · 4β 2 ω 2 A2 (ω) .
2R
Energy conservation means
PV (t) = PR (t) + Ė(t) ,
where
LQ̇2 Q2
E(t) = +
2 2C
is the energy in the inductor and capacitor. Since Q(t) is periodic, the average of Ė over a
cycle must vanish, which guarantees h PV (t) i = h PR (t) i.
What was not asked:
(d) What is the maximum charge Qmax on the capacitor plate if ω = 3000 s−1 ?
[10 points]
Solution : Kirchoff’s law gives for this circuit the equation

V0
Q̈ + 2β Q̇ + ω02 Q = cos(ωt) ,
L
with the solution
V0
Q(t) = Qhom (t) + A(ω) cos ωt − δ(ω) ,
L
where Qhom (t) is the homogeneous solution, i.e. the transient which we ignore, and
i−1/2
h
2 2 2 2 2 −1 2βω
A(ω) = (ω0 − ω ) + 4β ω , δ(ω) = tan .
ω02 − ω 2
Then
V0
Qmax = A(ω) .
L
Plugging in ω = 3000 s−1 , we have
−1/2 1
A(ω) = (52 − 42 )2 + 4 · 42 · 32 × 10−3 s2 = √ × 10−3 s2 .

8 13
Since V0 /L = 104 C/s2 , we have
5
Qmax = √ Coul .
4 13
[1] A point mass m slides frictionlessly, under the influence of gravity, along a massive ring
of radius a and mass M . The ring is affixed by horizontal springs to two fixed vertical
surfaces, as depicted in fig. 22.12. All motion is within the plane of the figure.
Figure 22.12: A point mass m slides frictionlessly along a massive ring of radius a and mass
M , which is affixed by horizontal springs to two fixed vertical surfaces.
(a) Choose as generalized coordinates the horizontal displacement X of the center of the
ring with respect to equilibrium, and the angle θ a radius to the mass m makes with respect
to the vertical (see fig. 22.12). You may assume that at X = 0 the springs are both
unstretched. Find the Lagrangian L(X, θ, Ẋ, θ̇, t).
[15 points]
The coordinates of the mass point are
x = X + a sin θ , y = −a cos θ .

2
T = 12 M Ẋ 2 + 12 m Ẋ + a cos θ θ̇ + 21 ma2 sin2 θ θ̇2
= 12 (M + m)Ẋ 2 + 12 ma2 θ̇2 + ma cos θ Ẋ θ̇ .

U = kX 2 − mga cos θ .
Thus, the Lagrangian is
L = 21 (M + m)Ẋ 2 + 12 ma2 θ̇2 + ma cos θ Ẋ − kX 2 + mga cos θ .
(b) Find the generalized momenta pX and pθ , and the generalized forces FX and Fθ
[10 points]
We have
∂L ∂L
pX = = (M + m)Ẋ + ma cos θ θ̇ , pθ = = ma2 θ̇ + ma cos θ Ẋ .
∂ Ẋ ∂ θ̇
For the forces,
∂L ∂L
FX = = −2kX , Fθ = = −ma sin θ Ẋ θ̇ − mga sin θ .
∂X ∂θ
(c) Derive the equations of motion.

[15 points]

d ∂L ∂L
= ,
dt ∂ q̇σ ∂qσ
for each generalized coordinate qσ . For X we have
(M + m)Ẍ + ma cos θ θ̈ − ma sin θ θ̇2 = −2kX .
For θ,
ma2 θ̈ + ma cos θẌ = −mga sin θ .
(d) Find expressions for all conserved quantities.

[10 points]
Horizontal and vertical translational symmetries are broken by the springs and by gravity,
respectively. The remaining symmetry is that of time translation. From dH ∂L
dt = − ∂t , we have
P
that H = σ pσ q̇σ − L is conserved. For this problem, the kinetic energy is a homogeneous
function of degree 2 in the generalized velocities, and the potential is velocity-independent.
Thus,
H = T + U = 21 (M + m)Ẋ 2 + 12 ma2 θ̇2 + ma cos θ Ẋ θ̇ + kX 2 − mga cos θ .

[2] A point particle of mass m moves in three dimensions in a helical potential

2πz
U (ρ, φ, z) = U0 ρ cos φ − .
b
We call b the pitch of the helix.
(a) Write down the Lagrangian, choosing (ρ, φ, z) as generalized coordinates.

[10 points]
The Lagrangian is

1 2 2 2 2
2πz
L= 2m ρ̇ + ρ φ̇ + ż − U0 ρ cos φ −
b
(b) Find the equations of motion.

[20 points]
Clearly
pρ = mρ̇ , pφ = mρ2 φ̇ , pz = mż ,
and

2 2πz 2πz 2πU0 2πz
Fρ = mρ φ̇ −U0 cos φ− , Fφ = U0 ρ sin φ− , Fz = − ρ sin φ− .
b b b b
Thus, the equation of motion are

2 2πz
mρ̈ = mρ φ̇ − U0 cos φ −
b

2πz
mρ2 φ̈ + 2mρ ρ̇ φ̇ = U0 ρ sin φ −
b

2πU0 2πz
mz̈ = − ρ sin φ − .
b b
(c) Show that there exists a continuous one-parameter family of coordinate transformations
which leaves L invariant. Find the associated conserved quantity, Λ. Is anything else
conserved?
[20 points]
Due to the helical symmetry, we have that
b
φ→φ+ζ , z→z+ ζ
2π
is such a continuous one-parameter family of coordinate transformations. Since it leaves

the combination φ − 2πz dL
b unchanged, we have that dζ = 0, and

∂ρ ∂φ ∂z
Λ = pρ + pφ + pz
∂ζ ζ=0 ∂ζ ζ=0 ∂ζ ζ=0
b
= pφ + p
2π z
mb
= mρ2 φ̇ + ż
2π
is the conserved Noether ‘charge’. The other conserved quantity is the Hamiltonian,

1 2 2 2 2
2πz
H = 2 m ρ̇ + ρ φ̇ + ż + U0 ρ cos φ − .
b
Note that H = T + U , because T is homogeneous of degree 2 and U is homogeneous of

degree 0 in the generalized velocities.
22.6 F07 Physics 110A Final Exam
[1] Two masses and two springs are configured linearly and externally driven to rotate with
angular velocity ω about a fixed point on a horizontal surface, as shown in fig. 22.13. The
unstretched length of each spring is a.
Figure 22.13: Two masses and two springs rotate with angular velocity ω.
(a) Choose as generalized coordinates the radial distances r1,2 from the origin. Find the
Lagrangian L(r1 , r2 , ṙ1 , ṙ2 , t).
[5 points]
The Lagrangian is
L = 12 m ṙ12 + ṙ22 + ω 2 r12 + ω 2 r22 − 12 k (r1 − a)2 − 12 k (r2 − r1 − a)2 .

(22.72)
(b) Derive expressions for all conserved quantities.

[5 points]
The Hamiltonian is conserved. Since the kinetic energy is not homogeneous of degree 2 in
the generalized velocities, H 6= T + U . Rather,
X
H= pσ q̇σ − L (22.73)
σ
1
ṙ12 + ṙ22 − 12 mω 2 r12 + r22 ) + 12 k (r1 − a)2 + 12 k (r2 − r1 − a)2 .

= 2m (22.74)
We could define an effective potential
Ueff (r1 , r2 ) = − 12 mω 2 r12 + r22 ) + 12 k (r1 − a)2 + 12 k (r2 − r1 − a)2 . (22.75)
Note the first term, which comes from the kinetic energy, has an interpretation of a fictitious
potential which generates a centrifugal force.
(c) What equations determine the equilibrium radii r10 and r20 ? (You do not have to solve
these equations.)
[5 points]
The equations of equilibrium are Fσ = 0. Thus,

∂L
0 = F1 = = mω 2 r1 − k (r1 − a) + k (r2 − r1 − a) (22.76)
∂r1
∂L
0 = F2 = = mω 2 r2 − k (r2 − r1 − a) . (22.77)
∂r2
(d) Suppose now that the system is not externally driven, and that the angular coordinate
φ is a dynamical variable like r1 and r2 . Find the Lagrangian L(r1 , r2 , φ, ṙ1 , ṙ2 , φ̇, t).
[5 points]
Now we have
L = 12 m ṙ12 + ṙ22 + r12 φ̇2 + r22 φ̇2 − 12 k (r1 − a)2 − 12 k (r2 − r1 − a)2 .

(22.78)
(e) For the system described in part (d), find expressions for all conserved quantities.
[5 points]
There are two conserved quantities. One is pφ , owing to the fact the φ is cyclic in the
Lagrangian. I.e. φ → φ + ζ is a continuous one-parameter coordinate transformation which
leaves L invariant. We have
∂L
= m r12 + r22 φ̇ .

pφ = (22.79)
∂ φ̇
The second conserved quantity is the Hamiltonian, which is now H = T + U , since T is
homogeneous of degree 2 in the generalized velocities. Using conservation of momentum,
we can write
p2φ
1
ṙ12 ṙ22 + 12 k (r1 − a)2 + 12 k (r2 − r1 − a)2 .

H= 2m + + (22.80)
2m(r12 + r22 )
Once again, we can define an effective potential,
p2φ
Ueff (r1 , r2 ) = + 12 k (r1 − a)2 + 12 k (r2 − r1 − a)2 , (22.81)
2m(r12 + r22 )
which is different than the effective potential from part (b). However in both this case and
in part (b), we have that the radial coordinates obey the equations of motion
∂Ueff
mr̈j = − , (22.82)
∂rj
Figure 22.14: A mass point m rolls inside a hoop of mass M and radius R which rolls
without slipping on a horizontal surface.
for j = 1, 2. Note that this equation of motion follows directly from Ḣ = 0.
[2] A point mass m slides inside a hoop of radius R and mass M , which itself rolls without
slipping on a horizontal surface, as depicted in fig. 22.14.
Choose as general coordinates (X, φ, r), where X is the horizontal location of the center of
the hoop, φ is the angle the mass m makes with respect to the vertical (φ = 0 at the bottom
of the hoop), and r is the distance of the mass m from the center of the hoop. Since the
mass m slides inside the hoop, there is a constraint:
G(X, φ, r) = r − R = 0 .
Nota bene: The kinetic energy of the moving hoop, including translational and rotational
components (but not including the mass m), is Thoop = M Ẋ 2 (i.e. twice the translational
contribution alone).
(a) Find the Lagrangian L(X, φ, r, Ẋ, φ̇, ṙ, t).

[5 points]
The Cartesian coordinates and velocities of the mass m are
x = X + r sin φ ẋ = Ẋ + ṙ sin φ + rφ̇ cos φ (22.83)

y = R − r cos φ ẏ = −ṙ cos φ + rφ̇ sin φ (22.84)
The Lagrangian is then

T U
z }| { z }| {
L =(M + 21 m)Ẋ 2 + 12 m(ṙ2 + r2 φ̇2 ) + mẊ(ṙ sin φ + rφ̇ cos φ) − mg(R − r cos φ) (22.85)
Note that we are not allowed to substitute r = R and hence ṙ = 0 in the Lagrangian prior
to obtaining the equations of motion. Only after the generalized momenta and forces are
computed are we allowed to do so.
(b) Find all the generalized momenta pσ , the generalized forces Fσ , and the forces of
constraint Qσ .
[10 points]
The generalized momenta are

∂L
pr = = mṙ + mẊ sin φ (22.86)
∂ ṙ
∂L
pX = = (2M + m)Ẋ + mṙ sin φ + mrφ̇ cos φ (22.87)
∂ Ẋ
∂L
pφ = = mr2 φ̇ + mrẊ cos φ (22.88)
∂ φ̇
The generalized forces and the forces of constraint are
∂L ∂G
Fr = = mrφ̇2 + mẊ φ̇ cos φ + mg cos φ Qr = λ =λ (22.89)
∂r ∂r
∂L ∂G
FX = =0 QX = λ =0 (22.90)
∂X ∂X
∂L ∂G
Fφ = = mẊ ṙ cos φ − mẊ φ̇ sin φ − mgr sin φ Qφ = λ =0. (22.91)
∂φ ∂φ
ṗσ = Fσ + Qσ . (22.92)
At this point, we can legitimately invoke the constraint r = R and set ṙ = 0 in all the pσ
and Fσ .
(c) Derive expressions for all conserved quantities.

[5 points]
There are two conserved quantities, which each derive from continuous invariances of the
Lagrangian which respect the constraint. The first is the total momentum pX :
FX = 0 =⇒ P ≡ pX = constant . (22.93)
The second conserved quantity is the Hamiltonian, which in this problem turns out to be
the total energy E = T + U . Incidentally, we can use conservation of P to write the energy
in terms of the variable φ alone. From
P mR cos φ
Ẋ = − φ̇ , (22.94)
2M + m 2M + m
we obtain
E = 12 (2M + m)Ẋ 2 + 12 mR2 φ̇2 + mRẊ φ̇ cos φ + mgR(1 − cos φ)

2
αP 2

2 1 + α sin φ
= 1
+ mR φ̇2 + mgR(1 − cos φ) , (22.95)
2m(1 + α) 2 1+α
where we’ve defined the dimensionless ratio α ≡ m/2M . It is convenient to define the
quantity
1 + α sin2 φ

2
Ω ≡ φ̇2 + 2ω02 (1 − cos φ) , (22.96)
1+α
p
with ω0 ≡ g/R. Clearly Ω 2 is conserved, as it is linearly related to the energy E:
αP 2
E= + 1 mR2 Ω 2 . (22.97)
2m(1 + α) 2
(d) Derive a differential equation of motion involving the coordinate φ(t) alone. I.e. your
equation should not involve r, X, or the Lagrange multiplier λ.
[5 points]
From conservation of energy,
d(Ω 2 ) 1 + α sin2 φ

α sin φ cos φ
= 0 =⇒ φ̈ + φ̇2 + ω02 sin φ = 0 , (22.98)
dt 1+α 1+α
again with α = m/2M . Incidentally, one can use these results in eqns. 22.96 and 22.98 to
eliminate φ̇ and φ̈ in the expression for the constraint force, Qr = λ = ṗr − Fr . One finds
φ̇2 + ω02 cos φ

λ = −mR
1 + α sin2 φ
( )
mR ω02
2
Ω 2 1 2
=− (1 + α) − 4 sin ( 2 φ) + (1 + α sin φ) cos φ . (22.99)
(1 + α sin2 φ)2 ω02
This last equation can be used to determine the angle of detachment, where λ vanishes and
the mass m falls off the inside of the hoop. This is because the hoop can only supply a
repulsive normal force to the mass m. This was worked out in detail in my lecture notes on
constrained systems.
[3] Two objects of masses m1 and m2 move under the influence of a central potential
1/4
U = k r1 − r2 .
(a) Sketch the effective potential Ueff (r) and the phase curves for the radial motion. Identify
for which energies the motion is bounded.
[5 points]
The effective potential is

`2
Ueff (r) = + krn (22.100)
2µr2
with n = 14 . In sketching the effective potential, I have rendered it in dimensionless form,
Ueff (r) = E0 Ueff (r/r0 ) , (22.101)

Figure 22.15: The effective Ueff (r) = E0 Ueff (r/r0 ), where r0 and E0 are the radius and
energy of the circular orbit.
−1 1 1
where r0 = (`2 /nkµ)(n+2) and E0 = `2 /µr02 , which are obtained from the results

2 + n
of part (b). One then finds
n x−2 + 2 xn
Ueff (x) = . (22.102)
n+2
Although it is not obvious from the detailed sketch in fig. 22.15, the effective potential does
diverge, albeit slowly, for r → ∞. Clearly it also diverges for r → 0. Thus, the relative
coordinate motion is bounded for all energies; the allowed energies are E ≥ E0 .
(b) What is the radius r0 of the circular orbit? Is it stable or unstable? Why?
[5 points]
For the general power law potential U (r) = krn , with nk > 0 (attractive force), setting
0 (r ) = 0 yields
Ueff 0
`2
− + nkr0n−1 = 0 . (22.103)
µr03
Thus,
1 4
`2 4`2

n+2 9
r0 = = . (22.104)
nkµ kµ
The orbit r(t) = r0 is stable because the effective potential has a local minimum at r = r0 ,
00 (r ) > 0. This is obvious from inspection of the graph of U (r) but can also be
i.e. Ueff 0 eff
computed explicitly:
00 3`2
Ueff (r0 ) = + n(n − 1)kr0n
µr04
`2
= (n + 2) 4 . (22.105)
µr0
00 (r ) > 0.
Thus, provided n > −2 we have Ueff 0
(c) For small perturbations about a circular orbit, the radial coordinate oscillates between
two values. Suppose we compare two systems, with `0 /` = 2, but µ0 = µ and k 0 = k. What
is the ratio ω 0 /ω of their frequencies of small radial oscillations?
[5 points]
0 (r), we expand r = r + η and find
From the radial coordinate equation µr̈ = −Ueff 0
00
µη̈ = −Ueff (r0 ) η + O(η 2 ) . (22.106)
The radial oscillation frequency is then

` 2 2 n n−2
ω = (n + 2)1/2 2 = (n + 2)1/2 n n+2 k n+2 µ− n+2 ` n+2 . (22.107)
µr0
The ` dependence is what is key here. Clearly

0 n−2
ω0 ` n+2
= . (22.108)
ω `
In our case, with n = 41 , we have ω ∝ `−7/9 and thus
ω0
= 2−7/9 . (22.109)
ω
(d) Find the equation of the shape of the slightly perturbed circular orbit: r(φ) = r0 +η(φ).
That is, find η(φ). Sketch the shape of the orbit.
[5 points]
We have that η(φ) = η0 cos(βφ + δ0 ), with
ω µr2 √
β= = 0 ·ω = n+2 . (22.110)
φ̇ `
Figure 22.16: Radial oscillations with β = 32 .
With n = 14 , we have β = 32 . Thus, the radial coordinate makes three oscillations for every
two rotations. The situation is depicted in fig. 22.21.
(e) What value of n would result in a perturbed orbit shaped like that in fig. 22.22?
[5 points]
Figure 22.17: Closed precession in a central potential U (r) = krn .

√
Clearly β = n + 2 = 4, in order that η(φ) = η0 cos(βφ+δ0 ) executes four complete periods
over the interval φ ∈ [0, 2π]. This means n = 14.
[4] Two masses and three springs are arranged as shown in fig. 22.18. You may assume
that in equilibrium the springs are all unstretched with length a. The masses and spring
constants are simple multiples of fundamental values, viz.
m1 = m , m2 = 4m , k1 = k , k2 = 4k , k3 = 28k . (22.111)
(a) Find the Lagrangian.

[5 points]
Figure 22.18: Coupled masses and springs.
Choosing displacements relative to equilibrium as our generalized coordinates, we have
T = 12 m η̇12 + 2m η̇22 (22.112)
and
U = 21 k η12 + 2k (η2 − η1 )2 + 14k η22 . (22.113)
Thus,
L = T − U = 12 m η̇12 + 2m η̇22 − 12 k η12 − 2k (η2 − η1 )2 − 14k η22 . (22.114)
You are not required to find the equilibrium values of x1 and x2 . However, suppose all the
unstretched spring lengths are a and the total distance between the walls is L. Then, with
x1,2 being the location of the masses relative to the left wall, we have
U = 12 k1 (x1 − a)2 + 12 k2 (x2 − x1 − a)2 + 12 k3 (L − x2 − a)2 . (22.115)
Differentiating with respect to x1,2 then yields
∂U
= k1 (x1 − a) − k2 (x2 − x1 − a) (22.116)
∂x1
∂U
= k2 (x2 − x1 − a) − k3 (L − x2 − a) . (22.117)
∂x2
Setting these both to zero, we obtain
(k1 + k2 ) x1 − k2 x2 = (k1 − k2 ) a (22.118)

−k2 x1 + (k2 + k3 ) x2 = (k2 − k3 ) a + k3 L . (22.119)
Solving these two inhomogeneous coupled linear equations for x1,2 then yields the equilib-
rium positions. However, we don’t need to do this to solve the problem.

[5 points]
We have
∂2T

m 0
Tσσ0 = = (22.120)
∂ η̇σ ∂ η̇σ0 0 4m
and
∂2U

5k −4k
Vσσ0 = = . (22.121)
∂ησ ∂ησ0 −4k 32k
(c) Find the eigenfrequencies ω1 and ω2 .

[5 points]
We have
mω 2 − 5k

2 4k
Q(ω) ≡ ω T − V =
4k 4mω 2 − 32k

λ−5 4
=k , (22.122)
4 4λ − 32
p
where λ = ω 2 /ω02 , with ω0 = k/m. Setting det Q(ω) = 0 then yields
λ2 − 13λ + 36 = 0 , (22.123)
the roots of which are λ− = 4 and λ+ = 9. Thus, the eigenfrequencies are
ω− = 2 ω0 , ω+ = 3 ω0 . (22.124)
(d) Find the modal matrix Aσi .

[5 points]
To find the normal modes, we set

! !
(±)
λ± − 5 4 ψ1
(±) =0. (22.125)
4 4λ± − 32 ψ2
This yields two linearly dependent equations, from which we can determine only the ratios
(±) (±)
ψ2 /ψ1 . Plugging in for λ± , we find
! !
(−) (+)
ψ1 4 ψ1 1
(−) = C− , (+) = C+ . (22.126)
ψ2 1 ψ2 −1
(i) (j)
We then normalize by demanding ψσ Tσσ0 ψσ0 = δij . We can practically solve this by
inspection:
20m |C− |2 = 1 , 5m |C+ |2 = 1 . (22.127)
We may now write the modal matrix,

1 2 1
A= √ 1 . (22.128)
5m 2 −1
(e) Write down the most general solution for the motion of the system.
[5 points]
The most general solution is

!
η1 (t) 4 1
= B− cos(2ω0 t + ϕ− ) + B+ cos(3ω0 t + ϕ+ ) . (22.129)
η2 (t) 1 −1
Note that there are four constants of integration: B± and ϕ± .

22.7 W08 Physics 110B Midterm Exam
[1] Two identical semi-infinite lengths of string are joined at a point of mass m which
moves vertically along a thin wire, as depicted in fig. 22.21. The mass moves with friction
coefficient γ, i.e. its equation of motion is
mz̈ + γ ż = F , (22.130)
where z is the vertical displacement of the mass, and F is the force on the mass due to
the string segments on either side. In this problem, gravity is to be neglected. It may be
convenient to define K ≡ 2τ /mc2 and Q ≡ γ/mc.
Figure 22.19: A point mass m joining two semi-infinite lengths of identical string moves
vertically along a thin wire with friction coefficient γ.
(a) The general solution with an incident wave from the left is written
(
f (ct − x) + g(ct + x) (x < 0)
y(x, t) =
h(ct − x) (x > 0) .
Find two equations relating the functions f (ξ), g(ξ), and h(ξ).
[20 points]
The first equation is continuity at x = 0:
f (ξ) = g(ξ) + h(ξ)
where ξ = ct ranges over the real line [−∞, ∞]. The second equation comes from Newton’s
2nd law F = ma applied to the mass point:
m ÿ(0, t) + γ ẏ(0, t) = τ y 0 (0+ , t) − τ y 0 (0− , t) .

22.7. W08 PHYSICS 110B MIDTERM EXAM 539
Expressed in terms of the functions f (ξ), g(ξ), and h(ξ), and dividing through by mc2 ,
this gives
f 00 (ξ) + g 00 (ξ) + Q f 0 (ξ) + Q g 0 (ξ) = − 12 K h0 (ξ) + 12 K f 0 (ξ) − 12 K g 0 (ξ).
Integrating once, and invoking h = f + g, this second equation becomes
f 0 (ξ) + Q f (ξ) = −g 0 (ξ) − (K + Q) g(ξ)
(b) Solve for the reflection amplitude r(k) = ĝ(k)/fˆ(k) and the transmission amplitude
t(k) = ĥ(k)/fˆ(k). Recall that
Z∞ Z∞
dk ˆ
f (ξ) = f (k) eikξ ⇐⇒ f (k) = dξ f (ξ) e−ikξ ,
ˆ
2π
−∞ −∞
et cetera for the Fourier

2 transforms.
2 Also compute the sum of the reflection and transmission
coefficients, r(k) + t(k) . Show that this sum is always less than or equal to unity, and
interpret this fact.
[20 points]
Using d/dξ −→ ik, we have
(Q + ik) fˆ(k) = −(K + Q + ik) ĝ(k) . (22.131)
Therefore,
ĝ(k) Q + ik
r(k) = =− (22.132)
ˆ
f (k) Q + K + ik
To find the transmission amplitude, we invoke h(ξ) = f (ξ) + g(ξ), in which case
ĥ(k) K
t(k) = =− (22.133)
fˆ(k) Q + K + ik
The sum of reflection and transmission coefficients is
2 2 2
r(k)2 + t(k)2 = Q + K + k

(22.134)
(Q + K)2 + k 2
Clearly the RHS of this equation is bounded from above by unity, since both Q and K are
nonnegative.
(c) Find an expression in terms of the functions f , g, and h (and/or their derivatives) for
the rate Ė at which energy is lost by the string. Do this by evaluating the energy current
on either side of the point mass. Your expression should be an overall function of time t.
[10 points]
Recall the formulae for the energy density in a string,

2
E(x, t) = 1
2 µ ẏ 2 (x, t) + 12 τ y 0 (x, t) (22.135)
and
jE (x, t) = −τ ẏ(x, t) y 0 (x, t) . (22.136)
The energy continuity equation is ∂t E + ∂x jE = 0. Assuming jE (±∞, t) = 0, we have
Z0− Z∞
dE ∂E ∂E
= dx + dx
dt ∂t ∂t
−∞ 0+
= −jE (∞, t) + jE (0+ , t) + jE (−∞, t) − jE (0− , t) . (22.137)
Thus,
dE 2 2 2
= cτ g 0 (ct) + h0 (ct) − f 0 (ct) (22.138)
dt
Incidentally, if we integrate over all time, we obtain the total energy change in the string:
Z∞
2 2 2
∆E = τ dξ g 0 (ξ) + h0 (ξ) − f 0 (ξ)

−∞
Z∞
dk 2QK k 2 fˆ(k)2 .

= −τ (22.139)
2π (Q + K)2 + k 2
−∞
Note that the initial energy in the string, at time t = −∞, is

Z∞
dk 2 ˆ 2
E0 = τ k f (k) . (22.140)
2π
−∞
If the incident wave packet is very broad, say described by a Gaussian f (ξ) = A exp(−x2 /2σ 2 )
with σK 1 and σQ 1, then k 2 may be neglected in the denominator of eqn. 22.139,
in which case
2QK
∆E ≈ − E ≥ − 12 E0 . (22.141)
(Q + K)2 0
22.7. W08 PHYSICS 110B MIDTERM EXAM 541
[2] Consider a rectangular cube of density ρ and dimensions a × b × c, as depicted in fig.

22.22.
Figure 22.20: A rectangular cube of dimensions a × b × c. In part (c), a massless torsional

fiber is attached along the diagonal of one of the b × c faces.
(a) Compute the inertia tensor Iαβ along body-fixed principle axes, with the origin at the
center of mass.
[15 points]
We first compute Izz :
Za/2 Zb/2 Zc/2

= ρ dx dy dz x2 + y 2 ) = 1
M a2 + b2 ,
CM

Izz 12 (22.142)
−a/2 −b/2 −c/2
where M = ρ abc. Corresponding expressions hold for the other moments of inertia. Thus,
 2
b + c2

0 0
1
I CM = 12 M 0 a2 + c2 0  (22.143)
0 0 2
a +b 2
(b) Shifting the origin to the center of either of the b×c faces, and keeping the axes parallel,
compute the new inertia tensor.
[15 points]
We shift the origin by a distance d = − 12 a x̂ and use the parallel axis theorem,
Iαβ (d) = Iαβ (0) + M d2 δαβ − dα dβ ,

(22.144)
resulting in
 2
b + c2

0 0
I= 0 4a2 + c2 0  (22.145)
0 0 2
4a + b 2
(c) A massless torsional fiber is (masslessly) welded along the diagonal of either b × c face.
The potential energy in this fiber is given by U (θ) = 21 Y θ2 , where Y is a constant and θ is
the angle of rotation of the fiber. Neglecting gravity, find an expression for the oscillation
frequency of the system.
[20 points]
Let θ be the twisting angle of the fiber. The kinetic energy in the fiber is
1
T = 2 Iαβ ωα ωβ
1 2
= 2 nα Iαβ nβ θ̇ , (22.146)
where
b ŷ c ẑ
n̂ = √ +√ . (22.147)
b2 + c2 b2 + c2
We then find
b2 c2
Iaxis ≡ nα Iαβ nβ = 13 M a2 + 16 M . (22.148)
b2 + c2
q
The frequency of oscillation is then Ω = Y /Iaxis , or
s
6Y b2 + c2
Ω= · 2 2 (22.149)
M 2a b + c2 + b2 c2
22.8. W08 PHYSICS 110B FINAL EXAM 543
22.8 W08 Physics 110B Final Exam
[1] Consider a string with uniform mass density µ and tension τ . At the point x = 0, the
string is connected to a spring of force constant K, as shown in the figure below.
Figure 22.21: A string connected to a spring.
(a) The general solution with an incident wave from the left is written
(
f (ct − x) + g(ct + x) (x < 0)
y(x, t) =
h(ct − x) (x > 0) .
Find two equations relating the functions f (ξ), g(ξ), and h(ξ). [10 points]
SOLUTION : The first equation is continuity at x = 0:
f (ξ) + g(ξ) = h(ξ)
where ξ = ct ranges over the real line [−∞, ∞]. The second equation comes from Newton’s
2nd law F = ma applied to the mass point:
τ y 0 (0+ , t) − τ y 0 (0− , t) − K y(0, t) = 0 ,
or
− τ h0 (ξ) + τ f 0 (ξ) − τ g 0 (ξ) − K f (ξ) + g(ξ) = 0

(b) Solve for the reflection amplitude r(k) = ĝ(k)/fˆ(k) and the transmission amplitude
t(k) = ĥ(k)/fˆ(k). Recall that
Z∞ Z∞
dk ˆ
f (ξ) = f (k) eikξ ⇐⇒ fˆ(k) = dξ f (ξ) e−ikξ ,
2π
−∞ −∞
et cetera for the Fourier transforms.

2 Also
2 compute the sum of the reflection and
transmission coefficients, r(k) + t(k) . [10 points]
SOLUTION : Taking the Fourier transform of the two equations from part (a), we have
fˆ(k) + ĝ(k) = ĥ(k)

ˆ iτ k ˆ
f (k) + ĝ(k) = f (k) − ĝ(k) − ĥ(k) .
K
Solving for ĝ(k) and ĥ(k) in terms of fˆ(k), we find
ĝ(k) = r(k) fˆ(k) , ĥ(k) = t(k) fˆ(k)
where the reflection coefficient r(k) and the transmission coefficient t(k) are given by
K 2iτ k
r(k) = − , t(k) =
K + 2iτ k K + 2iτ k
Note that
r(k)2 + t(k)2 = 1

which says that the energy flux is conserved.
(c) For the Lagrangian density

2 2 4
1 ∂y 1 ∂y 1 ∂y
L= 2µ − 2τ − 4γ ,
∂t ∂x ∂x
find the Euler-Lagrange equations of motion. [7 points]
SOLUTION : For a Lagrangian density L(y, ẏ, y 0 ), the Euler-Lagrange equations are

∂L ∂ ∂L ∂ ∂L
= + .
∂y ∂t ∂ ẏ ∂x ∂y 0
Thus, the wave equation for this system is
2
µ ÿ = τ y 00 + 3γ y 0 y 00
(d) For the Lagrangian density

2 2 2
∂ 2y

1 ∂y 1 2 1 ∂y 1
L= 2µ − 2α y − 2τ − 4β ,
∂t ∂x ∂x2
find the Euler-Lagrange equations of motion. [7 points]
SOLUTION : For a Lagrangian density L(y, ẏ, y 0 , y 00 ), the Euler-Lagrange equations are
∂ 2 ∂L

∂L ∂ ∂L ∂ ∂L
= + − 2 .
∂y ∂t ∂ ẏ ∂x ∂y 0 ∂x ∂y 00
The last term arises upon integrating by parts twice in the integrand of the variation of the
action δS. Thus, the wave equation for this system is
µ ÿ = −α y + τ y 00 − β y 0000
[2] Consider single species population dynamics governed by the differential equation
dN N2 HN
= γN − − ,
dt K N +L
where γ, K, L, and H are constants.
(a) Show that by rescaling N and t that the above ODE is equivalent to
du hu
= r u − u2 − .
ds u+1
Give the definitions of u, s, r, and h. [5 points]
SOLUTION : From the denominator u + 1 in the last term of the scaled equation, we see that
we need to define N = Lu. We then write t = τ s, and substituting into the original ODE
yields
L du L2 2 Hu
= γLu − u − .
τ ds K u+1
Multiplying through by τ /L then gives
du Lτ 2 τ H u
= γτ u − u − .
ds K L u+1
We set the coefficient of the second term on the RHS equal to −1 to obtain the desired
form. Thus, τ = K/L and
N Lt γK KH
u= , s= , r= , h=
L K L L2
(b) Find and solve the equation for all fixed points u∗ (r, h). [10 points]
SOLUTION : In order for u to be a fixed point, we need u̇ = 0, which requires

h
u r−u− =0
u+1
One solution is always u∗ = 0 . The other roots are governed by the quadratic equation
(u − r)(u + 1) + h = 0 ,
with roots at

u∗ =
p
1
2 r−1± (r + 1)2 − 4h
Figure 22.22: Bifurcation curves for the equation u̇ = ru − u2 − hu/(u + 1). Red curve:
hSN (r) = 41 (r + 1)2 , corresponding to saddle-node bifurcation. Blue curve: hT (r) = r,
corresponding to transcritical bifurcation.
(c) Sketch the upper right quadrant of the (r, h) plane. Show that there are four distinct
regions:
Region I : 3 real fixed points (two negative)

Region II : 3 real fixed points (one positive, one negative)
Region III : 3 real fixed points (two positive)
Region IV : 1 real fixed point
Find the equations for the boundaries of these regions. These boundaries are the
locations of bifurcations. Classify the bifurcations. (Note that negative values of u
are unphysical in the context of population dynamics, but are legitimate from a purely
mathematical standpoint.) [10 points]
SOLUTION : From the quadratic equation for the non-zero roots, we see the discriminant
vanishes for h = 14 (r + 1)2 . For h > 14 (r + 1)2 , the discriminant is negative, and there is
one real root at u∗ = 0. Thus, the curve hSN (r) = 14 (r + 1)2 corresponds to a curve of
saddle-node bifurcations. Clearly the largest value of u∗ must be a stable node, because
for large u the −u2 dominates on the RHS of u̇ = f (u). In cases where there are three
fixed points, the middle one must be unstable, and the smallest stable. There is another
bifurcation, which occurs when the root at u∗ = 0 is degenerate. This occurs at
p
r − 1 = (r + 1)2 − 4h =⇒ h=r .
Figure 22.23: Examples of phase flows for the equation u̇ = ru − u2 − hu/(u + 1). (a) r = 1,
h = 0.22 (region I) ; (b) r = 1, h = 0.5 (region II) ; (c) r = 3, h = 3.8 (region III) ; (d)
r = 1, h = 1.5 (region IV).
This defines the curve for transcritical bifurcations: hT (r) = r . Note that hT (r) ≤ hSN (r),
since hSN (r) − hT (r) = 14 (r − 1)2 ≥ 0. For h < r, one root is positive and one negative,
corresponding to region II.
The (r, h) control parameter space is depicted in fig. 22.22, with the regions I through IV
bounded by sections of the bifurcation curves, as shown.
(d) Sketch the phase flow for each of the regions I through IV. [8 points]
SOLUTION : See fig. 22.23.

[3] Two brief relativity problems:
(a) A mirror lying in the (x, y) plane moves in the ẑ direction with speed u. A monochro-
matic ray of light making an angle θ with respect to the ẑ axis in the laboratory
frame reflects off the moving mirror. Find (i) the angle of reflection, measured in the
laboratory frame, and (ii) the frequency of the reflected light. [17 points]
SOLUTION : The reflection is simplest to consider in the frame of the mirror, where p̃z → −p̃z
upon reflection. In the laboratory frame, the 4-momentum of a photon in the beam is
P µ = E , 0 , E sin θ , E cos θ ,

where, without loss of generality, we have taken the light ray to lie in the (y, z) plane, and
where we are taking c = 1. Lorentz transforming to the frame of the mirror, we have
P̃ µ = γE(1 − u cos θ) , 0 , E sin θ , γE(−u + cos θ) .

which follows from the general Lorentz boost of a 4-vector Qµ ,
Q̃0 = γQ0 − γuQk

Q̃k = −γuQ0 + γQk
Q̃⊥ = Q⊥ ,
where frame K̃ moves with velocity u with respect to frame K.
Upon reflection, we reverse the sign of P̃ 3 in the frame of the mirror:
P̃ 0µ = γE(1 − u cos θ) , 0 , E sin θ , γE(u − cos θ) .

Transforming this back to the laboratory frame yields
E 0 = P 00 = γ 2 E (1 − u cos θ) + γ 2 E u (u − cos θ)
= γ 2 E 1 − 2u cos θ + u2

P 01 = 0
P 02 = E sin θ
P 03 = γ 2 E u (1 − u cos θ) + γ 2 E (u − cos θ)
= −γ 2 E (1 + u2 ) cos θ − 2u

Thus, the angle of reflection is

03
0
P (1 + u2 ) cos θ − 2u
cos θ = 00 =
P 1 − 2u cos θ + u2
and the reflected photon frequency is ν 0 = E 0 /h, where
1 − 2u cos θ + u2

0
E = E
1 − u2
(b) Consider the reaction π + + n → K+ + Λ0 . What is the threshold kinetic energy of the
pion to create kaon at an angle of 90◦ in the rest frame of the neutron? Express your
answer in terms of the masses mπ , mn , mK , and mΛ . [16 points]
SOLUTION : We have conservation of 4-momentum, giving
Pπµ + Pnµ = PKµ + PΛµ .
Thus,
PΛ2 = (Eπ + En − EK )2 − (Pπ + Pn − PK )2
= (Eπ2 − Pπ2 ) + (En2 − Pn2 ) + (EK

2
− PK2 )
+ 2Eπ En − 2Eπ EK − 2En EK − 2Pπ · Pn + 2Pπ · PK + 2Pn · PK
= EΛ2 − PΛ2 = m2Λ .
Now in the laboratory frame the neutron is at rest, so
Pnµ = (mn , 0) .
Thus, Pπ · Pn = Pn · PK = 0. We are also told that the pion and the kaon make an angle
of 90◦ in the laboratory frame, hence Pπ · PK = 0. And of course for each particle we have
E 2 − P 2 = m2 . Thus, we have
m2Λ = m2π + m2n + m2K − 2mn EK + 2(mn − EK ) Eπ ,
or, solving for Eπ ,

m2Λ − m2π − m2n − m2K + 2mn EK
Eπ = .
2(mn − EK )
The threshold pion energy is the minimum value of Eπ , which must occur when EK takes
its minimum allowed value, EK = mK . Thus,
m2Λ − m2π − m2n − m2K + 2mn mK

Tπ = Eπ − mπ ≥ − mπ
2(mn − mK )
[4] Sketch what a bletch might look like. [10,000 quatloos extra credit]
[-50 points if it looks like your professor]
Figure 22.24: The putrid bletch, from the (underwater) Jkroo forest, on planet Barney.

Lecture Notes On Classical Mechanics

Enviado por

Dados do documento

Título original

Direitos autorais

Formatos disponíveis

Compartilhar este documento

Compartilhar ou incorporar documento

Opções de compartilhamento

Você considera este documento útil?

Este conteúdo é inapropriado?

Direitos autorais:

Formatos disponíveis

Lecture Notes On Classical Mechanics

Enviado por

Direitos autorais:

Formatos disponíveis

Lecture Notes on Classical Mechanics

Version 0.1 – Not for Distribution!

0.1 Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xvi

0.2 Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xvii

1.1 Introduction and Review . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

1.1.1 Newton’s laws of motion . . . . . . . . . . . . . . . . . . . . . . . . 2

1.1.2 Aside : inertial vs. gravitational mass . . . . . . . . . . . . . . . . . 3

1.2 Examples of Motion in One Dimension . . . . . . . . . . . . . . . . . . . . 4

1.2.1 Uniform force . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

1.2.2 Uniform force with linear frictional damping . . . . . . . . . . . . . 5

1.2.3 Uniform force with quadratic frictional damping . . . . . . . . . . . 6

1.2.4 Crossed electric and magnetic fields . . . . . . . . . . . . . . . . . . 7

1.3 Pause for Reflection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

2.1 Work-Energy Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

2.2 Conservative and Nonconservative Forces . . . . . . . . . . . . . . . . . . . 12

2.2.1 Example : integrating F = −∇U . . . . . . . . . . . . . . . . . . . 14

2.3 Conservative Forces in Many Particle Systems . . . . . . . . . . . . . . . . 16

2.4 Linear and Angular Momentum . . . . . . . . . . . . . . . . . . . . . . . . 16

2.5 Scaling of Solutions for Homogeneous Potentials . . . . . . . . . . . . . . . 18

2.5.1 Euler’s theorem for homogeneous functions . . . . . . . . . . . . . . 18

2.5.2 Scaled equations of motion . . . . . . . . . . . . . . . . . . . . . . . 18

2.6 Appendix I : Curvilinear Orthogonal Coordinates . . . . . . . . . . . . . . 20

2.6.1 Example : spherical coordinates . . . . . . . . . . . . . . . . . . . . 21

2.6.2 Vector calculus : grad, div, curl . . . . . . . . . . . . . . . . . . . . 21

2.6.3 Common curvilinear orthogonal systems . . . . . . . . . . . . . . . 23

3 One-Dimensional Conservative Systems 27

3.1 Description as a Dynamical System . . . . . . . . . . . . . . . . . . . . . . 27

3.1.1 Example : harmonic oscillator . . . . . . . . . . . . . . . . . . . . . 28

3.2 One-Dimensional Mechanics as a Dynamical System . . . . . . . . . . . . . 29

3.2.1 Sketching phase curves . . . . . . . . . . . . . . . . . . . . . . . . . 29

3.3 Fixed Points and their Vicinity . . . . . . . . . . . . . . . . . . . . . . . . . 30

3.3.1 Linearized dynamics in the vicinity of a fixed point . . . . . . . . . 31

3.4 Examples of Conservative One-Dimensional Systems . . . . . . . . . . . . . 33

3.4.1 Harmonic oscillator . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

3.4.3 Other potentials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

4.1 Damped Harmonic Oscillator . . . . . . . . . . . . . . . . . . . . . . . . . . 41

4.1.1 Classification of damped harmonic motion . . . . . . . . . . . . . . 42

4.1.2 Remarks on the case of critical damping . . . . . . . . . . . . . . . 44

4.2 Damped Harmonic Oscillator with Forcing . . . . . . . . . . . . . . . . . . 44

4.2.1 Resonant forcing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47

4.2.2 R-L-C circuits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48

4.3 General solution by Green’s function method . . . . . . . . . . . . . . . . . 52

4.4 General Linear Autonomous Inhomogeneous ODEs . . . . . . . . . . . . . . 53

4.5 Kramers-Krönig Relations (advanced material) . . . . . . . . . . . . . . . . 57

5.1 Snell’s Law . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59

5.2 Functions and Functionals . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61

5.3 Examples from the Calculus of Variations . . . . . . . . . . . . . . . . . . . 63

5.3.1 Example 1 : minimal surface of revolution . . . . . . . . . . . . . . 63

5.3.2 Example 2 : geodesic on a surface of revolution . . . . . . . . . . . 66

5.3.3 Example 3 : brachistochrone . . . . . . . . . . . . . . . . . . . . . . 67

5.3.4 Ocean waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68

5.4 Appendix : More on Functionals . . . . . . . . . . . . . . . . . . . . . . . . 70

6.1 Generalized Coordinates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77

6.2 Hamilton’s Principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78

6.2.1 Momentum conservation . . . . . . . . . . . . . . . . . . . . . . . . 78

6.2.2 Invariance of the equations of motion . . . . . . . . . . . . . . . . . 78

6.2.3 Remarks on the order of the equations of motion . . . . . . . . . . 79

6.2.4 Lagrangian for a free particle . . . . . . . . . . . . . . . . . . . . . . 79

6.3 Remarks on the Choice of Generalized Coordinates . . . . . . . . . . . . . . 80

6.4 How to Solve Mechanics Problems . . . . . . . . . . . . . . . . . . . . . . . 81

6.5.1 One-dimensional motion . . . . . . . . . . . . . . . . . . . . . . . . 81

6.5.2 Central force in two dimensions . . . . . . . . . . . . . . . . . . . . 82

6.5.3 A sliding point mass on a sliding wedge . . . . . . . . . . . . . . . . 83