Quantum Mechanics

Mathematical Structure
and
Physical Structure

Part 2
John R. Boccio
Emeritus Professor of Physics
Swarthmore College

Copyright c 2014 by Professor John Boccio

Do not reproduce or distribute copies of this book without permission.

June 25, 2014

Contents

11 Time-Dependent Perturbation Theory 921

11.1 Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 921
11.1.1 What is the physical meaning of this result? . . . . . . . . 927
11.2 Atomic Radiation and Selection Rules . . . . . . . . . . . . . . . 940
11.2.1 The Electric Dipole Approximation . . . . . . . . . . . . . 941
11.2.2 Induced Emission and Absorption . . . . . . . . . . . . . 943
11.3 A Real Physical Process - Ionization . . . . . . . . . . . . . . . . 946
11.3.1 Evaluation of the Matrix Element . . . . . . . . . . . . . 948
11.4 Adiabatic and Sudden Approximations . . . . . . . . . . . . . . . 950
11.5 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 958
11.5.1 Square Well Perturbed by an Electric Field . . . . . . . . 958
11.5.2 3-Dimensional Oscillator in an electric field . . . . . . . . 959
11.5.3 Hydrogen in decaying potential . . . . . . . . . . . . . . . 959
11.5.4 2 spins in a time-dependent potential . . . . . . . . . . . 959
11.5.5 A Variational Calculation of the Deuteron Ground State
Energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 960
11.5.6 Sudden Change - Dont Sneeze . . . . . . . . . . . . . . . 960
11.5.7 Another Sudden Change - Cutting the spring . . . . . . . 960
11.5.8 Another perturbed oscillator . . . . . . . . . . . . . . . . 961
11.5.9 Nuclear Decay . . . . . . . . . . . . . . . . . . . . . . . . 961
11.5.10 Time Evolution Operator . . . . . . . . . . . . . . . . . . 961
11.5.11 Two-Level System . . . . . . . . . . . . . . . . . . . . . . 962
11.5.12 Instantaneous Force . . . . . . . . . . . . . . . . . . . . . 962
11.5.13 Hydrogen beam between parallel plates . . . . . . . . . . 962
11.5.14 Particle in a Delta Function and an Electric Field . . . . 962
11.5.15 Nasty time-dependent potential [complex integration needed]963
11.5.16 Natural Lifetime of Hydrogen . . . . . . . . . . . . . . . . 964
11.5.17 Oscillator in electric field . . . . . . . . . . . . . . . . . . 964
11.5.18 Spin Dependent Transitions . . . . . . . . . . . . . . . . . 965
11.5.19 The Driven Harmonic Oscillator . . . . . . . . . . . . . . 966
11.5.20 A Novel One-Dimensional Well . . . . . . . . . . . . . . . 966
11.5.21 The Sudden Approximation . . . . . . . . . . . . . . . . . 967
11.5.22 The Rabi Formula . . . . . . . . . . . . . . . . . . . . . . 967

i
ii CONTENTS

11.5.23 Rabi Frequencies in Cavity QED . . . . . . . . . . . . . . 968

12 Identical Particles 971

12.1 Theoretical ideas . . . . . . . . . . . . . . . . . . . . . . . . . . . 971
12.2 Bosons with Spin = 0 . . . . . . . . . . . . . . . . . . . . . . . . 976
12.3 Spin = 1/2 Fermions . . . . . . . . . . . . . . . . . . . . . . . . . 978
12.4 The N-Electron Atom . . . . . . . . . . . . . . . . . . . . . . . . 984
12.5 The Helium Atom . . . . . . . . . . . . . . . . . . . . . . . . . . 990
12.6 Multielectron Atoms . . . . . . . . . . . . . . . . . . . . . . . . . 998
12.6.1 Screening . . . . . . . . . . . . . . . . . . . . . . . . . . . 1000
12.6.2 Shell Structure . . . . . . . . . . . . . . . . . . . . . . . . 1001
12.7 Angular Momentum Coupling . . . . . . . . . . . . . . . . . . . . 1005
12.7.1 LS Coupling . . . . . . . . . . . . . . . . . . . . . . . . . 1006
12.7.2 Hunds Rules . . . . . . . . . . . . . . . . . . . . . . . . . 1013
12.7.3 JJ-Coupling . . . . . . . . . . . . . . . . . . . . . . . . . . 1014
12.8 Spherical Harmonics Addition Theorem . . . . . . . . . . . . . . 1018
12.8.1 Orbital Angular Momentum . . . . . . . . . . . . . . . . . 1018
12.8.2 The Addition Theorem . . . . . . . . . . . . . . . . . . . 1021
12.9 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1023
12.9.1 Two Bosons in a Well . . . . . . . . . . . . . . . . . . . . 1023
12.9.2 Two Fermions in a Well . . . . . . . . . . . . . . . . . . . 1024
12.9.3 Two spin1/2 particles . . . . . . . . . . . . . . . . . . . 1024
12.9.4 Hydrogen Atom Calculations . . . . . . . . . . . . . . . . 1025
12.9.5 Hunds rule . . . . . . . . . . . . . . . . . . . . . . . . . . 1026
12.9.6 Russell-Saunders Coupling in Multielectron Atoms . . . . 1026
12.9.7 Magnetic moments of proton and neutron . . . . . . . . . 1027
12.9.8 Particles in a 3-D harmonic potential . . . . . . . . . . . . 1028
12.9.9 2 interacting particles . . . . . . . . . . . . . . . . . . . . 1028
12.9.10 LS versus JJ coupling . . . . . . . . . . . . . . . . . . . . 1029
12.9.11 In a harmonic potential . . . . . . . . . . . . . . . . . . . 1029
12.9.12 2 particles interacting via delta function . . . . . . . . . . 1029
12.9.13 2 particles in a square well . . . . . . . . . . . . . . . . . 1030
12.9.14 2 particles interacting via a harmonic potential . . . . . . 1030
12.9.15 The Structure of helium . . . . . . . . . . . . . . . . . . . 1030

13 Scattering Theory and Molecular Physics 1033

13.1 Scattering Theory . . . . . . . . . . . . . . . . . . . . . . . . . . 1033
13.1.1 Greens Functions . . . . . . . . . . . . . . . . . . . . . . . 1035
13.1.2 Cross Sections . . . . . . . . . . . . . . . . . . . . . . . . 1041
13.1.3 Partial Waves . . . . . . . . . . . . . . . . . . . . . . . . . 1042
13.1.4 The Optical Theorem . . . . . . . . . . . . . . . . . . . . 1047
13.1.5 Born Approximation . . . . . . . . . . . . . . . . . . . . . 1047
13.1.6 A General Property of Phase Shifts . . . . . . . . . . . . . 1050
13.1.7 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . 1051
13.1.8 Born Approximation Examples . . . . . . . . . . . . . . . 1055
CONTENTS iii

13.2 Molecular Physics . . . . . . . . . . . . . . . . . . . . . . . . . . 1059

13.2.1 General Properties of Molecules . . . . . . . . . . . . . . . 1059
13.2.2 The Born-Oppenheimer approximation: A toy version . . 1065
13.2.3 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . 1073
13.2.4 Vibrational and Rotational Levels of a Molecule . . . . . 1081
13.3 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1083
13.3.1 S-Wave Phase Shift . . . . . . . . . . . . . . . . . . . . . 1083
13.3.2 Scattering Slow Particles . . . . . . . . . . . . . . . . . . 1083
13.3.3 Inverse square scattering . . . . . . . . . . . . . . . . . . . 1083
13.3.4 Ramsauer-Townsend Effect . . . . . . . . . . . . . . . . . 1084
13.3.5 Scattering from a dipole . . . . . . . . . . . . . . . . . . . 1084
13.3.6 Born Approximation Again . . . . . . . . . . . . . . . . . 1084
13.3.7 Translation invariant potential scattering . . . . . . . . . 1084
13.3.8 ` = 1 hard sphere scattering . . . . . . . . . . . . . . . . . 1085
13.3.9 Vibrational Energies in a Diatomic Molecule . . . . . . . 1085
13.3.10 Ammonia Molecule . . . . . . . . . . . . . . . . . . . . . . 1085
13.3.11 Ammonia molecule Redux . . . . . . . . . . . . . . . . . . 1086
13.3.12 Molecular Hamiltonian . . . . . . . . . . . . . . . . . . . . 1087
13.3.13 Potential Scattering from a 3D Potential Well . . . . . . . 1087
13.3.14 Scattering Electrons on Hydrogen . . . . . . . . . . . . . 1088
13.3.15 Greens Function . . . . . . . . . . . . . . . . . . . . . . . 1088
13.3.16 Scattering from a Hard Sphere . . . . . . . . . . . . . . . 1089
13.3.17 Scattering from a Potential Well . . . . . . . . . . . . . . 1089
13.3.18 Scattering from a Yukawa Potential . . . . . . . . . . . . 1089
13.3.19 Born approximation - Spin-Dependent Potential . . . . . 1089
13.3.20 Born approximation - Atomic Potential . . . . . . . . . . 1090
13.3.21 Lennard-Jones Potential . . . . . . . . . . . . . . . . . . . 1090
13.3.22 Covalent Bonds - Diatomic Hydrogen . . . . . . . . . . . 1091
13.3.23 Nucleus as sphere of charge - Scattering . . . . . . . . . . 1091

14 Some Examples of Quantum Systems 1093

14.1 Coherent and Squeezed States . . . . . . . . . . . . . . . . . . . . 1093
14.2 Electron in a circular wire . . . . . . . . . . . . . . . . . . . . . . 1098
14.3 Spin-Orbit Coupling in Complex Atoms . . . . . . . . . . . . . . 1102
14.4 Zeeman Effect in Complex Atoms . . . . . . . . . . . . . . . . . . 1105
14.4.1 Method #1: Plausibility Derivation . . . . . . . . . . . . 1106
14.4.2 Method #2: Full Formal Derivation . . . . . . . . . . . . 1106
14.5 Neutron Interferometry . . . . . . . . . . . . . . . . . . . . . . . 1107
14.5.1 Neutron Interferences . . . . . . . . . . . . . . . . . . . . 1109
14.5.2 The Gravitational Effect . . . . . . . . . . . . . . . . . . . 1110
14.6 The Penning Trap . . . . . . . . . . . . . . . . . . . . . . . . . . 1112
14.6.1 Motion of an Electron in a Penning Trap . . . . . . . . . 1112
14.6.2 The Transverse Motion . . . . . . . . . . . . . . . . . . . 1114
14.6.3 Measurement of Electron Anomalous Magnetic Moment . 1115
14.7 Schrodingers Cat . . . . . . . . . . . . . . . . . . . . . . . . . . . 1116
iv CONTENTS

14.7.1 Schrodingers Cat - a more detailed presentation . . . . . 1117

14.7.2 Construction of a Schrodinger-Cat State . . . . . . . . . . 1126
14.7.3 Quantum Superposition Versus Statistical Mixture . . . . 1127
14.7.4 The Fragility of a Quantum Superposition . . . . . . . . . 1130
14.8 The Quantum Eraser . . . . . . . . . . . . . . . . . . . . . . . . . 1133
14.8.1 Magnetic Resonance . . . . . . . . . . . . . . . . . . . . . 1133
14.8.2 Ramsey Fringes . . . . . . . . . . . . . . . . . . . . . . . . 1135
14.8.3 Detection of the Neutron Spin State . . . . . . . . . . . . 1139
14.8.4 The Actual Quantum Eraser . . . . . . . . . . . . . . . . 1141

15 States and Measurement 1145

15.1 State Preparation/Determination
Density Matrix Revisited . . . . . . . . . . . . . . . . . . . . . . 1145
15.1.1 State Preparation . . . . . . . . . . . . . . . . . . . . . . 1145
15.1.2 State Determination . . . . . . . . . . . . . . . . . . . . . 1150
15.1.3 Spin State (S = 1/2) . . . . . . . . . . . . . . . . . . . . . 1152
15.1.4 Spin State (S =1) . . . . . . . . . . . . . . . . . . . . . . 1153
15.1.5 Composite Systems . . . . . . . . . . . . . . . . . . . . . . 1157
15.2 Measurement and Interpretation of States . . . . . . . . . . . . . 1163
15.2.1 Example of Spin Measurement . . . . . . . . . . . . . . . 1163
15.2.2 A General Theorem of Measurement Theory . . . . . . . 1165
15.2.3 The Interpretation of the State Vector . . . . . . . . . . . 1167
15.2.4 The Measurement Theorem for General States . . . . . . 1170
15.2.5 Which Wave Function? . . . . . . . . . . . . . . . . . . . 1171
15.2.6 Spin Recombination Experiment . . . . . . . . . . . . . . 1174
15.2.7 Joint and Conditional Probabilities . . . . . . . . . . . . . 1180
15.2.8 A General Experiment and Coupling to the Environment 1208
15.3 Kochen-Specker Theorem . . . . . . . . . . . . . . . . . . . . . . 1212
15.3.1 Theorem (due to Kochen and Specker) . . . . . . . . . . . 1217
15.4 The Logic of Quantum Propositions . . . . . . . . . . . . . . . . 1226
15.4.1 The Meaning of True . . . . . . . . . . . . . . . . . . . . 1226
15.4.2 Is False the Same as Not True? . . . . . . . . . . . . 1230
15.5 Quantum Mechanical Picture of the World . . . . . . . . . . . . . 1237
15.5.1 The Model . . . . . . . . . . . . . . . . . . . . . . . . . . 1237
15.5.2 The Question . . . . . . . . . . . . . . . . . . . . . . . . . 1241
15.5.3 The Geometrical Lemma . . . . . . . . . . . . . . . . . . 1243
15.5.4 The Theorem . . . . . . . . . . . . . . . . . . . . . . . . . 1249
15.5.5 The Discrete Version . . . . . . . . . . . . . . . . . . . . . 1252
15.6 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1256
15.6.1 Measurements in a Stern-Gerlach Apparatus . . . . . . . 1256
15.6.2 Measurement in 2-Particle State . . . . . . . . . . . . . . 1257
15.6.3 Measurements on a 2 Spin-1/2 System . . . . . . . . . . . 1258
15.6.4 Measurement of a Spin-1/2 Particle . . . . . . . . . . . . 1259
15.6.5 Mixed States vs. Pure States and Interference . . . . . . . 1259
15.6.6 Which-path information, Entanglement, and Decoherence 1261
15.6.7 Livio and Oivil . . . . . . . . . . . . . . . . . . . . . . . . 1262
CONTENTS v

15.6.8 Measurements on Qubits . . . . . . . . . . . . . . . . . . 1263

15.6.9 To be entangled.... . . . . . . . . . . . . . . . . . . . . . . 1263
15.6.10 Alice, Bob and Charlie . . . . . . . . . . . . . . . . . . . . 1263

16 The EPR Argument and Bell Inequality 1265

16.1 Hidden variables and Bells Inequalities-1st Try . . . . . . . . . . 1265
16.1.1 The Electron Spin . . . . . . . . . . . . . . . . . . . . . . 1265
16.1.2 Correlations Between the Two Spins . . . . . . . . . . . . 1266
16.1.3 A Simple Hidden Variable Model . . . . . . . . . . . . . . 1269
16.2 Bells Theorem and Experimental Results . . . . . . . . . . . . . 1271
16.3 The EPR(Einstein-Podolsky-Rosen) Argument-Quick Overview . 1273
16.3.1 The Bell Inequality again . . . . . . . . . . . . . . . . . . 1275
16.4 EPR and Bell - The Details . . . . . . . . . . . . . . . . . . . . . 1278
16.4.1 Single-Photon Interference . . . . . . . . . . . . . . . . . . 1279
16.4.2 Basic Formalism . . . . . . . . . . . . . . . . . . . . . . . 1293
16.5 Inseparable Photons (the EPR Paradox) including some history . 1296
16.5.1 The Philosophical Stakes in the Debate . . . . . . . . . . 1296
16.5.2 From Como to Brussels (1927-30) . . . . . . . . . . . . . . 1297
16.5.3 From Brussels to the EPR Paradox (1930-35) . . . . . . . 1298
16.5.4 Elementary Introduction to the EPR Paradox . . . . . . . 1299
16.5.5 The EPR Paradox (1935-52) . . . . . . . . . . . . . . . . 1301
16.5.6 The BCHSH Inequality (1964) . . . . . . . . . . . . . . . 1302
16.5.7 BCHSH Inequality(Bells inequality in real world) . . . . 1305
16.5.8 The Beginnings of the Experiment at Orsay (1976) . . . . 1309
16.6 The Principle of Non-Separability . . . . . . . . . . . . . . . . . . 1316
16.7 An Example and a Solution - Bells Theorem with Photons . . . 1318
16.8 Non-Locality, EPR and Bell - a last time . . . . . . . . . . . . . . 1322
16.8.1 The Bell Inequalities . . . . . . . . . . . . . . . . . . . . . 1326
16.9 Bayesian Probability in QM . . . . . . . . . . . . . . . . . . . . . 1328
16.9.1 Using Bayesian Ideas in Analysis of Experiments . . . . . 1328
16.9.2 Simple Example . . . . . . . . . . . . . . . . . . . . . . . 1330
16.9.3 Simple Ideas . . . . . . . . . . . . . . . . . . . . . . . . . 1331
16.9.4 More about the Greenberger-Horne-Zeilinger(GHZ) State 1339
16.10Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1341
16.10.1 Bell Inequality with Stern-Gerlach . . . . . . . . . . . . . 1341
16.10.2 Bells Theorem with Photons . . . . . . . . . . . . . . . . 1342
16.10.3 Bells Theorem with Neutrons . . . . . . . . . . . . . . . . 1343
16.10.4 Greenberger-Horne-Zeilinger State . . . . . . . . . . . . . 1344

17 Path Integral Methods 1345

17.1 Historical Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . 1345
17.2 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1345
17.3 Path Integrals in Quantum Mechanics . . . . . . . . . . . . . . . 1346
17.3.1 General ideas . . . . . . . . . . . . . . . . . . . . . . . . . 1346
17.3.2 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . 1351
17.4 The Classical Limit . . . . . . . . . . . . . . . . . . . . . . . . . . 1354
vi CONTENTS

17.4.1 Derivation of the Principle of Least Action . . . . . . . 1354

17.5 Topology and Path Integrals in Quantum Mechanics . . . . . . . 1357
17.5.1 Aharonov-Bohm Effect . . . . . . . . . . . . . . . . . . . . 1357
17.6 Evaluation of Path Integrals - Details . . . . . . . . . . . . . . . 1360
17.6.1 Propagator for a Free Particle . . . . . . . . . . . . . . . . 1361
17.6.2 Propagator for a Quadratic Lagrangian . . . . . . . . . . 1368
17.6.3 Propagator of a Harmonic Oscillator . . . . . . . . . . . . 1372
17.7 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1374
17.7.1 Path integral for a charged particle moving on a plane in
the presence of a perpendicular magnetic field . . . . . . . 1374
17.7.2 Path integral for the three-dimensional harmonic oscillator 1375
17.7.3 Transitions in the forced one-dimensional oscillator . . . . 1375
17.7.4 Greens Function for a Free Particle . . . . . . . . . . . . 1376
17.7.5 Propagator for a Free Particle . . . . . . . . . . . . . . . . 1376

18 Solid State Physics 1377

18.1 Crystal Structure and Symmetry . . . . . . . . . . . . . . . . . . 1377
18.1.1 Symmetry of the Crystal System . . . . . . . . . . . . . . 1378
18.2 Bloch Theorem, the Reciprocal Lattice and Brillouin Zones . . . 1391
18.2.1 Translation Operators in Configuration Space . . . . . . . 1391
18.2.2 Derivation of Blochs Theorem . . . . . . . . . . . . . . . 1393
18.3 Free-Electron and Weak-Binding
Approximations in 1-Dimension . . . . . . . . . . . . . . . . . . . 1408
18.3.1 The Free-Electron Approximation . . . . . . . . . . . . . 1408
18.4 Introduction to the Weak-Binding Approximation . . . . . . . . . 1417
18.5 The Kronig-Penney Model . . . . . . . . . . . . . . . . . . . . . . 1427
18.5.1 Exact Analysis . . . . . . . . . . . . . . . . . . . . . . . . 1427
18.6 Free-Electron and Weak-Binding
Approximations in 2-Dimensions . . . . . . . . . . . . . . . . . . 1434
18.6.1 The Free-Electron Approximation . . . . . . . . . . . . . 1435
18.7 Born-Oppenheimer description of two atoms in a combined oscil-
lator and lattice trap . . . . . . . . . . . . . . . . . . . . . . . . . 1457
18.7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . 1457
18.7.2 Lattice Hamiltonian . . . . . . . . . . . . . . . . . . . . . 1458
18.7.3 Relative- and Center-of-Mass Quasi-Momenta . . . . . . . 1460
18.7.4 Born-Oppenheimer Separation . . . . . . . . . . . . . . . 1464
18.7.5 Exact and Born-Oppenheimer Approximate Solutions . . 1469
18.7.6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . 1476
18.8 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1476
18.8.1 Piecewise Constant Potential Energy
One Atom per Primitive Cell . . . . . . . . . . . . . . . . 1476
18.8.2 Piecewise Constant Potential Energy
Two Atoms per Primitive Cell . . . . . . . . . . . . . . . 1477
18.8.3 Free-Electron Energy Bands for a Crystal with a Primitive
Rectangular Bravais Lattice . . . . . . . . . . . . . . . . . 1478
CONTENTS vii

18.8.4 Weak-Binding Energy Bands for a Crystal with a Hexag-

onal Bravais Lattice . . . . . . . . . . . . . . . . . . . . . 1479
18.8.5 A Weak-Binding Calculation #1 . . . . . . . . . . . . . . 1480
18.8.6 Weak-Binding Calculations with Delta-Function Potential
Energies . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1480
18.8.7 Is the spectrum of the harmonic crystal example really thin?1481
18.8.8 Are the limits really noncommutative in the harmonic
crystal example? . . . . . . . . . . . . . . . . . . . . . . . 1481
18.8.9 The Bogoliubov transformation in the harmonic crystal
example. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1482

19 Second Quantization 1483

19.1 Identical Particles Revisited . . . . . . . . . . . . . . . . . . . . . 1483
19.1.1 Indistinguishability . . . . . . . . . . . . . . . . . . . . . . 1483
19.1.2 State Vectors . . . . . . . . . . . . . . . . . . . . . . . . . 1484
19.2 Occupation Number Space . . . . . . . . . . . . . . . . . . . . . . 1485
19.2.1 Creation and Annihilation Operators . . . . . . . . . . . . 1485
19.2.2 An Example . . . . . . . . . . . . . . . . . . . . . . . . . 1493
19.3 Second Quantized Operators . . . . . . . . . . . . . . . . . . . . 1499
19.3.1 Example - Gas of N Non-Interacting Spin 1/2 Fermions . 1503
19.4 Pair Correlation Function . . . . . . . . . . . . . . . . . . . . . . 1506
19.5 Hanbury-Brown and Twiss Experiment . . . . . . . . . . . . . . 1511
19.6 The Hamiltonian . . . . . . . . . . . . . . . . . . . . . . . . . . . 1515
19.6.1 An Example . . . . . . . . . . . . . . . . . . . . . . . . . 1516
19.7 Bogoliubov Transformation, Quasiparticles and Superfluidity . . 1518
19.8 Spontaneous Symmetry Breaking in Quantum Mechanics . . . . 1523
19.8.1 The Harmonic Crystal . . . . . . . . . . . . . . . . . . . . 1524
19.8.2 The Thin Spectrum . . . . . . . . . . . . . . . . . . . . . 1525
19.8.3 Subtleties . . . . . . . . . . . . . . . . . . . . . . . . . . . 1530
19.9 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1532
19.9.1 Bogoliubov Transformations . . . . . . . . . . . . . . . . . 1532
19.9.2 Weakly Interacting Bose gas in the Bogoliubov Approxi-
mation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1532
19.9.3 Problem 19.9.2 Continued . . . . . . . . . . . . . . . . . . 1533
19.9.4 Mean-Field Theory, Coherent States and the Gross-Pitaevkii
Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . 1534
19.9.5 Weakly Interacting Bose Gas . . . . . . . . . . . . . . . . 1535
19.9.6 Bose Coulomb Gas . . . . . . . . . . . . . . . . . . . . . . 1535
19.9.7 Pairing Theory of Superconductivity . . . . . . . . . . . . 1535
19.9.8 Second Quantization Stuff . . . . . . . . . . . . . . . . . . 1536
19.9.9 Second Quantized Operators . . . . . . . . . . . . . . . . 1538
19.9.10 Working out the details in Section 19.8 . . . . . . . . . . 1538
viii CONTENTS

20 Relativistic Wave Equations

Electromagnetic Radiation in Matter 1541
20.1 Spin 0 particles: Klein-Gordon Equation . . . . . . . . . . . . . . 1541
20.1.1 How to find correct form of relativistic wave equation? . . 1542
20.1.2 Negative Energy States and Antiparticles . . . . . . . . . 1546
20.2 Physics of the Klein-Gordon Equation . . . . . . . . . . . . . . . 1549
20.3 Free Particles as Wave Packets . . . . . . . . . . . . . . . . . . . 1555
20.4 Bound State Problems . . . . . . . . . . . . . . . . . . . . . . . . 1559
20.4.1 Nonrelativistic Limit . . . . . . . . . . . . . . . . . . . . . 1562
20.5 Relativistic Spin 1/2 Particles - The Dirac Equation . . . . . . . 1563
20.5.1 Lorentz Transformation of Spin . . . . . . . . . . . . . . . 1563
20.6 The Dirac Equation . . . . . . . . . . . . . . . . . . . . . . . . . 1572
20.6.1 Nonrelativistic Limit . . . . . . . . . . . . . . . . . . . . . 1573
20.6.2 Currents and Continuity Equations . . . . . . . . . . . . . 1575
20.6.3 Free Particle Solutions . . . . . . . . . . . . . . . . . . . . 1577
20.6.4 More About Currents . . . . . . . . . . . . . . . . . . . . 1581
20.6.5 Non-relativistic Limit . . . . . . . . . . . . . . . . . . . . 1584
20.6.6 The Dirac Hydrogen Atom . . . . . . . . . . . . . . . . . 1586
20.7 Electromagnetic Radiation and Matter . . . . . . . . . . . . . . . 1606
20.7.1 Interacting with the Classical Radiation Field . . . . . . . 1606
20.7.2 Relation to Gauge Invariance . . . . . . . . . . . . . . . . 1608
20.7.3 Interactions . . . . . . . . . . . . . . . . . . . . . . . . . . 1609
20.7.4 Induced Absorption and Emission . . . . . . . . . . . . . 1610
20.7.5 Quantized Radiation Field and Spontaneous Emission . . 1613
20.8 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1618
20.8.1 Dirac Spinors . . . . . . . . . . . . . . . . . . . . . . . . . 1618
20.8.2 Lorentz Transformations . . . . . . . . . . . . . . . . . . . 1618
20.8.3 Dirac Equation in 1 + 1 Dimensions . . . . . . . . . . . . 1618
20.8.4 Trace Identities . . . . . . . . . . . . . . . . . . . . . . . . 1619
20.8.5 Right- and Left-Handed Dirac Particles . . . . . . . . . . 1619
20.8.6 Gyromagnetic Ratio for the Electron . . . . . . . . . . . . 1620
20.8.7 Dirac Schrodinger . . . . . . . . . . . . . . . . . . . . . 1620
20.8.8 Positive and Negative Energy Solutions . . . . . . . . . . 1620
20.8.9 Helicity Operator . . . . . . . . . . . . . . . . . . . . . . . 1620
20.8.10 Non-Relativisitic Limit . . . . . . . . . . . . . . . . . . . . 1620
20.8.11 Gyromagnetic Ratio . . . . . . . . . . . . . . . . . . . . . 1621
20.8.12 Properties of 5 . . . . . . . . . . . . . . . . . . . . . . . . 1621
20.8.13 Lorentz and Parity Properties . . . . . . . . . . . . . . . . 1621
20.8.14 A Commutator . . . . . . . . . . . . . . . . . . . . . . . . 1621
20.8.15 Solutions of the Klein-Gordon equation . . . . . . . . . . 1621
20.8.16 Matrix Representation of Dirac Matrices . . . . . . . . . . 1622
20.8.17 Weyl Representation . . . . . . . . . . . . . . . . . . . . . 1622
20.8.18 Total Angular Momentum . . . . . . . . . . . . . . . . . . 1623
20.8.19 Dirac Free Particle . . . . . . . . . . . . . . . . . . . . . . 1623
CONTENTS ix

21 Interacting Fields and Feynman Diagrams 1625

21.1 Relativistic Quantum Mechanics of Free Particles . . . . . . . . . 1625
21.1.1 Hilbert Space . . . . . . . . . . . . . . . . . . . . . . . . . 1625
21.1.2 Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1632
21.1.3 Structure of Hilbert Space . . . . . . . . . . . . . . . . . . 1637
21.2 Interacting Fields - Part 1 . . . . . . . . . . . . . . . . . . . . . . 1638
21.2.1 Physical System . . . . . . . . . . . . . . . . . . . . . . . 1638
21.2.2 Hilbert Space . . . . . . . . . . . . . . . . . . . . . . . . . 1639
21.3 Interacting Fields - Part 2 . . . . . . . . . . . . . . . . . . . . . . 1644
21.3.1 Interpolating Fields . . . . . . . . . . . . . . . . . . . . . 1644
21.4 Interacting Fields - Part 3 . . . . . . . . . . . . . . . . . . . . . . 1653
21.4.1 Feynman Rules . . . . . . . . . . . . . . . . . . . . . . . . 1653
21.5 Interacting Fields - Part 4 . . . . . . . . . . . . . . . . . . . . . . 1660
21.5.1 Feynman Propagator . . . . . . . . . . . . . . . . . . . . . 1660
21.5.2 Scattering Cross Section . . . . . . . . . . . . . . . . . . . 1661
21.5.3 Lifetime . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1666

Index 1669
Chapter 11
Time-Dependent Perturbation Theory

11.1 Theory
Time-independent or stationary-state perturbation theory, as we developed ear-
lier, allows us to find approximations for the energy eigenvalues and eigenvectors
in complex physical systems that are not solvable in closed form and where we
could write H in two parts as

H = H0 + V (11.1)

For these perturbation methods to work, V must be weak and time-independent.

We now turn our attention to the case

H = H0 + Vt (11.2)

where Vt is weak and time-dependent.

Examples might be the decays of an atomic system by photon emission or the

ionization of an atom by shining light on it.
 E
 (0))
We assume that at some time t0 the system has evolved into the state t ,
 E
 (0
i.e., the state t satisfies the time evolution equation

 (0) E 
 (0)
E
i~ t = H0 t t t0 (11.3)
t
It is a solution of the time-dependent Schrodinger equation with no perturbing
interactions before t0 where

H = H0 t t0 (11.4)

At time t0 we turn on the interaction potential (or perturbation) so that

H = H0 + Vt t t0 (11.5)

921
The new state of the system then satisfies

i~ |t i = H |t i = (H0 + Vt ) |t i t t0 (11.6)
t
with the boundary condition(initial value)
 E
 (0)
|t i = t at t=t0 (11.7)

As we said, we assume that the full time-dependent Schrodinger equation cannot

be solved in closed form and so we look for approximate solutions.

We let Vt be a small perturbation, i.e., we assume there is a natural small

parameter in Vt (as we saw in time-independent perturbation theory) and we
make an expansion of the solution in powers of Vt or this small parameter.

Since the effect of H0 will be much greater than the effect of Vt , most of the time
dependence comes from H0 . If we could neglect Vt , then since H0 is independent
of time, we would have the simple time dependence
i
 E
 (0)
|t i = e ~ H0 t t (11.8)

Let us assume that this is still approximately true and remove this known time
dependence from the solution. This should remove the major portion of the
total time dependence from the problem. We do this by assuming a solution of
the form
i
|t i = e ~ H0 t |(t)i (11.9)
and then determining and solving the equation for the new state vector |(t)i.

Substituting this assumption in our original equation, the equation for |(t)i is
then given by
 i H0 t   i 
i~ e ~ |(t)i = (H0 + Vt ) e ~ H0 t |(t)i
t
i i
H0 e ~ H0 t |(t)i + i~e ~ H0 t |(t)i
t
i i
= H0 e ~ H0 t |(t)i + Vt e ~ H0 t |(t)i

i~ |(t)i = V (t) |(t)i (11.10)
t
where
i i
V (t) = e ~ H0 t Vt e ~ H0 t (11.11)
The substitution has removed H0 from the equation and changed the time de-
pendence of the perturbing potential. We are in the so-called interaction picture
or representation where both the state vectors and the operators depend on time
as we discussed earlier in Chapter 6.

922
We develop a formal solution by integrating this equation of motion for the
state vector to get
Zt Zt

i~ |(t0 )idt0 = V (t0 ) |(t0 )idt0
t0
t0 t0

= i~(|(t)i |(t0 )i) (11.12)

so that the formal solution is given by

Zt
1
|(t)i = |(t0 )i + V (t0 ) |(t0 )idt0 (11.13)
i~
t0

This is an integral equation for |(t)i. We solve it as a power series in Vt by the

method of iteration.

The 0th order approximation is found by neglecting the perturbing potential.

We get
|(t)i = |(t0 )i (11.14)
The 1st order approximation is obtained by inserting the 0th order approxi-
mation into the full equation. We get
Zt
1
|(t)i = |(t0 )i + V (t0 ) |(t0 )idt0
i~
t0
t

Z
1
= 1 + V (t0 )dt0 |(t0 )i (11.15)
i~
t0

The 2nd order approximation is obtained by inserting the 1st order approxi-
mation into the full equation. We get
Zt
1
|(t)i = |(t0 )i + V (t0 ) |(t0 )idt0
i~
t0
0
Zt Zt
1 0
+ dt dt00 V (t0 )V (t00 ) |(t0 )i (11.16)
(i~)2
t0 t0

Notice that in all subsequent iterations the operators V (t0 ),V (t00 ),. . ., etc, always
occur in order of increasing time from right to left.

We can write the general result as

|(t)i = U (t, t0 ) |(t0 )i (11.17)

923
where

Zt Zt1 tZ
n1

X 1
U (t, t0 ) = I + n
dt1 dt2 ........ dtn V (t1 )V (t2 ).......V (tn ) (11.18)
n=1
(i~)
t0 t0 t0

The complete, formal solution to the problem is then given by

i i
|t i = e ~ H0 t |(t)i = e ~ H0 t U (t, t0 ) |(t0 )i (11.19)

so that
i
e ~ H0 t U (t, t0 ) = the total time development operator (11.20)

Before developing the detailed techniques of time-dependent perturbation the-

ory, let us spend some time with the operator U (t, t0 ) and discuss some of its
properties.

We first introduce the idea of a time-ordered product of operators as follows.

The symbol
   
A(t)B(t1 )C(t2 ).......X(tn ) T A(t)B(t1 )C(t2 ).......X(tn ) (11.21)
+

means the product of the operators where the operators are written from right
to left in order of increasing times, i.e.,

A(t)B(t0 ) t t0
  
0
A(t)B(t ) = (11.22)
+ B(t0 )A(t) t0 t

Now, we have using the time-ordered product definition

2 t
Zt Z Zt
V (t0 )dt0 = V (t0 )dt0 V (t00 )dt00


t0 t0 t0 +
+
Zt Zt
= dt0 dt00 (V (t0 )V (t00 ))+
t0 t0
0 00
Zt Zt Zt Zt
= dt0 dt00 V (t0 )V (t00 ) + dt00 dt0 V (t00 )V (t0 )
t0 t0 t0 t0
0
Zt Zt
=2 dt0 dt00 V (t0 )V (t00 ) (11.23)
t0 t0

924
and in general
t n t
Z Z Zt Zt
V (t0 )dt0 = V (t1 )dt1 V (t2 )dt2 ..... V (tn )dtn
t0 + t0 t0 t0 +
Zt Zt Zt
= dt1 dt2 ...... dtn (V (t1 )V (t2 )......V (tn ))+
t0 t0 t0
Zt Zt Zt
= n! dt1 dt2 ...... dtn V (t1 )V (t2 )......V (tn ) (11.24)
t0 t0 t0

because there are n! possible orderings of the n terms involved. This last form
is identical to the expression for U (t, t0 ) and thus we have
t n
Z
X 1 1
U (t, t0 ) = n n!
V (t0 )dt0
n=0
(i~)
t0 +

Zt
i
= exp V (t0 )dt0 (11.25)
~
t0 +

The last expression is just a convenient shorthand for the infinite sum. In order
to verify that this is in fact a solution of

i~ |(t)i = V (t) |(t)i (11.26)
t
we must prove that

i~ U (t, t0 ) |(t0 )i = V (t)U (t, t0 ) |(t0 )i (11.27)
t

i~ U (t, t0 ) = V (t)U (t, t0 ) (11.28)
t
Substituting, we have

Zt
i
i~ U (t, t0 ) = i~ exp V (t0 )dt0
t t ~
t0 +

Zt
i
= V (t) exp V (t0 )dt0 (11.29)
~
t0 +

In the differentiation we do not have to worry about the non-commutation of the

operators inside the time-ordered product since the order is already specified.

925
Since t is certainly the latest time in the time-ordered product and therefore
all the other operators will be on the right of V (t) we can pull it outside the
time-ordered product and write

Zt
i
i~ U (t, t0 ) = V (t) exp V (t0 )dt0 = V (t)U (t, t0 )
t ~
t0 +

as required.

The most important question (really the only question) that is usually asked in
quantum mechanics is the following:

Suppose that the system is initially in an eigenstate

|ni of H0 , i.e., H0 |ni = n |ni. What is the probability
that the system will be observed, after the perturbation
has had time to act, in a different (and thus orthogonal)
eigenstate of H0 , say |mi?

Alternatively, the question is sometimes posed this way:

What is the probability that the interaction causes the

system to make a transition from the state |ni to the
state |mi?

The probability amplitude for observing the system in the state |mi at time t
is given by

i
hm | t i = hm| e ~ H0 t U (t, t0 ) |(t0 )i
i
= hm| e ~ H0 t U (t, t0 ) |ni (11.30)

where

|(t0 )i = |ni (11.31)

is the initial state.

Setting t0 = 0 for simplicity and using the 1st order approximation for U (t, 0)
and also using

i
 i +  i + i
hm| e ~ H0 t = e ~ H0 t |mi = e ~ m t |mi = hm| e ~ m t (11.32)

926
we get

Zt
1 i
hm | t i = e ~ m t dt0 hm| V (t0 ) |ni
i~
0
Zt
1 i m t i i
= e ~ dt0 hm| e ~ H0 t Vt0 e ~ H0 t |ni
i~
0
Zt
1 i m t i
= e ~ dt0 e ~ (m n )t hm| Vt0 |ni (11.33)
i~
0

The probability of the transition is then

 t 2
Z 
2 1 i
Pnm (t) = |hm | t i| = 2  dt0 e ~ (m n )t hm| Vt0 |ni
 
(11.34)
~  
0

The simplest example is when Vt is not a function of t, or Vt = V . We then

have
 2  2
t
Vt0 |ni Z
  
hm|
2  dt0 e ~i (m n )t 

Pnm (t) = |hm | t i| = 2
(11.35)
~ 



0

If we define = m n , then we have

2  1 e ~i t 2 
  !2
 2 sin t
2~
Pnm (t) = hm| Vt0 |ni   = hm| Vt0 |ni (11.36)
     
  /2

for the transition probability.

11.1.1 What is the physical meaning of this result?

We must be very careful when we use the words

the perturbation causes a transition

between eigenstates of H0

What this means physically is that the system has absorbed from the perturbing
field (or emitted to it) the energy difference = m n and therefore the system
has changed its energy.

Does the statement also mean that the state vector has changed from an initial
value |(0)i = |ni to a final value |(t)i = |mi?

927
We can get a better feeling for the correct answer to this question by deriving
the result in a different manner.

We have

i~ |t i = H |t i = (H0 + Vt ) |t i (11.37)
t
and
H0 |ni = n |ni (11.38)
As in our development of time-independent perturbation theory, we let

Vt = g Ut (11.39)

where g is a small parameter.

The set of eigenvectors {|ni} is a complete set and therefore we can use it as a
basis for the space and, in particular, we can write
i
X
|t i = an (t)e ~ n t |ni (11.40)
n

The reason for pulling out the phase factors will be clear shortly.

It is clear that if g = 0, then this is the correct general solution with

an (t) = an (0) = constant (11.41)

The phase factors we pulled out represent the time dependence due to H0 and
this is, by assumption, the major time dependence in the system.

If g is small we expect the time dependence of an (t), which is due to the per-
turbation to be weak or that
dan (t)
is small (11.42)
dt
It is in this sense that we can propose to use perturbation theory on the system.

Using the eigenbasis expansion we have

X  dan (t) 
i
i~ + n an (t) e ~ n t |ni
n
dt
X  i
= n an (t) + g Ut an (t) e ~ n t |ni (11.43)
n

Applying the linear functional hm| from the left and using the orthonormality
relation
hm | ni = mn (11.44)

928
we get

dam (t) X
i~ =g hm| Ut |ni eimn t an (t)
dt n
X
= Vmn (t)eimn t an (t) (11.45)
n

where
m n
mn = (11.46)
~
This is an exact equation. It implies that the time dependence of an (t) is due
entirely to Vt (because we explicitly extracted out the dependence due to H0 ).
This is the interaction picture that we had earlier.

Exactly Solvable 2-State Example

Consider a 2-state system with

eit
     
E1 0 0 V11 V12
H0 = , V (t) = = (11.47)
0 E2 eit 0 V21 V22

In the interaction picture, as derived above, we have

dcm (t) X X i
i~ = Vmn (t)eimn t cn (t) , |t i = cn (t)e ~ n t |ni (11.48)
dt n n

or
dc1 (t) E1 E2
i~ = ei[+ ~ ]t c2 (t) (11.49)
dt
dc2 (t) E1 E2
i~ = ei[ ~ ]t c1 (t) (11.50)
dt
i i
|t i = c1 (t)e ~ E1 t |1i + c2 (t)e ~ E2 t |2i (11.51)

We can write these equations as

ei[21 ]t
 
dC(t) i 0
= C(t) (11.52)
dt ~ ei[21 ]t 0

where  
c1 (t) E2 E1
C(t) = , 21 = (11.53)
c2 (t) ~
We can find an exact solution. With initial conditions c1 (0) = 1 and c2 (0) = 0
we get
2 2 2
|c1 (t)| = ~ (21 )2
2
sin2 t = 1 |c2 (t)| (11.54)
2
+ 4

929
where
2
2 ( 21 )
2 = + (11.55)
~2 4
A graph of these functions is shown in Figure 11.1 below.

Figure 11.1: Exact Solution

A straightforward calculation gives

2
2 ( 21 )
|c1 (t)|min = 2 4 2
(11.56)
( 21 ) + ~2

At resonance, = 21 , we have
2
= , |c1 (t)|min = 0 (11.57)
~
as shown in Figure 11.2 below.

Figure 11.2: At Resonance

The amplitude as a function of is shown in Figure 11.3 below.

930
 Figure 11.3: Amplitude versus

where = full width at half maximum = 4/~. The amplitude is peaked at

resonance and the width is proportional to (the strength of the perturbation).

This periodically forced 2state system is a basic problem - it demonstrates the

fundamental features of absorption and emission.

We now return to the full, general equations and look for a perturbation solution.
Now we assume (power series)

an (t) = a(0) (1) 2 (2)

n + gan + g an + ...... (11.58)

Substituting and arranging the terms in a power series in g we have

! !
(0) (1)
dan 0 dam X i
t (0)
g + i~ hm| Ut |ni e ~ mn
an g 1 + .....+
dt dt n
!
(r+1)
dam X i
i~ hm| Ut |ni e ~ mn t a(r)
n g r + ....... = 0
dt n

or looking at each order separately we have

(0)
dan
0th order =0
dt
(1)
dam X i
1st order i~ = hm| Ut |ni e ~ mn t a(0)
n
dt n
.......
.......
(r+1)
dam X i
(r + 1)st order i~ = hm| Ut |ni e ~ mn t a(r)
n
dt n
.......
(0)
Note that the coefficients an follow from the initial condition
X
|(0)i = a(0)
n |ni (11.59)
n

The solution proceeds as follows:

931
 initial condition a(0)
n

a(0) (1) st
n an using the 1 order equation
........
a(r) (r+1)
n an using the (r + 1)st order equation

Now consider the following example. We assume that

H = H0 t0 (11.60)

where
H0 |ni = n |ni (11.61)
and during the time interval 0 t T a perturbation Vt is applied to the
system and the an (t) change with time.

Finally, for t T the perturbation is turned off and an (t) = an (T ).

The probability that, as a result of the perturbation, the energy of the system
becomes r , is given by
 2
X 
2 ~i i t 2
|hr | t i| =  ai (t)e hr | ii = |ar (t)| (11.62)
 
 
i

and as t we get
2 2
|hr | t i| = |ar (T )| (11.63)
Now to 1st order we have
(1)
dar X i
i~ = hr| Ut |ni e ~ rn t a(0)
n (11.64)
dt n

If |(0)i = |ii, then (

1 n=i
a(0)
n = (11.65)
0 n 6= i
This gives
(1)
dar i
i~ = hr| Ut |ii e ~ ri t (11.66)
dt
Integrating we have
ZT
1 i
a(1)
r (T ) = hr| Ut |ii e ~ (r i )t dt (11.67)
i~
0

and
ar (T ) = a(0) (1)
r (T ) + gar (T ) (11.68)

932
which is identical to our earlier result as t .

Now let return to our question. Has the state changed also?

In the example we found that the perturbation produces a final state |t i for
t T which to 1st order is
i
X
|t i = an (t)e ~ n t |ni (11.69)
n

This is a coherent (definite relative phases) superposition of eigenvectors of H0 .

This is NOT a stationary state. Interference effects between the terms in the
sum are detectable. They do not, however, affect
2
|ar (T )| = probability that the energy changes to r (11.70)

Thus, the perturbation does not cause a jump from one stationary state |ii of
H0 to another |ri, but instead it produces a non-stationary state.

The conventional language of quantum mechanics produces this ambiguity be-

tween the two statements

the energy is r and the state is |ri

For the state

i
X
|t i = an (t)e ~ n t |ni
n

it is correct to say

the probability of the energy being r is |ar (T )|2

or

P rob(E = r | |i) = |ar (T )|2

The state, however, is |t i and NOT |ri.

An example
Suppose we perturb an oscillator with a decaying electric field of the form
t
Vt = qE0 xe t0 (11.71)

To 1st order, starting with the initial state |ni with energy

1
n = ~(n + ) (11.72)
2

933
we have
Zt
1
|(t)i = |ni + dt0 V (t) |ni (11.73)
i~
0

where
i i
V (t) = e ~ H0 t Vt e ~ H0 t (11.74)
We let n = 0 (the ground state) for this example. We then have

Zt
1 i t i
|(t)i = |0i + dt0 e ~ H0 t (qE0 x)e e ~ H0 t |0i (11.75)
i~
0

Using
i
e ~ H0 t |0i = ei 2 t |0i (11.76)
r
~
x |0i = |1i (11.77)
2m
i 3
e ~ H0 t |1i = ei 2 t |0i (11.78)

we get (letting t )
r Z
1 ~ 0 t
|(t)i = |0i + (qE0 ) dt0 eit |1i (11.79)
i~ 2m
0

and finally,
 2
q 2 E02
Z 
2 0 t 
 dt0 eit 

P01 = |h1 | (t)i| =
2m~ 



0
q 2 E02 2
= (11.80)
2m~ 2 2 + 1
We now return to the earlier general result (11.36) we derived for the probability,
namely,
2 sin t 2
!

2~
P0n (t) = hn| Vt0 |0i (11.81)
 
/2
In Figure 11.4 below we plot this function.

934
 Figure 11.4: Probability(0,n) versus Delta

The height of the central peak is proportional to t2 and the location of the first
zero is at
2~
= (11.82)
t
so that the width of the peak decreases as 1/t.

The formula implies that for very short times

P0n t2 for all n (11.83)

As t , however, the probability is largest for those states whose energy lies
under the sharp bump near = 0 or those states with whose energy lies under
the peak around 0 . Now the energy n 0 lies under the sharp bump when
2~
|| = |n 0 | < (11.84)
t
The area under the bump is proportional to t and the rest of the area oscillates
in time around zero. This latter feature means that if n 6= 0 , the transition
probability oscillates in time with a repetition time of
2~
(11.85)
|n 0 |
The case, where we are looking for a transition to a single state, is, thus, only
valid in perturbation theory for very small time t. Otherwise the condition that

935
the
P00 (t) 1 (11.86)
will not be true and perturbation theory breaks down. We also note that the
probability cannot grow larger than one or that, after a while, the higher-order
effects of the perturbation which we have neglected so far must become impor-
tant and prevent the probability from exceeding one.

The condition that tells us whether a transition probability to a state with

an energy appreciably different than the original energy is the same condition
in time-independent perturbation theory that tells whether the state vector
changes appreciably from the unperturbed state, namely
 
 hn| V |0i 
 << 1 (11.87)
 
 n 0 


Physically, a more interesting case occurs when the state |ni is one of a contin-
uum of energy states, or it lies in a group of very closely spaced levels.

In this case we ask a different experimental question, namely,

What is the probability that the system makes

a transition to a small group of states near
|ni (or has energy near n )?

Since the area under the bump near = 0 or n 0 is proportional to t, we

expect that the transition probability to a small group of states near 0 will
grow linearly with t and thus

P0n (t)
= transition rate = = constant as t (11.88)
t
Quantities that we measure are related to the transition rate and this result
says that these measurements will make sense.

Let us now carry out this derivation in detail.

To calculate this transition rate we must sum P0n over the group of final
states. We assume that | hb| Vt0 |0i |2 is relatively constant over the small group
of states near |ni (has a weak energy dependence).

We then have
h i 2
(n 0 )t
X  2 Z sin ~
P0n (t) = hn| Vt0 |0i dn (n ) h (11.89)
  i
(n 0 )
n
ingroup group 2

936
where

(n ) = number of states per unit energy

(n )dn = number of states in the interval dn

Now in the limit t

h i 2
(n 0 )t
sin ~ 2t
h i (0 n ) (11.90)
(n 0 ) ~
2

i.e., in general, for a sequence of functions

sin2 t
t () = (11.91)
2 t
we have that (
t
=0
t () = 1
(11.92)
2 t 6= 0

and
Z
lim dt ()F () = F (0) (11.93)
t

Therefore,
lim t () = () (11.94)
t

Using this result, we have

X
P0n (t) =t (11.95)
n

and thus
2  2
= transition rate = hn| V |0i (n )n =0 (11.96)

~
which is called Fermis Golden Rule.

We now consider a perturbation that depends explicitly on time. In particular,

suppose we have a harmonic perturbation of the form

V t it
Vt = V et cos t = + eit


e e (11.97)
2

and |(t0 )i = |0i, where we let t0 . The et factor is necessary to make

the mathematical operations valid in the limit. It is equivalent for small to
turning the perturbation on slowly. In the end we will let 0.

937
We have
Z0
1 i n t
hn | (t)i = hn | 0i + e ~ dt0 hn|V (t0 ) |0i
i~

" #
i(n 0 ~) ~t t
e t
e ei(n 0 +~) ~
= + hn| V |0i (11.98)
2 0 n + ~ + i~ 0 n ~ + i~

Thus, the probability of the energy being n at the time t is

2
|hn | (t)i|
1 1
( )
e2t  2
(0 n +~)2 +(~)2
+
(0 n ~)2 +(~)2
= hn| V |0i (11.99)

2it
e
4 +2Re (0 n +~+i~)( 0 n ~+i~)

The first term comes from the eit part of Vt (positive frequency) and the
second term comes from the eit part of Vt (negative frequency). The last term
represents interference effects.

Since
P0n (t) = 0n t (11.100)
we have
dP0n (t)
0n = (11.101)
dt
and thus

0n (11.102)
h i
1 1
2t 
e   2
(0 n +~)2 +(~)2 + (0 n ~)2 +(~)2
(1 cos 2t)
= hn| V |0i
 h i
2it
4 e
+2 sin 2t (0 n +~+i~)(
0 n ~+i~)

The sin t and cos t terms arise from the interference term. In the limit 0
and assuming that |ni is in the continuum part of spectrum, we have

1. the first two terms are not equal to zero only if n 0 = ~

2. the sin t and cos t terms average to zero if we assume that 0n is

dP0n (t)/dt averaged over a few cycles of Vt

which gives the result

 2
V |0i
 
2 hn| 
0n = [(n 0 ~) + (n 0 + ~)] (11.103)
~ 4
The positive and negative frequency parts act independently and the interference
averages to zero.

938
Thus, the eit part produced a E > 0 process (absorption), while the eit
part produced a E < 0 process (emission).

To enhance our understanding of time-dependent perturbation theory, we look

at a variation of this harmonic perturbation. Suppose we have a harmonic
perturbation of the form

Vt = V eit + V + eit (11.104)

which is only applied for a finite time interval 0 t T . If we start with energy
i in the state |ii, then at any time t T
(1)
af (t) = 1st order amplitude for the state |(t)i
to have energy f (be in state |f i ?) f 6= i

is given by

ZT ZT
(1) 1 i(f i )t 1
af (T ) = hf | V |ii e dt + hf | V + |ii ei(f i +)t dt (11.105)
i~ i~
0 0

and
 2
 (1)
af (T ) = probability that the final energy will be f (11.106)


As an example we consider spin resonance (we solved this problem exactly ear-
lier).

We consider a spin = 1/2 particle in a static magnetic field B0 (in the zdirection).
This says that the unperturbed Hamiltonian is
1
H0 = ~B0 z (11.107)
2
This operator has the eigenvectors and eigenvalues
   
1 0 1
|+i = , |i = ,  = ~B0 (11.108)
0 1 2

We now perturb the system with another magnetic field B1 , which is rotating
in the x y plane with angular velocity . This implies that
1
Vt = ~B1 (cos ti + sin tj)
2
eit
 
1 1 0
= ~B1 [x cos t + y sin t] = ~B1
2 2 eit 0
1  
= ~B1 V eit + V + eit (11.109)
2

939
where    
0 1 + 0 0
V = , V = (11.110)
0 0 1 0
We choose the initial state to be

|ii = |ni (spin up in the z-direction) (11.111)

The first order perturbation theory approximation for this result is

2  2

 (1) 1 1
af (T ) = sin2 (0 + )T (11.112)

0 + 2

where
0 = B0 , 1 = B1 (11.113)
and we have used
1
h| V |+i = 0, h| V + |+i = ~B1 (11.114)
2
f i
f i = = 0 (11.115)
~
When is the first order perturbation theory result valid?

If we compare the exact result with perturbation theory by expanding the exact
result in a power series, we find that the two results agree exactly if
 
 1 
 0 +  << 1 (11.116)
 

which corresponds B1  B0 (as long as 0 + 6= 0). When 0 + = 0 we have

a phenomenon called resonance. The exact solution gives

2 1
|af (T )| = sin2 T (11.117)
2
2
where 2 = (0 + ) + 12 and perturbation theory gives
 2  T 2
 (1) 1
af (T ) = (11.118)

2

Thus, the results agree only if |1 T |  1 or if the perturbation only acts for a
short time.

11.2 Atomic Radiation and Selection Rules

We now apply time-independent perturbation theory to the absorption and
emission electromagnetic radiation by matter.

940
The Hamiltonian for an electron in an atom interacting with and electromagnetic
field is  2
~
p~op qc A
H = + q + U (11.119)
2me
where q = e, U = the potential energy function that binds the electrons in
the atom, and A~ and are the vector and scalar potentials associated with the
electromagnetic field.

These potentials imply the electric and magnetic fields

~
~ = 1 A ,B
E ~ =A
~ (11.120)
c t
~ = 0, the B
Note that if A ~ = 0 and

Z~r
= ~ r, t)d~r
E(~ (11.121)
0

We rewrite H as
H = H0 + V (11.122)
where
p~2op
H0 = + U = Hamiltonian for the atom with no electromagnetic field
2me
(11.123)
and
2
q  ~ p~op + q
~+A
 
~A

~ + q
V = p~op A A (11.124)
2me c 2me c2
is the perturbation due to the presence of the electromagnetic field, i.e., the
term V tells us how the atom interacts with the electromagnetic field.

11.2.1 The Electric Dipole Approximation

The typical wavelength of visible electromagnetic radiation is 5000A and the
typical dimension of an atom is a few A. This implies that the electromagnetic
fields are approximately constant over the volume of the atom.

~ |B|,
In Gaussian units |E| ~ but the force due to B ~ (v/c) the force due
~
to E. Thus, magnetic effects are negligible in most atoms compared to electric
effects. We therefore assume
~ constant over the volume of the atom
1. E
~ can be neglected
2. B

941
This is the so-called electric dipole approximation.

~ = 0 and as we said above (11.121), B

In this approximation, we have A ~ = 0 and

Z~r
= ~ r, t)d~r
E(~ (11.125)
0

This last integral is independent of path since

~
~ = 1 B = 0
E (11.126)
c t
~ constant over the volume of the atom, we get
Since we are assuming that E
~
= ~r E(t) (11.127)

This adds a perturbation of the form

~
V = q~r E(t) (11.128)

to the Hamiltonian.

In some derivations the quantity

2
q  ~ p~op + q
~+A
 
~A

~ + q
V = p~op A A (11.129)
2me c 2me c2
is chosen as the perturbation. The next step would be to expand in powers of
the potentials. Since E~ = 0 for the radiation field, we can choose = 0 and
A ~ = 0 (if these relations were not true we could make them so with a gauge
transformation).

~ The first term in the

Therefore, we end up with an expansion in powers of A.
expansion is
q  ~ p~op = q A
~+A

~ p~op
VAp = p~op A (11.130)
2me c me c
since
~A
p~op A ~ p~op A
~=0 (11.131)
In this case, in the electric dipole approximation
~ constant
E , ~ =0
B
Z~r
~ r, t) = c
A(~ ~ r, t)dt
E(~
0
(~r, t) = 0

942
 ~ r, t) = E(~
If E(~ ~ r)eit , then A
~ = eE/i.
~ Using
p~op ih i
= H0 , ~rop (11.132)
m ~
we get
q h i
~
VAp = H0 , ~rop E (11.133)
~
We can then calculate matrix elements
q ~
hm| VAp |ni = hm| H0~rop ~rop H0 |ni E
~
q ~ |ni
= (m n ) hm| ~rop E
~
mn
= hm| V |ni (11.134)

where
m n
mn = (11.135)
~
~ pop ) differ from the matrix elements of
Thus, the matrix elements of VAp (from A~
V (from ) by the factor mn /. This implies different transition probabilities in
first order except at resonance where mn / = 1. The reason for the differences
is as follows:
1. we assumed that the perturbation = 0 for t 0 , t T
~ changing discontinuously
2. there is no physical problem with V = q~r E
~ p~op changes discontinuously, then the relation
3. however, if A
~
~ = 1 A
E (11.136)
c t
~ fields
generates spurious function type E
It is clear that one must exercise great care in choosing a starting point for
perturbation theory.

11.2.2 Induced Emission and Absorption

We now look at the physics connected with the harmonic perturbation. Physi-
cally, it represents an electromagnetic wave interacting with the atom.

We consider the perturbing potential

(
~ 0 (eit + eit ) 0 < t < T
q~r E
V = (11.137)
0 t < 0, t > T

where E ~ 0 = a constant vector which tells us the strength and polarization of

the electromagnetic field.

943
This perturbing potential corresponds to monochromatic(single wavelength)
electromagnetic radiation.

For an initial state |(0)i = |ii where H0 |ii = i |ii the probability, at any time
t T , that the atom will have a final energy f is
 2
 (1)
Pif (T ) = af (T ) (11.138)


to first order, where we have from our earlier derivation

 
~ 0 |ii 
hf | q~r E 1 ei(f i )T 1 ei(f i +)T

(1)
af (T ) = + (11.139)
~ f i f i +
If f > i this gives the probability amplitude for absorbing radiation and if
f < i this gives the probability amplitude for emitting radiation.

In the limit T , as we saw earlier, Pif (t) = if t which gives

 2
2 hf | ~

r E~ 0 |ii
2q
if = [(i f ~) + (i f + ~)] (11.140)
~ 4
In this expression

(i f ~) absorption
(i f + ~) emission

We then have the transition rates

 2

2| ~
r E~ 0 |ii
2q hf
aif = eif = (11.141)
~ 4
These expressions are zero/nonzero or transitions are not-allowed/allowed de-
pending on the matrix element of the perturbation, i.e.,

if for i f ~ 0 |ii = 0, then the transition i f is not allowed

hf | ~r E

These relationships between the quantum numbers of the initial and final states
that tell us whether or not a transition is allowed are called selection rules.

To determine the selection rules for one-electron atoms we only need to consider
matrix elements of the form
~ 0 |n`m` ms i
hn0 `0 m`0 ms0 | ~r E (11.142)

or we need to look at three matrix elements, namely,

hn0 `0 m`0 ms0 | x |n`m` ms i , hn0 `0 m`0 ms0 | y |n`m` ms i and hn0 `0 m`0 ms0 | z |n`m` ms i

944
Now

z = r cos rY10 = x3 (11.143)

x = r sin cos r(Y11 + Y1,1 ) = x1 (11.144)
y = r sin sin r(Y11 Y1,1 ) = x2 (11.145)

A typical matrix element, therefore, will have a term like the following:
Z Z
hsms0 | sms i Rn0 `0 (r)Rn` (r)r3 dr Y`0 m`0 Y`m` Y1m d (11.146)

where m = 1, 0.

The radial integral would equal zero only by accident implying that it is not
part of the general selection rules, which must come from the other terms.

The term hsms0 | sms i = ms0 ms gives us a simple selection rule (this simple
rule arises here because the interaction does not depend on spin). We have

ms0 = ms the ms = 0 SELECTION RULE (11.147)

The rest of the selection rules come from the angular integration terms
Z
Y`0 m`0 Y`m` Y1m d (11.148)

This integral equals 0 unless ` + `0 + 1 = an even integer. This rule follows from
parity considerations. For any angular integration over all angles to be nonzero,
the integrand must be even under the parity operation.

Now
Y`m (1)` Y`m (11.149)
under the parity operation. Therefore
0
Y`0 m`0 Y`m` Y1m (1)`+` +1 Y`0 m`0 Y`m` Y1m (11.150)

which gives the stated rule.

The integration says that we must have

|`0 `| 1 `0 + `

This corresponds to thinking of the integrand as made up of two states, namely,

h` 0 m`0 | ( |`m` i |1mi)

Our angular momentum addition rules say that

|`m` i |1mi = |` + 1i |`i |` 1i

945
and the selection rule then follows from the orthogonality condition.

The two `rules when combined imply the selection rule

m` = m`0 m` = 1,0 SELECTION RULE (11.151)

Thus, for transitions within the electric dipole approximation, as defined above,
we have the SELECTION RULES

m` = m`0 m` = 1 , 0 (11.152)
0
` = ` ` = 1 (11.153)
ms = 0 = s (11.154)

The derivation is more complex for multi-electron atoms due the complexity of
the wave function (see next chapter), but it can be shown that, in general, the
SELECTION RULES in the electric dipole approximation are

1. parity changes
P
2. ( `i ) = 1

3. S = 0

4. MS = 0

5. L = 1, 0 (possibility of 0 which is not allowed for 1-electron atoms)

6. ML = 1, 0

7. J = 1, 0

8. MJ = 1, 0

9. J = 0 J = 0 is strictly forbidden

11.3 A Real Physical Process - Ionization

We now carry out the calculations of the transition rate for the ionization(transition
to the continuum) of hydrogen by electromagnetic radiation.

The initial state of the electron is the ground state of hydrogen

e2
|ii = |100i with energy i = 100 = (11.155)
2a0

The final state of the electron in the ionization process is a free particle state(ionized
electron)
*E ~2 k 2
|f i =  k with energy f = k = (11.156)

2m

946
We then have  E  E
p~op ~k = ~~k ~k (11.157)
 

which says this is a momentum eigenstate also (for free particles [H, p~op ] = 0
and momentum and energy have the same eigenstates). The momentum is given
by p~ = ~~k. Since f > i this is an absorption process and thus the transition
rate is given by
P ~ (t)
0~k = transitionrate = lim 0k
t t
2 D~  2
=  k  V |100i (k 100 ~) (11.158)

~
where = frequency of the electromagnetic radiation and

V = e~r E~ (electric dipole approximation)

E~ = the electric field vector

We define

d = rate of transition into a small solid angle d

X
= 0~k (11.159)
~
k in d

For convenience we use a common trick and assume that the universe is a large
box (side = L, volume = L3 ). This allows us to normalize the plane wave states
associated with the free electron. We have
D  E
~
~r  ~k = Aeik~r (11.160)

D  E Z D  ED  E
~k  ~k = 1 = d3~r ~k  ~r ~r  ~k
Z Z
~ ~
= A2 d3~reik~r eik~r = A2 d3~r = A2 L3 (11.161)

or
1
A= (11.162)
L3/2
Now there are
d3~k L3 L3 mk
L3 = dk 2
dk = ddk (11.163)
(2)3 (2)3 (2)3 ~2

states in the volume d3~k of phase space. This implies that there are

L3 mk
(11.164)
(2)3 ~2
states per unit energy per unit solid angle.

947
Therefore,
Z
X L3 mk e2 D~  ~ 2
d = 0~k d dk  k  ~r E |100i (k 100 ~)

3
(2) ~2 2~
~
kind 0

Doing the integration (using the delta function) we get

L3 mke2 D~  ~ 2
d = d k ~
r E |100i (11.165)

16 2 ~3
  

where
~2 k 2
k = 100 + ~ = (11.166)
2m
 1/2
2m 1
k= 2 (11.167)
~ a0

11.3.1 Evaluation of the Matrix Element

We have
D  Z D  E
~k  ~r E~ |100i = d3~r ~k  ~r h~r| ~r E~ |100i
Z
1 ~
= 3/2 d3~reik~r~r E~ h~r | 100i
L
Z
1 ~ ~ 10 (r)Y00
= 3/2 d3~reik~r~r ER
L
Z
2 ~ r
= 3/2 d3~reik~r e a0 ~r E~ (11.168)
a0 L 3/2

We can arbitrarily choose ~k = kz which gives

Z
D  2 r
~k  ~r E~ |100i =
3/2 3/2
ddrr2 eikr cos a0 ~r E~ (11.169)
a0 L

using ~k R
~ = kz = kr cos . Now in spherical-polar coordinates we have

E~ = E (sin cos ex + sin sin ez + cos ez ) (11.170)

~r = r (sin cos ex + sin sin ey + cos ez ) (11.171)
so that
E~ ~r = Er (cos cos + sin sin cos( )) (11.172)
We then have
D 
~k  ~r E~ |100i

Z2 Z Z
2 r
= 3/2
d sin d drr3 eikr cos a0 [cos cos + sin sin cos( )]
a0 L3/2
0 0

948
Since
Z2 Z2
d cos = d sin =0 (11.173)
0 0
the integration wipes out the cos( ) term. Letting x = cos we then
have
Z1 Z

D 
~k  ~r E~ |100i = 4E cos r
drr3 eikrx a0 xdx
3/2
(11.174)
a0 L3/2
1 0

Now
Z Z
r
3 ikrx a0 3! 6
drr e = drr3 er = = 4 (11.175)
4 1
0 0 ikx + a0

Therefore, we have
Z1
D 
~k  ~r E~ |100i = 4E cos 6x
4 dx

3/2 3/2 
a0 L ikx + 1
1 a0

4E cos 16ka50
= (11.176)
3/2
a0 L3/2 (1 + k 2 a20 )3

and D 
 ~ ~
2 4096 2 E 2 cos2 a70 k 2
 k  ~r E |100i = (11.177)

L3 (1 + k 2 a20 )6
where
2m 1
k2 = 2 (11.178)
~ a0
Finally, we get
L3 mke2
Z Z D  2
ionization = d = d~k  ~k  ~r E~ |100i (11.179)
  
16 2 ~3
where
d~k = integration over the angles of ~k
(varies direction of arbitrary z - axis)

However, varying the z-direction is the same as keeping E~ fixed and integrating
over dE~ . Therefore, we have
L3 mke2 4096 2 E 2 a70 k 2
Z
ionization = dE~ cos2
16 2 ~3 L3 (1 + k 2 a20 )6
64me2 E 2 a70 k3
= (11.180)
3~3 (1 + k 2 a20 )6

949
Now
2ma20
1 + k 2 a20 = (11.181)
~
and letting
~
0 = (11.182)
2ma20
we find that  1/2
1
1 + k 2 a20 = and k = 1 (11.183)
0 a0 0
and we get
3/2
64e2 E 2 a30  0 6

ionization = 1 (11.184)
3~ 0
Thus, there exists a threshold energy for this process, i.e., it cannot occur unless
the energy of the photon is greater than a minimum amount, which makes
physical sense. In particular, we must have 0 so that

~0 = |100 | = Eionization = minimum = threshold (11.185)

We also get the correct 6th power term in the answer which agrees with exper-
iment.

11.4 Adiabatic and Sudden Approximations

In standard time-dependent perturbation theory, we assume that the time-
dependent perturbation is weak. An alternative approach, where we assume
the time-dependence is slow, is called the adiabatic approximation.

Suppose that H = H(g(t)), where g(t) tells us the dependence on time. This
might correspond to a variation in time of some parameters. We still have

d
i~ |(t)i = H(g(t)) |(t)i (11.186)
dt
and at any instant of time we have

H(g(t)) |n(g(t))i = En (g(t)) |n(g(t))i (11.187)

where n(g(t)) represents the quantum numbers describing the instantaneous

state vector.

Let us assume that the instantaneous eigenvectors always form a complete set
so that we can write
X
|(t)i = n (t)ein (t) |n(g(t))i (11.188)
n

950
where we have generalized the phase factor
i
e ~ n t (11.189)

that appeared in a similar expression in our earlier derivations to include the

term
Zt
1
n (t) = En (g(t0 ))dt0 (11.190)
~
0

which is called the dynamical phase.

Inserting this expression for the state vector |(t)i into the time-dependent
Schrodinger equation we get
d X X
ih n (t)ein (t) |n(g(t))i = H(g(t)) n (t)ein (t) |n(g(t))i
dt n n
X dn (t) X dein (t)
i~ ein (t) |n(g(t))i + ih n (t) |n(g(t))i
n
dt n
dt
X
in (t) d
+ ih n (t)e |n(g(t))i
n
dt
X X
= n (t)ein (t) H(g(t)) |n(g(t))i = n (t)ein (t) En (g(t)) |n(g(t))i
n n
X dn (t) dn (t) in (t)
X
i~ ein (t) |n(g(t))i + ih e |n(g(t))i
n (t)i
n
dt n
dt
X d X
+ ih n (t)ein (t) |n(g(t))i = n (t)ein (t) En (g(t)) |n(g(t))i
n
dt n

Now
Zt
dn (t) 1 d 1
= En (g(t0 ))dt0 = En (g(t)) (11.191)
dt ~ dt ~
0

Therefore, we get
X dn (t) X d
ein (t) |n(g(t))i + n (t)ein (t) |n(g(t))i = 0 (11.192)
n
dt n
dt

Applying the linear functional

hm| = hm(g(t))| (11.193)

from the left we get

X dn X d
ein hm | ni + n ein hm| |ni = 0 (11.194)
n
dt n
dt

951
Using
hm | ni = mn (11.195)
we have
dm im X d
e = n ein hm| |ni (11.196)
dt n
dt
dm X d
= n ei(n m ) hm| |ni (11.197)
dt n
dt

Now taking the time derivative of the eigenvalue equation we have

dH d dEn d
|ni + H |ni = |ni + En |ni (11.198)
dt dt dt dt
Again, applying the linear functional hm| from the left we get

dH d dEn d
hm| |ni + hm| H |ni = hm | ni + En hm| |ni (11.199)
dt dt dt dt
For m 6= n, using hm| H = hm| Em , we have

dH d d
hm| |ni + Em hm| |ni = En hm| |ni (11.200)
dt dt dt
d hm| ddt
H
|ni
hm| |ni = (11.201)
dt En Em
Thus, we finally have

dm X hm| ddt
H
|ni
= n ei(n m ) (11.202)
dt n
E m En

We choose the initial state to be one of the instantaneous eigenvectors

|(0)i = |n(g(0))i (11.203)

which implies that

n (0) = 1
m (0) = 0 m 6= n

Therefore, for m 6= n at small t we have

dm hm| ddt
H
|ni
ei(n m )t (11.204)
dt Em En
We now assume that
dH
hm| |ni and Em En (11.205)
dt

952
have slow time dependence and that to this order of approximation we can write
t
ei(n m )t = ei(Em En ) ~ (11.206)

which is what we would have if there was no extra time dependence.

We then get
hm| ddt
H
|ni h i(Em En )t i
m (t) i~ e 1 (11.207)
(Em En )2
This implies that
m (t) remains small for m 6= n (11.208)
The adiabatic theorem assumes that in the case where the system starts in an
eigenstate |ni at t = 0, i.e.,

m (t) = 0 m 6= n (11.209)

and that
|(t)i = ein t |n(g(t))i (11.210)
which says that if the system was in the eigenstate |ni at t = 0, i.e.,

H(g(0)) |ni = H0 |ni = n |ni (11.211)

then at a later time t, it is still in the same eigenstate |n(g(t))i of the new
Hamiltonian H(g(t)), i.e.,

H(g(t)) |n(t)i = En (t) |n(t)i (11.212)

This result is independent of the size of the perturbation. It depends only on

the change in time being slow.

This means that if we start with a particle in the ground state of a harmonic
oscillator potential
1
V = k(0)x2 0 (k(0), x) (11.213)
2
and assume that
k(0) k(T ) (11.214)
slowly, the particle ends up in the ground state of the harmonic oscillator po-
tential
1
V = k(T )x2 0 (k(T ), x) (11.215)
2
to within a phase factor.

The opposite result comes from the so-called sudden approximation , where the
change occurs so fast that no changes of the state vector are possible.

Since the state vector does not change at all, if you are in the ground state and

953
a sudden change in the parameters occurs, then you remain in the ground state
for the old parameters. This is not the ground state with new parameters. It is
some linear combination of the new states.

Let us look at the adiabatic approximation in another way. We consider a time

dependent part of the Hamiltonian of the form
H(t) = H0 + H 0 , H 0 = V f (t) (11.216)
where f (t) has the form shown in Figure 11.5 below.

Figure 11.5: Time Dependence

We assume that the particle starts out in the nth eigenstate of H0 = H(0)
|(0)i = |nii (11.217)
where the subscripts are
i initial parameters
f final parameters
The state vectors is assumed to change in time to |(t)i.

If V is small, we can write, using first order time-independent perturbation

theory
X Vkm
|mif = |mii + |kii (11.218)
Em Ek
k6=m

where
Vkm = i hk| V |mii (11.219)
On the other hand, first order time-dependent perturbation theory implies that
i
X
|(t)i = n (t)e ~ En t |nii (11.220)
n

954
with

Zt
i
n (t) = 1 Vnn f (t0 )dt0 (11.221)
~
0
Zt
i t0
m (t) = Vmn f (t0 )ei(Em En ) ~ dt0 , m 6= n (11.222)
~
0

Now we are assuming that df /dt is small, we integrate by parts to get

Zt
i t0
m (t) = Vmn f (t0 )ei(Em En ) ~ dt0
~
0
Zt
Vmn d h i(Em En ) t0 i 0
= f (t0 ) e ~ dt
Em En dt0
0
Vmn h 0 i(Em En ) t0 it
= f (t )e ~
Em En 0
t
df (t0 ) i(Em En ) t0 0
Z
Vmn
+ e ~ dt (11.223)
Em En dt0
0

Using f (0) = 0 and neglecting the last term because df /dt is small gives

Vnn
X Vqn i
|(T )i = (1 i ) |nii |qi e ~ En T (11.224)
~ Eq En i
q6=n

where
ZT
= f (t)dt (11.225)
0

Therefore

X V kn V nn
X Vqn i
f hn | (T )i = i hn| +
i hk|
(1 i ) |nii |qi e ~ En T
En Ek ~ Eq En i
k6=n q6=n

2
Vnn X |Vqn | e ~i En T
= 1 i + (11.226)
~ (En Eq )2
q6=n

955

X V km Vnn
X Vqn i
f hm | (T )i = i hm| +
i hk|
(1 i ) |nii |qi e ~ En T
Em Ek ~ Eq En i
k6=m q6=n

V nn Vnm
X V nk V km e ~i En T
= i +
~(Em En ) (En Ek )(Em Ek )
m6=k6=n
(11.227)

Note that all the first order terms cancel in the last expression. If we only keep
terms to first order (which is consistent with the derivation) we then have
(
1 i Vnn

k=n
f hk | (T )i = (11.228)
~
0 k 6= n

which implies that

2
2 |Vnn | 2
|f hn | (T )i| = 1 + 1 to first order
~2
2
|f hm | (T )i| = 0 to first order m 6= n

which is the adiabatic approximation.

Therefore, the adiabatic approximation says:

if the Hamiltonian H(t) changes slowly in time,

then there will be no transitions from the
eigenstate |nii of H(0) to a different
eigenstate |mif of H(t)

In time-dependent perturbation theory, we assume the perturbation is turned

on and off and thus H(0) = H(T ), i.e., we have the same unperturbed Hamil-
tonian at the end.

The transitions in that case are from one eigenvector of the unperturbed Hamil-
tonian to another eigenvector of the same unperturbed Hamiltonian.

This implies a first-order transition amplitude and hence a second-order transi-

tion probability.

In the adiabatic approximation, however, we have a second-order transition am-

plitude (m 6= n) and hence a fourth-order transition probability. That is why
we can assume that the transition probability for m 6= n is equal to zero.

The first derivation gives the adiabatic approximation for any size perturbation.
In the second derivation,however, we not only assumed a slow change in time,

956
but also assumes a small perturbation so that we could use first order pertur-
bation theory.

What happens in the second derivation if the perturbation is not small?

The way to handle this is to divide the time interval (0, T ) into N subintervals
such that the perturbation V is small within any subinterval. In fact, it is of
O(V /N ). Thus, if N is large, V is small.

We then apply our arguments to each subinterval. If the transition amplitude is

first-order in the perturbation, then the total transition amplitude behaves like
 
V
N V (11.229)
N
with each of the N steps giving a contribution proportional to V . This says
that the net result is of order V , and thus, if V is large, the transition amplitude
will be large.

However, the transition amplitude is second-order and thus the total transition
amplitude behaves like
 2
V V2
N 0asN (11.230)
N N

Therefore, the transition amplitude (m 6= n) is zero independentof thesize of V .

An Example
Let us consider a 1dimensional square well where
(
0 |x| a2
V (x) = (11.231)
|x| > a2

The eigenfunctions and energies are

(
cos nx
2a n = 1, 3, 5, .......
n (x) = (11.232)
sin nx
2a n = 2, 4, 6, .......
a
for |x| 2 and zero otherwise and

2 ~2 n2
En = n = 1, 2, 3, ....... (11.233)
8ma2
Suppose that we change the size of the well and ask what happens to the ground
state in the sudden and adiabatic approximations.

957
Sudden
x
1 (x) = cos before (11.234)
2a
leads to
x
(x) = cos after (no change in the wave function) (11.235)
2a
However, after the change we have new eigenfunctions and energies
(
0 cos nx
4a n = 1, 3, 5, .......
n (x) = nx
(11.236)
sin 4a n = 2, 4, 6, .......
a
for |x| 2 and zero otherwise and
2 ~2 n 2
En0 = n = 1, 2, 3, ....... (11.237)
8ma2
The state of the system is still an eigenstate of the old well and, thus, is not an
eigenstate of the new well. In fact, we have
x X
(x) = cos = bn n0 (x) (11.238)
2a n

Adiabatic
x
1 (x) = cos before (ground state of old well) (11.239)
2a
x
10 (x) = cos after (ground state of new well) (11.240)
4a
The state of the system is an eigenstate of the new well and, thus, is not an
eigenstate of the old well any longer. In fact, we have
x X
n0 (x) = cos = bn n (x) (11.241)
4a n

It is a superposition of the old energy eigenstates.

11.5 Problems
11.5.1 Square Well Perturbed by an Electric Field
At time t = 0, an electron is known to be in the n = 1 eigenstate of a
1dimensional infinite square well potential
(
for |x| > a/2
V (x) =
0 for |x| < a/2
At time t = 0, a uniform electric field of magnitude E is applied in the direction
of increasing x. This electric field is left on for a short time and then removed.
Use time-dependent perturbation theory to calculate the probability that the
electron will be in the n = 2, 3 eigenstates at some time t > .

958
11.5.2 3-Dimensional Oscillator in an electric field
A particle of mass M , charge e, and spin zero moves in an attractive potential

V (x, y, z) = k x2 + y 2 + z 2


(11.242)

(a) Find the three lowest energy levels E0 , E1 , E2 and their associated degen-
eracy.

(b) Suppose a small perturbing potential Ax cos t causes transitions among

the various states in (a). Using a convenient basis for degenerate states,
specify in detail the allowed transitions neglecting effects proportional to
A2 or higher.

(c) In (b) suppose the particle is in the ground state at time t = 0. Find the
probability that the energy is E1 at time t.

11.5.3 Hydrogen in decaying potential

A hydrogen atom (assume spinless electron and proton) in its ground state is
placed between parallel plates and subjected to a uniform weak electric field
(
0 for t < 0
E~ =
E~0 et for t > 0

Find the 1st order probability for the atom to be in any of the n = 2 states after
a long time.

11.5.4 2 spins in a time-dependent potential

Consider a composite system made up of two spin = 1/2 objects. For t < 0, the
Hamiltonian does not depend on spin and can be taken to be zero by suitably
adjusting the energy scale. For t > 0, the Hamiltonian is given by
 
4 ~ ~
H = S1 S2
~2

Suppose the system is in the state |+ i for t 0. Find, as a function of time,

the probability for being found in each of the following states |+ +i, | +i and
| i.

(a) by solving the problem exactly.

(b) by solving the problem assuming the validity of 1st order time-dependent
perturbation theory with H as a perturbation switched on at t = 0. Under
what conditions does this calculation give the correct results?

959
11.5.5 A Variational Calculation of the Deuteron Ground
State Energy
Use the empirical potential energy function

V (r) = Aer/a

where A = 32.7 M eV , a = 2.18 1013 cm, to obtain a variational approxima-

tion to the energy of the ground state energy of the deuteron (` = 0).

Try a simple variational function of the form

(r) = er/2a

where is the variational parameter to be determined. Calculate the energy

in terms of and minimize it. Give your results for and E in M eV . The
experimental value of E is 2.23 M eV (your answer should be VERY close! Is
your answer above this? [HINT: do not forget about the reduced mass in this
problem]

11.5.6 Sudden Change - Dont Sneeze

An experimenter has carefully prepared a particle of mass m in the first excited
state of a one dimensional harmonic oscillator, when he sneezes and knocks the
center of the potential well a small distance a to one side. It takes him a time
T to blow his nose, and when he has done so, he immediately puts the center
back where it was. Find, to lowest order in a, the probabilities P0 and P2 that
the oscillator will now be in its ground state and its second excited state.

11.5.7 Another Sudden Change - Cutting the spring

A particle is allowed to move in one dimension. It is initially coupled to two
identical harmonic springs, each with spring constant K. The springs are sym-
metrically fixed to the points a so that when the particle is at x = 0 the
classical force on it is zero.

(a) What are the energy eigenvalues of the particle when it is connected to
both springs?

(b) What is the wave function in the ground state?

(c) One spring is suddenly cut, leaving the particle bound to only the other
one. If the particle is in the ground state before the spring is cut, what is
the probability that it is still in the ground state after the spring is cut?

960
11.5.8 Another perturbed oscillator
Consider a particle bound in a simple harmonic oscillator potential. Initially(t <
0), it is in the ground state. At t = 0 a perturbation of the form
H 0 (x, t) = Ax2 et/
is switched on. Using time-dependent perturbation theory, calculate the prob-
ability that, after a sufficiently long time (t  ), the system will have made a
transition to a given excited state. Consider all final states.

11.5.9 Nuclear Decay

Nuclei sometimes decay from excited states to the ground state by internal
conversion, a process in which an atomic electron is emitted instead of a photon.
Let the initial and final nuclear states have wave functions i (~r1 , ~r2 , ..., ~rZ ) and
f (~r1 , ~r2 , ..., ~rZ ), respectively, where ~ri describes the protons. The perturbation
giving rise to the transition is the proton-electron interaction,
Z
X e2
W =
j=1
|~r ~rj |

where ~r is the electron coordinate.

(a) Write down the matrix element for the process in lowest-order perturba-
tion theory, assuming that the electron is initially in a state characterized
by the quantum numbers (n`m), and that its energy, after it is emitted,
is large enough so that its final state may be described by a plane wave,
Neglect spin.
(b) Write down an expression for the internal conversion rate.
(c) For light nuclei, the nuclear radius is much smaller than the Bohr radius
for a give Z, and we can use the expansion
1 1 ~r ~rj
+ 3
|~r ~rj | r r
Use this expression to express the transition rate in terms of the dipole
matrix element
Z
X
d~ = hf | ~rj |i i
j=1

11.5.10 Time Evolution Operator

A one-dimensional anharmonic oscillator is given by the Hamiltonian
H = ~ a a + 1/2 + a aa

where is a constant. First compute a+ and a in the interaction picture and

then calculate the time evolution operator U (t, t0 ) to lowest order in the per-
turbation.

961
11.5.11 Two-Level System
Consider a two-level system |a i , |b i with energies Ea , Eb perturbed by a
jolt H 0 (t) = U (t) where the operator U has only off-diagonal matrix elements
(call them U ). If the system is initially in the state a, find the probability Pab
that a transition occurs. Use only the lowest order of perturbation theory that
gives a nonzero result.

11.5.12 Instantaneous Force

Consider a simple harmonic oscillator in its ground state. An instantaneous
force imparts momentum p0 to the system. What is the probability that the
system will stay in its ground state?

11.5.13 Hydrogen beam between parallel plates

A beam of excited hydrogen atoms in the 2s state passes between the plates
of a capacitor in which a uniform electric field exists over a distance L. The
hydrogen atoms have a velocity v along the xaxis and the electric field E~ is
directed along the zaxis as shown in the figure.

Figure 11.6: Hydrogen beam between parallel plates

~
All of the n = 2 states of hydrogen are degenerate in the absence of the field E,
but certain of them mix (Stark effect) when the field is present.

(a) Which of the n = 2 states are connected (mixed) in first order via the
electric field perturbation?
(b) Find the linear combination of the n = 2 states which removes the degen-
eracy as much as possible.
(c) For a system which starts out in the 2s state at t = 0, express the wave
function at time t L/v. No perturbation theory needed.
(d) Find the probability that the emergent beam contains hydrogen in the
various n = 2 states.

11.5.14 Particle in a Delta Function and an Electric Field

A particle of charge q moving in one dimension is initially bound to a delta
function potential at the origin.

962
 Figure 11.7: Electric Field

From time t = 0 to t = it is exposed to a constant electric field E0 in the

xdirection as shown in the figure above. The object of this problem is to find
the probability that for t > the particle will be found in an unbound state
with energy between Ek and Ek + dEk .

(a) Find the normalized bound-state energy eigenfunction corresponding to

the delta function potential V (x) = A(x).

(b) Assume that the unbound states may be approximated by free particle
states with periodic boundary conditions in a box of length L. Find the
normalized wave function of wave vector k, k (x), the density of states as
a function of k, D(k) and the density of states as a function of free-particle
energy Ek , D(Ek ).

(c) Assume that the electric field may be treated as a perturbation. Write
down the perturbation term in the Hamiltonian, H1 , and find the matrix
element of H1 between the initial and the final state h0| H1 |ki.

(d) The probability of a transition between an initially occupied state |Ii and
a final state |F i due to a weak perturbation H1 (t) is given by
 t 2
Z 
1 0
hF | H1 (t0 ) |Ii eiF I t dt0 
 
PIF (t) = 2 
~  

where F I = (EF EI )/~. Find an expression for the probability P (Ek )dEk
that the particle will be in an unbound state with energy between Ek and
Ek + dEk for t > .

11.5.15 Nasty time-dependent potential [complex integra-

tion needed]
A one-dimensional simple harmonic oscillator of frequency is acted upon by
a time-dependent, but spatially uniform force (not potential!)

(F0 /m)
F (t) = , < t <
2 + t2

963
At t = , the oscillator is known to be in the ground state. Using time-
dependent perturbation theory to 1st order, calculate the probability that the
oscillator is found in the 1st excited state at t = +.

Challenge: F (t) is so normalized that the impulse

Z
F (t)dt

imparted to the oscillator is always the same, that is, independent of ; yet
for >> 1/, the probability for excitation is essentially negligible. Is this
reasonable?

11.5.16 Natural Lifetime of Hydrogen

Though in the absence of any perturbation, an atom in an excited state will stay
there forever(it is a stationary state), in reality, it will spontaneously decay to the
ground state. Fundamentally, this occurs because the atom is always perturbed
by vacuum fluctuations in the electromagnetic field. The spontaneous emission
rate on a dipole allowed transition from the initial excited state |e i to all
allowed ground states |g i is,
4 3 X  2
= k |e i
hg |~d
3~ g

where k = eg /c = (Ee Eg )/~c is the emitted photons wave number.

Consider now hydrogen including fine structure. For a given sublevel, the spon-
taneous emission rate is
4 3 X  0 0 0 0 ~ 2
(nLJMJ )(n0 L0 J 0 ) = k hn L J MJ | d |nLJMJ i

3~ 0
MJ

(a) Show that the spontaneous emission rate is independent of the initial MJ .
Explain this result physically.
(b) Calculate the lifetime ( = 1/) of the 2P1/2 state in seconds.

11.5.17 Oscillator in electric field

Consider a simple harmonic oscillator in one dimension with the usual Hamil-
tonian
p2 m 2 2
H = + x
2m 2
Assume that the system is in its ground state at t = 0. At t = 0 an electric field
E~ = E x is switched on, adding a term to the Hamiltonian of the form

H 0 = eE x

964
 (a) What is the new ground state energy?
(b) Assuming that the field is switched on in a time much faster than 1/,
what is the probability that the particle stays in the unperturbed ground
state?

11.5.18 Spin Dependent Transitions

Consider a spin= 1/2 particle of mass m moving in three kinetic dimensions,
subject to the spin dependent potential
1 2
V1 = k |i h| |~r|
2
where k is a real positive constant, ~r is the three-dimensional position operator,
and {|i , |+i} span the spin part of the Hilbert space. Let the initial state of
the particle be prepared as
|0 i = |i |0i
where |0i corresponds to the ground state of the harmonic (motional) potential.
(a) Suppose that a perturbation
W = ~ (|i h+| + |+i h|) ICM
where ICM denotes the identity operator on the motional Hilbert space,
is switched on at time t = 0.

Using Fermis Golden Rule compute the rate of transitions out of |0 i.

(b) p
Describe qualitatively
p the evolution induced by W , in the limits 
k/m and  k/m. HINT: Make sure you understand part(c).
(c) Consider a different spin-dependent potential
V2 = |+i h+| + (~x) + |i h| (~x)
where (~x) denote the motional potentials
(
+ |x| < a
+ (~x) =
0 |x| a
(
0 |x| < a
(~x) =
+ |x| a
and a is a positive real constant. Let the initial state of the system be
prepared as
|0 i = |i |00 i
where |00 i corresponds to the ground state of (~x). Explain why Fermis
Golden Rule predicts a vanishing transition rate for the perturbation W
specified in part (a) above.

965
11.5.19 The Driven Harmonic Oscillator
At t = 0 a 1dimensional harmonic oscillator with natural frequency is driven
by the perturbation
H1 (t) = F xeit
The oscillator is initially in its ground state at t = 0.
(a) Using the lowest order perturbation theory to get a nonzero result, find
the probability that the oscillator will be in the 2nd excited state n = 2
at time t > 0. Assume 6= .
(b) Now begin again and do the simpler case, = . Again, find the prob-
ability that the oscillator will be in the 2nd excited state n = 2 at time
t>0
(c) Expand the result of part (a) for small times t, compare with the results
of part (b), and interpret what you find.

In discussing the results see if you detect any parallels with the driven
classical oscillator.

11.5.20 A Novel One-Dimensional Well

Using tremendous skill, physicists in a molecular beam epitaxy lab, use a graded
semiconductor growth technique to make a GaAs(Gallium Arsenide) wafer con-
taining a single 1-dimensional (Al,Ga)As quantum well in which an electron is
confined by the potential V = kx2 /2.
(a) What is the Hamiltonian for an electron in this quantum well? Show that
2
0 (x) = N0 ex /2 is a solution of the time-independent Schrodinger
equation with this Hamiltonian and findp the corresponding eigenvalue.
Assume here that = m/~, = k/m and m is the mass of the
electron. Also assume that the mass of the electron in the quantum well
is the same as the free electron mass (not always true in solids).
(b) Let us define the raising and lowering operators a and a+ as
   
+ 1 d 1 d
a = y , a = +y
2 dy 2 dy
p
where y = m/~x. Find the wavefunction which results from operating
on 0 with a+ (call it 1 (x)). What is the eigenvalue of 1 in this quantum
well? You can just state the eigenvalue based on your knowledge - there
is no need to derive it.
(c) Write down the Fermis Golden Rule expression for the rate of a transition
(induced by an oscillating perturbation from electromagnetic radiation)
occuring between the lowest energy eigenstate and the first excited state.
State the assumptions that go into the derivation of the expression.

966
 (d) Given that k = 3.0 kg/s2 , what photon wavelength is required to excite
the electron from state 0 to state 1 ? Use symmetry arguments to decide
whether this is an allowed transition (explain your reasoning); you might
want to sketch 0 (x) and 1 (x) to help your explanation.
(e) Given that

a |i = | 1i , a+ |i = + 1 | + 1i

evaluate the transition matrix element h0| x |1i. (HINT: rewrite x in terms
of a and a+ ). Use your result to simplify your expression for the transition
rate.

11.5.21 The Sudden Approximation

Suppose we specify a three-dimensional Hilbert space HA and a time-dependent
Hamiltonian operator

1 0 0 0 0 1
H(t) = 0 2 0 + (t) 0 0 0
0 0 3 1 0 2

where and (t) are real-valued parameters (with units of energy). Let (t)
be given by a step function
(
t0
(t) =
0 t>0

The Schrodinger equation can clearly be solved by standard methods in the

intervals t = [, 0] and t = (0, +], within each of which H remains constant.
We can use the so-called sudden approximation to deal with the discontinuity
in H at t = 0, which simply amounts to assuming that

|(0+ )i = |(0 )i

Suppose the system is initially prepared in the ground state of the Hamiltonian
at t = 1. Use the Schrodinger equation and the sudden approximation to
compute the subsequent evolution of |(t)i and determine the function

f (t) = h|(0)i | |(t)ii , t0

Show that |f (t)|2 is periodic. What is the frequency? How is it related to the
Hamiltonian?

11.5.22 The Rabi Formula

Suppose the total Hamiltonian for a spin1/2 particle is

H = [B0 Sz + b1 (cos (t)Sx + sin (t)Sy )]

967
which includes a static field B0 in the z direction plus a rotating field in the
x y plane. Let the state of the particle be written

|(t)i = a(t) |+z i + b(t) |z i

with normalization |a|2 + |b|2 = 1 and initial conditions

a(0) = 0 , b(0) = 1

Show that
(b1 )2
 
tp 2
2
|a(t)| = 2 sin2 + (b1 )2
+ (b1 )2 2

where = B0 . This expression is known as the Rabi Formula.

11.5.23 Rabi Frequencies in Cavity QED

Consider a two-level atom whose pure states can be represented by vectors in a
two-dimensional Hilbert space HA . Let |gi and |ei be a pair of orthonormal basis
states of HA representing the ground and excited states of the atom, respec-
tively. Consider also a microwave cavity whose lowest energy pure states can be
described by vectors in a three-dimensional Hilbert space HC . Let {|0i , |1i , |2i}
be orthonormal basis states representing zero, one and two microwave photons
in the cavity.

The experiment is performed by sending a stream of atoms through the mi-

crowave cavity. The atoms pass through the cavity one-by-one. Each atom
spends a total time t inside the cavity (which can be varied by adjusting the
velocities of the atoms). Immediately upon exiting the cavity each atom hits a
detector that measures the atomic projection operator Pe = |ei he|.

Just before each atom enters the cavity, we can assume that the joint state of
that atom and the microwave cavity is given by the factorizable pure state

|(0)i = |gi (c0 |0i + c1 |1i + c2 |2i)

where |c0 |2 + |c1 |2 + |c2 |2 = 1

(a) Suppose the Hamiltonian for the joint atom-cavity system vanishes when
the atom is not inside the cavity and when the atom is inside the cavity
the Hamiltonian is given by
   
HAC = ~ |ei hg| |0i h1| + 2 |1i h2| + ~ |gi he| |1i h0| + 2 |2i h1|

Show that while the atom is inside the cavity, the following joint states

968
 are eigenstates of HAC and determine the eigenvalues:

|E0 i = |gi |0i

1
|E1+ i = (|gi |1i + |ei |0i)
2
1
|E1 i = (|gi |1i |ei |0i)
2
1
|E2+ i = (|gi |2i + |ei |1i)
2
1
|E2 i = (|gi |2i |ei |1i)
2
Then rewrite |(0)i as a superposition of energy eigenstates.

(b) Use part (a) to compute the expectation value

hPe i = h(t)| Pe I C |(t)i

as a function of atomic transit time t. You should find your answer is of

the form X
hPe i = P (n) sin2 [n t]
n

where P (n) is the probability of having n photons in the cavity and n is

the nphoton Rabi frequency.

969
970
Chapter 12
Identical Particles

12.1 Theoretical ideas

We now apply these quantum mechanical methods we have developed to multi-
electron atoms.

We will create a model to handle these atoms that follows from the one-electron
case we just considered. These systems are very complex and all the results
that we derive will be approximations.

If we consider an atom or system of N particles, the wave function

(1, 2, 3, 4, ......, N, t) = (~r1 s1 , ~r2 s2 , ~r3 s3 , ~r4 s4 , ......, ~rN sN , t)

= h1, 2, 3, 4, ....., N, t | i (12.1)
th
where ~rj sj = (radius vector,spin) of the j particle

describing the system will be a function of 3N spatial coordinates, time and

all of the particle spin variables. The 3N spatial coordinates form a multi-
dimensional configuration space.

The Hamiltonian of the system is given by

H = T + V (12.2)

where
N N 
~2 2
X X 
T = Tj = j and V = V (~r1 , ~r2 , ......, ~rN , t) (12.3)
j=1 j=1
2mj

The time-dependent Schrodinger equation is

H(1, 2, 3, 4, ......, N, t) = i~ (1, 2, 3, 4, ......, N, t) (12.4)
t

971
The probability density is defined as

(1, 2, 3, 4, ......, N, t) = (1, 2, 3, 4, ......, N, t)(1, 2, 3, 4, ......, N, t) (12.5)

so that

(1, 2, 3, 4, ......, N, t)d3~r1 d3~r2 .......d3~rN = probability of finding

particle 1 at ~r1 in d3~r1 , and particle 2 at ~r2 in d3~r2 ,
..........., and particle N at ~rN in d3~rN all at time time t

We assume that the N -particle wave function is normalized.

The energy eigenstates or stationary states are solutions of

HE (1, 2, 3, 4, ......, N ) = EE (1, 2, 3, 4, ......, N ) (12.6)

which implies the time dependence

E
(1, 2, 3, 4, ......, N, t) = E (1, 2, 3, 4, ......, N )ei ~ t (12.7)

where E is the energy of the system.

When we consider an N electron atom, the system really has N + 1 particles

(we must include the nucleus). However, the nucleus is so much more massive
than the electrons that we can make the approximation that it has infinite mass
and is fixed.

We put the nucleus of charge Ze at the origin and define

~rj = position vector of the j th electron

rjk = |~rj ~rk | = separation between the j th and k th electrons

The potential energy is

N N X N
X Ze2 X e2
V (1, 2, 3, ....., N ) = + (12.8)
j=1
rj r
j=1 i>j ij

= Coulomb energy between nucleus and electrons

+ Coulomb energy between electrons

We will assume no spin-dependence or time dependence in the potential energy.

All electrons in the atom are considered to be identical or indistinguishable.

This means that there are no interactions that can, in any way, distinguish
them from each other.

Alternatively, we can say that, if we interchange the coordinates and spins of

972
two particles, then it is not possible to determine via any physical measurement
that any change was made in the system.

This says that all measurable quantities or the operators representing them must
remain unchanged by the interchange of indistinguishable particles.

In particular, the Hamiltonian must remain unchanged, i.e., we must have

H(1, 2, 3, 4, ..., j, k, ..., N ) = H(1, 2, 3, 4, ..., k, j, ..., N ) (12.9)

This property of H is called exchange symmetry. Operators that have this

property are symmetric functions of their indices 1, 2, 3, 4, ....., N and they are
called symmetric operators.

Now, every symmetry of a physical system must be represented by an operator

that commutes with H. In this case, we introduce the particle interchange or
permutation operator Pij such that

Pij (1, 2, 3, ........, i, j, ......, N ) = h1, 2, 3, ........, i, j, ......, N | Pij |i

= h1, 2, 3, ........, j, i, ......, N | i = (1, 2, 3, ........, j, i, ......, N )

In words, we say

Pij (1, 2, 3, ........, i, j, ......, N ) gives the amplitude for

finding the j th particle at ~ri with spin si
and ith particle at ~rj with spin sj

Now, the transformed Hamiltonian operator is given by

H 0 = Pij H Pij1 = H (by assumption) (12.10)

This implies that h i

Pij H = H Pij H, Pij = 0 (12.11)
as we expected. The same result holds for all symmetric operators.

Now, suppose that the state vector |i is an eigenvector of the symmetric, N

particle H with energy E. We then have

H |i = E |i (12.12)
H Pij |i = Pij H |i = E Pij |i (12.13)

which says that

Pij |i is also an eigenvector of H with the same energy (12.14)

This holds for any pair (i, j). So H and Pij share a common eigenbasis as
expected. This phenomenon is called exchange degeneracy.

973
For simplicity, we assume that N = 2. We then have

H(1, 2) and (1, 2) (12.15)

and h i
H, P12 = 0 (12.16)

What are the simultaneous eigenfunctions? We have

P12 (1, 2) = (2, 1)

2
P12 (1, 2) = P12 (2, 1) = (1, 2)

which says that

2
P12 = I (12.17)
and that P12 has eigenvalues 1. Now if

H(1, 2) = E(1, 2) (12.18)

then
H(2, 1) = E(2, 1) (12.19)
and these two state functions are degenerate. Then, we can write

S (1, 2) = (1, 2) + (2, 1) symmetric wave function

A (1, 2) = (1, 2) (2, 1) antisymmetric wave function

which are the simultaneous eigenfunctions with

HS = ES HA = EA (12.20)
P12 S = +S P12 A = A (12.21)

It is an experimental fact that the behavior of wave functions under pairwise

particle interchange depends only on the kind of particles involved, in particular
on their spin.

All known particles divide themselves in to two classes:

1. Bosons particles with integer spin, s = 0, 1, 2, 3, 4, ......

2. Fermions particles with half-integer spin, s = 1/2, 3/2, 5/2, ......

and

1. Fermions have antisymmetric wave functions under particle interchange

2. Bosons have symmetric wave functions under particle interchange

974
This relationship between spin and wave function symmetry cannot be proved
in non-relativistic quantum mechanics. It can, however, be proved if we add rel-
ativity and construct the relativistic waves equations for bosons and fermions.

As we shall see, this symmetry/antisymmetry connection of spin and wave func-

tions will generalize to more complex systems with more particles.

Before proceeding to study real atoms with N electrons, let us see what we can
learn from a one-dimensional systems containing either two identical bosons or
two identical fermions.

The general Hamiltonian for a one-dimensional two-particle system is

H = H1 + H2 + U (x1 x2 ) (12.22)
2 2
~
H1 = + V (x1 ) (12.23)
2m x21
~2 2
H2 = + V (x2 ) (12.24)
2m x22

where
U (x1 x2 ) = the particle - particle interaction (12.25)

We will assume that U (x1 x2 ) is small enough that we can apply perturbation
theory. We then use direct product states and write

H = H0 + U (12.26)
H1 n(0)
1
(x1 ) = En(0)
1
n(0)
1
(x1 ) (12.27)
H2 n(0)
2
(x2 ) = En(0)
2
n(0)
2
(x2 ) (12.28)
(0)
Hn1 n2 (x1 , x2 ) = H0 n(0)1
(x1 )n(0)
2
(x2 )
(0) (0)
= (H1 + H2 )n1 (x1 )n2 (x2 )
= (En(0)
1
+ En(0)
2
)n(0)
1
(x1 )n(0)
2
(x2 )
= En(0)
1 n2
n(0)
1 n2
(x1 , x2 ) (12.29)

We will construct the unperturbed(zero order) eigenfunctions and energies from

these direct product states.

For the moment, we will also ignore spin.

The simple direct product states will not work for a description of the two
particle system since the eigenfunctions of H0 must be either symmetric or
antisymmetric under particle interchange.

975
The correct choice is S or A where
1 h i
n(0)S
1 n2
= n(0) 1
(x1 )n(0)
2
(x2 ) + n(0)
1
(x2 )n(0)
2
(x1 ) (12.30)
2
1 h i
n(0)A
1 n2
= (0)
n1 (x 1 ) (0)
n2 (x 2 ) (0)
n1 (x 2 ) (0)
n2 (x 1 ) (12.31)
2
(0) (0) (0)
Both of these states have energy En1 n2 = En1 + En2 .

12.2 Bosons with Spin = 0

We assume that s1 = s2 = 0. This says that there are no new degrees of freedom
and hence no reason to change the wave functions.

Indistinguishable bosons of spin = 0 require a symmetric wave function and

thus we choose as the properly symmetrized zero-order wave functions
1 h i
n(0)S
1 n2
= n(0)
1
(x1 )n(0)
2
(x2 ) + n(0)
1
(x2 )n(0)
2
(x1 ) (12.32)
2
The ground state corresponds to n1 = n2 = 1 or
(0)S (0) (0)
11 = 1 (x1 )1 (x2 ) (12.33)
In perturbation theory, the first order energy is then
D   E
(0) (0)S   (0)S
E11 = 2E1 + 11  U (x1 x2 ) 11
(0)
= 2E1 (12.34)
Z Z Z Z
+ dx1 dx2 dx01 dx02

D   E ED D 
(0)S   (0)  (0) (0)  (0)
11  1 (x2 ) 1 (x1 ) 1 (x1 ) 1 (x2 )

 E ED D   E
 (0)  (0) (0)  (0)   (0)S
U (x1 x2 ) 1 (x01 ) 1 (x02 ) 1 (x01 ) 1 (x02 ) 11

Now
D D   E E
(0)  (0)  (0)  (0)
1 (x1 ) 1 (x2 ) U (x1 x2 ) 1 (x01 ) 1 (x02 )


= U (x1 x2 )(x1 x01 )(x2 x02 ) (12.35)

which implies that
(0)
E11 = 2E1
Z Z D   E E
(0)S   (0)  (0)
+ dx1 dx2 11  1 (x2 ) 1 (x1 )

D D   E
(0)  (0)   (0)S
U (x1 x2 ) 1 (x1 ) 1 (x2 ) 11 (12.36)

976
or
Z Z  2  2
(0)  (0)  (0)
E11 = 2E1 + dx1 dx2 1 (x1 ) U (x1 x2 ) 1 (x2 ) (12.37)
 

For later use we define the general direct integral

Z Z  2  2
Jn1 n2 = dx1 dx2 n(0) (x1 ) U (x1 x2 ) n(0) (x2 ) (12.38)
   
1 2

In this case, we have

(0)
E11 = 2E1 + J11 (12.39)
Now we look at the first excited state of this system. We assume that for the
zero-order states, the first excited state corresponds to n1 = 1 and n2 = 2.
Therefore, the zero-order symmetric wave function for the first excited state is

(0)S 1 h (0) (0) (0) (0)

i
12 = 1 (x1 )2 (x2 ) + 1 (x2 )2 (x1 ) (12.40)
2
and the first order energy is
D   E
(0) (0) (0)S   (0)S
E12 = E1 + E2 + 12  U (x1 x2 ) 12 (12.41)

Using the same procedure as before we get

(0) (0)
E12 = E1 + E2 + J12 + K12 (12.42)

where
Z Z
Kn1 n2 = dx1 dx2 n(0)
1
(x1 )n(0)
2
(x1 )U (x1 x2 )n(0)
2
(x2 )n(0)
1
(x2 ) (12.43)

is called the exchange integral.

Now let us look at a possible physical meaning of these direct and exchange
integrals.

We define
 2
 (0)
n1 (x1 ) = 1 = probability density for particle 1 in state n1 (12.44)


and
 2
 (0)
n2 (x2 ) = 2 = probability density for particle 2 in state n2 (12.45)


977
Therefore, the direct integrand takes the form

1 2 U (r12 ) (12.46)

To see what this means let

e2
U (r12 ) = (12.47)
r12
which corresponds to a repulsive Coulomb potential. The direct integral is then
Z Z
(e1 )(e2 )
dx1 dx2 (12.48)
r12

This represents the total energy of two classical charge distributions interacting
with the potential energy U (r12 ).

The exchange integral, however, has no such classical counterpart. It is the

result of symmetrizing the wave function and therefore arises because of the
invariance of H with respect to particle interchange.

The energy level diagram to first order might look like Figure 12.1 below.

Figure 12.1: Typical Boson Energy Level Diagram

The more interesting case is a two spin = 1/2 fermion system (since electrons
are spin = 1/2 fermions).

12.3 Spin = 1/2 Fermions

The particles now have internal degrees of freedom. The single particle state
vectors must now have both spatial and spin parts

|spacei |spini (12.49)

978
For example,  E
 (0)
n1 |+i1 (12.50)

(0)
presents a fermion in the n1 spatial state with spin up.

We write the corresponding wave function as

D  E
x1  n(0) |+i1 = n(0) (x1 )(1) (12.51)

1 1

and so on, where we define the labels (j) = |+ij and (j) = |ij .

We must choose the antisymmetric combination for the zero-order wave func-
tions. We have 4 possible direct product states given n1 and n2 , i.e.,

1 (1, 2) = n(0)
1
(x1 )n(0)
2
(x2 )(1)(2) (12.52)
2 (1, 2) = n(0)
1
(x1 )n(0)
2
(x2 )(1)(2) (12.53)
(0) (0)
3 (1, 2) = n1 (x1 )n2 (x2 )(1)(2) (12.54)
4 (1, 2) = n(0)
1
(x1 )n(0)
2
(x2 )(1)(2) (12.55)

These are not antisymmetric, however. A useful operator allows us to construct

antisymmetric states. Consider the operator

1
R = (1 P12 ) (12.56)
2

Now for any function A(1, 2) we have

1 1
RA(1, 2) = (1 P12 )A(1, 2) = [A(1, 2) A(2, 1)] (12.57)
2 2
p
which is antisymmetric. The factor 1/ (2) keeps the state normalized. We
now use R to construct four antisymmetric states from the four direct product
states (12.52).

(0) 1
n1 n2 ++ (x1 , x2 ) = (1 P12 )1 (1, 2)
2
1 h (0) i
= n1 (x1 )n(0) 2
(x 2 )(1)(2) (0)
n2 (x 1 ) (0)
n1 (x 2 )(1)(2) (12.58)
2

(0) 1
n1 n2 + (x1 , x2 ) = (1 P12 )2 (1, 2)
2
1 h (0) i
= n1 (x1 )n(0) 2
(x 2 )(1)(2) (0)
n2 (x 1 ) (0)
n1 (x 2 )(2)(1) (12.59)
2

979
 (0) 1
n1 n2 + (x1 , x2 ) = (1 P12 )3 (1, 2)
2
1 h (0) i
= n1 (x1 )n(0) 2
(x2 )(1)(2) n(0)
2
(x1 )n(0)
1
(x2 )(1)(2) (12.60)
2

(0) 1
n1 n2 (x1 , x2 ) = (1 P12 )4 (1, 2)
2
1 h (0) i
= n1 (x1 )n(0) 2
(x2 )(1)(2) (0)
n2 (x 1 ) (0)
n1 (x 2 )(1)(2) (12.61)
2

where the subscripts imply

+ + means both spins up

+ - or - + means one spin up and one spin down
- - means both spins down

Each of these wave functions is antisymmetric and each is an eigenfunction of

H0 (since H0 does not contain any spin dependent terms) with the same energy.
This implies that, at this point, we have a 4-fold degenerate zero-order system
(0) (0)
with energy En1 + En2 .

We could use these states as the zero-order wave function to start perturbation
theory. It would be like doing the spin-orbit calculation using the |`sm` ms i ba-
sis, rather than the |`sjmj i basis where Hso is diagonal. It is always important
to choose zero-order wave functions, if it is not to difficult to do, that incorpo-
rate as much of the symmetry of the system as possible. In other words, choose
zero-order wave functions that are simultaneous eigenstates of the maximal set
of commuting observables. This will hopefully produce a diagonal perturbation
matrix or at least so many zeros that it is easy to diagonalize the rest of the
matrix.

In this case, we not only have [H, P12 ] = 0 which told us to choose antisymmetric
zero-order states, but we also have [H, S ~op
2
] = 0 and [H, Sz ] = 0 where

~op = S
S ~1,op + S
~2,op = the total spin angular momentum
Sz = S1z + S2z = the z - component of the total spin angular momentum

Therefore we should choose antisymmetric state functions which are also eigen-
~op
functions of H0 , S 2
and Sz as our zero-order states.

From our earlier work we know that the possible values of the total spin are

980
 ~ 2 and Sz are
S = 0, 1 and the state vectors that are eigenstates of S op

|1, 1i = (1)(2) = 11 (12.62)

1
|1, 0i = ((1)(2) + (2)(1)) = 10 (12.63)
2
|1, 1i = (1)(2) = 1,1 (12.64)
1
|0, 0i = ((1)(2) (2)(1)) = 00 (12.65)
2
Notice that the 1,m=1,0 are symmetric under P12 and 00 is antisymmetric.

Therefore, we will maintain overall antisymmetry by writing the wave functions

as products of spatial wave function and spin functions such that the spatial
function is symmetric when combined with 00 and the spatial function is an-
tisymmetric when combined with 1,m=1,0 .

The symmetric spatial function is

1 h i
n(0)S
1 n2
= n(0)1
(x1 )n(0)
2
(x2 ) + n(0)
1
(x2 )n(0)
2
(x1 ) (12.66)
2
and the antisymmetric spatial function is
1 h (0) i
n(0)A
n
1 2
= n 1
(x 1 )n
(0)
2
(x 2 ) (0)
n 1
(x 2 ) (0)
n 2
(x 1 ) (12.67)
2
The four zero-order wave functions, which are now eigenfunctions of P12 , H0 ,
~op
S 2
and Sz are then

(0) 1 h i
n1 n2 00 = n(0)1
(x 1 ) (0)
n2 (x 2 ) + (0)
n1 (x 2 ) (0)
n2 (x 1 ) 00 (12.68)
2
(0) 1 h i
n1 n2 1ms = n(0) 1
(x1 )n(0)2
(x2 ) n(0)1
(x2 )n(0)2
(x1 ) 1ms ms = 1, 0
2
(12.69)
Notice that if we have identical spatial states, i.e., n1 = n2 , the S = 1 states
vanish identically. This says that two fermions in an S = 1 spin state cannot be
in the same spatial state(the wavefunction vanishes). This is the first example
of a general principle we will discuss later called the Pauli Exclusion Principle.

An alternative way to find these zero-order wave functions is to go back to first

principles and use CG coefficients. For example
X
n(0)
1 n2 sms
= ams1 ,ms2 n(0)
1 n2 ms1 ms2
(12.70)
ms1 ,ms2
ms1 +ms2 =ms

where
ams1 ms2 = hs1 s2 ms1 ms2 | s1 s2 sms i (12.71)

981
Now   
11  11
ms ms  11 = ms1 , 21 ms2 , 21 (12.72)
22 1 2  22
which implies that
(0) (0)
n1 n2 11 = n1 n2 ++ (12.73)
as written above.

Similarly, for s = 1, ms = 0, the only nonzero CG coefficients are

     
1 1 1 1  1 1 1 1 1 1 1  1 1
 10 = = 10 (12.74)
222 2 22 2 22 22  22
which implies that
(0) 1 (0) 1 (0)
n1 n2 10 = n1 n2 + + n1 n2 +
2 2
1 h (0) i
= n1 (1)n2 (2) n(0)
(0)
1
(2)n(0)
2
(1) 10 (12.75)
2
as written above. We now have the appropriate zero-order wave functions and
can apply perturbation theory to the two fermion system.

As with the two boson case, the zero-order ground state for two fermions corre-
(0)
sponds to n1 = n2 = 1 with zero-order energy 2E1 . Since the S = 1 or triplet
states have identically zero state functions in the case (since the spatial function
(0)
are antisymmetric), we have 11,1ms = 0. The unperturbed ground state must
(0)
then have S = 0, Sz = 0 or it is 11,00 (1, 2). This involves a singlet state with
ms = 0 only
1
00 = ((1)(2) (1)(2)) (12.76)
2
In this state, the particle spins are always opposite or antiparallel.

The ground state energy to first order is

D   E
(0) (0)   (0)
E11 = 2E1 + 11,00  U (x1 x2 ) 11,00
(0)
= 2E1 + J11 (12.77)

which is the same energy as in the two boson system(we are assuming the same
Hamiltonian applies).

The spatial part of the wave function is the same also, namely,
(0) (0)
1 (1)2 (2) (12.78)

We must use a symmetric spatial wave function here because the spin vector is
antisymmetric in the ground state of two fermions. The presence of the spin

982
internal degrees of freedom(and the Pauli principle) has a more dramatic effect
on the first excited state for two fermions.

We again assume that the first excited state corresponds to n1 = 1, n2 = 2.

This gives the energy to first order as
D   E
(0) (0) (0)  (0)
E12 = E1 + E2 + 12,sms  U (x1 x2 ) 12,sms (12.79)


We can write the energy this way, i.e., we do not need to write a 4 4 matrix
hU i and diagonalize it because the hU i matrix is already diagonal in this basis
due the orthogonality of the spin functions and the fact that the perturbing
potential does not depend on spin. This first order energy is different for the
triplet and singlet states. If we do the integrals (they are the same as the boson
case) we get
(0) (0)
E12 = E1 + E2 + J12 K12 (12.80)
where
+ singlet s = 0, ms = 0
triplet s = 1, ms = 1, 0
All the triplet states have the same energy because they have the same spatial
wave function and the perturbing potential does not depend on spin.

We thus get the energy level structure shown in Figure 12.2 below.

Figure 12.2: Typical Fermion Energy Level Diagram

The energies now depend on the total spin S even though the Hamiltonian H
does not explicitly depend on spin. A very dramatic effect!!

This level splitting is not due to any additional terms added to the Hamiltonian
such as Hso or HZeeman . This effect is strictly due to symmetry requirements.
The requirement of symmetry or antisymmetry forced on the spatial wave func-
tions by the symmetry or antisymmetry of the spin vectors causes this level

983
splitting. The entire effect is due to the invariance of the Hamiltonian under
pairwise particle interchange.

Physically, we can argue as follows:

1. symmetric spatial functions are large for x1 x2 , while antisymmetric
spatial functions are 0 for x1 x2
2. U (x1 x2 ) is expected to be largest for x1 x2
3. this implies that for S = 1 fermions hU i is relatively small while for S = 0
fermions hU i is relatively large
4. two identical fermions with antiparallel spins have a large probability of
being close together they attract each other
5. two identical fermions with parallel spins have zero probability of being
close together they repel each other
This repulsion is spin dependent and not due to the Coulomb repulsion between
the electrons.

The first order energy for the singlet state is larger than for the triplet states
because the repulsive interaction is enhanced in the singlet state. This overall
effect is called spin pairing and it is a purely quantum mechanical effect.

12.4 The N-Electron Atom

We now extend our discussion to a system with N electrons (fermions). We
write
H = H0 + H 0 (12.81)
where
N 
~2 2
X 
H0 = + V (~ri ) (12.82)
i=1
2me i
N X
X N
H 0 = U (~ri ~rj ) (12.83)
i=1 j>i

The energy eigenstates for the N electron atom are solutions of the time inde-
pendent Schrodinger equation
HE = EE (12.84)
where E = E (1, 2, 3, 4, ......., N ) and 1 = (~r1 , s1 ) and so on.

The indistinguishability of the N electrons implies that

h i
H, Pij = 0 i, j = 1, 2, 3, 4, ......, N ; i 6= j (12.85)

984
where the Pij interchange all attributes of the electrons, i.e., both the spatial
and spin degrees of freedom.

The wave function must be antisymmetric under pairwise electron interchange

Pij E = E i, j = 1, 2, 3, 4, ......, N ; i 6= j (12.86)

The general problem of N interacting electrons is very complex. At this stage we

only want to extract general properties that will also hold in real 3dimensional
atomic systems. It turns out to be instructive to consider the case of non-
interacting electrons - the so-called independent particle model. In this model
we neglect the electron-electron interactions and look only at the zeroth order.

In particular, we consider N identical non-interacting particles in a potential

well V (~r). The Hamiltonian for any particle in the well is

p~2k,op
H0 (k) = + V (~rk ) (12.87)
2m
where
H0 (k)n (~rk ) = n n (~rk ) n = 0, 1, 2, 3, 4, ...... (12.88)
Thus, any single particle sees the energy level structure as shown in Figure 12.3
below.

Figure 12.3: Single Particle Energy Level Structure

The N-particle Hamiltonian is then

H = H0 (1) + H0 (2) + H0 (3)....... + H0 (N ) (12.89)

with solutions given by

H(1, 2, 3, ......, N ) = E(1, 2, 3, ......, N ) (12.90)

where
(1, 2, 3, ......, N ) = a (1)b (2).......n (N ) (12.91)
and
E = a + b + ...... + n (12.92)

985
This solution implies that

particle 1 is in state a with energy a

particle 2 is in state b with energy b
....................................
....................................
....................................
particle N is in state n with energy n

Electrons have spin = 1/2. Thus, corresponding to any single particle energy
level, say a, there are two possible single particle states, namely,

a (1)(1) and a (1)(1) (12.93)

From now on when we write a (1), where the subscript a will be understood to
also include the spin information.

The simple product state solutions are not physically admissible solutions since
they are not antisymmetric under particle interchange for any two particles.

All such states with particles interchanged pairwise have the same energy. In
fact, any permutation of the indices produces a state with the same energy. We
need to construct a completely antisymmetric linear combination of all of these
solutions.

If these were bosons we would have to construct a completely symmetric linear

combination of all these solutions.

If we define

(1, 2, 3, 4, ......, N ) = a permutation of the particles (12.94)

then, the completely symmetric state is easy to construct. It is

X
S (1, 2, 3, ....., N ) = (1, 2, 3, 4, ......, N ) (12.95)

where the sum means a sum over all possible permutations or arrangements.
There are N ! such permutations.

986
Examples

N = 2 N! = 2
S (1, 2) = 1 (1)2 (2) + 1 (2)2 (1)
N = 3 N! = 6
S (1, 2, 3) = 1 (1)2 (2)3 (3) + 1 (2)2 (1)3 (3) + 1 (3)2 (2)3 (1)
+ 1 (1)2 (3)3 (2) + 1 (3)2 (1)3 (2) + 1 (2)2 (3)3 (1)

How do we construct a completely antisymmetric state? Let us define a general

permutation operator by (illustrate for N = 5)

13452 (1, 2, 3, 4, 5) = (3, 1, 4, 5, 2)

23451 (1, 2, 3, 4, 5) = (2, 3, 4, 5, 1)

Any such permutation operator can be written as the product of the 2particle
interchange operators Pij , i.e.,

(2, 3, 1) = 231 (1, 2, 3)

= P12 (3, 2, 1) = P12 P13 (1, 2, 3)

Thus, any permutation can be written in terms of an odd or even number of

pair interchanges or pair permutations and we call it an odd or even permutation
accordingly. All pair permutations are odd.

Therefore, for a completely antisymmetric state we must have

(
+A if is an even permutation

A= (12.96)
A if is an odd permutation

We therefore form a completely antisymmetric state as follows. We let

(
+1 if is an even permutation
(1) = (12.97)
1 if is an odd permutation

and then

Examples

N = 2 N! = 2
A (1, 2) = 1 (1)2 (2) 1 (2)2 (1)
N = 3 N! = 6
A (1, 2, 3) = 1 (1)2 (2)3 (3) 1 (2)2 (1)3 (3) 1 (3)2 (2)3 (1)
1 (1)2 (3)3 (2) + 1 (3)2 (1)3 (2) + 1 (2)2 (3)3 (1)

It is clear that if any two states are identical (put 2 = 3 above), then A is
identically = 0 as it should be for fermions.

987
This implies that we can put at most 2 electrons in each energy level of the
potential well. The two electrons in the k th level would then have wave functions

k and k (12.98)

i.e., they must have opposite spins. This says that N spin = 1/2 fermions must
occupy at least N/2 different states in the well.

This is very different than for bosons where all the N bosons can be in any
energy level.

Another way to write the completely antisymmetric wave function for fermions
is the so-called Slater determinant
 
 a (1) a (2) . a (N ) 
 
 (1) b (2) . b (N )
A (1, 2, 3, ....., N ) =  b

 (12.99)
 . . . . 

 n (1) n (2) . n (N ) 

The last thing we must do is to normalize these state vectors.

X Z
hA | A i = d3~r1 .......d3~rN
s1 ,s2 ,....,sN
X 0
(1) (1) [a (1)....n (N )] [0 a (1)....n (N )]
0

Now if 6= 0 , then [a (1)....n (N )] and [0 a (1)....n (N )] are orthogonal and

the integration for that term is zero.

Therefore, we get
X Z X 2 2
hA | A i = d3~r1 .......d3~rN |a (1)| .... |n (N )| (12.100)
s1 ,s2 ,....,sN

But
XZ 2
d3~rk |k (j)| = 1 (12.101)
sk

so we finally get
X
hA | A i = 1 = number of possible permutations = N ! (12.102)

and therefore, the properly normalized completely antisymmetric wave function

is  
 a (1) a (2) . a (N ) 
 
1  (1) b (2) . b (N ) 
A (1, 2, 3, ....., N ) =  b (12.103)
N!  . . . . 

 n (1) n (2) . n (N ) 

988
In a similar manner
N!
hS | S i = (12.104)
Na !.......Nn !
where
Nk = the number of times the single particle state k occurs
What is the difference between the ground state of N fermions and N bosons?

For N bosons, all N particles occupy the lowest level 0 and the wavefunction
is
S (1, 2, 3, ...., N ) = 0 (1)0 (2)0 (3).........0 (N ) (12.105)
with energy
E0 = N 0 (12.106)
This is true no matter how large N might be, even for macroscopic systems
where N 1023 . As we shall see in later discussions, this is one of the physi-
cal requirements for phenomena like superconductivity, superfluidity and Bose-
Einstein condensation.

Such a state is not allowed for fermions however. We must have

N even N odd
2 in 0 2 in 0
2 in 1 2 in 1
....... .......
2 in N 1 2 in N 1
2 2

2 in N 1 in N +1
2 2

This difference for systems with even or odd numbers of fermions will lead to
dramatic physical consequences later for some atomic systems.

The ground state energy for N fermions is

2(0 + 1 + ....... + N ) for N even
2

2(0 + 1 + ....... + N 1 ) + N +1 for N odd

2 2

Either of these two energies is always greater than the N boson ground state
energy.

The extra energy is called the zero point energy and it arises from particle inter-
change invariance or it arises from the Pauli Exclusion Principle which states

No two identical fermions in a physical system

can have the same set of quantum numbers

It is equivalent to the antisymmetry of the wave function requirement for fermions.

989
12.5 The Helium Atom
We now consider the simplest multielectron atom, namely, helium, which has
two electrons. The Hamiltonian is
p~21,op Ze2 p~22,op Ze2 e2
H = H(1) + H(2) + V = + + (12.107)
2m r1 2m r2 |~r1 ~r2 |

where

H(i) = hydrogen atom Hamiltonian with nuclear charge Ze (instead of e)

V = electrostatic repulsion between the electrons

We start by neglecting the electrostatic repulsion between the electrons. This

gives us a zero-order solution that we can use in perturbation theory. This is
equivalent to the independent particle model we just discussed.

Since H = H(1) + H(2) in this model, we can write

|i = |n1 `1 m1 i |n2 `2 m2 i (12.108)

where

H(1) |n1 `1 m1 i = En(0)

1
|n1 `1 m1 i and H(2) |n2 `2 m2 i = En(0)
2
|n2 `2 m2 i
H |i = (H(1) + H(2)) |i = (H(1) + H(2)) |n1 `1 m1 i |n2 `2 m2 i
= En(0)
1 n2
|i = (En(0)
1
+ En(0)
2
) |n1 `1 m1 i |n2 `2 m2 i = (En(0)
1
+ En(0)
2
) |i

and
Z 2 e2
En(0) = (Z = 2 for helium) (12.109)
2a0 n2
We will be working out the numbers in this problem so that we can compare our
results to experiment. The zero order energies are shown in Table 12.1 below:
(0) (0)
n1 n2 En1 n2 (Ry) En1 n2 (eV )
1 1 -8 -108.8
1 2 -5 -68.0
1 3 -40/9 -64.4
1 .. .. ..
1 .. -4 -54.5
2 2 -2 -27.2

Table 12.1: Zero Order Energies

where
e2
1 Ry(Rydberg) = = 13.6 eV (12.110)
2a0

990
The ground state energy is
(0) (0)
Egs = E11 = 2E1 = 8 Ry (12.111)

and the energy of the system when one electron has been ionized (no longer
bound) is
(0) (0)
Eion = E1 + E = 4 Ry (12.112)
Therefore, it requires the addition of 4 Ry to create singly ionized helium. Notice
that the (2, 2) state has an energy greater than Eion , which implies that it is
not a bound state of the helium atom. All the states (1, n) are bound states.
The energy level spectrum looks as shown in Figure 12.4 below.

Figure 12.4: Helium Energy level Spectrum

Since the particles are electrons we must antisymmetrize the wave functions.
We have two spin = 1/2 fermions. The spin functions are
(
|1, (1, 0)i symmetric
|s, ms i = (12.113)
|0, 0i antisymmetric

The spatial part of the wave function must be of opposite symmetry to the spin
functions so that the product is antisymmetric. By convention we label the

991
states as follows:

Parahelium
(symmetric space part)00
(|100i |100i) |00i
1
[|100i |2`mi + |2`mi |100i] 00
2
and so on.

Orthohelium
(antisymmetric space part)1ms
1
[|100i |2`mi |2`mi |100i] 1m
2
and so on.

These are the zero-order wave functions. We now handle the

e2 e2
= (12.114)
|~r1 ~r2 | r12
term by perturbation theory.

The first order ground state energy correction is

e2
E = h100| h100| |100i |100i h00 | 00i
r12
2 2
|100 (~r1 )| |100 (~r2 )|
Z Z
= e2 d3~r1 d3~r2 (12.115)
r12
where  3/2
1 Z Zr
100 (~r) = e a0 (12.116)
a0
Therefore,
 3 Z Z Z Z
1 Z 2Zr1 2Zr2 1
E = e2 dr1 r12 e a0
dr2 r22 e a0
d1 d2 (12.117)
2 a0 r12
0 0

Even though this calculation does not give a very accurate result, it is still very
instructive to learn the tricks necessary to evaluate the integrals.

We first need to find a useful expression for 1/r12 . We have

p
r12 = |~r1 ~r2 | = (~r1 ~r2 ) (~r1 ~r2 ) (12.118)
2
r12 = (~r1 ~r2 ) (~r1 ~r2 ) = r12 + r22 2~r1 ~r2
= r12 + r22 2r1 r2 cos (12.119)

992
where
= angle between ~r1 and ~r2 (12.120)
Therefore,
1 1
= 1/2
(12.121)
r12 (r12 + r22 2r1 r2 cos )
In the subsequent development, we let the larger of r1 , r2 be called r> and the
smaller be called r< . We then have
1 1
=
r12   2 1/2
r> 1 2 rr<
>
cos + r<
r>

 2 !  2 !2
1 1 r< r< 3 r< r<
= + 2 cos 2 cos
r> 2r> r> r> 8r> r> r>
 2 !3
15 r< r<
+ 2 cos ... (12.122)
48r> r> r>
or
"  2   #
1 1 r< r< 3 2 1
= 1+ cos + cos + ... (12.123)
r12 r> r> r> 2 2
"  2 #
1 r< r<
= P0 (cos ) + P1 (cos ) + P2 (cos ) + ... (12.124)
r> r> r>

Therefore,
 
1 1 X r<
= P (cos ) (12.125)
r12 r> r>
=0
Now, the addition theorem for spherical harmonics, which is proved at the end
of this chapter, gives

4 X

P (cos ) = Ym (1 )Ym (2 ) (12.126)
2 + 1
m=

Therefore, we finally have

 
1 1 X r< 4 X

= Ym (1 )Ym (2 ) (12.127)
r12 r> r> 2 + 1
=0 m=

Now, the factor Z Z

1
d1 d2 (12.128)
r12
contains terms like Z Z

d1 d2 Ym (1 )Ym (2 ) (12.129)

993
and Z Z
dYm () dYm ()Y00 () = 0 m0 (12.130)

Therefore, the only term that contributes from the sum is = m = 0 and we
get Z Z
1 1
d1 d2 = (12.131)
r12 r>
and therefore we have
 3 Z Z
1 Z 2Zr1 2Zr2 1
E = 2 e 2
dr1 r12 e a0
dr2 r22 e a0
(12.132)
a0 r>
0 0

or
 3 Z Zr1
1 Z 2Zr1 2Zr2 1
E = 2 e 2
dr1 r12 e a0
dr2 r22 e a0
a0 r1
0 0
 3 Z Z
1 Z 2Zr1 2Zr2 1
+ 2 e 2
dr1 r12 e a0
dr2 r22 e a0
(12.133)
a0 r2
0 r1

which gives

5 Ze2
E = = J1s,1s = J10,10 = 2.5 Ry = 34 eV (12.134)
8 a0
for Z = 2.

The ground state energy corrected to first order is then

(0)
E11 = E11 + E = 74.8 eV = 5.5 Ry (12.135)

The experimental value is

(E11 )exp t = 78.975 eV = 5.807 Ry (12.136)

This first order result is amazingly good for this complex system!

Now we deal with the first excited state.

The first order energy shifts are once again given by standard perturbation
theory since the hV i matrix is diagonal in this basis due to the orthonormality
of the spin vectors and the fact that V is independent of spin.

We thus have
e2
Z Z
s,t 1 2
En` = d3~r1 d3~r2 |100 (1)n`0 (2) 100 (2)n`0 (1)| (12.137)
2 r12

994
where s, t singlet, triplet S = 0, 1 , +. As shown before, we need only
calculate the m = 0 case because [L~ op , V ] = 0 where

~ op = L
L ~ 1,op + L
~ 2,op = total orbital angular momentum (12.138)

which implies that the result is independent of m.

Therefore
Z Z
s,t 2 2 1
En` = e2 d3~r1 d3~r2 |100 (1)| |n`0 (2)| (12.139)
r12
Z Z
1
+ e2 d3~r1 d3~r2 100 (1)n`0 (2)100 (2)n`0 (1) = Jn` Kn`
r12
where

Jn` = electrostatic repulsion between two charge distributions

2 2
|100 (1)| and |n`0 (2)| = the direct integral

and

Kn` = the exchange integral which arises from

antisymmetrization of the wave function

with

+ = singlet and = triplet

A convenient way of representing this result is as follows.

~op = S
S ~1,op + S
~2,op

~1,op S
~2,op = S~op
2 ~1,op
2 ~2,op
2 3
2S S S = ~2 (S(S + 1) ) (12.140)
2
(
1
2S ~2,op = ~2 + 2 triplet
~1,op S (12.141)
23 singlet
and therefore
s,t 1 ~1,op S
~2,op )Kn`
En` = Jn` (1 + 4S (12.142)
2~2
The calculation results (in eV) are shown in Table 12.2 below and the energy
levels are shown in Figure 12.5 below. Not bad!!

995
 State 1s2s 1s2p
n` 10,20 10,20 10,21 10,21
singlet triplet singlet triplet
0th order -68.0 -68.0 -68.0 -68.0
J 11.4 11.4 13.2 13.2
K 1.2 1.2 0.9 0.9
1st order -55.4 -57.8 -53.9 -55.7
Eexpt -58.4 -59.2 -57.8 -58.0
error 3.0=5.1% 1.4=2.4% 3.9=6.7% 2.3=4.0%

Table 12.2: Calculation Results

Figure 12.5: Helium Energy Levels

For comparison, we will also calculate the ground state energy using the vari-
ational method. We neglect spin in this case. The simplest choice of a trial
function is the product of two hydrogen atom wave functions as in (12.143)
below, which would be an exact solution if the electron-electron repulsion was
neglected.
 3/2  3/2
1 Z
Zr1 1 Z Zr2
(~r1 , ~r2 ) = e a0 e a0
(12.143)
a0 a0

Since we expect the true wave function to be approximately represented by the

above function, we change Z to and thus obtain the best possible value for
E0 for this type of trial function.

996
We do the calculation as follows. We write
p~21,op Ze2 p~22,op Ze2 e2
H = HZ (1) + HZ (2) + V = + +
2m r1 2m r2 r12
2 2 2 2 2 2
p~1,op e p~2,op e ( Z)e ( Z)e e2
= + + + +
2m r1 2m r2 r1 r2 r12
( Z)e2 ( Z)e2 e2
= H (1) + H (2) + + + (12.144)
r1 r2 r12
Now
(0)
H (1)100 (1) = E100 ()100 (1) = 2 100

(1) (12.145)
(0) 2
H (2)100 (2) = E100 ()100 (2) = 100 (2) (12.146)

in Rydbergs. Therefore

( Z)e2
f () = 22 + h100

(1)| |100 (1)i
r1
( Z)e2 e2
+ h100 (2)| |100 (2)i + h()| |()i (12.147)
r2 r12
But

( Z)e2 ( Z)e2
h100 (1)| |100 (1)i = h100 (2)| |100 (2)i
r1 r2
( Z)e2
= h100| |100i (12.148)
r
Therefore,

1 e2
f () = 22 + 2e2 ( Z) h100| |100i + h100| h100| |100i |100i (12.149)
r r12
Using some earlier calculations we get
5
f () = 22 + 4( Z) + (12.150)
4
Minimizing
df 5
= 0 = 2 + 2Z (12.151)
d 8
or
5
=Z (12.152)
16
and
5 5 5 5
E0variational = f (Z ) = 2(Z )2 = 2Z 2 + Z 2( )2 (12.153)
16 16 4 16

997
The first two terms are just the first order perturbation theory result. The third
term lowers the energy relative to perturbation theory.

For Z = 2, we get

E0variational = 5.7 Ry = 77.48 eV

E0experimental = 78.975 eV
E0perturbation theory = 74.8 eV

Even with the simple trial function, we get a significantly better result using the
variational method. The reduction in the value of Z represents the effect of the
inner electron screening the outer electron so it see a smaller nuclear charge.

12.6 Multielectron Atoms

We now return to the case of N electrons (N > 2). We have

H = E (12.154)

where = all quantum numbers needed to specify the N -electron state and the
Hamiltonian H is
N  N X N
~2 2 Ze2 e2
X  X
H = i + (12.155)
i=1
2me ri r
i=1 j>i ij

For the moment we are neglecting many small(weak) interactions (spin-orbit,

etc). We are also not including the electromagnetic field at this stage. We will
consider it later when we talk about time-dependent perturbation theory and
we will see that its presence leads to instability of atoms with respect to photon
absorption/emission.

The electrons are all indistinguishable, which says that

h i
H, Pij = 0 i, j = 1, 2, 3, ....., N ; i 6= j (12.156)

This implies, since electrons are fermions, that the wave functions must be
completely antisymmetric, i.e.,

Pij = i, j = 1, 2, 3, ....., N ; i 6= j (12.157)

The full Hamiltonian is much too complex to solve exactly. We will approach
the solution as a series of increasingly better approximations and obtain a qual-
itative picture of the energy level structure of these complex atoms.

Since the difficulties arise from the e2 /rij terms that represent the electron-
electron repulsion, we start with a model where each electron moves indepen-
dently of all the other electrons (an independent particle model). In this model

998
each electron will be described by a single-particle wavefunction called an or-
bital.

This leads us to write an approximate Hamiltonian in terms of the single particle

Hamiltonians
~2 2
H0i = Vi (~ri ) (12.158)
2me i
where we assume that the potential energy of the ith electron Vi (~ri ) does not
depend on the coordinates of the other N 1 electrons. We then have the
approximate Hamiltonian for the entire system
N N 
~2 2
X X 
H0A = H0i = i Vi (~ri ) (12.159)
i=1 i=1
2me

This Hamiltonian is separable, i.e., we can assume that the system wavefunction
is a product of single-particle wavefunctions or orbitals.

= 1 (~r1 )2 (~r2 )3 (~r3 )........N (~rN ) (12.160)

where the subscript k represents all applicable single particle quantum numbers
for the k th electron, that is,

k = (ni `i m`i msi ) (12.161)

Each single particle wave function is a product of the form

= (spatial wave function)(spin vector) (12.162)

We will assume that the system wavefunction is a completely antisymmetric

combination of the product states .

Each term in H0A has an eigenvalue equation of the form

~2 2
 
Vi (~ri ) ni `i m`i msi = Eni `i m`i msi ni `i m`i msi (12.163)
2me i
and there are N such equations.

To solve these equations, we must know the potential energy functions Vi (~ri ). As
a first approximation within the orbital approximations, we ignore the electron-
electron repulsion so that the electrons only interact with the nucleus and we
have
Ze2
Vi (~ri ) = V (ri ) = (12.164)
ri
In this approximation, all the other electrons do not matter at all and each
electron satisfies
~2 2 Ze2
 
i ni `i m`i msi (~ri ) = Eni `i m`i msi ni `i m`i msi (~ri ) (12.165)
2me ri

999
This is a hydrogen atom with charge Ze. The single-particle wavefunctions are
given by
n`m` ms (~r) = Rn` (r)Y`m` (, )sms = (~r) (12.166)
where
me Z 2 e4
Enk nk = 1, 2, 3, ...... (12.167)
2~2 nk
The wave function corresponding to a set of orbitals (1 , 2 , ......., N ) is then
properly antisymmetrized by writing it as
 
 1 (1) 2 (1) .. N (1) 
 
1  1 (2) .. .. .. 
=   (12.168)
N!  .. .. .. .. 

 1 (N ) .. .. N (N ) 

where
E = E1 + E2 + ..... + EN (12.169)
We certainly can write down an answer in this approximation but the result,
not surprisingly, is terrible. Any real electron is dramatically affected by the
others, even when there is only one other electron as we saw in helium.

We move on by making the next incrementally better approximation. This

involves the concept of screening.

12.6.1 Screening
Any electron, on the average, if it is far from the nucleus, does not feel all of
the nuclear charge Ze and hence has a weaker Coulomb attraction then we have
assumed. This is clear in the helium variational calculation where we found that
the best value of the charge variational parameter Z 0 was
5
Z0 =Z (12.170)
16
This implies that, on the average, the distant electrons are shielded or screened
from the nucleus by the other electrons.

What is the simplest correction that we can make to take this effect into account
for multi-electron atoms and still leave us with solvable equations?

Suppose we write
Ze2
Vi (ri ) = + Vief f (ri ) (12.171)
ri
where Vief f (ri ) includes the screening effects of the other N 1 electrons. An
important feature of this assumption is that Vief f (ri ) is independent of (, ).
This says that the angular part of the wave function is still
Y`i m`i (, ) (12.172)

1000
The radial function, however, now satisfies a modified equation
 
1 d 2 d `(` + 1) 2me
r + 2 (Eni Vi (ri )) Rni `i (ri ) = 0 (12.173)
ri2 dri i dri ri2 ~

This is called the central field approximation. We still have N difficult equations
to solve.

Hartree proposed the following solution using a successive approximations tech-

nique.

1. an initial potential function is guessed (a very educated guess)

2. this potential function is used to derive new wave functions

3. the new wave functions generate a new potential energy function

4. the procedure is continued until the final wave functions determine a self-
consistent potential, i.e., it stops changing as we iterate

The Hartree method is equivalent to a variational calculation, where the trial

function is taken to be a simple product of single-particle orbitals and the vari-
ation is performed by varying each orbital in an arbitrary way.

Using single-particle wave functions, however, we are still neglecting the corre-
lations between the electrons. Although the simple single-particle orbital prod-
uct functions ignore antisymmetry, some effect of the Pauli exclusion principle
(PEP) can be included in the calculations by choosing the single-particle quan-
tum numbers so they do not violate the PEP.

If we make the calculation more complicated, we can include antisymmetry by

using Slater determinant wave functions. This is called the Hartree-Fock the-
ory. Correlation effects arising from the 1/rij terms can then be added using
perturbation theory. At the level of this text, we assume that this can be done
(see Bethe/Jackiw for details).

12.6.2 Shell Structure

The hydrogen atom solution exhibited a kind of shell structure. We found that
the energy levels were given by

Z 2 e2
En = (12.174)
2a0 n2

and each level had a degeneracy equal to n2 arising from the allowed ranges

` = 0, 1, 2, ......, n 1
m` = `, ........, `

1001
We say that each n value defines a shell with energy En and within each shell
we have subshells defined by `. Thus,

n = 1 ` = 0 1ssubshell
n = 2 ` = 0, 1 2s, 2psubshells
n = 3 ` = 0, 1, 2 3s, 3p, 3dsubshells

We can generalize this idea to N electron atoms.

We assume that the atom consists of shells (n) and subshells (n`). Electrons
are placed into these shells so that we do not violate the PEP, that is, since the
electrons are fermions only two electrons can be in each energy level. We define
in this model

hrin0 = radius of a shell

hrin` = radius of a subshell

and we have
En` = En`0 (degenerate) (12.175)
(this is not true in the central field approximation).

For n > 1, the sorbital has a nonzero probability near the origin r = 0
(the nucleus). This implies that it penetrates the n = 1 shell more than the
corresponding porbitals do. This implies that the s subshell electrons feel a
stronger nuclear charge then the p subshell electrons.

Therefore, we expect in this model that the energy levels will look like Figure
12.6 below.

Figure 12.6: Expected Level Structure

which is the screening effect.

Complex screening arguments of this type lead to the Aufbau principle, which
tells us how electrons fill shells.

1002
We see how it works by figuring out the ground state of an N electron atom.
The ground state corresponds to that state where all the lowest energy levels
are filled with a maximum of two electrons per level. It is clear that this is the
state of lowest energy.

The single-particle Hamiltonian in the central field approximation commutes

with `iz and siz which implies that the energy levels are independent of m`i and
msi . Each energy level is therefore characterized by the 2N quantum numbers

ni , `i i = 1, 2, 3, ......, N (12.176)

We define the electronic configuration of an atomic state as the set of quan-

tum numbers (ni , `i ) for all the electrons in the atom. We use the symbolic
representation

shell label = (n label)(` label)number of electrons

(12.177)

i.e.,

ground state of hydrogen = 1s1 = 1s

(a superscript 1 is always understood)
ground state of helium = 1s2
ground state of lithium = 1s2 2s

The electronic configuration of helium is an example of a closed or full shell.

The 1s subshell has the maximum number electrons as allowed by the PEP, i.e.,
1 1
n = 1, ` = 0, m` = 0, s = , ms = (12.178)
2 2
The Aufbau principle says that we fill the shells so that we obey the PEP or in
the order
1s, 2s, 2p, 3s, 3p, 3d, 4s, ..............
with energy increasing from left to right.

The screening arguments of the type we just discussed imply that for a given n
(a given shell) the energy order is s, p, d, ..... and generally the energy of a shell
increases with n = the principal quantum number. The closed shells correspond
to

s shell maximum number of electrons = 2

= 2(2` + 1)
p shell maximum number of electrons = 6
d shell maximum number of electrons = 10

and so on.

1003
Electrons in the shell beyond the last closed shell are called valence electrons.

Much of the form and shape of the periodic table is determined by the Aufbau
principle. For instance

number of valence electrons chemical properties

The valence electrons are the ones that participate in bonding and chemical
reactions.

This implies that

carbon 1s2 2s2 2p2
and
silicon 1s2 2s2 2p6 3s2 3p2
which each have two p valence electrons should have similar chemical properties,
which is the case.

As with all simple principles of this type, anomalies and breakdowns soon ap-
pear. For the Aufbau principle this occurs at the n = 3 shell.

In real atoms, when the 3d and 4s subshells are partially full, the 4s level fills
ups before the 3d level. This means that
potassium 1s2 2s2 2p2 3s2 3p6 4s
and not 1s2 2s2 2p6 3s2 3p6 3d
The 4s state has a larger probability of being near r = 0 then the 3d state and
hence its energy is lower.

For neutral atoms, an experimental ordering scheme is

shell increasing energy

n=1 1s
n=2 2s 2p
n=3 3s 3p 3d
n=4 4s 4p 4d 4f
n=5 5s 5p 5d
n=6 6s 6p
At the level of this text we will not be doing any actual energy calculations.

1004
12.7 Angular Momentum Coupling
The N electrons each have spin and orbital angular momentum and thus have
associated magnetic moments.

For the full Hamiltonian H including the electron-electron repulsion terms we

have h i
H, ~`i,op 6= 0 i = 1, 2, 3, ......, N (12.179)
However, h i
~ op = 0
H, L (12.180)
where
N
* X
Lop = ~`i,op = the total orbital angular momentum (12.181)
i=1

~ is conserved.
Therefore, the individual `i are not conserved, but L

Thus, the electron-electron repulsion or electrostatic terms in the Hamiltonian

couple the electron orbital angular momenta.

In addition, we must add in spin-orbit and other magnetic interactions (spin-

spin, etc).

The spin-orbit interaction leads to terms of the form ~`i ~si and thus couple a
particles orbital and spin angular momentum leading to (ji , mji ) values, where
~ji = ~`i + ~si as we saw earlier in hydrogen.

In most light atoms, the magnetic interactions are usually weaker than the
electrostatic interactions, i.e., electrostatic 1 eV and spin-orbit 104
105 eV .

The angular momentum coupling in a light atom goes like:

1. the orbital angular momenta ~`i couple to form a total orbital angular
momentum
N
* X
Lop = ~`i,op (12.182)
i=1

2. the spin angular momenta ~si couple to form a total spin angular momen-
tum
XN
~op =
S ~si,op (12.183)
i=1

These two couplings occur when we include the electrostatic interactions

in the Hamiltonian H.

1005
 ~ and S
3. the weaker magnetic interactions then couple L ~ to form the total
angular momentum of the atom

J~ = L+
~ S~ (12.184)

This coupling scheme or order where the electrostatic interactions dominate the
magnetic interactions is called LS or Russell-Saunders coupling.

In heavy atoms, the spin-orbit magnetic interactions dominate the electrostatic

interactions and we get an alternative coupling scheme:
1. each electrons ~`i and ~si couple via the magnetic interactions to form
~ji = ~`i + ~si
= the total angular momentum for the ith electron

2. the electrostatic interactions then couple the ~ji to form

N
X
J~op = ~ji,op (12.185)
i=1

This scheme is called jj-coupling.

We will now investigate the energy level structure in detail for these two different
schemes.

Our discussion of helium has shown that exchange symmetry, which requires
that the wave functions are completely antisymmetric, has dramatic observable
consequences. We saw a spin-spin correlation energy that is characterized by
the rule:

There is a tendency for electrons with parallel

spins to repel (avoid) each other. This fact,
together with the electrostatic repulsion between
electrons implies a strong exchange correlation
that cause the spins to tend to align with each other

12.7.1 LS Coupling
In this regime we have the observables and quantum numbers as shown in Table
12.3 below:

1006
 Operator Quantum Number
~2
L L
op
J~op
2
J
~2
S S
op
Lz ML
Sz MS

Table 12.3: LS Coupling Quantum Numbers

When we discussed the spin-orbit interaction in hydrogen we found

1. when we neglect Hso we can use either |n, L, S, ML , MS i or |n, L, S, J, MJ i
as basis states
2. when we add in Hso , ML and MS are no longer conserved (not good
quantum numbers) and therefore we must use |n, L, S, J, MJ i as basis
states
Now in the orbital approximation we have
N
X N
X
ML = m `i , MS = msi (12.186)
i=1 i=1

and
Lz = ~ML Sz = ~MS
What are the possible L, S values? We can use our addition of angular momen-
tum rules to find out.

Consider two pelectrons , i.e., an vp2 configuration. We have

`1 = `2 = 1 L = 0, 1, 2
1
s1 = s2 = S = 0, 1
2
and

for a given L ML = L, ........, L

for a given S MS = S, ........, S

and

M J = ML + MS
J = |L S| , ......., L + S

Therefore we get the possibilities shown in Table 12.4 below:

1007
 L S J State(s)
1
0 0 0 S0
3
0 1 1 S1
1
1 0 1 P1
3
1 1 0,1,2 P0,1,2
1
2 0 2 D1
3
2 1 1,2,3 D1,2,3

Table 12.4: Possible States

We will discuss the state notation shortly.

For a closed shell we must have L = S = 0 or we would violate the PEP. For
example,

s2 `1 = `2 = 0 L = 0
1
s1 = s2 = S = 0 or 1
2
However, there is only one way to choose the m quantum numbers without
violating the PEP which is shown in Table 12.5 below.

m`1 m`2 ms1 ms2

0 0 1/2 -1/2

Table 12.5: s2 m-values

We therefore have ML = MS = 0 (only possibility) which implies that

L = S = J = 0 1 S0 state (12.187)

For

p6 `1 = `2 = `3 = `4 = `5 = `6 = 1
1
s1 = s2 = s3 = s4 = s5 = s6 =
2
Once again it turns out there is only one way to choose the values without vio-
lating the PEP. This is shown in Table 12.6 below.

1008
 m`1 m `2 m`3 m`4 m`5 m`6 ms1 ms2 ms3 ms4 ms5 ms6
1 1 0 0 -1 -1 1/2 -1/2 1/2 -1/2 1/2 -1/2

Table 12.6: p6 m-values

We therefore have ML = MS = 0 (only possibility) which implies that

L = S = J = 0 1 S0 state (12.188)

This result is true for all closed shells.

In the presence of Hso the energy levels will depend on L, S and J but not on
ML , MS or MJ , which is why we label their atomic terms by
2S+1
LJ (12.189)

where
S, P, D, F, ......... means L = 0, 1, 2, 3, ..... (12.190)
The superscript 2S + 1 s the multiplicity of the level (singlet, doublet, triplet,
etc).

If we ignore Hso then we have (2S + 1)(2L + 1) degeneracy for a given level.

2S+1
Adding Hso splits the J states. Each term LJ remains 2J + 1 degenerate
(the MJ values).

This degeneracy is removed by an external magnetic field which splits the 2J +1

MJ levels(Zeeman effect).

How do we determine the ground state for a particular atom in this scheme?

First, we fill up as many closed shells as possible. The remaining (valence)

electrons determine the ground state configuration.

Let us consider carbon which has two equivalent (same subshell) 2p-electrons in
the unfilled shell. We have

2p2 `1 = `2 = 1 L = 0, 1 or 2
1
s1 = s2 = S = 0 or 1
2
We get Table 12.7 below by applying these rules:

1009
 L S J Term Sublevels
1 1
0 0 0 S S0
1 1
1 0 1 P P1
1 1
2 0 2 D D2
3 3
0 1 1 S S1
3 3
1 1 0,1,2 P P0,1,2
3 3
2 1 1,2,3 D D1,2,3

Table 12.7: Possible States from LS Rules

Not all of these sublevels are allowed by the PEP however. To see this we must
look at the individual electron quantum numbers. Table 12.8 below shows those
m`1 , m`2 , ms1 , ms2 values allowed by the PEP (i.e., no two electrons have the
same set of quantum numbers).

Before proceeding to the table, in this case, we can use symmetry arguments
to determine the allowed levels. In the special case of only two electrons in an
unfilled shell, we can easily determine the symmetry of the spin vectors

S = 0 antisymmetric spin function

S = 1 symmetric spin function

We also know the symmetry of the spatial state in general. The symmetry fol-
lows from the symmetry of the angular part of the 2electron wave function.
Since we have a central field approximation, the angular part of the wave func-
tion is given by the YLML spherical harmonics. The radial function is always
symmetric. The symmetry of the spherical harmonics is (1)L . Therefore,

L = odd antisymmetric space function

L = even symmetric space function

The product of the spin vector and the spatial function must always be anti-
symmetric. Therefore we have

S = 0 always combines with even L

S = 1 always combines with odd L

This method is only simple to carry out for 2electron unfilled shells. In the
case of carbon we get the allowed states

L = 2, S = 0 5 states = (2L + 1)(2S + 1)

L = 1, S = 1 9 states
L = 0, S = 0 1 states

for a total of 15 allowed states. The individual quantum numbers table corre-
sponding to these 15 states is

1010
 Entry m` = 1 m` = 0 m` = +1 ML MS
1 -2 0
2 0 0
3 2 0
4 -1 1
5 -1 0
6 -1 0
7 -1 -1
8 0 1
9 0 0
10 0 0
11 0 -1
12 1 1
13 1 0
14 1 0
15 1 -1

Table 12.8: Individual Quantum Numbers

Any other combinations will violate the PEP. This table can be constructed just
using the PEP.

We now construct an implied terms table which tells us how many states exist
with a particular pair of (ML , MS ) values. It is shown as Table 12.9 below.

ML /MS 1 0 -1
2 0 1 0
1 1 2 1
0 1 3 1
-1 1 2 1
-2 0 1 0

Table 12.9: Implied terms

We use this table to determine which atomic terms are allowed for carbon. The
steps are as follows:
1. Consider the largest possible values of L and S, L = 2,S = 1 which
correspond to the 3 D terms.

Now if a 3 D atomic term existed , then we would necessarily have ML =

2,MS = 1 terms. However, there are no such terms, which implies that
the 3 D term is not allowed and thus the sublevels 3 D1 , 3 D2 , 3 D3 are ruled
out by the PEP.

1011
 2. We now look at the next largest values, namely, L = 2,S = 0 or the 1 D
term. A L = 2,S = 0 term requires ML = 2,MS = 0 terms which do exist.
Therefore the 1 D term and the sublevel 1 D2 exist. This has J = 2 and
thus 2J + 1 = 5 MJ levels. This accounts for 5 of the 15 entries in the
table.

3. We subtract these 5 states to get a second-implied terms table

ML /MS 1 0 -1
2 0 0 0
1 1 1 1
0 1 2 1
-1 1 1 1
-2 0 0 0

Table 12.10: Implied terms

4. We now look at the next largest values, namely, L = 1,S = 1 or the 3 P

term. Since entries with ML = 1, 0, +1 and MS = 1, 0, +1 still exist
in the new table, the 3 P term and the sublevels 3 P2 , 3 P1 , 3 P0 are allowed.
These correspond to a total of (2L + 1)(2S + 1) = 5 + 3 + 1 = 9 states.

5. We subtract these 9 states to get the third-implied terms table

ML /MS 1 0 -1
2 0 0 0
1 0 0 0
0 0 1 0
-1 0 0 0
-2 0 0 0

Table 12.11: Implied terms

Only one state is left with ML = MS = 0, which is a 1 S atomic term.

Therefore the 1 S0 sublevel is allowed.

This accounts for all the 15 entries in the table. No more states are allowed,
which means that the 3 S and 1 P atomic terms and their associated sublevels are
forbidden by the PEP. This result agrees with the allowed states we obtained
from symmetry arguments.

We always need to use the implied-terms tables in the general cases (more than
2 electrons in an unfilled shell) because the corresponding symmetry arguments

1012
are very complex to apply.

Finally, for carbon we have the 1 S, 3 P and 1 D allowed by the PEP.

This result is true for all atoms with 2 equivalent pelectrons outside closed
subshells.

To complete the picture, we must now determine how the allowed terms are
ordered in energy.

12.7.2 Hunds Rules

Each sublevel in carbon 1 S0 , 3 P2 , 3 P1 , 3 P0 and 1 D2 has a different energy when
Hso is included in H.

A set of rules exists for qualitatively ordering the levels. They are called Hunds
rules.

Hunds rules apply when we are ordering the energy levels and sublevels for
equivalent electrons in the ground state.

Rule 1 Terms in the ground state configuration with maximum multiplicity

2S + 1 lie lowest in energy.

This follows from the fact that same spins (unpaired spins) repel and different
spins (paired spins) attract.

High multiplicity implies a greater number of electrons with parallel spin than
low multiplicity in multielectron atoms. Since parallel spin electrons avoid each
other, the e2 /rij effect decreases and the energy of high multiplicity states lies
below that of low multiplicity states.

Rule 2 Of several levels with the same multiplicity S, the one with maximum
L lies lowest in energy

In some sense, the maximum L state implies that all electrons are orbiting in
the same direction. These electrons tend to remain separated from each other
and so have a lower energy than those orbiting in the opposite direction, which
get close to each other some of the time.

Rule 3 Of several sublevels with the same multiplicity S and same L

1. the sublevel with the minimum value of J lies lowest in energy if the shell
is less than half-filled. These are called regular multiplets
2. the sublevel with the maximum value of J lies lowest in energy if the shell
is more than half-filled. These are called inverted multiplets

1013
This results follows from Hso and the fact that e2 /r increases as r .
Applying Hunds rules to an np2 configuration we get the energy level scheme
in Figure 12.7 below.

Figure 12.7: np2 Level Scheme

Hunds rules are not perfect since they are based on the orbital approximation.

To determine an electronic configuration, we must specify how the electrons are

placed into subshells. It turns out there is a phenomenon called configuration
interaction or configuration mixing which forces the quantum mechanical state
to sometimes be a mixture of more than one configuration.

12.7.3 JJ-Coupling
In heavy atoms, the magnetic interactions which couple the ~`i and ~si together
into the ~ji , dominate over the electrostatic interactions which led to LS coupling.
The configurations are then better described by the so-called jj-coupling scheme.

Since si = 1/2 for all electrons, we have for `i 1

1
ji = `i
2
mji = ji , ......., ji

~
The individual ~ji then couple together to give the total J.

In a two-electron configuration the levels are labelled by J, j1 , j2 , MJ where

J = |j1 j2 | , ......., j1 + j2
MJ = J, ......., J

1014
Let us consider the P b (lead) atom, which has np2 valence electrons (built on
many closed shells). We have

`1 = `2 = 1
1 3 1 3
j1 = , and j2 = ,
2 2 2 2
The possible total J values are then
3 3
= 3, 2, 1, 0
2 2
1 3
= 2, 1
2 2
1 1
= 1, 0
2 2
Not all of these states are allowed by the PEP. For example,
3 3
J = 3, MJ = 3 j1 = j2 = , mj1 = mj2 =
2 2
`1 = `2 = 1, m`1 = m`2 = 1
1 1
s1 = s2 = , ms1 = ms2 =
2 2
Both electrons need to have identical quantum numbers for this state to exist.
Thus, this state is not allowed. In a similar manner,
3
J =3 , j1 = j2 =
2
1
J =1 , j1 = j2 =
2
can be shown to be forbidden by the PEP. Therefore we have
3 3
= 2, 0
2 2
1 3
= 2, 1
2 2
1 1
=0
2 2
Usually, the level with the lowest J for a given pair (j1 , j2 ) has the lowest energy
(this is not a strict rule).

For medium weight atoms, neither LS nor jj coupling is valid.

There is a connection between the levels in the two schemes as illustrated by

the energy level diagram in Figure 12.8 below.

1015
 Figure 12.8: LS - jj Energy Level Connection

The connection between the two schemes is clear.

The spacing between J-levels in LS coupling is given as follows.

D E 1
Hso = C [(J(J + 1) L(L + 1) S(S + 1)] (12.191)
2

For the same L, S we have

1
EJ+1 EJ = C [(J + 1)(J + 2) J(J + 1)]
2
= C(J + 1) (12.192)

This says that the spacing between consecutive levels of a fine structure mul-
tiplet is proportional to the larger J value involved. This is the Lande interval
rule.

We end this discussion with an example of two electrons that are not equiva-
lent(in different shells). The discussion is more straightforward since we do not
have to worry about the PEP (all possibilities are allowed).

We consider two electrons in a 4p4d configuration. In the LS coupling scheme

1016
we have:

`1 = 1 , `2 = 2 L = 1, 2, 3
1
s1 = s2 = S = 0, 1
2
3 1 J = 4, 3, 2
3 0 J = 3 7 states
2 1 J = 3, 2, 1 16 states
2 0 J = 2 5 states
1 1 J = 2, 1, 0 9 states
1 0 J = 1 3 states

or

J = 4 in 1 level
J = 3 in 3 levels
J = 2 in 4 levels
J = 1 in 3 levels
J = 0 in 1 level

Thus, we have 12 total levels. The LS coupling energy level diagram is shown
in Figure 12.9

Figure 12.9: LS 4p4d Energy Levels

1017
In the jj-coupling scheme the energy leveldiagram is shown in Figure 12.10

Figure 12.10: jj 4p4d Energy Levels

Notice that the final J values are identical, but their arrangement in energy is
very different.

12.8 Spherical Harmonics Addition Theorem

In Chapter 9 we defined the properties of the spherical harmonics. We found
the following results.

12.8.1 Orbital Angular Momentum

Abstractly,
h i h i
Li , Lj = i~ijk Lk and L2op , Lj = 0 (12.193)
L2op |`mi = ~2 `(` + 1) |`mi and L3 |`mi = ~m |`mi (12.194)
L = Lx iLy (12.195)
integer
`= 0 (12.196)
2
For a given value of `, m takes on the 2` + 1 values
m = `, ` + 1, ` + 2, ........., ` 2, ` 1, `

1018
In ordinary 3dimensional space, if we define

Y`m (, ) = h | `mi = spherical harmonic (12.197)

then we have the defining equations for the Y`m (, ) given by

~ 2op |`mi = L
h| L ~ 2op h | `mi = L
~ 2op Y`m (, )
= ~2 `(` + 1) h | `mi = ~2 `(` + 1)Y`m (, ) (12.198)
h| L3 |`mi = L3 h | `mi = L3 Y`m (, )
= ~m h | `mi = ~mY`m (, ) (12.199)

The general result is

s `m
(1)` 2` + 1 (` + m)! eim

d
Y`m (, ) ` (sin )2` (12.200)
2 `! 4 (` m)! (sin )m d cos

Some examples are:

1
Y00 = (12.201)
4
r r
3 3 i
Y10 = cos , Y1,1 = e sin (12.202)
4 8
r r
5 15
Y20 = (3 cos2 1) , Y2,1 = sin cos ei (12.203)
16 8
r
15
Y2,2 = sin2 e2i (12.204)
32

Some Properties
Complex Conjugate

Y`,m (, ) = (1)m Y`,m (, ) (12.205)

Under the parity operation

~r ~r or r r, , +

which says that

eim eim eim = (1)m eim

sin sin( ) sin
cos cos( ) cos

which imply that

Y`,m (, ) (1)` Y`,m (, ) (12.206)

1019
Therefore,

if ` is even, then we have an even parity state

if ` is odd, then we have an odd parity state

Since they form a complete set, any function of (, ) can be expanded in terms
of the Y`,m (, ) (the Y`,m (, ) are a basis), i.e., we can write
X
f (, ) = f`m Y`,m (, ) (12.207)
`,m

where
Z2 Z
f`m = d sin dY`0 m0 (, )f (, ) (12.208)
0 0

and we have used the orthonormality relation

Z2 Z
d sin dY`0 m0 (, )Y`m (, ) = `0 ` m0 m (12.209)
0 0

The spherical harmonics also satisfy these relations:

Closure:
X
`
X
( 0 )( 0 )
Y`m (, )Y`m (0 , 0 ) = (r, r0 ) (12.210)
sin
`=0 m=`

i.e., the solid angle delta function is equal to zero unless the two vectors r(, ) , r0 (0 , 0 )
coincide. It has the property
Z
f (r0 )(r, r0 )d0 = f (r) (12.211)

for any function f (~r) of the spatial direction specified by , .

Recursion:
1/2
L Y`m = [`(` + 1) m(m 1)] Y`,m1
1/2
= [(` m)(` + 1 m)] Y`,m1 (12.212)
 1/2
(` + 1 + m)(` + 1 m)
cos Y`m = Y`+1,m
(2` + 1)(2` + 3)
 1/2
(` + m)(` m)
+ Y`1,m (12.213)
(2` + 1)(2` 1)

1020
12.8.2 The Addition Theorem
Consider two coordinate systems xyz and x0 y 0 z 0 . The addition theorem is the
formula expressing the eigenfunction P` (cos 0 ) of the angular momentum Lz0
about the z 0 axis in terms of the eigenfunctions Y`,m (, ) of Lz . . See Figure
12.11 below for orientations.

Figure 12.11: Angles Used in the Addition Theorem

The angles and are the azimuth and the polar angles of the z 0 axis in
the Cartesian xyz coordinate frame. They are also the first two Euler angles
specifying the orientation of the Cartesian coordinate system x0 y 0 z 0 with respect
to xyz. The third Euler angle is left unspecified here and the x0 and y 0 axes
are not shown. The projections of the z 0 axis and the radius vector on the xy
plane are dashed lines.

As we can see the position vector ~r has angular coordinates , and 0 , 0 in

the two coordinate systems.

The direction of the z 0 axis in space is specified by its polar angle and its
azimuth angle with respect to the xyz system.

1021
Since P` is an eigenfunction of L~ 2 , only spherical harmonics with the same
op
subscript ` can appear in the expansion.

An interchange of , and , is equivalent to the transformation 0 and

must leave the expansion unchanged because P` (cos 0 ) is an even function of
0 . This means that P` (cos 0 ) must be a function of .

All of these requirements are satisfied only if we write

`
X
P` (cos 0 ) = cm Y`,m (, )Y`,m (, ) (12.214)
m=`

We determine the coefficients cm using the conditions

Lz0 P` (cos 0 ) = 0 (12.215)

We also use the identity (from rotation of a vector component)

Lz0 = sin cos Lx + sin sin Ly + cos Lz

1 1
= sin ei L+ + sin ei L + cos Lz (12.216)
2 2
and invoke the linear independence of the spherical harmonics to obtain (after
some algebra)
cm1 = cm cm = (1)m c0 (12.217)
so that we only need to determine c0 . We specialize to = 0 or 0 = so that
we have the relations
r
2` + 1
Y`m (0, ) = m0 (12.218)
4
r
2` + 1
Y`0 (, ) = P` (cos ) (12.219)
4
We then have
` ` r
X X 2` + 1
P` (cos ) = cm Y`,m (0, )Y`,m (, ) = cm m0 Y`,m (, )
4
m=` m=`
r r r
2` + 1 2` + 1 2` + 1
= c0 Y`,0 (, ) = c0 P` (cos ) (12.220)
4 4 4
or
4
c0 = (12.221)
2` + 1
and we end up with the addition theorem
`
4 X
P` (cos 0 ) =
cm Y`m (, )Y`,m (, ) (12.222)
2` + 1
m=`

1022
where 0 = angle between the directions (, ) and (, ).

If we combine the closure relation with the addition theorem we get the identity

X
(2` + 1)P` (r r0 ) = 4(r, r0 ) (12.223)
`=0

Since we can write

(r r0 )
(~r ~r0 ) = (r, r0 ) (12.224)
r2
we then have the identity

(r r0 ) X 2` + 1
(~r ~r0 ) = P` (r r0 ) (12.225)
r2 4
`=0

Another useful relation is

X
eikz = (2` + 1)i` j` (kr)P` (cos ) (12.226)
`=0

or, in general
X
`
~
X

eik~r = 4 i` j` (kr)Y`m (~k , ~k )Y`m (~r , ~r ) (12.227)
`=0 m=`

12.9 Problems
12.9.1 Two Bosons in a Well
Two identical spin-zero bosons are placed in a 1dimensional square potential
well with infinitely high walls, i.e., V = 0 for 0 < x < L, otherwise V = . The
normalized single particle energy eigenstates are
r
2
un (x) = sin (nx/L)
L

(a) Find the wavefunctions and energies for the ground state and the first two
excited states of the system.

(b) Suppose that the two bosons interact with each other through the per-
turbing potential
H 0 (x1 , x2 ) = LV0 (x1 x2 )
Compute the first-order correction to the ground state energy of the sys-
tem.

1023
12.9.2 Two Fermions in a Well
Two identical spin1/2 bosons are placed in a 1dimensional square potential
well with infinitely high walls, i.e., V = 0 for 0 < x < L, otherwise V = . The
normalized single particle energy eigenstates are
r
2
un (x) = sin (nx/L)
L

(a) What are the allowed values of the total spin angular momentum quantum
number, J ? How many possible values are there fore the zcomponent
of the total angular momentum?

(b) If single-particle spin eigenstates are denoted by |i = u and |i = d,

construct the two-particle spin states that are either symmetric or anti-
symmetric. How many states of each type are there?

(c) Show that the j = 1, m = 1 state must be symmetric. What is the

symmetry of the J = 0 state?

(d) What is the ground-state energy of the two-particle system, and how does
it depend on the overall spin state?

12.9.3 Two spin1/2 particles

The Hamiltonian for two spin1/2 particles, one with mass m1 and the other
with m2 , is given by
!
p~2 p~2 1 S ~1 S~2
H = 1 + 2 + Va (r) + 2
Vb (r)
2m1 2m2 4 ~

where |~r| = ~r1 ~r2 , |~r| = r and

( (
0 for r < a 0 for r < b
Va (r) = , Vb (r) =
V0 for r > a V0 for r > b

with b < a and V0 very large (assume V0 is infinite where appropriate) and
positive.

(a) Determine the normalized position-space energy eigenfunction for the ground
state. What is the spin state of the ground state? What is the degeneracy?

(b) What can you say about the energy and spin state of the first excited
state? Does your result depend on how much larger a is than b? Explain.

1024
12.9.4 Hydrogen Atom Calculations
We discuss here some useful tricks for evaluating the expectation values of cer-
tain operators in the eigenstates of the hydrogen atom.

(a) Suppose we want to determine h1/rin`m . We can interpret h/rin`m as the

1st order correction due to the perturbation /r (same dependence on r
as the potential energy). Show that this problem can be solved exactly
by just replacing e2 by e2 everywhere in the original solution. So, the
exact energy is

2
m e2
E () =
2n2 ~2
the 1st order correction is the term linear in , that is,

me2
E (1) = = h/rin`m
n2 ~2
Therefore we get
me2 1
h1/rin`m = = 2
n 2 ~2 n a0
We note (for later use) that
 
(0) (1) dE
E() = E +E + .... = E( = 0) + + ....
d =0

so that one way to extract E (1) from the exact answer is to calculate
 
dE

d =0


(b) Evaluate, in a manner similar to part (a), p~2 /2 n`m by considering the
Hamiltonian
p2 Ze2 p2
H = +
2 r 2


(c) Consider now /r2 n`m . In this case, an exact solution is possible since
the perturbation just modifies the centrifugal term as follows:

~2 `(` + 1) ~2 `0 (`0 + 1)
+ =
2mr2 r2 2mr2
where `0 is a function of . Now go back to the original hydrogen atom
solution and show that the dependence of E on `0 () is

mZ 2 e4
E(`0 ) = = E() = E (0) + E (1) + ....
2~2 (k+ `0 + 1)2

1025
 Then show that
     0
dE dE d`
/r2 n`m = E (1)


=
d =0 d`0 `0 =` d `0 =`

= 3 2
n a0 (` + 1/2)
or

2 1
1/r n`m = 3 2
n a0 (` + 1/2)


(d) Finally consider /r3 n`m . Since there is no such term in the hydrogen
Hamiltonian, we resort to different trick. Consider the radial momentum
operator  
1
pr = i~ +
r r
Show that in terms of this operator we may write the radial part of the
Hamiltonian
~2
 
1 2
r
2m r2 r r
as
p2r
2m
Now show that
h[H, pr ]i = 0
in the energy eigenstates. Using this fact, and by explicitly evaluating the
commutator, show that

3 Z
2
1/r n`m = 1/r n`m
a0 `(` + 1)
and hence
Z3
1/r3


=
n`m n3 a30 `(` + 1)(` + 1/2)

12.9.5 Hunds rule

Explain on the basis of Hunds rules why the ground state of carbon is 3 P0 and
that of oxygen is 3 P2 .

12.9.6 Russell-Saunders Coupling in Multielectron Atoms

Consider a configuration of k equivalent p electrons outside a closed shell, which
we denote simply by pk , i.e., carbon= p2 , nitrogen= p3 and oxygen= p4 .
(a) Use the implied-terms method to determine all the terms that can arise
from p3 . Which of them will have the lowest energy?
(b) Repeat this calculation for p4 and show that we get the same result as for
p2

1026
12.9.7 Magnetic moments of proton and neutron
The magnetic dipole moment of the proton is
e
p = gp Sp
2mp

with a measured magnitude corresponding to a value for the gyromagnetic ratio

of
gp = 2 (2.792847337 0.000000029)

We have not studied the Dirac equation yet, but the prediction of the Dirac
equation for a point spin1/2 particle is gp = 2. We can understand the fact
that the proton gyromagnetic ratio is not two as being due its compositeness,
i.e., in a simple quark model, the proton is made up of three quarks, two ups
(u), and a down (d). The quarks are supposed to be point spin1/2, hence,
their gyromagnetic ratios should be gu = gd = 2 (up to higher order corrections,
as in the case of the electron). Let us see if we can make sense out of the proton
magnetic moment.

The proton magnetic moment should be the sum of the magnetic moments of
its constituents, and any moments due to their orbital motion in the proton.
The proton is the ground state baryon, so we assume that the three quarks are
bound together (by the strong interaction) in a state with no orbital angular
momentum. The Pauli principle says that the two identical up quarks must
have an overall odd wave function under interchange of all quantum numbers.
We must apply this rule with some care since we will be including color as one
of these quantum numbers.

Let us look at some properties of color. It is the strong interaction analog

of electric charge in the electromagnetic interaction. However, instead of one
fundamental dimension in charge, there are three color directions, labeled as
red (r), blue (b), and green (g). Unitary transformations in this color space(up
to overall phases) are described by elements of the group SU (3), the group of
unimodular 3 3 matrices (electromagnetic charge corresponds to the group
U (1) whose elements are local phase changes). Just like combining spins, we
can combine these three colors according to a Clebsch-Gordon series, with the
result
3 3 3 = 10 8 8 1

These are different rules than for the addition of spin case because that case
uses the rotation group instead. We do not need to understand all aspects of
the SU (3) group for this problem. The essential aspect here is that there is a
singlet in the decomposition, i.e., it is possible to combine three colors in a way
as to get a color singlet state or a state with no net color charge. These turn
out to be the states of physical interest for the observed baryons according to a
postulate of the quark model.

1027
 (a) The singlet state in the decomposition above must be antisymmetric under
the interchange of any two colors. Assuming this is the case, write down
the color portion of the proton wave function.
(b) Now that you know the color wave function of the quarks in the proton,
write down the spin wave function. You must construct a total spin state
|1/2, 1/2i total spin angular momentum state from three spin1/2 states
where the two up quarks must be in a symmetric state.
(c) Since the proton is uud and its partner the neutron (the are just two states
of the same particle) is ddu and mp ' mn , we can make the simplifying
assumption that mu ' md . Given the measured value of gp , what does
you model give for mu ? Remember that the up quark has electric charge
2/3 and the down quark has electric charge 1/3, in units of positron
charge.
(d) Finally, use your results to predict the gyromagnetic moment of the neu-
tron(neutron results follows from proton results by interchanging u and d
labels) and compare with observation.

12.9.8 Particles in a 3-D harmonic potential

A particle of mass m moves in a 3dimensional harmonic oscillator well. The
Hamiltonian is
p~ 2 1
H = + kr2
2m 2
(a) Find the energy and orbital angular momentum of the ground state and
the first three excited states.
(b) If eight identical non-interacting (spin 1/2) particles are placed in such
a harmonic potential, find the ground state energy for the eight-particle
system.
(c) Assume that these particles have a magnetic moment of magnitude . If a
magnetic field B is applied, what is the approximate ground state energy
of the eight-particle system as a function of B (what is the effect of a closed
shell?). Determine the magnetization E/B for the ground state as a
function of B. What is the susceptibility? Dont do any integrals.

12.9.9 2 interacting particles

Consider two particles of masses m1 6= m2 interacting via the Hamiltonian
p21 p2 1 1 1 2
H = + 2 + m1 2 x21 + m2 2 x22 + K (x1 x2 )
2m 2m 2 2 2
(a) Find the exact solutions.
(b) Sketch the spectrum in weak coupling limit K << 2 where = reduced
mass.

1028
12.9.10 LS versus JJ coupling
Consider a multielectron atom whose electron configuration is
1s2 2s2 2p6 3s2 3p6 3d10 4s2 4p4d
(a) To what element does this configuration belong? Is it the ground state or
an excited state? Explain.
(b) Suppose that we apply the Russell-Saunders coupling scheme to this atom.
Draw and energy level diagram roughly to scale for the atom, beginning
with the single unperturbed configuration energy and taking into account
the various interactions one at a time in the correct order. Be sure to
label each level at each stage of your diagram with the appropriate term
designation, quantum numbers and so on.
(c) Suppose instead we apply pure jjcoupling to the atom. Starting again
from the unperturbed n = 4 level, draw a second energy level diagram.
[HINT: Assume that for a given level (j1 , j2 ), the state with the lowest J
lies lowest in energy]

12.9.11 In a harmonic potential

Two identical, noninteracting spin= 1/2 particles of mass m are in a one dimen-
sional harmonic oscillator potential for which the Hamiltonian is
p21x 1 p2 1
H= + m 2 x21 + 2x + m 2 x22
2m 2 2m 2
(a) Determine the ground-state and first-excited state kets and the corre-
sponding energies when the two particles are in a total spin= 0 state.
What are the lowest energy states and the corresponding kets for the
particles if they are in a total spin= 1 state?
(b) Suppose that the two particles interact with a potential energy of interac-
tion (
V0 |x1 x2 | < a
V (|x1 x2 |) =
0 elsewhere
Argue what the effect will be on the energies that you determined in (a),
that is, whether the energy of each state moves up, moves down, or remains
unchanged.

12.9.12 2 particles interacting via delta function

Two particles of mass m are placed in a rectangular box of sides a > b > c in the
lowest energy state of the system compatible with the conditions below. The
particles interact with each other according to the potential V = A(~r1 ~r2 ).
Using first order perturbation theory calculate the energy of the system under
the following conditions:

1029
 (a) particles are not identical
(b) identical particles of spin= 0
(c) identical particles of spin= 1/2 with spins parallel

12.9.13 2 particles in a square well

Two identical nonrelativistic fermions of mass m, spin= 1/2 are in a 1dimensional
square well of length L with V infinitely large outside the well. The fermions
are subject to a repulsive potential V (x1 x2 ), which may be treated as a
perturbation.
(a) Classify the three lowest-energy states in terms of the states of the indi-
vidual particles and state the spin of each.
(b) Calculate to first-order the energies of the second- and third- lowest states;
leave your result in the form of an integral. Neglect spin-dependent forces
throughout.

12.9.14 2 particles interacting via a harmonic potential

Two particles, each of mass M are bound in a 1dimensional harmonic oscillator
potential
1
V = kx2
2
and interact with each other through an attractive harmonic force F12 = K(x1
x2 ). Assume that K is very small.
(a) What are the energies of the three lowest states of this system?
(b) If the particles are identical and spinless, which of the states of (a) are
allowed?
(c) If the particles are identical and have spin= 1/2, which of the states of (a)
are allowed?

12.9.15 The Structure of helium

Consider a Helium atom in the 1s2p configuration. The total angular momen-
tum is L = 1 (a P state). Due to the Fermi-Pauli symmetry this state splits
into singlet and triplet multiplets as shown below.
where the superscripts 1 and 3 represent the spin degeneracy for the singlet/triplet
respectively.
(a) Explain qualitatively why the triplet state has lower energy.

Now include spin-orbit coupling described by the Hamiltonian HSO =

f (r)L S, where L and S are the total orbital and spin angular momentum
respectively.

1030
 Figure 12.12: Fermi-Pauli Splittings

(b) Without the spin-orbit interaction, good quantum numbers for the angu-
lar momentum degrees of freedom are |LML SMS i. What are the good
quantum numbers with spin-orbit present?

(c) The energy level diagram including spin-orbit corrections is sketched be-
low.

Figure 12.13: Including Spin-Orbit

Label the states with appropriate quantum numbers. NOTE: Some of the
levels are degenerate; the sublevels are not shown.

1031
1032
Chapter 13
Scattering Theory and Molecular Physics

13.1 Scattering Theory

One of the main techniques used by physicists to obtain information about
the structure of all forms of matter from elementary particles to solids is the
scattering of particles.

A particle is a localized region in space and time that contains energy and
momentum. In quantum mechanics a good representation of a particle is a wave
packet as we saw earlier. We will assume that the potential that is responsible
for the scattering effects is of short range in space.

We also assume that the spatial extent of the wave packet is small compared to
the spatial dimensions of the laboratory, i.e., the detectors , etc, but that it is
large compared to the size of the scattering region.

We imagine an experimental setup as shown below in Figure 13.1:

Figure 13.1: Scattering Experimental setup

1033
The incident wavepacket is emitted by a source at time t0 in a region where the
potential is zero (or negligible) and we detect the scattered wave packet (at the
detector) in another region where the potential is zero (or negligible).

The wave packet representing the incident particle is given by the expression

d3~k i~k~r
Z
(~r, t0 ) = e a~k (13.1)
(2)3

where a~k has a maximum near ~k0 . Our earlier stationary phase arguments then
say that this wave packet (region in space and time where ||2 is nonzero) travels
with a velocity(group) ~~k0 /m towards the target.

The scattering problem is to determine

(~r, t) = wave function at the time t

after the particle has interacted
with the target (and we are far
from the target once again)

The solution method is as follows:

1. We solve for the exact eigenstates ~k (~r) of the potential V (~r) using the
Schrodinger equation
 2 
~ 2
+ E~k ~k (~r) = V (~r)~k (~r) (13.2)
2m

where
~2 k 2
E~k = (13.3)
2m
and all values of E~k 0 are allowed eigenvalues.

2. These eigenstates are a complete set so we can always write

d3~k
Z
(~r, t0 ) = ~ (~r)b~k (13.4)
(2)3 k

where
~2 k 2
0 (13.5)
2m
for all the ~k (~r) in the expansion.

These assumptions imply that only the incident (incoming) and scat-
tered(outgoing) waves appear and that no bound states contribute since
they would have E < 0 and thus fall off exponentially.

1034
 3. Since we now have an expansion in energy eigenstates it is trivial to in-
corporate the time evolution of (~r, t). We get
d3~k
Z E~
i ~k (tt0 )
(~r, t) = ~
k (~
r )b~
k e (13.6)
(2)3
This is the formal solution to the scattering problem.

13.1.1 Greens Functions

We now need to determine the exact eigenstates ~k (~r). We will use a Greens
function approach similar to the one we introduced earlier for 1dimensional
transmission/reflection problems.

The Greens function for the free particle Schrodinger equation is given by the
differential equation
 2 
~ 2
+ E~k G(~r, k) = (~r) (13.7)
2m
In terms of this function, the solution to the full Schrodinger equation is
Z
~k (~r) = 0 (~r) + d3~r 0 G(~r ~r 0 , k)V (~r 0 )~k (~r 0 ) (13.8)

where
~2 2
 
+ E~k 0 (~r) = 0 (13.9)
2m
and 0 (~r) is the free-particle wave function (solution of the homogenous equation
that results when V = 0). We then have
~
0 (~r) = eik~r (13.10)
Direct substitution shows that we have a general solution to the Schrodinger
equation.
 2   2 
~ 2 ~ 2
+ E~k ~k (~r) = + E~k 0 (~r)
2m 2m
Z  2 
3 0 ~
+ d ~r + E~k G(~r ~r0 , k)V (~r0 )~k (~r0 )
2
2m
Z
= d3~r 0 (~r ~r 0 )V (~r 0 )~k (~r 0 ) = V (~r)~k (~r)

How do we interpret this solution in the context of the scattering problem?

We identify the following terms

0 (~r) = incident wave
Z
d3~r 0 G(~r ~r 0 , k)V (~r 0 )~k (~r 0 ) = scattered wave

1035
In addition, we then say

V (~r 0 )~k (~r 0 ) = potential amplitude evaluated at ~r 0

= source of the scattered wave
G(~r ~r 0 , k) = amplitude of the scattered wave at ~r due
to a unit source ((~r 0 )) at ~r 0

The integral is then the total scattered wave (adds up all waves coming from all
parts of the target region) and is equal to the sum over all source points ~r 0 .

For this interpretation to make physical sense G(~r ~r 0 , k) must generate out-
going waves only!

We can determine the Greens function, as we did earlier, using Fourier trans-
forms and complex contour integration. As before, the choice of the contour
will be equivalent to choosing boundary conditions for the differential equation
and thus completing the solution of the problem.

In this case we must choose a boundary condition (contour of integration) that

generates outgoing waves only.

Let us now determine the Greens function and do the integrals.

We write (using Fourier transforms)

d3 p~ i~p~r
Z
G(~r, k) = e G(~ p) (13.11)
(2)3
d3 p~ i~p~r
Z
(~r) = e (13.12)
(2)3

Substitution then gives

~2 2 d3 p~
  Z  2 
~
+ E~k G(~r, k) = + E~k ei~p~r G(~
2
p)
2m (2)3 2m
d3 p~ ~2 p2 ~2 k 2 i~p~r
Z  
= + e G(~ p)
(2)3 2m 2m
d3 p~ i~p~r
Z
= e (13.13)
(2)3

This implies that

1
G(~
p) = ~2
(13.14)
2m (k 2 p2 )

1036
Therefore, we obtain
d3 p~ i~p~r
Z
1
G(~r, k) = e ~2
(2)3 2m (k 2 p2 )

Z2 Z1 Z
2m eipr cos 2
= 3 2 d d cos p dp
8 ~ p2 k 2
0 1 0
Z Zipr
m 1 p
= 2 2 dp dxex
2 ~ ir p k2
2
0 ipr
Z
m 1 p
eipr eipr dp


=
2 2 ~2 ir p2 k 2
0
Z
m 1 peipr
= 2 2 dp (13.15)
2 ~ ir p2 k 2

The integrand has poles at p = k. If we choose a contour as shown in Figure

13.2 below

Figure 13.2: Integration Contour

then we have

peipr zeizr
Z Z
m 1
2 2 dp + dz = 2iResidue(+k)
2 ~ ir p2 k 2 z 2 k2
semicircle

Now
zeizr
Z
dz 0 as the radius of the semicircle (13.16)
z2 k2
semicircle

due to the eipr term which behaves like eImag(p)r 0 if Imag(p) > 0 as it
does on contour C.

1037
The residue at +k is
keikr eikr
= (13.17)
2k 2
Therefore,
m 1 eikr m eikr
G(~r, k) = 2i = (13.18)
2 2 ~2 ir 2 2~2 r
This physically represents an outgoing spherical wave as is required by the
boundary conditions of the scattering problem.

Putting everything together we have

Z ik|~ r0 |
r ~
i~ m 3 0e
~k (~r) = e k~
r
d ~r V (~r0 )~k (~r0 ) (13.19)
2~2 |~r ~r0 |

Note that since the time dependence of ~k (~r 0 ) is eiE~k t/~ , we are adding up
terms of the form
0 E~
ei(k|~r~r | ~
k t)
(13.20)
|~r ~r 0 |

which are outgoing waves with the appropriate inverse square relationship built
in!

The magnitude of the outgoing momentum is

~k = same as the incoming momentum

(only the direction has changed)

Therefore, potential scattering of this type is elastic scattering.

Since we are assuming that the detectors are far away from the target, we can
look at solutions where |~r| >> |~r 0 |. In this case, we have
p
1/2
k |~r ~r 0 | = k(~r ~r 0 ) (~r ~r 0 ) = k r2 + r 02 2~r ~r 0
 1/2

1/2 ~r
k r2 2~r ~r 0 = kr 1 2 ~r 0
r
 
~r
kr 1 ~r 0 = kr ~k 0 ~r 0 (13.21)
r

where
~k 0 = kr = k ~r = wave vector seen at the detector (13.22)
r

1038
Therefore, at the detector we have

~ eikr
~k (~r) = eik~r + f~ (, ) (13.23)
r k
where (, ) = direction of vecr and
Z
m ~ 0
f~k (, ) = d3~r 0 eik~r V (~r 0 )~k (~r 0 ) (13.24)
2~2
= the scattering amplitude
The term
eikr
f~ (, ) (13.25)
r k
implies

amplitude that particle
amplitude that incident

will reach r after being
particle will be scattered
scattered at the target (13.26)
with k in the direction
outgoing wave + inverse
(, )
rsquared effect
Using the exact solution for ~k (~r)
Z ik|~ r 0|
r ~
m 3 0e ~
~k (~r) + d ~r V (~r 0 )~k (~r 0 ) = eik~r (13.27)
2~2 |~r ~r 0 |
~
we can substitute for eik~r to get
ik|~ r 0|
" #
d3~k r ~
Z Z
m 3 0e 0 0
(~r, t0 ) = ~k (~r) + d ~r V (~r )~k (~r ) a~k (13.28)
(2)3 2~2 |~r ~r 0 |

Now we assumed that a~k has a maximum near ~k0 which implies  that inside the
brackets [....] we can write ~ (~r ) ~ (~r ) and since k0 >> ~k ~k0  we get
0 0  
k k0

~k ~k ~k0 ~k
k= = k0 ~k (13.29)
k k0
Therefore, the last term becomes

d3~k
Z
~ 0
a~ eikk0 |~r~r | ~k0 (~r 0 ) = (k0 |~r ~r 0 | , t0 )~k0 (~r 0 ) = 0 (13.30)
(2)3 k
 
This is zero because at t = t0 , (k0 ~r r~0  , t0 ) = 0 since k0 |~r r~0 | is to the
 
right of the potential.

Therefore,
d3~k
Z
(~r, t0 ) = ~ (~r)a~k (13.31)
(2)3 k

1039
which implies a~k = b~k , i.e., the expansion coefficients are the same whether we
expand in plane waves or the exact energy eigenstates.

Thus, we can write

d3~k
Z E~
k
(~r, t) = 3
~k (~r)a~k ei ~ (tt0 ) (13.32)
(2)

For ~r far from the target, we can use the asymptotic form for ~k (~r) to get
 
E~
i ~ r ~k (tt0 )
k~
d3~k
Z
e
(~r, t) = 0 (~r, t) + a~ f~k (, ) (13.33)
(2)3 k r
where
 
E~
i ~ r ~k (tt0 )
k~
d3~k
Z
e
0 (~r, t) = a~ (13.34)
(2)3 k r
= wave packet at time t if V = 0

Finally, putting in the approximations

f~k f~k0 and k ~k k0 (13.35)

we get
f~k0 (, )
(~r, t) = 0 (~r, t) + 0 (k0 r, t) (13.36)
r
The meaning of these terms is as follows:
1. (~r, t) = (no scattering term) + (scattered wave)
2. the scattered term includes

0 (k0 r, t) = amplitude to be at k0 r at time t

= value of wave function at this point;
all the potential did was to bend the
particle path towards ~r
f~k0 (, )
probability amplitude that the
r
potential did the bending

This result fails

1. if a scattering resonance is present since this implies that the wave packet
experiences strong deformations
2. if the potential has a long range effect

1040
13.1.2 Cross Sections
The quantity used to connect theory to experiment in scattering experiments is
the differential scattering cross section.

It is defined as follows: if

1. Nin = number of incident particles per cm2

2. dN (, ) = number of particles scattered inot a solid angle d centered

at (, )

then
d dN (, )
= (13.37)
d Nin d
In terms of a single particle, d/d is the total probability that the particle is
scattered into a unit solid angle divided by the total probability that the particle
crosses a unit area in front of the target.

The same assumptions are made as earlier, i.e., the incident packet size  size
of the target and the detector is far from the incident bean and the target.

Now the total probability of being scattered into an infinitesimal solid angle
d at ~r is the rate that probability strikes an area r2 d in the detector plane
integrated over time.

This is given by
Z
dt (velocity) (area at detector) (probability of scattering)

 2
Z 
~ (, )
  2
~k0 2
fk0  
= dt ( ) (r d) ( k r, t) (13.38)

0 0
m r2
 

or the total probability of being scattered into d is

Z 

 2 ~k 2
0
f~k0 (, ) d dt 0 (k0 r, t) (13.39)
   
m

The total probability that crosses a unit area at ~r0 in front of the target in the
incident beam is
Z Z  2
~k0
dt(probability flux) = dt 0 (k0 r, t) (13.40)
 
m

1041
If we assume that the wave packet does not spread between ~r0 and ~k0 r (this
just says a particle remains a particle during the duration of the experiment),
then we have
d  2
= ratio of these last two terms = f~k0 (, ) (13.41)
 
d
This result does not depend on the details of the incident wave packet.

Finally, the total cross section , where

Z
d
= d (13.42)
d
all
solid
angles

= total probability of being scattered divided by the

total probability that the particle crossed a unit area
in front of the target
Z  2
= d f~k0 (, ) (13.43)
 

Since experimentalists can measure and d/d, the theorist needs to be able
to calculate them given the potential function.

13.1.3 Partial Waves

We now look at the special case where V (~r) = V (r) is spherically symmetric.
In this case, the angular momentum of the incident particle is conserved, i.e.,
h i h i
~ 2op = 0 = H, Lz
H, L (13.44)

This means we should only consider stationary state solutions that are also
eigenstates of the angular momentum.

We start off with an identity(see Addition Theorem discussion in Chapter 12).

If we choose ~k = kz, then we can write

~
X
eik~r = eikr cos = eikz = i` (2` + 1)P` (cos )j` (kr) (13.45)
`=0

This is an expansion of a plane wave in spherical harmonics (or angular momen-

tum eigenfunctions). The expansion coefficients are spherical Bessel functions.
Physically, the expansion implies that an plane wave of infinite spatial extent
contains all possible orbital angular momentum values.

1042
In this expansion
r
4
P` (cos ) = Y`0 (, ) (13.46)
2` + 1
= Legendre polynomial of order `

and
1
(h` (kr) + h` (kr))
j` (kr) = (13.47)
2
where the h` (kr) are Hankel functions.

Since the incident plane wave is independent of (rotation about the direction
of ~k), the scattered wave must be invariant under rotations about the direction
of ~k 0 , which implies that
f~k (, ) f~k () (13.48)
For functions of , the Legendre polynomials are a complete set so we can expand
f~k () as

X
f~k () = (2` + 1)f` P` (cos ) (13.49)
`=0

Each ` term is called a partial wave and f` is the partial wave scattering ampli-
tude.

For similar reasons, the stationary state wave functions can be written as

X
~k (~r) = i` (2` + 1)P` (cos )R` (r) (13.50)
`=0

where
d2
 
`(` + 1) 2m
+ k2 rR` (r) = V (r)rR` (r) (13.51)
dr2 r2 ~2
In the limit of large r (where the detector is located), assuming that r2 V (r) 0,
we get  2 
d 2 `(` + 1)
+k rR` (r) = 0 (13.52)
dr2 r2
which is Bessels equation.

Therefore, near the detector (r ) we can write R` (r) as a linear combination

of the h` (kr) and h` (kr). In particular, we assume the form

R` (r) = B` [h` (kr) + S` (E)h` (kr)] (13.53)

The h` (kr) term represents an incoming spherical wave and the h` (kr) term
represents an outgoing spherical wave. We have therefore assumed that the
effect of the potential will be to modify only the outgoing wave.

1043
In the absence of scattering, i.e., V = 0, we have
1
R` (r) = j` (kr) = [h` (kr) + h` (kr)] (13.54)
2
which implies that
1
B` = and S` = 1 (13.55)
2
When V (r) 6= 0 only S` changes.

We will be considering only elastic scattering in this discussion.

In the Schrodinger picture, we can talk about probability densities and proba-
bility flows or currents. In non-relativistic quantum mechanics we have a prob-
ability interpretation for the wave function which implies that
2
probability density = P (~r, t) = |(~r, t)| (13.56)
and
Z
2
|(~r, t)| d3~r = 1 (13.57)

Since there are no sources or sinks of probability in non-relativistic quantum

mechanics (no particle creation or annihilation processes) we must have a con-
tinuity equation of the form
P
+ J~ = 0 (13.58)
t
where
J~ = probability current (13.59)
Using (13.56) and the time-dependent Schrodinger equation we get
~
J~ = [ ] (13.60)
2im
Proof:
P 2
i~ = i~ |(~r, t)| = i~ + i~
t t t  2t
~2 2
  
~
= +V + 2 V
2m 2m
~  ~
2 2 =

= [ ]
2im 2im
= J~
Since as many particles enter a sphere of radius r surrounding the target as leave
the sphere (this is the meaning of elastic scattering), the radial component of
the current must be zero or
 
~ R` (r) R` (r)
Jr (r) = R` (r) R` (r) =0 (13.61)
2im r r

1044
Substitution of (13.53) implies that we must have

|S` (E)| = 1 (13.62)

which says that we can always write

S` (E) = e2i` (E) where ` is real (13.63)

` is called the phase shift and 2` equals the difference in phase between the
~
outgoing parts of the wave function K r) and the incident plane wave eik~r .
~ (~

The phase shift contains the all of the effects of the potential on the wave
function.

Since the ` (E) are constants for a given scattering process (since E = constant),
we can determine their values at any point in the scattering process.

Let us look at r very large where V (r) = 0. We can then write

1X `
i (2` + 1)P` (cos ) h` (kr) + e2i` h` (kr)
 
~k (~r) =
2
`
1X `
i (2` + 1)P` (cos ) h` (kr) + h` (kr) + (e2i` 1)h` (kr)
 
=
2
`
1X `
= i (2` + 1)P` (cos ) [h` (kr) + h` (kr)]
2
`
1X `
i (2` + 1)P` (cos ) (e2i` 1)h` (kr)
 
+
2
`
X
`
= i (2` + 1)P` (cos )j` (kr)
`
1X `
i (2` + 1)P` (cos ) (e2i` 1)h` (kr)
 
+
2
`

i~
k~
r 1X `
i (2` + 1)P` (cos ) e2i` 1 h` (kr)


=e + (13.64)
2
`

But we know that

`
ei(kr 2 )
h` (kr) as r (13.65)
ikr
Substituting this result and using

~ f () ikr
~k (~r) = eik~r + e
r

1045
we get
1 X
(2` + 1)P` (cos ) e2i` 1


f () = (13.66)
2ik
`
1X
= (2` + 1)P` (cos )ei` sin ` (13.67)
k
`

where
f` = e2i` 1 = ei` sin ` (13.68)
= the partial wave scattering amplitude
We obtain the total cross section using
Z
2
= |f ()| d (13.69)

X X (2` + 1) (2`0 + 1) Z2 Z1
i` i`0
= e e sin ` sin `0 d dxP` (x)P`0 (x)
k k
` `0 0 1

Now
Z1
``0
dxP` (x)P`0 (x) = (13.70)
2` + 1
1

which implies that

4 X 2
X
= (2` + 1) sin ` = ` (13.71)
k2
` `

where
4
(2` + 1) sin2 `
` = (13.72)
k2
is the partial wave cross section for scattering particles in angular momentum
`states.

Since sin2 ` 1, we have the limit

4
` (2` + 1) (13.73)
k2
Classically, the maximum ` value that contributes to the sum is given by
`max = ka (13.74)
where a is the range of the potential (i.e., for, V (r) = 0 we have no scattering).

Quantum mechanically, for r > a, only the term

~2 `(` + 1)
(13.75)
2mr2

1046
acts. For energy
~2 k 2
E= (13.76)
2m
the classical turning radius is
p
`(` + 1)
rcl = (13.77)
k
For r < rcl , the wave function falls off exponentially. If rcl > a, then the particle
feels nothing from the potential, which says that the particle is scattered only
if p
rcl a or `(` + 1) ` ka (13.78)
Note that there are interference effects between different partial waves d/d
but not in .

13.1.4 The Optical Theorem

We earlier derived the relations
1X
f () = (2` + 1)P` (cos )ei` sin ` (13.79)
k
`

4 X
= (2` + 1) sin2 ` (13.80)
k2
`

These lead to another important result.

1X
Imag(f ()) = (2` + 1)P` (cos ) sin2 ` (13.81)
k
`
1X 1X
Imag(f (0)) = (2` + 1)P` (1) sin2 ` = (2` + 1) sin2 ` (13.82)
k k
` `

or
4
= Imag(f (0)) (13.83)
k
which is the optical theorem.

13.1.5 Born Approximation

In some special cases, i.e., special potential functions, large ` or sma;; k for ` 1,
we can find an approximate analytic expression for ell . In general, however, we
must solve the problem numerically.

One such analytical method goes as follows.

1047
We have the general relations
Z
m ~0 0
f~k (, ) = 2
d3~r 0 eik ~r V (~r 0 )~k (~r 0 ) (13.84)
2~
X
~k (~r) = i` (2` + 1)P` (cos )R` (r) (13.85)
`=0

Using the relation

X
`
~ 0 ~
r0
X
eik = 4 (1)` Y `m (~k 0 )j` (kr)Y`m (~r 0 ) (13.86)
`=0 m=`

we get
Z X
`
2m X
f~k (, ) = d3~r0 (1)` Y `m (~k 0 )j` (kr 0 )Y`m (~r 0 )V (~r 0 )
~2
`=0 m=`

X 0
i` (2`0 + 1)P`0 (cos )R`0 (r) (13.87)
`0 =0

Now
Z Z2 Z1 Z
3 0
d ~r d~r 0 d(cos ~r 0 ) r 02 dr 0 (13.88)
0 1 0

and

~r 0 integration m0
~r 0 integration ``0

This implies (using ~k 0 = ) that we get

Z
2m X
f () = 2 (2` + 1)P` (cos ) r2 drV (r)j` (kr)R` (r) (13.89)
~
`=0 0

But we also have, in general, that

1X
f () = (2` + 1)P` (cos )ei` sin ` (13.90)
k
`

which then says that

Z
i` 2mk
e sin ` = 2 r2 drV (r)j` (kr)R` (r) (13.91)
~
0

1048
If we suppose that the potential is weak for a particular partial wave, then we
have very small scattering effects on R` (r). We then have

R` (r) j` (kr) (13.92)

and
ei` sin ` ` (13.93)
or
Z
2mk
` = 2 r2 drV (r)j`2 (kr) (13.94)
~
0

This is the so-called Born Approximation for the phase shift ` .

When the potential has a small effect on all partial waves we can derive another
form of the Born approximation. In this case, we assume the scattered wave is
small compared to the incident wave. This implies that
~
~k (~r) eik~r (13.95)

and
Z
m ~ ~0 0
f () = d3~r 0 ei(kk )~r V (~r 0 ) (13.96)
2~2
m
= V~ 0 ~ (13.97)
2~2 k k
where
V~k 0 ~k = Fourier transform of V (~r) (13.98)

Now we have elastic scattering (|~k| = k = |~k 0 |)which implies that

 
~q = ~k ~k 0  (13.99)
 
 2
q 2 = ~k ~k 0  = k 2 + k 02 2~k ~k 0 (13.100)
 

= 2k 2 2k 2 cos = 2k 2 (1 cos ) (13.101)

 2

= 2k sin (13.102)
2

where is the scattering angle (angle between ~k and ~k 0 ) as shown in Figure

13.3 below.

1049
 Figure 13.3: Elastic Scattering Angles

When is the Born approximation valid?

For short range potentials, the wave function ~k (~r) only contributes to the
scattered wave at small distances. We can therefore use a small distance ap-
proximation for ~k (~r)
Z ikr 0
i~ m 3 0e ~ 0 ~
~k (~r) e k~
r
d ~r V (r0 )eik~r eik~r (13.103)
2~ r0
only if
   
 m Z ikr 0   m Z ikr 0 0
3 0e 0 i~ r 0
k~ 02 0e 0 sin kr 
d ~r V (r )e = r dr V (r ) 1
 
 2~ r0   2~ r0 kr 0 


This corresponds to weak potentials and large k (or large energy).

13.1.6 A General Property of Phase Shifts

Let us assume the potential
(
0 r>b
V (r) = (13.104)
v(r) r b

The general results we will obtain are valid as long as r2 V (r) 0 as r .

For r > b, we have

1
`> (r) = h` (kr) + e2i` h` (kr)

(13.105)
2
For r < b, we must solve the Schrodinger equation with ? present. We assume
that has been done and we obtained the solution `< (r).

At r = b, both the wave function and its derivative must be continuous, which
implies that the logarithmic derivative is continuous
 <   > 
d` (r) d` (r)
dr dr
r=b r=b
` = = (13.106)
`< (b) `> (b)

1050
or using our earlier expression for `> (r) we have

 !
r h` (kr) + e2i` h` (kr)
` = (13.107)
h` (kr) + e2i` h` (kr)
r=b

Since h` (kr) = j` (kr) + i` (kr), this gives

 dj (kr) 
2 `
dr j
` ` (kr) 2i
e2i` 1 = = (13.108)
` h` (kr) dh`dr(kr) cot ` 1
r=b

or !
d` (kr)
dr ` ` (kr)
cot ` = dj` (kr)
(13.109)
dr ` j` (kr) r=b

13.1.7 Examples
Scattering from a Hard Sphere
In this case we have (
0 r>b
V (r) = (13.110)
rb
This is a repulsive potential. It implies that

`< (r) = 0 `< (b) = `> (b) = 0 (13.111)

and
` = (13.112)
This gives
` (kb)
cot ` = (13.113)
j` (kb)
For ` = 0 we have
0 (kb) cos kb
cot 0 = = = cot kb (13.114)
j0 (kb) sin kb
or
0 = kb (13.115)
therefore, for a repulsive potential 0 < 0.

What about ` ? Consider low energy or small k. We then have for kb << 1,
(2`1)!!
` (kb) (kb)`+1
cot ` = (kb)`
= (kb)2`1 (2` 1)!!(2` + 1)!! (13.116)
j` (kb)
(2`+1)!!

1051
where
(2` + 1)!! = 1 3 5 (2` + 1) (13.117)
Therefore, we have
sin ` (kb)2`+1 (13.118)
Thus, we can neglect all terms where ` 6= 0 in the low energy limit and write
1 1
f () = P0 (cos )ei0 sin 0 = ei0 sin 0 (13.119)
k k
d 2 sin2 0
= |f ()| = = b2 (13.120)
d k2
sin2 0
= 4 = 4b2 (13.121)
k2
These results for the hard sphere show that, for low energy, the total cross
section is
4 ( classical scattering cross section) (13.122)
due to quantum interference effects and the scattering is isotropic.

Scattering from a Potential Well

In this case (
0 r>b
V (r) = (13.123)
V0 rb
The solution for positive energy inside the well is
`< (r) = j` (qr) (13.124)
where r
2m
q= (V0 + E) (13.125)
~2
Therefore, the boundary conditions at the well edge give
dj (qr)
!
`
dr
` = (13.126)
j` (qr)
r=b

Now
sin sin cos
j0 () = , j1 () = 2 (13.127)

cos cos sin
0 () = , 1 () = 2 (13.128)

Therefore,
1
0 = q cot qb
b
0 b = qb cot qb 1

1052
which gives
d0 (kr)
!
dr 0 0 (kr) kb sin kb + qb cot qb cos kb
cot 0 = dj0 (kr)
= (13.129)
0 j0 (kr) kb cos kb qb cot qb sin kb
dr r=b

For kb << 1 we get

qb cot qb
cot 0 = (13.130)
kb(1 qb cot qb)
or
kb0 b k0 b2
tan 0 = = (13.131)
qb cot qb 1 + 0 b
In a similar manner, we can derive

(kb)3 1 1 b
tan 1 = (13.132)
3 2 + 1 b

For low energy (k 0) and a deep well (V0 large) we get

 2
tan qb
= 4b2 1 (13.133)
qb

This calculation fails if either

1 + 0 b = 0 or 2 + 1 b = 0 (13.134)

which corresponds to the ` = 0 or the ` = 1 partial wave being in resonance

with the potential.

In general, this resonance condition is given by

` + 1 + ` b = 0 (13.135)

which implies that

 
1
cot ` = 0 tan ` = ` (k) = n+ (13.136)
2

and thus the partial wave scattering cross section takes on its maximum value,
namely,
4(2` + 1)
` (k) = (13.137)
k2
For other k values (off-resonance) we have

` (kb)2`+1 (13.138)

This says that the resonance is very sharp for large ` values.

1053
We can understand this effect better by looking at other features. As we saw,
for small k, cot ` is very large since it is proportional to k 2`1 . Therefore,
4 4 1
` = (2` + 1) sin2 ` = 2 (2` + 1) (13.139)
k2 k 1 + cot2 `
will be proportional to k 4` . In this case, except for the s-wave (` = 0), ` will
be small.

However, near a resonance energy Er where

` + 1 + b` (Er ) = 0 (13.140)
cot ` goes to zero and ` will be proportional to k 2 and this is quite large (an
effect that is called resonance).

Near Er we can make a Taylor expansion and write

 
`
` + 1 + b` (Er ) ` + 1 + b` (Er ) + (E Er )b + ....
E E=Er
 
`
(E Er )b (13.141)
E E=Er

and
` b` (Er ) 2` + 1 (13.142)
After some algebra, these relations imply that
2(E Er )
cot ` (13.143)
k
where
2k 2`+1 b2`
k = 2 `

(13.144)
[(2` 1)!!] E E=Er

Now ` /E < 0, which implies that k > 0. Therefore, near resonance, the
partial wave cross section ` is
2
4 k2k
` = (13.145)
4(E Er )2 + 2k
When
k
E Er = (13.146)
2
the partial wave cross section is equal to one-half its maximum value. Therefore,
= k (Er ) is the width of the resonance at half-maximum.

This functional form is called the Breit-Wigner formula for a scattering reso-
nance. The curves are shown in Figure 13.4 below.

1054
 Figure 13.4: ` versus E ; ` versus E

which are standard resonance type curves.

13.1.8 Born Approximation Examples

Yukawa Potential
In this case
er
V (r) = a (13.147)
r
Although this potential is nonzero out to infinity, it is effectively short-range
because it falls off exponentially. The effective range is usually given as 1/.

The Fourier transform of this potential is easily calculated to be

4a
V (~k ~k 0 ) =  (13.148)
~ ~ 0 2

k k  + 2

where
 2
~ ~ 0 2
 
k k  = 2k sin (13.149)
2

1055
This gives
2
2 2
d m  ~ ~ 0 2 m 4a
 
= V (k k ) =

2
d (2~2 )2 2
(2~ )2 
~ ~ 0 
k k  + 2
a2
= 2 (13.150)
~2 2
4Ek sin2 2 + 2m

By pure coincidence, if we let 0 so that the Yukawa potential turns into

the Coulomb potential we get the result
d a2
= (13.151)
d 16Ek2 sin4
2

which is the correct Rutherford cross section!

This is an accident and higher order terms will completely change this result.
The higher order terms that we have neglected in the Born approximation can-
not be ignored for a long-range potential like the Coulomb potential.

Later we will show how to handle Coulomb scattering correctly.

Scattering from Two Delta Functions

Here we have a free particle with momentum p = ~k parallel to the zaxis
scattering from the potential
V (~r) = v0 [(~r z) (~r + z)] (13.152)

Using ~q = ~k ~k 0 as defined earlier, we have, in the Born approximation,

Z
m mv0 i~qz
d3~r V (~r)ei~q~r = ei~qz


f () = 2 2
e
2~ 2~
mv0
= i sin |qz | (13.153)
~2
where

qz = q sin = 2k sin2 (13.154)
2 2
is the projection of ~q on the zaxis.

The scattering cross section is

d 2 m2 v 2
= |f ()| = 2 40 sin2 qz
d  ~
m2 v02


= 2 4 sin 2k sin2
2
(13.155)
~ 2
which corresponds to the standard two-source interference pattern as expected.

1056
Spin Dependent Potential
We now consider the scattering of a particle of mass m and momentum p = ~k
through an angle by the spin-dependent potential
2
V (~r) = er [A + B~ ~r] (13.156)

where
~ = (x , y , z ) (13.157)
We assume that the initial spin is along the incident direction and we sum
over all final spins, which corresponds to measuring only the direction of the
scattered particles and not their spins. Note that this potential violates parity.

We use the Born approximation. The Fourier transform of the potential function
is given by Z
V (~q) = d3~r V (~r)ei~q~r (13.158)

where
~q = ~k ~k 0 (13.159)
We get
Z
2
V (~q) = d3~r er [A + B~ ~r] ei~q~r
 3/2  
q2 iB
= e 4 A+ ~ ~q
2
 3/2  
q2 iB
= e 4 A+ [q+ + q + + qz z ] (13.160)
2
where
1
q = qx iqy and = (x iy ) (13.161)
2
Since the initial spin is oriented along ~k, which we choose to be in the zdirection,
we also quantize the final spin along the same axis (use the same representa-
tion).

How do the different terms scatter spins?

1. The term A is spin-independent. This implies that the final spin state is
the same as the initial spin state and the value of this term is A.
2. The term Bqz z is a diagonal operator in this representation. This implies
that the final spin state is the same as the initial spin state and the value
of this term is Bqz . We only get the plus sign because we assumed spin
up in the initial state.
3. The term Bq+ flips the spin from +z to z. This implies the value is
Bq+ in a final state with the spin flipped(lowered).

1057
 4. The term Bq + gives a zero contribution since the initial spin cannot be
raised.

Therefore, if the final state is |+zi we get a contribution

i
A+ Bqz (13.162)
2

and if the final state is |zi we get a contribution

i
B(qx + iqy ) (13.163)
2

To get the total contribution to f () we must square each separate term (these
are not indistinguishable processes) and then sum over the final states. We get
 2  2 2 2
A + i Bqz  +  i B(qx + iqy ) = A2 + B q
   
(13.164)
 2   2  42

Therefore, the differential cross section is

" #
~2 B 2 (~k ~k 0 )2
Z 
d 02 0 (~ k 0 )2 /2
k~
k 2 k 02 2


= k dk e A +
d 4~v3 2m 42
2 m2 k2 (1cos )/2 2 B 2 k 2 (1 cos )
 
= 4 3e A + (13.165)
4~ 22

Scattering from Atomic Electrons

The atomic potential seen by an incoming electron can be represented by the

function
T (~r 0 )d3~r 0
Z
V (r) = Ze2 (13.166)
|~r ~r 0 |

where
T (~r 0 ) = nuclear (~r 0 ) + electronic (~r 0 )
= (~r 0 ) (~r 0 )

We then have

mZe2 T (~r 0 )ei~q~r 3 3 0

Z Z
f () = d ~rd ~r
2~2 |~r ~r 0 |
Z i~q(~r~r 0 )
mZe2
Z
0 i~ r0 3 0
q ~ e
= T (~
r )e d ~
r d3~r (13.167)
2~ 2 |~r ~r 0 |

1058
In the last integral ~r 0 is a constant. Therefore we can write
0
ei~q(~r~r ) 3 ei~q~x 3
Z Z
d ~r = d ~x
|~r ~r 0 | x
Z2 Z Z
= lim d sin d ex eiqx cos xdx
0
0 0 0
Z Z0
= 2 lim ex xdx d(cos )eiqx cos
0
0
Z
eiqx eiqx
 
x
= 2 lim e xdx
0 iqx
0
Z
4
= lim ex sin qxdx
q 0
0
4 q 4
= lim 2 2
= 2 (13.168)
q 0 q + q

This trick of introducing the integrating factor ex is a very useful procedure

for evaluating this type of integral. We then have

2mZe2
Z
0
f () = 2 2
T (~r 0 )ei~q~r d3~r 0
~ q
2mZe2
Z
0
= 2 2
[(~r 0 ) (~r 0 )] ei~q~r d3~r 0
~ q
2mZe2
= [1 F (q)] (13.169)
~2 q 2

where
Z
0
F (q) = (~r 0 )ei~q~r d3~r 0 (13.170)

= the atomic scattering function energy of the nuclei

It is a measure of the amount of shielding of the nuclear charge by the electrons

in the atom. It is the Fourier transform of the electron charge density.

13.2 Molecular Physics

13.2.1 General Properties of Molecules
Molecules are made up of nuclei and electrons. Since they break up into atoms
if enough energy is supplied, we can regard them as bound states of atoms.

1059
The determination of the electronic energy levels of molecules is significantly
more complicated than in atomic systems since the potential felt by the electrons
does not exhibit many of the simplifying features of the atomic case.

On the other hand, nuclei are very massive compared to the electrons
me
103 105 (13.171)
mN
and this can be used to simplify our work.

First, we must determine to what extent we need to take nuclear motion into
account in our calculations. On the average, the nuclei move more slowly than
the electron, have a small zero point energy and are well localized. We can pic-
ture the nuclei in a molecule as having classical equilibrium positions (minimum
potential energy points) and that they oscillate slowly about these positions. In
the meantime, the electrons are moving rapidly through the nuclear potentials.
The electrons effectively see a static potential and the electronic wave functions
can only be adiabatically changed by the slow nuclear vibrations. The nuclei
have translational, vibrational and rotational motions that we must take into
account.

What are the typical energies associated with these motions?

In a molecule of size a, the typical momentum is of order ~/a (due to the

uncertainty principle). This implies that
Eel = typical electronic energy
p2 ~2
= a few eV
2m 2ma2
To estimate the nuclear vibrational motion relative to the equilibrium positions,
we assume a harmonic oscillator potential. The potential energy associated with
each mode of vibration is then
1
M 2 R2 (13.172)
2
where
M = the nuclear mass
R = amplitude of the oscillations
= vibrational frequency
For R = a, the vibrational energy is approximately the same as the electronic
energy. We see this as follows. We have
1 ~2
M 2 a2
2 2ma2
 m 1/2 ~2

M ma2

1060
Therefore, the total vibrational energy is
 m 1/2
Evib = ~ Eel 0.01 0.1 eV (13.173)
M
The rotational energy for an angular momentum `~ is

~2 `(` + 1) ~2 m
Erot = Eel (13.174)
2I M a2 M
Therefore, we have
Emol = Eel + Evib + Erot (13.175)
where the relative sizes are given by
 m 1/2 m
Eel : Evib : Erot = 1 : : (13.176)
M M
In this model of a molecule we will use

H = Te + TN + Vee + VeN + VN N (13.177)

where
2
X p~op,i
Te = = kinetic energy of the electrons
2m
electrons
i
2
X p~op,
TN = = kinetic energy of the nuclei
2m
nuclei

Vee = repulsive electron - electron Coulomb interaction

VeN = attractive electron - nucleus Coulomb interaction
VN N = repulsive nucleus - nucleus Coulomb interaction

Since the nuclear kinetic energy is proportional to 1/M , it is relatively small

and we can treat it as a perturbation using the small perturbation parameter
 m 1/4
(13.178)
M

Thus, we start the calculation by neglecting TN (neglecting the nuclear motion).

In this approximation, the nuclear coordinates R ~ are fixed, which implies that
they are no longer dynamical variables (observables) but only parameters. The
Schrodinger equation is then

~ = (n (R)
(Te + Vee + VeN )n (~r, R) ~ VN N (R))
~ n (~
~
r, R) (13.179)
~ n (~r, R)
= Eel (R) ~ (13.180)

1061
Here we have defined

~ = NN interaction energy + the electronic energy eigenvalue

n (R)
~ + VN N (R)
= Eel (R) ~
~ = wave function of the electrons (remember R
n (~r, R) ~ is fixed)
~r represents the coordinates of all the electrons

Now the wave function for the whole molecule satisfies the Schrodinger equation

~ = E(~r, R)
(Te + TN + Vee + VeN + VN N )(~r, R) ~ (13.181)

~ form a complete set of basis functions. Therefore we can

The solutions n (~r, R)
write

X
~ =
(~r, R) ~ n (~r, R)
n (R) ~ (13.182)
n

~ = expansion coefficients. Our task is to find the n (R).

where the n (R) ~

Substituting we get

X X
~ m (~r, R)
(Te + TN + Vee + VeN + VN N )m (R) ~ =E ~ m (~r, R)
m (R) ~
m m
X X
~ m (~r, R)
(TN + VN N )m (R) ~ + ~ m (~r, R)
(Te + Vee + VeN )m (R) ~
m m
X
=E ~ m (~r, R)
m (R) ~
m
X X
~ m (~r, R)
(TN + VN N )m (R) ~ + ~ VN N (R))
(m (R) ~ ~
m (R)m (~
~
r, R)
m m
X
=E ~ m (~r, R)
m (R) ~
m
X X
~
(TN + m (R)) ~
m (R)m (~
~ =E
r, R) ~ m (~r, R)
m (R) ~
m m

1062
 ~ and integrate over all electron positions
If we multiply on the left by n (~r, R)
using the orthonormality of the m (~r, R)~ we get
XZ
~ TN + m (R))
d3 rn (~r, R)( ~ ~
m (R)m (~
~
r, R)
m
X Z
=E ~
m (R) ~ m (~r, R)
d3 rn (~r, R) ~
m
XZ X
~ TN m (R)
d3 rn (~r, R) ~ m (~r, R)
~ + ~ m (R)
m (R) ~ nm
m m
X
=E ~ nm
m (R)
m
XZ
d 3 ~ TN m (R)
rn (~r, R) ~ m (~r, R)
~ ~ n (R)
+ n (R) ~
m
~
= En (R)

~ and thus it acts on both

Now TN contains 2nd derivatives with respect to the R
~ ~
functions m (R) and m (~r, R). Using

2 () = (2 ) + 2 + (2 ) (13.183)

we have
XZ X ~2
~
d3 rn (~r, R) ~ m (~r, R)
2R~ m (R) ~ + n (R)
~ n (R)
~
m
2M

~
= En (R)
X ~ 2 X Z
~ m (~r, R)
d3 rn (~r, R)[ ~ 2 m (R)
~

2M m
~ m (~r, R)
+ 2m (R) ~ + m (R) ~ 2 m (~r, R)]
~ + n (R)
~ n (R)
~ = En (R)
~
2
X ~ X Z
~
2 m (R) ~ m (~r, R)
d3 rn (~r, R) ~

2M m
X ~2 X Z
+ ~
d3 rn (~r, R)[2 ~
m (R) m (~
~
r, R)

2M m
~ 2 m (~r, R)]
+ m (R) ~ + n (R)
~ n (R)
~ = En (R)
~

X ~2 X
~ nm
2 m (R)

2M m
X ~2 X Z
+ ~
d3 rn (~r, R)[2 ~
m (R) m (~
~ + m (R)
r, R) ~ 2 m (~r, R)]
~

2M m
~ n (R)
+ n (R) ~ = En (R)
~

1063
X ~2
~
2R~ n (R)

2M

X ~2 X Z
+ ~
d3 rn (~r, R)[2 ~
m (R) m (~
~ + m (R)
r, R) ~ 2 m (~r, R)]
~

2M m
~ n (R)
+ n (R) ~ = En (R)
~
~
TN n (R)
X ~2 X Z
+ ~
d3 rn (~r, R)[2 ~
m (R) m (~
~ + m (R)
r, R) ~ 2 m (~r, R)]
~

2M m

~ n (R)
+ n (R) ~ = En (R)
~

or X
~
(TN + n (R)) ~ ~
n (R) = En (R)
~
Anm m (R) (13.184)
m
where
~
Anm m (R)
X ~2 Z
= ~
d3 rn (~r, R)[2 ~
~ m (R) R
~ m (~
~ + m (R)
r, R) ~
~ 2 m (~r, R)]
R ~
R

2M

The Anm term mixes different n and m, which corresponds to different electronic
~
wave functions in (~r, R).

How large is this term? The term

~2   2
~ m ~ 2 n (~r)
2R~ n (R)
2M M 2m ~r
m
electron kinetic energy
M
Since this is very much less than the spacing between different n levels, it implies
that we will have negligible mixing of different n states. Now
~ harmonic oscillator wave function
m (R)
~ R
(R ~ )2 M
0
e 2~

~ M m (R)
~ 0  M m (R)
 
~ R
R~ m (R) ~ R ~
~ ~
where typical nuclear displacement from equilibrium. But
1
M 2 2 ~ (13.185)
2
This implies the first term in Anm is of order
 m 1/2
~ spacing between n- levels (13.186)
M

1064
Therefore we can neglect all of the Anm terms. We thus have
~
(TN + n (R)) ~ ~
n (R) = En (R) (13.187)

This is a simple Schrodinger equation for the n (R)~ , which are the expansion
~ ~
coefficients and therefore we can write down (~r, R) once we know the n (~r, R).

~ is the total electron energy and it has become the effective

The term n (R)
potential energy for the nuclear motion in this approximation.

The effect of the electrons is to couple together the nuclei with rubber bands
~
whose force constants depend on the electronic state. Thus, for each n (~r, R),
~ ~
we have an n (R) which implies a n (R). To lowest order in
 m 1/2
(13.188)
M
we have no mixing of different n-levels.

We can therefore write the stationary states of the molecule as

~ = n (R)
n() (~r, R) ~ n (~r, R)
~ (13.189)

with an energy En of the molecule, where labels different solutions to

~
(TN + n (R)) ~ ~
n (R) = En n (R) (13.190)

This approximation is the so-called Born-Oppenheimer approximation.

13.2.2 The Born-Oppenheimer approximation: A toy ver-

sion
Before proceeding with more complicated examples, we illustrate the Born-
Oppenheimer approximation, which, as we have described, is central to the
physics and chemistry of molecules and solids, by a one-dimensional toy model
that is easily solved.

Review of method and motivation

As we said earlier, molecules and solids are describable in terms of the motion
of the electrons and nuclei of the constituent atoms. Although this description
is immensely complicated, the fact that nuclei are much heavier than electrons
makes it possible to adopt an approach in which the nuclei, in the lowest order
of approximation, are taken to be at rest, and their motion is considered subse-
quently in higher orders.

This toy model is a simplification that makes the solution more transparent and
analytically tractable.

1065
Review of prior derivation

For molecules the underlying Hamiltonian of the system is

X p2 X Z e2 X e2
i
H= +
i
2m ,i
|R ri | i>j |ri rj |
X P2 X Z Z e2
+ + (13.191)

2M > |R R |

where the upper-case letters, R and P, denote the position and momentum
operators of the nuclei which are indexed by Greek letters. Lower-case letters,
r and p, and Latin subscripts are reserved for the electrons; m is the mass of an
electron, M is the mass of the th nucleus, e is the electron charge, and +Z e
is the charge on the th nucleus. The Schrodinger equation for the stationary
states of the molecule is

2 X X Z e2
~
X e2
2i +
2m i ,i
|R ri | i>j |ri rj |
#
X ~2 X Z Z e2
2
+ + ({r}, {R}) = E({r}, {R})

2M > |R R |
(13.192)

The first step in the Born-Oppenheimer approach consists in asserting that the
kinetic energy term associated with the heavy particles may be neglected in the
lowest approximation and their coordinates treated as parameters (and not as
dynamical variables). We write the full Hamiltonian as H = H0 + TR where
H0 is the electronic Hamiltonian in which the nuclear coordinates appear as
parameters. The eigenfunctions and eigenvalues of H0 describe the electrons for
fixed values of the coordinates of the nuclei:

H0 n ({r}; {R}) = n ({R})n ({r}; {R}) (13.193)

We write {R} after a semicolon to indicate that {R} appear only as parameters.
The eigenvalues n ({R}) depend on the particular values of {R} of the nuclei
and yield energy surfaces that are labeled by n, the electronic quantum numbers.
Because the states n ({r}; {R}) are a complete set of orthonormal vectors, we
can use them as a basis to write the full wave function as
X
({r}; {R}) = n ({R})n ({r}; {R}) (13.194)
n

with the understanding that the summation includes integra- tion over con-
tinuum states. If we substitute the expansion (13.194) into Eq. (13.192) and
integrate over the electronic coordinates, we obtain a set of coupled equations

1066
describing the dynamics of the heavy nuclei, namely,

" #
X ~2
2 + n ({R}) E n ({R})

2M
X Z 
3
n ({r}; {R}) m ({r}; {R})i d ri
,m
X Z
m ({R}) + n ({r}; {R})
,m

~2
  
2 3
m ({r}; {R})i d ri }) m ({R (13.195)
2M

where m 6= n. The terms on the left-hand side describe the motion driven by
the kinetic energy operator of the nuclei on a given electronic energy surface
n ({R}), whereas those on the right-hand side represent jumps from one energy
sur- face to another ( n m ) resulting in the coupling of the different electronic
states.

Equation (13.195) represents a formidable set of coupled equations which involve

no approximations. The Born-Oppenheimer approximation asserts that the
terms on the right-hand side of Eq. (13.195) may be neglected. Hence, the
equation for the heavy nuclei, which are confined to the nth electronic energy
surface, becomes

" #
X ~2
2
+ n ({R}) E n ({R}) = 0 (13.196)

2M

Equation (13.196) is tantamount to the statement that the different electronic

energy surfaces n ({R}) do not intersect or come close to each other in the
relevant region of parameter space and are sufficiently distant in energy so that
they are not appreciably connected by the kinetic motion of the nuclei. That
is, the time scale of nuclear motions is slow compared to the movements of
the light electrons or, in other words, the amplitudes of the nuclear motion are
negligible in comparison with the equilibrium internuclear distances. Once this
simplification is made, the motion of the nuclei can be further investigated via
Eq. (13.196) to obtain the vibrations about the minima of n ({R}) and the
rotations of the molecules.

Real molecular systems are very complicated except for simple examples such
as the hydrogen molecular ion H2+ (see below). To illustrate the basic ideas
in a simple setting, we discuss a one-dimensional model of the actual physical
system.

1067
The Toy Model
Imagine a system of two particles: one light (of mass m) with coordinate x and
the other heavy (mass M ) located at X. The particles are confined between
two impenetrable walls a distance L apart, that is, L/2 x, X L/2. We as-
sume that these two particles interact with each other via an attractive Dirac
delta function interaction of strength . The Hamiltonian of the system within
the allowed range is thus
p2 P2
H= + (x X) (13.197)
2m 2M
where p and P are the momenta of the two particles. The stationary states are
given by
~2 2 ~2 2
 
(x X) (x, X) = E(x, X) (13.198)
2m x2 2M X 2
We follow the Born-Oppenheimer approach and first solve for the energy levels
of the light particle, En (X), treating X as a parameter, that is,

~2 2
 
(x X) (x; X) = (x; X) (13.199)
2m x2
or
d2
 
2
+ k (x; X) = (x X)(x; X) (13.200)
dx2
where k 2 2m/~2 and 2m/~2 . The wave function satisfies the
boundary conditions, (x = L/2, X) = 0. Thus for x < X the solution is
A sin k(x + L/2) while for x > X we must have B sin k(L/2 x), where A and
B are constants. The continuity of the wave function at x = X implies that
   
L L
A sin k + X = B sin k X (13.201)
2 2
The first derivative of the wave function must be discontinuous so as to yield
the delta function in Eq. (13.200), whose integral between x = X and x = X +
yields the condition 0 (x = X + ); X) 0 (x = X ; X) = (x = X; X), or
     
L L L
Bk cos k X Ak cos k + X = A sin k +X (13.202)
2 2 2
If we eliminate A and B from Eqs. (13.201) and (13.202), we find the eigenvalue
condition for k (and hence  = ~2 k 2 /2m):
   
L L
k sin kL = sin k + X sin k X (13.203)
2 2
For a given and L, Eq. (13.203) is to be solved for k as a function of X
to determine the energy surfaces (here curves as there is only one parameter

1068
X). We observe that for X = L/2 the right-hand side of the eigenvalue con-
dition, Eq.(13.203) vanishes and sin kL = 0 has the solutions k = n/L and
n = ~2 n2 2 /(2mL)2 . In other words, the infinite repulsion at the walls makes
the wave function of the light particle (and hence the probability of finding it)
zero at X = L/2, and hence it must disregard the presence of the heavy par-
ticle if the latter is located at either of these two places.

As we consider values of X away from X = L/2, the value of k must be

determined by solving Eq. (13.203) and would be less than or equal to n/L
because of the attractive delta function potential. Furthermore, the energy
curves belonging to different n cannot cross each other as X varies, because
such one-dimensional systems cannot possess any degeneracy. Thus we may use
the integer n to label the states corresponding to the motion of the light parti-
cles analogous to the set of quantum numbers used to designate the state of an
electron in a molecule. Moreover, the eigenvalue condition in Eq. (13.203) is
symmetric under the transformation X X, and hence it suffices to examine
the range of X between 0 to L/2.

In the above discussion we have tacitly assumed that  > 0 by taking trigono-
metric functions as the solutions of Eq. (13.200), which generally holds except
for the lowest state (and that too only if is large enough). The condition for
crossover from positive to negative  is easily derived by noting that if  becomes
negative, it must pass through zero and hence we can consider the eigenvalue
condition, Eq. (13.203), for small k to obtain the critical potential strength
L
= c = L2
(13.204)
4 X2
If > c for a range of X values, we must use the hyperbolic solutions of the
Schrodinger equation with || =  = ~2 k 2 /2m, and the eigenvalue condition
becomes
   
L L
k sinh kL = sinh k + X sinh k X (13.205)
2 2
Equation (13.205) admits only one solution for the corresponding k. The reason
is physically transparent. Note that for L ,c 0, which means that as
the walls recede to infinity, the attractive delta function potential no matter
how weak will have a solution with a negative energy eigenvalue (as can be seen
by solving the delta function potential without walls). Hence, in the presence
of the repulsive walls, this energy eigenvalue can be negative provided the walls
are not too close and the potential not too weak.

The nature of the energy surfaces is easily revealed by analyzing k as a function

of X as given by Eq. (13.203). The location of the extrema is obtainable from
dk k sin 2kX
= L
=0 (13.206)
dX sin kL + kL cos kL + X sin 2kX 2 sin kL

1069
which yields 2kX = N (N = 0, 1, 2, ....). For the lowest positive energy state,
n = 1, there will be only one extremum at X = 0 (corresponding to N = 0), and
this energy must be a minimum as can be seen from the following argument. At
X = L/2 we have k = /L, and as we move the heavy particle away from the
wall, the energy and hence k must de- crease because of the attractive nature
of the interaction. Thus for the n = 1 energy surface, X = 0 is a minimum.
For the first excited electronic state, n = 2, we have three extrema at N = 0
and 1. The point X = 0 (corresponding to N = 0) is a maximum, because
the vanishing of the wave function at x = 0 with X =0 (the wave function is
an odd function) implies that the derivative is continuous and hence the energy
corresponds to k = 2/L (as if the light particle disregards the presence of the
heavy one). As the heavy particle moves away from X = 0, the energy must
decrease due to the onset of the nonzero attraction. Thus the n = 2 energy
surface gives rise to a double well. Similarly, it is easy to see that for the second
excited electronic state, there are three minima at X = 0 and X = /kand
two maxima at X = /2k.

The wave functions for the ground and first excited electronic states are shown
in Fig. 13.5

Figure 13.5: Wave function for the light particle when the heavy one is located
at X = L/4 taking = 2/L

with the heavy particle located at X = L/4 and the strength of the delta func-
tion corresponding to = 2/L. We have taken L = 1. Attention is drawn
to the discontinuity in the derivative of the wave function of the light particle

1070
at the location of the heavy particle (indicated by the arrow). The shape and
symmetry of the wave functions are clearly affected by the location of the heavy
particle and the strength of the delta function interaction.

Now that we have solved for the motion of the light particle with the heavy
particle at X, we may, in the spirit of the Born-Oppenheimer approach, turn
our attention to the slower vibrational motion of the heavy particle governed
by the potential surfaces, n (X), obtained from  = ~2 k 2 /2m, where k are so-
lutions (labeled by the quantum number n) of Eq. (13.203). The vibrational
energies may be determined numerically given the values of the relevant param-
eters (m, M, , L). However, these energies can be estimated via the harmonic
approximation by using the curvature of the potential at the minima

d2 (X) ~2 k 2

4k cos 2kX
=
dX 2 2m C
2

2 sin 2kX(sin 2kX + 2kX cos 2kX)
+ (13.207)
C2

where the dimensionless quantity C = kL cos kL+X sin 2kX+(1L/2) sin kL.

For convenience we choose the value = 2/L. For the ground state electronic
configuration, the minimum of the potential (X) = 1 (X) is at X = 0 and the
corresponding solution of Eq. (13.203) yields kL = 2.33 (and cos kL = 0.69).
At the minimum of the potential, the curvature is

d2 (X) 4~2 k 2
= (13.208)
dX 2 mL2 cos kL
leading to the classical angular frequency
r
d2 (X)/dX 2
= (13.209)
M
The ensuing spectra is a ladder with spacing ~, and accordingly, the vibrational
energy levels are approximately given by
s
~2 16m k 2 L2
   
1 1
~ + = + (13.210)
2 2mL2 M | cos kL| 2

where is the vibrational quantum number and takes integer values 0, 1, 2, ... .

To make contact between the toy model and the realistic situation of molecules,
we take L to be of the order of chemical bond lengths (L 1A) and m and
M to be the masses of the electron and the proton, respectively, m = 1030 kg
and M 2000m. The separations between the electronic levels are thus of
(~2 /mL2 ) 6 eV , and the vibrational level spacings are
the order of p
~ ~2 /(mL2 ) m/M 0.1 eV .

1071
In Figures. 13.6 and 13.7 the potential energy surfaces (in units of ~2 /2mL2 )
of the ground and first excited electronic states are plotted against X (in units
of L) for = 2.0/L.

Figure 13.6: Potential energy surface (in units of ~2 /2mL2 of the ground elec-
tronic state (n = 1) plotted against X (in units of L) for = 2.0/L. We have
shown the vibrational bound states for M 2000m. Note the decreasing sep-
aration between the adjacent levels, particularly for higher excitations, which
reflects the anharmonicity of the potential surface.

Figure 13.7: Potential energy surface of the first excited (n = 2) electronic

state for the same parameters as in Figure 13.6. Each energy level is almost
doubly degenerate which undergoes splitting due to tunneling between the wells
through the potential barrier. Splitting, though not discernible for the lower
states, is clearly evident for the highest vibrational pair of levels.

1072
It can be seen that for M = 2000m, the energy separation of the lower vibra-
tional levels is almost 0.26(in units of ~2 /2mL2 ). Thus a quadratic fit to the
potential surface of the electronic ground state is very good at least for the
lower vibrational levels (the agreement is within 5%). Similarly, for the first
excited electronic state (n = 2), we have a double well potential surface and
the curvature at the minima both correspond to an energy separation of 0.56.
Indeed the results of a numerical calculation reveal that the almost equidistant
vibrational level structure is obtained for the low lying levels, but the energy
gaps reduce with increasing vibrational energy as the anharmonicity of the po-
tential surface becomes important. Another feature of the first excited state,
which is typical of a double well potential, is the occurrence of a doublet due to
a degenerate pair of levels corresponding to each well split due to their coupling
from tunneling through the barrier. This split- ting is too small to be seen in
Figure 13.7 for the lower vibrational excitation and barely discernible in the
next to highest vibrational level (shown by the thickening of this line in the
diagram) and is clearly evident in the highest vibrational doublet. Hence, we
see that many features of molecular systems apart from the methodology of the
Born-Oppenheimer adiabatic approximation find a simple illustration in the toy
model that we have discussed.

13.2.3 Examples
Hydrogen Molecular Ion (H2+ )
In the ionized H2 molecule, a single electron moves in the attractive potential
of two protons A and B as shown in Figure 13.8 below.

Figure 13.8: Hydrogen Molecular Ion Configuration

1073
The Hamiltonian, in the Born-Oppenheimer approximation, is

~2 2 e2 e2 e2
H =   +  (13.211)
2m ~r R~ A  ~r R~ B  R
~A R~ B 

We will solve the problem using the variational method.

As a trial function we use a linear combination of hydrogen atom 1s states for

separate protons:
= C [A (~r) B (~r)] (13.212)

where we have chosen the symmetric and antisymmetric combinations only be-
cause the potential is symmetric about the midpoint of the molecule

~A + R
R ~B
(13.213)
2
The wave functions are
1 |~ ~ |
r R A 1 |~ ~ |
r RB
A (~r) = e a and B (~r) = e a (13.214)
a3 a3

The C are normalization factors given by

Z Z
2 2
1= d3 r | (~r)| = C2
d3 r |A (~r) B (~r)|
Z  
2 2
= C d3 r |A (~r)| + |B (~r)| 2A (~r)B (~r)
2

2
= C (2 + 2S(R))

where
 
~ ~ 
R = R A RB  (13.215)
R2
Z  
R R
S(R) = d3 rA (~r)B (~r) = 1 + + 2 e a (13.216)
a 3a

The expectation value of H in these trial states is

D E 1  
H = hA| H |Ai + hB| H |Bi 2 hA| H |Bi (13.217)
2 2S

where Z
hA| H |Ai = d3 rA (~r)HA (~r) (13.218)

and so on.

1074
Now

Z 2 2 2 2
~ e e e
hA| H |Ai = d3 rA (~r) 2   +  A (~r)
2m ~ ~  ~
~r RA  ~r RB  RA RB 
  ~ 

Z 2 2
~ e
= d3 rA (~r) 2   A (~r)
2m ~r R ~ A 

Z 2 Z 2
e e
+ d3 rA (~r)   A (~r) d3 rA (~r)   A (~r)
R~A R ~ B  ~r R
 ~ B 

e2 e2 e2
     
R 2R R 2R
= E1 + 1 1+ e a = E1 + 1+ e a
R R a R a
= hB| H |Bi

where

E1 = ground state energy of a hydrogen atom = 1 Ry (13.219)

In a similar way

~2 2 e2 e2 e2
Z
hA| H |Bi = d3 rA (~r)    +  B (~r)
2m ~r R~ A  
r R~  ~
R ~
R

~ B  A B

e2 e2
   
R R
= E1 + S(R) 1+ e a
R a a

Putting everything together we get

D E
(R) = H (13.220)

e2 e2 e2
       
1 R 2R R R
= E1 + 1+ e a E1 + S(R) 1+ e a
1S R a R a a

A plot of (R) versus R/a is shown in Figure 13.9 below.

The top curve represents (R) and the bottom curve is + (R). It is clear that
+ (R) has a minimum and (R) does not.

Therefore, the symmetric wave function leads to binding between the nuclei and
the antisymmetric wave function does not.

In the region between the nuclei

+ (~r) >> (~r) (13.221)

1075
 Figure 13.9: (R) versus R/a

In fact, (~r) is zero on the bisecting plane. This implies that + (~r) gives
the larger probability for the electron to be between the nuclei. This is the
binding mechanism. In this state the electron has the greatest attraction from
both protons.

This theory implies

+ (~r) minimum nuclear separation = 1.3A

minimum energy = 1.76 eV

1076
The experimental results are

nuclear separation = 1.06A

minimum energy = 2.8 eV

A single electron wave function = C [A (~r) B (~r)] is a molecular orbital.

Our assumption of a linear combination or these atomic orbitals comprises the
LCAO model.

+ (~r) state binding binding orbital

(~r) state E > 0 always antibinding orbital

Hydrogen Molecule (H2 )

We now have two electrons. The Hamiltonian for the two electrons in the H2
molecule is
~2 2 ~2 2 e2 e2
H = 1 2   
2m 2m ~r1 R~ A  ~r1 R~ B 

e2 e2 e2 e2
  + +   (13.222)
~r2 R~ A  ~r2 R~ B  |~r1 ~r2 | R~A R~ B 

There are two different variational approaches to the solution of this problem.

Molecular Orbital Method

We make the assumption that the wave function of the two electrons in the
ground state is

symm (1, 2) = + (1)+ (2)spinsinglet (13.223)

1
= (A (1) + B (1)) (A (2) + B (2)) singlet
2(1 + S(R))

This wave function has a symmetric spatial part and an antisymmetric spin
part so that the total wave function for the two electrons is antisymmetric. The
spatial part is the product of the H2+ molecular wave functions. The spatial
part has been chosen to be a binding orbital since it is maximum in between the
nuclei. Therefore, we expect this spatial wave function to represent the ground
state of the H2 molecule. The choice of the singlet spin function (paired spins)
is solely to insure antisymmetry.

The triplet spin state is symmetric an requires an antisymmetric spatial part,

i.e.,
1
anti (1, 2) = [+ (1) (2) (1)+ (2)] triplet (13.224)
2

1077
This spatial part is an antibinding orbital since it is zero in between the nuclei
and thus has a higher energy than the other wave function.

These assumptions clearly have defects:

1. At small proton separations the wave function is the product of two 1s
hydrogen wave functions instead of approximating a helium wavefunction,
i.e., the system is like helium with Z = 2 at small separations, but our
approximation assumes two electrons in Sstates with Z = 1.
2. At larger proton separations there is little probability that the two elec-
trons will both be near the same proton. If we expand out the spatial part
of the wave function we have
(A (1)A (2) + B (1)B (2)) + (A (1)B (2) + A (2)B (1))
the first term says the two electrons are near the same proton. The last
term implies that one electron is near each proton. The presence of the first
term with the same amplitude as the second term is a poor approximation.
This last argument simply reflects the fact that at large proton separations
H + H is more energetically favorable then p + H
The assumed wave function will still give a reasonable upper bound for the
binding energy since it depends only on the behavior of the system near the
actual nuclear separations, which is in between the above two extremes.

The Heitler-London or Valence Bond Method

In this method we leave out the A (1)A (2) and B (1)B (2) terms. We choose
1
s (1, 2) = p (A (1)B (2) + A (2)B (1)) sin g (13.225)
2(1 + S 2 )
1
t (1, 2) = p (A (1)B (2) A (2)B (1)) trip (13.226)
2(1 S 2 )
In this case, for large separations we have separated hydrogen atoms, which is
the correct physics. These wave functions exhibit the same small separation
problems as in the molecular orbital method, however.

The expectation value of H in these states is

D E
(R) = H

hAB| H |ABi + hBA| H |BAi hAB| H |BAi hBA| H |ABi

=
2 2S 2
hAB| H |ABi hAB| H |BAi
= (13.227)
1 S2

1078
where we have used

hAB| H |ABi = hBA| H |BAi

hAB| H |BAi = hBA| H |ABi

The + sign is the singlet value and the sign is the triplet value. The matrix
elements are given by

Z
hAB| H |ABi = d3 r1 d3 r2 A (1)B (2)HA (1)B (2) (13.228)
Z
hAB| H |BAi = d3 r1 d3 r2 A (1)B (2)HA (2)B (1) (13.229)

The Schrodinger equation for each of the 1S wave functions (A and B ) is of

the form

2 2
~ 21  e  A (~r1 ) = E1 A (~r1 ) (13.230)
2m ~ A 
~r1 R


We therefore get for the matrix elements

hAB| H |ABi = 2E1 + Q (13.231)

2
e
hAB| H |BAi = S 2 (2E1 + ) + Q (13.232)
R

where

Q = total Coulomb energy

Z 2 2 2 2
e e e e
= d3 r1 d3 r2 A
2 2
(1)B (2)   + 
|~r1 ~r2 | ~r R ~   ~  ~
~r2 RA  RA R ~ B 
1 B

2
e2 e2
Z Z  
3 2 e 3 3 2 2
= 2 d r1 A (1) 
  + d r1 d r2 A (1) B (2) +
~r1 R ~ B  |~r1 ~r2 | R
2
~ +e
= VC (R) (13.233)
R

1079
Q = total exchange energy

e2 e2 e2
Z
= d3 r1 d3 r2 A (1)B (2)A (2)B (1)   
|~r1 ~r2 | ~r R ~ ~
~r2 RA 
  
1 B

Z 2
e
= 2 d3 r1 A (1)B (1)  
~ A 
~r1 R
e2 e2
Z  
+ d3 r1 d3 r2 A (1)B (2)A (2)B (1) + S2
|~r1 ~r2 | R
~ + S2 e2
= Vex (R) (13.234)
R
The exchange term is a measure of the overlap of the wave functions, weighted
by the potential energies. It is the result of an interplay between quantum
mechanics and the PEP.

Putting everything together we have

   
e2 2

VC Vex e 2 VC + R Vex + S 2 eR
= 2E1 + + = 2E1 + (13.235)
1 S2 R 1 S2
We have

e2 e2
VC + > 0 always and Vex + S 2 < 0 generally (13.236)
R R
which implies that + < . In fact, one finds that

2E1 has no minimum

+ 2E1 has a minimum 30 eV deep at R = 1.5a0

as shown in Figure 13.10 below.

Figure 13.10: Energy versus R

1080
This implies that the molecule binds in the singlet state but not in the triplet
state. One finds that the strength of the binding is proportional to the overlap
of the two electron states .

The triplet wave function spatial part vanishes if both electrons are in between
the protons implying a smaller overlap and thus no binding.

13.2.4 Vibrational and Rotational Levels of a Molecule

We found earlier that the nuclear wave functions satisfy an equation of the form
~
(TN + n (R)) ~ ~
n (R) = En (R) (13.237)

if we consider a two-atom molecule like HCL, this equation gives the rotational
and vibrational energy levels.

If we denote the relative coordinate of the nuclei by ~r, we have

~2 2
 
+ (r) (~r) = E(~r) (13.238)
2m
where
M1 M2
r = |~r| and m = = reduced mass (13.239)
M1 + M2
We have used the fact that the effective potential depends only on r.

This represents a rotationally symmetric problem (central force field). There-

fore, we assume
(~r) = Rn` (r)Y`m (, ) (13.240)
Substitution gives

~2
 2
~2 `(` + 1)
  
d 2 d
+ + (r) + Rn` (r) = ERn` (r) (13.241)
2m dr2 r dr 2mr2
Defining
~2 `(` + 1)
Vef f (r) = (r) + (13.242)
2mr2
and using
un` (r) = rRn` (r) (13.243)
we have
~2 d2
 
+ Vef f (r) un` (r) = Eun` (r) (13.244)
2m dr2
For small values of `, Vef f (r) has a minimum that depends on ` (say at r = r` ).
In the neighborhood of this minimum, we can expand Vef f (r) in a Taylor series
1
Vef f (r) = Vef f (r` ) + m`2 (r r` )2 + ..... (13.245)
2

1081
where
d2 Vef f
 
m`2 = (13.246)
dr2 r=r`

For small displacements from equilibrium, we have simple harmonic motion.

Letting q = r r` we get

~2 d2 ~2 `(` + 1) 1
 
2 2
+ (r` ) + + m` q un` (r) = Eun` (r) (13.247)
2m dq 2 2mr`2 2

where we have put r = r` (small displacement) in

~2 `(` + 1)
(r) and (13.248)
2mr2
We end up with the harmonic oscillator equation where the energy is

~2 `(` + 1)
E = (r` ) + + ~` (n + 1/2) (13.249)
2I`
with
I` = mr`2 = effective moment of inertia (13.250)
The corresponding stationary states are
  (  2 )
q 1 q
un` = An Hn exp (13.251)
q0` 2 q0`

where  1/2
~
q0` = (13.252)
m`
Although the wave functions do not satisfy the requirement u(r = 0) = 0, the
energy eigenvalues are still reasonably accurate in this approximations, i.e.,
(  2 )
1 r`
un` (0) Hn (0) exp (13.253)
2 q0`

With r`  q0` , u(0) is very small. The energy eigenvalues contain contributions
from

1. the effective electronic energy

2. the rotational energy

3. the vibrational energy

The rotational levels have 0.1 1 cm which corresponds to the far infrared
and microwave regions. The vibrational levels have 2 103 3 103 cm
which corresponds to the infrared region.

1082
13.3 Problems
13.3.1 S-Wave Phase Shift
We wish to find an approximate expression for the s-wave phase shift, 0 , for
scattering of low energy particles from the potential
C
V (r) = , C>0
r4
(a) For low energies, k 0, the radial Schrodinger equation for ` = 0 may be
approximated by (dropping the energy term):
 
1 d 2 d 2mC inside
2 r + 2 4 R`=0 (r) = 0
r dr dr ~ r
By making the transformations

1 i 2mC
R(r) = (r) , r=
r ~ x
show that the radial equation may be solved in terms of Bessel functions.
Find an approximate solution, taking into account behavior at r = 0.
outside
(b) Using the standard procedure of matching
this to R`=0 (r) at r = a
(where a is chosen such that ~a  2mC and ka  1) show that

2mC
0 = k
~
which is independent of a.

13.3.2 Scattering Slow Particles

Determine the total cross section for the scattering of slow particles (ka < 1)
by a potential V (r) = C(r a).

13.3.3 Inverse square scattering

Particles are scattered from the potential
g
V (r) =
r2
where g is a positive constant.
(a) Write the radial wave equation and determine the regular solutions.
(b) Prove that the phase shifts are given by
s
2
1 1 2g
` = `+ `+ + 2
2 2 2 ~

1083
 (c) Find the energy dependence of the differential cross section for a fixed
scattering angle.

(d) Find ` for 2g/~2  1 and show that the differential cross section is

3 g2
 
d
= 2 cot
d 2~ E 2

where E is the energy of the scattered particle.

(e) For the same potential, calculate the differential cross section using the
Born approximation and compare it with the above results. Why did this
happen?

13.3.4 Ramsauer-Townsend Effect

What must V0 a2 be for a 3-dimensional square well potential in order that
the scattering cross section be zero in the limit of zero bombarding energy
(Ramsauer-Townsend effect)?

13.3.5 Scattering from a dipole

Consider an electric dipole consisting of two electric charges e and e at a
mutual distance 2a. Consider also a particle of charge e and mass m with an
incident wave vector ~k perpendicular to the direction of the dipole, i.e., choose
the incident particle along the zaxis ~k = kz and the dipole set along he xaxis
or the charges are at ax.

Calculate the scattering amplitude in Born approximation, Find the directions

at which the differential cross-section is maximum.

13.3.6 Born Approximation Again

(a) Evaluate, in the Born approximation, the differential cross section for the
scattering of a particle of mass m by a delta-function potential V (~r) =
B(~r).

(b) Comment on the angular and velocity dependence.

(c) Find the total cross section.

13.3.7 Translation invariant potential scattering

Show that if the scattering potential has a translation invariance property
~ = V (~r), where R
V (~r + R) ~ is a constant vector, then the Born approxima-
tion scattering vanishes unless ~q R~ = 2n, where n is an integer and ~q is the
momentum transfer. This corresponds to scattering from a lattice. For any vec-
tor R~ of the lattice, the set of vectors ~k that satisfy ~k R
~ = 2n constitutes the

1084
reciprocal lattice. This prove then shows that the scattering amplitude vanishes
unless the momentum transfer ~q is equal to some vector of the reciprocal lattice.
This is the Bragg-Von Laue scattering condition.

13.3.8 ` = 1 hard sphere scattering

Consider the hard sphere potential of the form
(
0 r > r0
V (r) =
r < r0

where kr0  1. Write the radial Schrodinger equation for ` = 1, and show that
the solution for the p-wave scattering is of the form
  
sin(kr) cos(kr)
k1 (r) = rRk1 (r) = A cos(kr) + a + sin(kr)
kr kr

where A and a are constants. Determine 1 (k) from the condition imposed on
k1 (r0 ). Show that in the limit k 0, 1 (k) (kr0 )3 and 1 (k)  0 (k).

13.3.9 Vibrational Energies in a Diatomic Molecule

The nuclei of a diatomic molecule are moving in a potential field given as
      2 
a0 a0 2 ~
Vef f (R) = 2D + J(J + 1)
R R 2R2

Express this potential near its minimum by a harmonic oscillator potential and
determine the vibrational energies of the molecule.

13.3.10 Ammonia Molecule

In the ammonia molecule, N H3 , the three hydrogen atoms lie in a plane at
the vertices of an equilateral triangle. The single nitrogen atom can lie either
above or below the plane containing the hydrogen atoms, but in either case the
nitrogen atom is equidistant from each of the hydrogen atoms (they form an
equilateral tetrahedron). Let us call the state of the ammonia molecule when
the nitrogen atom is above the plane of the hydrogen atoms |1i and let us call
the state of the ammonia molecule when the nitrogen atom is below the plane
of the hydrogen atoms |2i.

How do we determine the energy operator for the ammonia molecule? If these
were the energy eigenstates, they would clearly have the same energy (since
we cannot distinguish them in any way). So diagonal elements of the energy
operator must be equal if we are using the (|1i , |2i) basis. But there is a
small probability that a nitrogen atom above the plane will be found below
the plane and vice versa (called tunneling). So the off-diagonal element of

1085
the energy operator must not be zero, which also reflects the fact that the
above and below states are not energy eigenstates. We therefore arrive with the
following matrix as representing the most general possible energy operator for
the ammonia molecule system:
 
E0 A
H =
A E0

where E0 and A are constants.

(a) Find the eigenvalues and eigenvectors of the energy operator. Label them
as (|Ii , |IIi)

(b) Let the initial state of the ammonia molecule be |Ii, that is, |(0)i = |Ii.
What is |(t)i, the state of the ammonia molecule after some time t?
What is the probability of finding the ammonia molecule in each of its
energy eigenstates? What is the probability of finding the nitrogen atom
above or below the plane of the hydrogen atoms?

(c) Let the initial state of the ammonia molecule be |1i, that is, |(0)i = |1i.
What is |(t)i, the state of the ammonia molecule after some time t?
What is the probability of finding the ammonia molecule in each of its
energy eigenstates? What is the probability of finding the nitrogen atom
above or below the plane of the hydrogen atoms?

13.3.11 Ammonia molecule Redux

Treat the ammonia molecule shown in the figure

Figure 13.11: Ammonia Molecule

as a symmetric rigid rotator. Call the moment of inertia about the zaxis I3
and the moments about the pairs of axes perpendicular to the zaxis I1 .
~ I3 and I1 .
(a) Write down the Hamiltonian of this system in terms of L,
h i
(b) Show that H, Lz = 0

1086
 (c) What are the eigenstates and eigenvalues of the Hamiltonian?
(d) Suppose that at time t = 0 the molecule is in the state
1 1
|(0)i = |0, 0i + |1, 1i
2 2
What is |(t)i?

13.3.12 Molecular Hamiltonian

A molecule consists of three atoms located on the corners of an equilateral
triangle as shown below

Figure 13.12: A Molecule

The eigenstates of the molecule can be written as linear combinations of the

atomic states |i i , i = 1, 2, 3, such that
(
if i = j
hi | H |j i =
if i 6= j

where H is the Hamiltonian and , > 0. Now define an operator R such that

R |1 i = |2 i , R |2 i = |3 i , R |3 i = |1 i

(a) Show that R commutes with H and find the eigenvalues and eigenvectors
of R.
(b) Find the eigenvalues and eigenvectors of H.

13.3.13 Potential Scattering from a 3D Potential Well

A 3D stepwise constant potential is given by
(
V1 0 < |~r| < R1
V (~r) =
V2 R1 < |~r| < R2

and zero outside R2 .

(1) Calculate the differential cross-section in the Born approximation as a
function of the momentum transfer q, where ~q = ~k 0 ~k.

1087
 (2) Verify your expression is correct by showing that it reduces to the result
for the spherical square well when V1 = V2 , i.e., calculate separately the
spherical square well result and set V1 = V2 V0 .

(3) Plot the result of the square well(part(2)) as a function of qR2 over a suf-
ficient region to understand its behavior (i.e., qR2 0 , qR2 1 , qR2 
1). Note and explain any noteworthy features.

(4) Now plot not simply versus q, but versus , 0 < < , for four represen-
tative values of the energy. Use atomic scales: R2 = 3aB , V0 = 1 Ry. You
have to decide on relevant energies to plot; it should be helpful to plot on
the same graph with different line styles or colors.

(5) Return to the potential of part(1). Let R2 = 21/3 R1 , so there is an equal

volume inside R1 and between R1 and R2 . Then set V1 = V2 , this means
the volume integral of the potential vanishes, and also that it has strong
~r dependence (step function). Determine the differential cross-section in
this case. Plot versus energy and angle.

(6) Finally, consider a Gaussian potential that has the same range parameter
and the same volume integral as the simple square well of part(2). Cal-
culate the differential cross-section in this case. Can you identify possible
effects due to the sharp structure (discontinuity) that occurs in only one
of them.

13.3.14 Scattering Electrons on Hydrogen

From measurements of the differential cross section for scattering electrons off
protons (in atomic hydrogen) it was found that the proton had a charge density
given by
(r) = aebr
where a and b are constants.

(a) Find a and b such that the proton charge equals e, the charge on the
electron.

(b) Show that the proton mean square radius is given by

2  12
r = 2
b

1/2
(c) Assuming a reasonable value for r2 calculate a in esu/cm2 .

13.3.15 Greens Function

Consider a particle of mass m which scatters off a potential V (x) in one dimen-
sion.

1088
 (a) Show that the free-particle Greens function for the time-independent
Schrodinger equation with energy E and outgoing-wave boundary con-
ditions is
Z
1 eikx
GE (x) = dk 2 k2
2 E ~2m + i

with a positive infinitesimal.

(b) Write that the equation for the energy eigenfunction corresponding to an
incident wave traveling in the positive xdirection. Using this equation
find the reflection probability in the first Born approximation for the po-
tential (
V0 |x| < a/2
V (x) =
0 |x| > a/2
For what values of E do you expect this to be a good approximation?

13.3.16 Scattering from a Hard Sphere

In this case we have (
0 r>b
V (x) =
rb
which is a repulsive potential. Determine the low energy differential and total
cross sections. Discuss your results.

13.3.17 Scattering from a Potential Well

In this case (
0 r>b
V (x) =
V0 rb
Determine 0 , the total cross section and the existence of resonances.

13.3.18 Scattering from a Yukawa Potential

Use the Born approximation to determine the differential cross section for a
Yukawa potential
er
V (~r) = a
r
Discuss the limit 0.

13.3.19 Born approximation - Spin-Dependent Potential

Use the Born approximation to determine the differential cross section for the
spin-dependent potential
2
V (~r) = er [A + B~ ~r]

1089
13.3.20 Born approximation - Atomic Potential
Use the Born approximation to determine the differential cross section for the
atomic potential seen by an incoming electron, which can be represented by the
function
0
T ( r )d3~r0
Z
V (r) = Ze2
|~r ~r0 |
where
T (~r0 ) = nuclear (~r0 ) + electronic (~r0 ) = (~r0 ) (~r0 )

13.3.21 Lennard-Jones Potential

Consider the Lennard-Jones potential (shown below) used to model the binding
of two atoms into a molecule.

Figure 13.13: Lennard-Jones Potential

It is given by
C12 C6
6
V (r) =
r12 r
(a) Near the minimum r0 , the potential looks harmonic. Including the first
anharmonic correction, show that up to a constant term
1
V (x) = m 2 x2 + x3
2
1/6
where r0 = (2C12 /C6 ) , x = r r0 , m 2 /2 = V 00 (r0 ) and = V 000 (r0 )/6.

p2
Let us write H = H0 + H1 , where H0 = 2m + 12 m 2 x2 and H1 = x3 .
(b) What is the small parameter of the perturbation expansion?
(c) Show that the first energy shift vanishes(use symmetry).
(d) Show that the second energy shift (first nonvanishing correction) is
 2

3 + 3
(a + a ) |ni
2 ~
 
(2) 2m X hm| 
En =
~ nm
m6=n

1090
 (e) Evaluate the matrix elements to show that

2 ~2 (n 2)(n 1)(n) (n + 3)(n + 2)(n + 1) 9n3 9(n + 1)3

 
En(2) = + + +
m3 4 3 3 1 1
2 2
 
~ 15 7
= 3 4 (n + 1/2)2 +
m 4 16

6
(f) Consider carbon C-C bonds take Lennard-Jones parameters C6 = 15.2 eV A
12
and C12 = 2.4 104 eV A . Plot the potential and the energy levels from
the ground to second excited state including the anharmonic shifts.

13.3.22 Covalent Bonds - Diatomic Hydrogen

Consider the simplest neutral molecule, diatomic hydrogen H2 , consisting of
two electrons and two protons.

Figure 13.14: Diatomic Molecule

(a) Classically, where would you put the electrons so that the nuclei are at-
tracted to one another in a bonding configuration? What configuration
maximally repels the nuclei (anti-bonding)?

(b) Consider the two-electron state of this molecule. When the nuclei are far
enough apart, we can construct this state out of atomic orbitals and spins.
Write the two possible states as products of orbital and spin states. Which
is the bonding configuration? Which is the anti-bonding?

(c) Sketch the potential energy seen by the nuclei as a function of the inter-
nuclear separation R for the two different electron configurations. Your
bonding configuration should allow for bound-states of the nuclei to one
another. This is the covalent bond.

13.3.23 Nucleus as sphere of charge - Scattering

To a first approximation, the potential that a charged particle feels from a
hydrogen atom can be thought of as due to a positive point charge at the origin
(the proton) plus a uniform region of negative charge occupying a sphere of
radius a0 (the so-called electron cloud).

1091
(a) Calculate, in the Born approximation, the differential cross section for
scattering a charged particle from the hydrogen atom as modeled above
(neglect the recoil of the hydrogen atom).
(b) What is the form of the differential cross section for low energy? Compare
with the pure Coulomb cross section.

(c) Show that the differential cross section becomes more and more like a
pure Coulomb cross section as the energy of the incident particle increases.
Explain why this happens.

1092
Chapter 14
Some Examples of Quantum Systems

14.1 Coherent and Squeezed States

We have derived relationships between the non-Hermitian operators a and a
and the position and momentum operators
r r
~ m~ +
x = (a+ + a) , p = i (a a) (14.1)
2m 2

Working with the coherent states defined by

a |i = |i (14.2)

we found

1 2 X m
|i = e 2 || |mi (14.3)
m=0 m!

where
 2
  = N = h| Nop |i (14.4)

Let us now derive some important relations.

r
~
h| x |i = h| (a+ + a) |i
2m
r r
~ 2~
= ( + ) = Real() (14.5)
2m m
r
~m
h| p |i = i h| (a+ a) |i
2

r
~m
=i ( ) = 2m~Imag() (14.6)
2

1093
 ~
h| x2 |i = h| (a+ a+ + aa+ + a+ a + aa) |i
2m
~
= h| (2 + + + 2 + 1) |i
2m
2 ~
= h| x |i + (14.7)
2m

~m
h| p2 |i = h| (a+ a+ aa+ a+ a + aa) |i
2
~m
= h| (2 + 2 1) |i
2
2 ~m
= h| p |i + (14.8)
2
Using these relations we have

2 2 ~
(x) = h| x2 |i h| x |i = (14.9)
2m
2 2 ~m
(p) = h| p |i h| p |i = (14.10)
2
Hence
~
xp = (14.11)
2
which says that coherent states are minimum uncertainty states.

Now let us find the differential equation satisfied by hx | i and determine its
solution. We have
r r !
1 2m p 2
hx| a |i = hx | i = hx| x |i
2 ~ i m~
r r  !
1 2m 1 2 d
= x i~ hx | i
2 ~ i m~ dx

or   r
~ d 2~
x + hx | i = hx | i (14.12)
m dx m
which has the solution (check by substitution)
" #
2
(x hxi) i
hx | i = C exp + hpi x (14.13)
4(x)2 ~
r !2
m 2~
= C 0 exp x (14.14)
2~ m

1094
For a fixed oscillator mode, specified by a given value of m, the coherent states
are the manifold of those minimum uncertainty states that have definite values
of
p x and p. If m = 1, then the uncertainties in x and p are both equal to
~/2.

We can construct other minimum uncertainty states with a narrower x, the

so-called squeezed states, for the same oscillator by defining a new set of raising
and lowering operators.
r r
m 0 m 0
   
p + p
b = x + i , b = x i (14.15)
2~ m 0 2~ m 0

where we introduce an arbitrarily chosen positive parameter 0 . Now we have

h i m 0  i i

b, b+ = [x, p] + [p, x]
2~ m 0 m 0
i i
= [x, p] = i~ = 1 (14.16)
~ ~
We also have
r
m 0
 
p
b = x + i
2~ m 0
r r r !
m 0 ~ + i ~m +
= (a + a ) + i (a a)
2~ 2m m 0 2
r r r r !
m 0 ~ m 0 i ~m
= + i a+
2~ 2m 2~ m 0 2
r r r r !
m 0 ~ m 0 i ~m
+ i a
2~ 2m 2~ m 0 2
r r ! r r !
1 0 1 0
= a+ + + a
2 0 2 0
= a + a+ (14.17)

where
r r ! r r !
1 0 1 0
= , = + (14.18)
2 0 2 0

and therefore
b+ = a+ + a (14.19)
since and are real. Algebra also shows that
1 0 0
 

2 2 = + 0 0 +2+2 =1 (14.20)
4

1095
We now invert the transformation. We have

b b+ = 2 a + a+ a+ 2 a = (2 2 )a = a (14.21)

which then implies that

b+ b = a+ (14.22)
We define the eigenstates of the lowering operator b by

b |i = |i (14.23)

These new states are also minimum uncertainty states for x and p. We want to
calculate
2 2 2 2
(x) = x2 hxi , (p) = p2 hpi



(14.24)
Now we have
r
~
a + a+


x =
2m
r  
~
= ( ) b + ( ) b+
2m
r  
~
= ( ) b + b+ (14.25)
2m
and
r
~m
a+ a


p = i
2
r  
~m
=i ( + ) b+ ( + ) b
2
r  
~m
=i ( + ) b+ b (14.26)
2
These equations imply that earlier derivation we did is the same with different
multiplicative factors so that

2 ~ 2
(x) = ( ) (14.27)
2m
2 ~m 2
(p) = ( + ) (14.28)
2
and therefore

2 2 ~2 2 2 ~2 2
2 ~2
(x) (p) = ( ) ( + ) = 2 =
4 4 4

It turns out that the operators a and b are related by a unitary transformation,
i.e.,
b = U aU + (14.29)

1096
where  
2
a a+2


U = exp (14.30)
2
and
e = + (14.31)

Proof:
   
2
U aU + = exp a a+2 a exp a2 a+2 = eB aeB



(14.32)
2 2

Using the identity derived earlier we have

h i 1h h ii 1 h h h iii
eB aeB = a + B, a + B, B, a + B, B, B, a + .... (14.33)
2 6
Now with
2
a a+2


B = (14.34)
2
we have
h i  
a2 , a a+2 , a = a+2 , a = a+
  

B, a =
2 2
h h ii 
B, B, a = a2 , a+ = 2 a

h h h iii2
B, B, B, a = 3 a+

and so on.

Therefore

2 3
U aU + = a + a+ + a + a+ + .......
2 6 
2 4 3 5
  
+
= a 1 + + + .... + a + + + .... (14.35)
2! 4! 3! 5!

If we define

2 4 3 5
   
= 1+ + + .... , = + + + .... (14.36)
2! 4! 3! 5!

so that
e = + (14.37)

we have
U aU + = a + a+ = b (14.38)

1097
Finally, we have
a |coherenti = |coherenti
b |squeezedi = |squeezedi
U a |coherenti = U |coherenti
U aU + U |coherenti = U U + U |coherenti = U |coherenti
bU |coherenti = U |coherenti = U |coherenti
bU |coherenti = U |coherenti

which says that U |coherenti = |squeezedi. Thus, U transforms a coherent

state into a squeezed state.

14.2 Electron in a circular wire

We now consider a loop of thin wire in the shape of a circle of radius R as
in Figure 14.1 below. A constant magnetic field perpendicular to the plane
of the loop produces a magnetic flux passing through the loop. Imagine that
the wire contains only one electron which is free to move. This electron has a
wavefunction () which depends only on the angular coordinate . We neglect
all interactions between the electron spin and the magnetic field as well as all
magnetic fields produced by the electron itself.

Figure 14.1: Circular Wire Configuration

We first determine the energies and energy eigenfunctions for a nonrelativistic

electron of mass m moving on this ring. In particular we want to find out how
ground state energy of the electron depends on the value of the applied magnetic
field in this approximation.

We have
~ = Bez
A , B = constant (14.39)

1098
In cylindrical coordinates (r, , z), we can choose

rB ~ = rB e
Ar = Az = 0 , A = A (14.40)
2 2
The Schrodinger equation for the electron is

1  e ~ 2
p~ A = E , e<0 (14.41)
2m c
We then let
ie ~ r
R
Ad~
= 0 e
~c
~
r (14.42)
~ (equivalent to a gauge transformation).
which should get rid of the effect of A
We have
ie ~ r
R
 e ~  e ~  0 ~c Ad~
p~ A = p~ A e ~r
c c
ie ~ r
 R
~ Ad~ e~
0 ~c ~
= e r A
i c
ie ~ r  ie R ~ 
R
Ad~ ~ ~ Ad~
r e~
0
0 A
~c ~c
= e ~r + e ~r
i i c
ie ~ r
R
Ad~ ~e~ e~
0 + A
~c
=e ~
r A
i c c
ie ~ r ie ~ r
R R
Ad~ ~ Ad~
0 = e ~r p~ 0
~c ~c
= e ~r
i
Similarly,
ie ~ r
R
 e ~ 2 Ad~
p~2 0
~c
p~ A = e ~r (14.43)
c
so that the Schrodinger equation becomes
1 2 0
p~ = E 0 (14.44)
2m
Since the electron is confined to a loop of radius R, we have
ie ~ r ie
R R
Ad~ A Rd ie
0 0
= 0 ()e ~c A R
~c ~c
= () = ()e ~
r = ()e ~
r (14.45)

Therefore we have
1 2 0 ~2 d2 0 () 0 ~2
p~ () = = E () = C 2 0 () (14.46)
2m 2mR2 d2 2mR2 1
which has the solution
2
 
ie i C1 + eBR
0 () = eiC1 () = eiC1 e ~c A R = e 2~c
(14.47)

1099
Now imposing single-valuedness, we have
() = ( + 2) (14.48)
2 2 2
     
i C1 + eBR i C1 + eBR 2i C1 + eBR
e 2~c
=e 2~c
e 2~c
(14.49)
which says that
eBR2
C1 + = n = 0, 1, 2, ...... (14.50)
2~c
or 2
eBR2 ~2 eBR2

C1 = n En = n (14.51)
2~c 2mR2 2~c
If we define 0 = ~c/e = unit of flux, and remembering that the flux through
the loop is = R2 B we have
2
~2


En = n+ (14.52)
2mR2 0
which says that the dependence of En on the external magnetic field B or flux
is parabolic. A plot is shown in Figure 14.2 below.

Figure 14.2: En versus B

Since n is an integer, the ground state energy Eg is given by

2
~2 eBR2


Eg = n (14.53)
2mR2 2~c

1100
where n is the integer nearest to eBR2 /2~c or near
e
= <0 (14.54)
0 c~
Note that n < 0 since e < 0.

Now imagine that we start with the wire in its ground state in the presence of
a magnetic flux . If the magnetic field is turned off determine the current in
the loop. Assume that R = 2 cm and = 0.6 gauss cm2 .

Suppose that we start with a state En which is the ground state. n will remain
the same when the magnetic field is turned off. Therefore, () = Cein and
the electric current is
e~
J~ = ( ) (14.55)
2mi
which follows from
2
J~ + || = 0 (14.56)
t
We then get
e~ 2 ne~
J~ = (in) e = e (14.57)
2mi R mR
Now, if S = the cross-section area of the thin wire, the normalization constant
is Z
2 2 1
d`dS = 2R |C| S = 1 |C| = (14.58)
2RS
Then, Z
I = current = J~ dS ~ = ne~ |C|2 S = ne~ (14.59)
mR 2mR2
where we have assumed that J~ is constant throughout the thin wire.

Since the electron is initially in the ground state, this implies that En is the
minimum energy and we have
e
n = greatest integer not greater than = (14.60)
0 c~
or

greatest integer not greater than1 (14.61)
0
For a macroscopic system we have n  1 and we can use
e ne~ e2
n I= 2
(14.62)
c~ 2mR 4mcR2
For R = 2 cm and = 0.6 gauss cm2 we have (using SI units)
(1.6 1019 )2 (0.6 104 ) 104
I= = 1.1 1014 amp (14.63)
4 2 (2 102 )2 (0.9 1030 )

1101
14.3 Spin-Orbit Coupling in Complex Atoms
Spin-orbit coupling is strictly an internal effect arising from the interaction
between the electron spin and the effective magnetic field due to the apparent
nuclear motion. In analogy with the one-electron atom, we can write for the
N electron atom
N
X
Hso = ~i S
i (~ri )L ~i (14.64)
i=1

where i (~ri ) is defined in a manner similar to the one-electron case, assuming

an effective potential field can be defined for each electron.

For the case of weak spin-orbit, let us use both classical and quantum mechanical
arguments to determine the first-order correction to the energy.

Classical Argument:

Vector Model for Combining Angular Momentum: The L ~ S

~ interaction causes
~ ~
L and S to exert torques on each other via their magnetic moments. This means
that neither of these quantities are independently constants of the motion.

However, if ~external = 0, then J 2 is a constant of the motion, where J~ = L

~ + S.
~

In this model, there exist two extreme orientations of these vectors for a single
electron as shown in Figure 14.3 below.

Figure 14.3: Extreme Orientations

~ and S
and as shown in Figure 14.4 below the L ~ vectors precess about the vector
J~ = L
~ +S~ at the same angular velocity.

1102
 Figure 14.4: Precessing Vectors

The quantum conditions now apply to J 2 and Jz instead of L2 , Lz and S 2 and

Sz separately. This means that we define the state |J, MJ i such that

J2 |J, MJ i = J(J + 1)~2 |J, MJ i (14.65)

Jz |J, MJ i = MJ ~ |J, MJ i (14.66)

where J = |L + S| , |L + S| 1, ......, |L S| and each J value has 2J + 1 MJ

values.
P~ ~
Now consider the interaction term of the form Li Si . In this model
i

~ i precesses rapidly about L

each L ~
~i precesses rapidly about S
each S ~

This means that on the average

~ i = i L
L ~ , ~ i = i S
S ~ (14.67)

so that X X
~i S
L ~i = L
~ S
~ , = i i (14.68)
i i

This means that (effectively)

~ S
~=  ~2 ~ 2 ~ 2 
Hso = L J L S (14.69)
2
as we assumed.

1103
More formally (using the Wigner-Eckart theorem) we have
X
hLSJM | Hso |LSJM i = C (LSML0 MS0 )C(LSML MS )
0 0
ML ,MS ,ML ,MS
0 0
ML +MS =ML +MS

hLSML0 MS0 | Hso |LSML MS i

X
= C (LSML0 MS0 )C(LSML MS )
0 0
ML ,MS ,ML ,MS
0 0
ML +MS =ML +MS
X
hLSML0 MS0 | ~i S
i (r)L ~i |LSML MS i
i
X
= C (LSML0 MS0 )C(LSML MS )

0 0
ML ,MS ,ML ,MS
0 0
ML +MS =ML +MS
X
~ i |LML i hSMS0 | S
i (r) hLML0 | L ~i |SMS i
i
X
= C (LSML0 MS0 )C(LSML MS )
0 0
ML ,MS ,ML ,MS
0 0
ML +MS =ML +MS
X D ED E
~ ~
i (r) L Li L S Si S
i
~ S
hLSML0 MS0 | L ~ |LSML MS i (14.70)

where the last step involves two uses of the Wigner-Eckart theorem. We then
have
X
hLSJM | Hso |LSJM i = C (LSML0 MS0 )C(LSML MS )
0 0
ML ,MS ,ML ,MS
0 0
ML +MS =ML +MS
X
~ S
i (r)i i hLSML0 MS0 | L ~ |LSML MS i
i
X
= C (LSML0 MS0 )C(LSML MS )
0 0
ML ,MS ,ML ,MS
0 0
ML +MS =ML +MS

~ S
hLSML0 MS0 | L ~ |LSML MS i
~ S
= hLSJM | L ~ |LSJM i (14.71)

~ S.
so that effectively, Hso = L ~

We also note that since

 
~ S~ E(LSJ) = A(LS) J(J + 1) L(L + 1) S(S + 1)
Hso = L (14.72)
2

1104
so that
 
J(J + 1) L(L + 1) S(S + 1)
E(L, S, J) E(L, S, J 1) = A(LS)
2
 
(J 1)(J) L(L + 1) S(S + 1)
A(LS)
2
A(LS) 2
J + J J 2 + J = A(LS)J


= (14.73)
2

which is called the Lande interval rule.

14.4 Zeeman Effect in Complex Atoms

The electronic spin and orbital angular momenta in a complex atom give rise to
a magnetic moment that we can write, by analogy with the one-electron atom,
as
N N 
~ i =
X X 
M~ = M ~ i + 2S
L ~i (14.74)
i=1
~ i=1

In an external magnetic field B ~ = Bz z, the total Hamiltonian becomes H =

H0 + Hso + HB , where the term containing B is

~ B
HB = M ~ = Mz Bz (14.75)

Let us now use classical precession arguments to derive an approximate operator

expression for Mz . We will assume both a weak spin-orbit interaction and a weak
magnetic field interaction, but take the spin-orbit interaction to be dominant.
Remember, in the case of weak spin-orbit interaction and a weak magnetic field,
we can use the precession picture of the section 14.3. The same type of vector
diagrams can be used to deal with summations like
X X
~i L
L ~ and ~i S
S ~ (14.76)
i i

By vector addition we have J~ = L~ +S~ with both L

~ and S
~ precessing rapidly
~
about J. Also by vector addition we have

~ + gS S
~= e ~ ~

~ =
~L +
~ S = gL L (L + 2S) (14.77)
2mc

~ is not parallel to J~ since

Clearly,

e
gS = = 2gL (14.78)
mc

1105
In addition, ~ Therefore,
~ is precessing rapidly about J.

J~ ~ + 2S)
B (L ~ J~ B (L~ + 2S)
~ (L~ + S)
~
~
ef f ective = = =
J  ~ J  ~ J
2 2 ~ ~
B L + 2S + 3L S B L2 + 2S 2 + 32 (J 2 L2 S 2 )


= =
~ J
~ J
2 2 2
B 3J L + S J 2 L2 + S 2
 
B
= J = J 1 + (14.79)
~ 2J 2 ~ 2J 2
or
J 2 L2 + S 2
 
B ~

~ ef f ective = J 1+ (14.80)
~ 2J 2
Now we use the Wigner-Eckert theorem.

14.4.1 Method #1: Plausibility Derivation

We have
H = ~ ~
B (14.81)
and
e ~ ~ = e (J~ + S)
~ = GJ~

~=
(L + 2S) (14.82)
2mc 2mc
by the Wigner-Eckart theorem. Therefore,

e ~ ~ ~ ~ e J~ J~ + S ~ J~
GJ~ J~ = (J J + S J) G =
2mc 2mc J~ J~
~ ~ ~ ~ ~ ~
e J(J + 1) + JJ+S2SLL
=
2mc J(J + 1)
 
e J(J + 1) + S(S + 1) L(L + 1)
= 1+ (14.83)
2mc 2J(J + 1)

14.4.2 Method #2: Full Formal Derivation

The Zeeman effect Hamiltonian is given by
!
eB X X eB
Hzeeman = Liz + 2 Siz = (Lz + 2Sz ) (14.84)
2mc i i
2mc

Therefore we need to evaluate the matrix element

hLSJM | (Lz + 2Sz ) |LSJM i

= hLSJM | Jz |LSJM i + hLSJM | Sz |LSJM i
= M ~ + hLSJ kSk LSJi hJM | Jz |JM i
= M ~ (1 + hLSJ kSk LSJi) (14.85)

1106
Now
~ J~ |LSJM i = hLSJ kSk LSJi hLSJM | J~ J~ |LSJM i
hLSJM | S
= J(J + 1)~2 hLSJ kSk LSJi (14.86)

or
~ J~ |LSJM i
hLSJM | S
hLSJ kSk LSJi = (14.87)
J(J + 1)~2
But we have

~ J~ |LSJM i = 1 hLSJM | J 2 + S 2 L2 |LSJM i

hLSJM | S
2
~2
= (J(J + 1) + S(S + 1) L(L + 1)) (14.88)
2
Putting it all together

J(J + 1) + S(S + 1) L(L + 1)

hLSJ kSk LSJi = (14.89)
2J(J + 1)

so that

hLSJM | (Lz + 2Sz ) |LSJM i = M ~(1 + hLSJ kSk LSJi)

 
J(J + 1) + S(S + 1) L(L + 1)
= M~ 1 +
2J(J + 1)
(14.90)

Therefore, the first-order correction to the energy level E(L, S, J) due to the
perturbation HB is
 
eB J(J + 1) + S(S + 1) L(L + 1)
hHzeeman i = M~ 1 +
2mc 2J(J + 1)
= 0 M g(LSJ) (14.91)

where
J(J + 1) + S(S + 1) L(L + 1)
g(LSJ) = 1 + (14.92)
2J(J + 1)
is the so-called Lande g-factor.

14.5 Neutron Interferometry

In the late 1970s several neutron interference experiments which are of funda-
mental importance in quantum mechanics and which settled debates started in
1930s, were carried out by Overhauser and collaborators.

In this section we investigate the effects on a neutron interference pattern of the

1107
gravitational field.

We mentioned some aspects of this type of experiment in Chapter 16. Here we

go into more detail.

We consider an interferometer made of three parallel equally spaced crystalline

silicon strips as shown in Figure 14.5 below.

Figure 14.5: Experimental Setup

The incident neutron beam is assumed to be monochromatic. C2 and C3 are

neutron counters.

For a particular value of the angle of incidence , called the Bragg angle, a plane
wave
inc = ei(~p~rEt)/~ (14.93)
where E is the energy of the neutrons and P~ their momentum, is split by
the crystal into two outgoing waves which are symmetric with respect to the
direction perpendicular to the crystal, as shown in Figure 14.6 below.

Figure 14.6: Splitting Waves

1108
The transmitted wave and the reflected wave have complex amplitudes which
can be written respectively as
= cos , = i sin real (14.94)
so that 0
I = ei(~p~rEt)/~ , II = ei(~p ~rEt)/~ (14.95)
0
where |~
p| = |~
p | since the neutrons scatter elastically on the nuclei of the crystal.
The transmission and reflection coefficients are
2 2
T = || , R = || with T + R = 1 (14.96)
In this interferometer setup the incident neutron beam is horizontal. It is split
by the interferometer into a set of beams, two of which recombine and interfere
at point D. The detectors C2 and C3 count the outgoing neutron fluxes. The
neutron beam velocity corresponds to a de Broglie wavelength = 1.445A and
the neutron mass is M = 1.675 1027 kg.

For calculational simplicity we are using monochromatic plane waves to rep-

resent the neutron beams; they are, however, quasi-monochromatic with finite
extension in directions transverse to the beams.

14.5.1 Neutron Interferences

The measured neutron fluxes are proportional to the intensities of the waves that
reach the counters. We define the intensity of the incoming wave to be 1 (units
are arbitrary). For C2 the beams ABDC2 and ACDC2 interfere. Omitting the
propagation factors, at C2 we have the amplitude
A2 = 2 + 3 = (2 + 2 ) (14.97)
Similarly, for ABDC2 and ACDC3 ,
A3 = 2 2 (14.98)
The intensities at the two counters are then
I2 = R 4R2 T , I3 = 4R2 T (14.99)
Suppose that we create a phase shift of the wave propagating along AC, i.e.,
in C the wave function is multiplied by ei .

The new amplitudes at the detectors are

A2 = 2 ei + 3 = (2 ei + 2 ) , A3 = 2 (1 + ei ) (14.100)
and the new intensities become
I2 = R 2R2 T (1 + cos ) , I3 = 2R2 T (1 + cos ) (14.101)
Note that I2 + I3 does not depend on the phase shift . This is because of the
conservation of the total number of particles arriving at D.

1109
14.5.2 The Gravitational Effect
The phase difference between the beams ACD and ABD is created by rotating
the interferometer by an angle around the direction of the incident beam as
shown in Figure 14.7 (on the left)below.

Figure 14.7: Geometrical Considerations

Now let d be the distance between the silicon strips(we neglect their thickness
in this discussion). We also define L as the side of ABCD and H as its height
as shown in Figure 14.7 (on the right) above. We then have(simple geometry)
that
d
L= , H = 2d sin = Bragg angle (14.102)
cos
Experimentally, the values of d and are d = 3.6 cm and = 22.1 .

For an angle we define the gravitational potential V to be V = 0 along AC

and V = V0 along BD.

Since there is no recoil energy of the silicon atoms to be taken into account, the
neutron total energy (kinetic + potential) is a constant of the motion in all of
the process. The energies are given by

p2 (p p)2
EAC = = EBD = + M gH sin (14.103)
2M 2M
M 2 gH sin
p (14.104)
p

where p is the difference in the neutron momentum.

The velocity 2gH is of order 0.5 m/s and the neutron velocity is

h
v= 2700 m/s (14.105)
M

1110
The change in the velocity v is therefore very small, i.e.,
gH
v = 2 104 m/s for = (14.106)
v 2
Now the gravitational potential varies in exactly the same way along AB and
CD. The neutron state in both cases is a plane wave with momentum p = h/
just before A or C. The same Schrodinger equation is used to determine the wave
function at the end of the segments. This implies that the phases accumulated
along the two segments AB and CD are equal.

When comparing segments AC and BD, the previous reasoning does not apply,
since the initial state of the neutron is not the same for the two segments. The
initial state is eipz/~ for AC and ei(pp)z/~ for BD. After traveling over a
distance L = AC BD, the phase difference between the two paths is

Lp M 2 gd2
= = tan sin (14.107)
~ ~2
The variation with of the experimentally measured intensity I2 in the counter
C2 is shown schematically in Figure 14.8 below (the data does not display a
minimum exactly at = 0 because of calibration difficulties).

Figure 14.8: Variation of Intensity with

From the previous result, we have

2 1 = Ag(sin 2 sin 1 ) (14.108)

1111
where
M 2 d2
A= tan (14.109)
~2
Therefore,
2 1
g= (14.110)
A(sin 2 sin 1 )
In the actual data there are 9 oscillations, i.e., 2 1 = 18 between 1 = 32
and 2 = +24 , which gives g = 9.8 m/s2 . This clearly shows that the neutron
interference effects are directly the result of the difference in the gravitational
potential along two arms of the interferometer.

14.6 The Penning Trap

A Penning trap allows one to confine electrons in a finite spatial region and
then allow the accurate measurement of various properties. It involves the
superposition of a uniform magnetic field B~ directed along the zaxis and a
quadrupole electric field which derives from an electrostatic potential of the
form
= K(2z 2 x2 y 2 ) (14.111)

where K is a positive constant.

An electron of charge q(q > 0) and mass m is placed in such a device. We

~ and its momentum operator by p~. The Hamiltonian
denote its spin operator by S
of the electron in the above superposition of fields is

1  2
H = ~ r) + V (~r) + (1 + a) q S
p~ + q A(~ ~ B
~ (14.112)
2m m
V (~r) = = m02 (2z 2 x2 y 2 )/4 (14.113)
= electrostatic potential energy (14.114)
~ r) = B
A(~ ~ ~r/2 (14.115)
= vector potential (14.116)

The constant a 1.16 103 is the gyromagnetic anomaly of the electron

magnetic moment.

14.6.1 Motion of an Electron in a Penning Trap

We set c = qB/m, where B is the magnitude of the magnetic field, and we
assume that this cyclotron frequency c is much larger than 0 .

~ r) = A(~
We note that p~ A(~ ~ r) p~ = L
~ B/2
~ = Lz B/2 and A
~ 2 = B 2 (x2 + y 2 )/4.

1112
We then get

H = Hz + Ht + Hs (14.117)
p2z 1
Hz = + m02 z 2 (14.118)
2m 2
p2 p2y 1 1
Ht = x + + m2 (x2 + y 2 ) + c Lz (14.119)
2m 2m 2 2
Hs = (1 + a)c Sz (14.120)

~ is the orbital angular

with 2 = (c2 202 )/4 ' c /2 02 /2c and L
momentum of the electron.

The eigenstates of Hs are the eigenstates of Sz , |i with energy eigenvalues

(1 + a)~c /2 = ~s , s = (1 + a)c /2 (14.121)

where we have used

Sz |i = ~/2 |i (14.122)

Since Hz , Ht and Hs act on different variables, they must commute. An eigen-

basis of H can be constructed using the eigenstates (z), (x, y) and |i of
Hz , Ht and Hs , respectively. The corresponding eigenvalues are the sum of the
individual eigenvalues.

In order to calculate the motion along the zaxis, we introduce the creation
and annihilation operators
   
1 i 1 i p
az = z + pz , a+
z = z pz , = m~0
2 ~ 2 ~

We then have
 
 +
 1 2 i i 1
az , az = [z, z] + [pz , z] [z, pz ] 2 2 [pz , pz ]
2 ~ ~ ~
 
1 i i
= + [i~] [i~] = 1 (14.123)
2 ~ ~

Thus, we have the same mathematical system as the harmonic oscillator so that

Hz = ~0 (Nz + 1/2) , Nz = a+
z az (14.124)
Nz |Nz i = Nz |Nz i , Nz = 0, 1, 2, 3, ...... (14.125)
Hz |Nz i = ENz |Nz i , ENz = ~0 (Nz + 1/2) (14.126)

1113
14.6.2 The Transverse Motion
We now investigate the x y motion governed by the Hamiltonian Ht . If we
define the right- and left-circular creation and annihilation operators
 
1 i
ar = (x iy) + (px ipy ) (14.127)
2 ~
 
1 i
al = (x + iy) + (px + ipy ) (14.128)
2 ~
where is a real constant, then we can show (in same way as above) that
+
ar , a+ , [ar , al ] = 0 = ar , a+
     
r = 1 = al , al l (14.129)

Defining
Nr = a+
r ar , Nl = a+
l al (14.130)
we have
 
1 1 2Lz
Nr = a+
r ar
= 2 2 2 2 2
(x + y ) + 2 2 (px + py ) 2 + (14.131)
4 ~ ~
 
+ 1 2 2 2 1 2 2 2L z
Nl = al al = (x + y ) + 2 2 (px + py ) 2 (14.132)
4 ~ ~
and thus
Lz = xpy y px = ~(Nr Nl ) (14.133)
and  
1 1
Nr + Nl = 2 (x2 + y 2 ) + (p 2
+ p 2
) 1 (14.134)
2 2 ~2 x y

If we define 2 = m/~ we then have

~c
Ht = ~(Nr + Nl + 1) + (Nr Nl ) (14.135)
2
or
Ht = ~c0 (Nr + 1/2) ~m (Nl + 1/2) (14.136)
where
 
c 1
q
c0
= += 2 2
c + c 20 ' c (14.137)
2 2
2
 
c 1
q
m = = c c 20 ' 0 << c
2 2 (14.138)
2 2 2c
Since this is just the difference of two oscillators we have the energy eigenvalues

Ecm = ~c0 (Nc + 1/2) ~m (Nm + 1/2) (14.139)

We also have from earlier

(1 + a)~c /2 = ~s , s = (1 + a)c /2 (14.140)

1114
Thus, the energy eigenvalues of H are

E = ~0 (Nz + 1/2) + ~c0 (Nc + 1/2) ~m (Nm + 1/2) + ~s (14.141)

where Nz , Nc and Nm are integers 0 and = 1.

Note that the magnetron motion Em corresponds to an inverted harmonic oscil-

lator, so that its spectrum has no mower bound in the harmonic approximation
used in the example. Consequently, when the system is coupled to a heat bath
and relaxes towards thermal equilibrium, it should cascade down the ladder
of levels of the magnetron motion, thus increasing the size of the orbit of the
trapped particle in the xyplane. Fortunately, the characteristic time corre-
sponding to the decay of the system in this way is very long, and the electron
can be confined around the center of the trap for a long time.

14.6.3 Measurement of Electron Anomalous Magnetic Mo-

ment
The electric quadrupole field is such that ~0 = 2.58 107 eV . The magnetic
field is B = 5.87 T . The system is placed in liquid helium at 4.2 K. We then
have (using q~/2m = 5.79 105 eV )

~c = 6.8 104 eV ~c0

~m = 4.9 1011 eV

In liquid helium, kT = 3.5 104 eV and the longitudinal and magnetron level
spacings are much smaller than the thermal fluctuations. Thus, a classical
description of these two motions is appropriate. In contrast, a few quanta of
oscillation are thermal excited for the cyclotron motion since kT ~ c. Now,
the electron anomaly is a 0.00116. Therefore we can draw the relative position
of the four energy levels

Nz = 0; Nm = 0; Nc = 0, 1 and = 1 (14.142)

The level configuration is shown in Figure 14.9 below.

Figure 14.9: Energy Levels

1115
The splitting E between the level Nc = 0, = +1 and the level Nc = 1, = 1
is proportional to the anomaly a. We have E = a~c = 5 107 eV , where
we have neglected the difference between c and 0c which is 7.9 1011 eV .
The splitting corresponds to a frequency = E/~ = 191 M Hz.

14.7 Schrodingers Cat

We first just describe a simple version.

Suppose that a cat within a closed box would be killed by a |i particle but not
by a |i particle. Now consider the effect of the state |i + Ket, which can
easily be produced by a properly oriented Stern-Gerlach device.

Suppose that a particle in the state |i + Ket hits the cat and that the state
of the (spin + cat) makes a transition to

|i |dead cati + |i |living cati (14.143)

which is a pure state.

When is it decided whether the cat is alive or dead?

Just when the observer opens the cats box?

An objective statement independent of the conscious

mind of the observer would be impossible.

What is the consequence of including the observer herself in the quantum me-
chanical description?

According to the point of view presented, the cat(together with the mechanism
for killing the cat, which was not mentioned above) is linked to other macro-
scopic objects. These are influenced differently in the two final states so that
their respective wave functions do not overlap. For everything that follows, the
macroscopic consequences are not recorded; the trace is taken over them. The
final state of the cat is described by a mixture of states corresponding to a dead
cat and a living cat: the cat is either dead or living and not in a pure state

|dead cati + |living cati (14.144)

which would include both possibilities.

1116
14.7.1 Schrodingers Cat - a more detailed presentation
The superposition principle states that if |a i and |b i are two possible states
of a quantum system, the quantum superposition
1
(|a i + |b i) (14.145)
2
is also an allowed state for this system. This principle is essential in explain-
ing quantum interference phenomena. However, when it is applied to large or
macroscopic objects, it leads to paradoxical situations where a system can be
in a superposition of states which is classical self-contradictory.

The most famous example is Schrodingers cat paradox where the cat is in a
superposition of the dead and alive states. The purpose of this discussion is to
show that such superposition of macroscopic states are not detectable in prac-
tice. They are extremely fragile, and very weak coupling to the environment
suffices to destroy the quantum superposition of the two states |a i and |b i.

The Quasi-Classical States of a Harmonic Oscillator

We consider the high energy excitations of a one-dimensional harmonic oscillator
or mass m and frequency . The Hamiltonian is written
p2 1
H = + m 2 x2 (14.146)
2m 2
We denote the eigenstates of H by {|ni} where the energy eigenvalues are given
by
H |ni = En |ni = ~(n + 1/2) |ni (14.147)

Preliminaries
We introduce the operators
p
X = m/~x , P = p/ m~ (14.148)

and the annihilation and creation operators

1   1  
a = X + iP , a+ = X iP , N = a+ a (14.149)
2 2

The commutator [x, p] = i~ leads to the commutators [X, P ] = i and [a, a+ ] = 1

and the relations

H = ~(N + 1/2) , N |ni = n |ni (14.150)

We also have the relations

P = i , X = i (14.151)
X P

1117

a |ni = n |n 1i , a+ |ni = n + 1 |n + 1i (14.152)
We can use these relations to derive the ground state wave function in the
position representation as follows:
1  
0 = hX| a |0i = hX| X + iP |0i
2
 
1 i
= X hX | 0i + i hX | 0i
2 2 X
 
2
X+ hX | 0i = 0 hX | 0i = AeX /2 = 0 (X)
X
2
0 (x) = Aemx /2~
(14.153)

Similarly, we can derive its the ground state wave function in the momentum
representation as follows:
1  
0 = hP | a |0i = hP | X + iP |0i
2
1 i
= i hP | 0i + P hP | 0i
2 P 2
 
2
P+ hP | 0i = 0 hP | 0i = AeP /2 = 0 (P )
P
2
0 (p) = Aep /2m~
(14.154)

These two wave functions are related by the Fourier transform, that is,
Z
2 2
0 (p) = ep /2m~
emx /2~ ipx/~
e dx

Z
0 (x)eipx/~ dx

The Quasi-Classical States

The eigenstates of the operator a are called quasi-classical states, for reasons
we will now discuss.

Since we are considering the question: what are the eigenstates of the lowering
operator a? We can write

a |i = |i where = || ei (14.155)

where |i is the eigenvector of a and is the eigenvalue, which is not necessarily

real since a is not Hermitian.

Since the vectors |ni are eigenvectors of a Hermitian operator, they form a

1118
orthonormal complete set and can be used as an orthonormal basis for the
vector space. We can then write

X
|i = bm |mi (14.156)
m=0

where

X
X
hk | i = bm hk | mi = bm km = bk (14.157)
m=0 m=0

Now
hn 1| a |i = hn 1 | i = bn1 (14.158)
and using
a+ |n 1i = n |ni hn 1| a = n hn| (14.159)
we have
hn 1| a |i = n hn | i = nbn (14.160)
or

bn = bn1 (14.161)
n
This says that
2
b1 = b0 , b2 = b1 = b0 (14.162)
1 2 2!
or
n
bn = b0 (14.163)
n!
We thus get the final result

X m
|i = b0 |mi (14.164)
m=0 m!

Let us now normalize this state (choose b0 ). We have

X
2
X m k
h | i = 1 = |b0 | hk | mi
m=0 k=0 m! k!
X 2
2
X m k X ||
= |b0 | km = |b0 |2
m=0 k=0 m! k! m=0
m!
2 2
= |b0 | e|| (14.165)

which says that

1 2
b0 = e 2 || (14.166)
and thus

12 ||2
X m
|i = e |mi (14.167)
m=0 m!

1119
Now

hn | i = probability amplitude that the system in the state

|i will be found in the state |ni

We have

1 2 X m 1 2
n
hn | i = e 2 || hn | mi = e 2 || (14.168)
m=0 m! n!

which then says that

2 2n
e|| ||
2 eN N n
Pn = |hn | i| = =
n! n!
= probability amplitude that the system in the state
|i will be found in the state |ni

where we have defined N = ||2 . We note that

h| a+ a |i = 2  h | i = 2  = N = h| Nop |i
   
(14.169)

or N = the average value or expectation value of the Nop operator in the state
|i. This type of probability distribution is called a Poisson distribution, i.e., the
state |i has the number states or energy eigenstates distributed in a Poisson
manner.

Since the states |ni are energy eigenstates, we know their time dependence, i.e.,
En
|n, ti = ei ~ t
|ni (14.170)

Therefore, we have for the time dependence of the state |i

21 ||2
X m 1 2 X
m Em
|, ti = e |m, ti = e 2 || ei ~ t |mi (14.171)
m=0 m! m=0 m!

This simple operation clearly indicates the fundamental importance of the en-
ergy eigenstates when used as a basis set.

If we are able to expand an arbitrary vector

representing some physical system in the energy
basis, then we immediately know the time dependence
of that state vector and hence we know the time
dependence of all the probabilities associated
with the state vector and the system.

Now let us try to understand the physics contained in the |i state vector. In a

1120
given energy eigenstate the expectation value of the position operator is given
by

r
~
hn, t| x |n, ti = hn, t| (a + a+ ) |n, ti
2m0
r
~ En En
= hn| ei ~ t
(a + a+ )ei ~ t
|ni
2m0
r
~
= hn| (a + a+ ) |ni
2m0

r
~

= hn| n |n 1i + n + 1 |n + 1i = 0
2m0

i.e., it is equal to zero and is a constant.

On the other hand, in the state |i we find

r
~ XX (Em Ek )
h, t| x |, ti = bm bk e i ~ t
hm| (a + a+ ) |ki (14.172)
2m0 m
k

Now
 
hm| (a + a+ ) |ki = hm| k |k 1i + k + 1 |k + 1i

= km,k1 + k + 1m,k+1 (14.173)

Using this result we have

!

r
~ X (Ek1 Ek ) X (Ek+1 Ek )
h, t| x |, ti = bk1 bk ke i ~ t
+ bk+1 bk k + 1e i ~ t
2m0
k=1 k=0

!

r
~ X X
= bk1 bk kei0 t + bk+1 bk k + 1ei0 t
2m0
k=1 k=0

!

r
~ X X
= bk bk+1 ke i0 t
+ bk+1 bk k + 1e i0 t
2m0
k=0 k=0

!
k k+1 i0 t X k+1 k
r
~ X
= b2 ke + k + 1ei0 t
2m0 0
p p
k=0
(k + 1)!k! k=0
(k + 1)!k!
r
~ X 1 2k
b20 ei0 t + ei0 t


= || (14.174)
2m0 k!
k

1121
Now using = ||ei we get
r
X || 2k
~
h, t| x |, ti = b20 2 || Real(ei ei0 t )
2m0 k!
k
X ||2k
= 2x0 || cos(0 t )(b20 )
k!
k
r
~
= 2x0 || cos(0 t ) , x0 = (14.175)
2m0

The expectation value in the state |i behaves like that of a classical oscillator.

Before proceeding with the discussion, we will repeat the derivation using an
alternate but very powerful technique.

Using the Translation Operator

In general, a displaced state |i is given in terms of the displacement operator
(in one dimension) by
i
|i = e ~ p |0i (14.176)
For the harmonic oscillator system
r
1 m~
p = (a a+ ) (14.177)
i 2

If we choose |0i to be the ground state of the oscillator, then we have for the
corresponding displaced ground-state
m +
|i = e 2~ (a a) |0i (14.178)

By Glaubers theorem
e(A+B) = eA eB e 2 [A,B ]
1
(14.179)
we have
m m m
(a+ a) a+ a 1 m
[a+ ,a]2
e 2~ =e 2~ e 2~ e2 2~

m m
a+ a 14 m 2
~
=e 2~ e 2~ e (14.180)

and thus m m
a+ a 41 m 2
~
|i = e 2~ e 2~ e |0i (14.181)
Now
m 2
 r   r !
m 1 m
e 2~ a
|0i = I + a + a + ..... |0i = |0i
2~ 2 2~

1122

using a |0i = 0. Similarly, using (a+ )n |0i = n! |ni we have

m 2
r r
 !
+ m + 1 m
e 2~ a
|0i = I + a + a+ + ..... |0i
2~ 2 2~
r r 2
m 1 m
= |0i + |1i + |2i + ....
2~ 2 2~
p m
n
X
= 2~ |ni (14.182)
n=0 n!

or

p m
n
41 m 2
X
|i = e ~ 2~ |ni (14.183)
n=0 n!

Thus,

X
|i = bn |ni (14.184)
n=0

where

N n
e 2 N 2 N m 2
bn = , = (14.185)
n! 2 4~

or

Pn = probability of finding the system in the state |ni

2 eN N n
= |bn | = (14.186)
n!

which is a Poisson distribution. Thus, we obtain the coherent states once again.

Let us now return to the original discussion. In the state |i we have

a |i = |i h| a+ = h| (14.187)

1123
so that

   
2
hEi = h| H |i = ~ h| N + 1/2) |i = ~ || + 1/2) (14.188)
r r
~ ~
hxi = +
h| (a + a ) |i = ( + ) (14.189)
2m 2m
r r
m~ + m~
hpi = i h| (a a ) |i = i ( ) (14.190)
2 2
~ 2
(x)2 = h| (a + a+ )2 |i hxi
2m
~ h 2
i ~ 2
= ( + ) + 1 ( + )
2m r 2m
~ ~
= x = (14.191)
2m 2m
2 m~ + 2 2
(p) = h| (a a ) |i hpi
2
m~ h 2
i m~
2
= ( ) 1 + ( )
2 r 2
m~ m~
= p = (14.192)
2 2

Therefore, the Heisenberg inequality becomes an equality in this case

~
xp = (14.193)
2

independent of the value of .

We can find the wave functions corresponding to |i using the earlier method.
We have in the position representation:

1  
hX| a |i = hX | i = hX| X + iP |i
2
 
1 i
= X hX | i + i hX | i (14.194)
2 2 X
 
1
X+ hX | i = hX | i
2 X

2)2 /2
hX | i = Ae(X = (X) (14.195)

1124
and in the momentum representation:
1  
hP | a |i = hP | i = hP | X + iP |i
2
1 i
= i hP | i + P hp | i (14.196)
2 P 2
 
1
P+ hP | i = hP | i
2 P

2)2 /2
hP | i = A0 e(P +i = (P ) (14.197)

Suppose that at time t = 0, the oscillator is in a quasi-classical state |(0)i =

|0 i with 0 = ei where is a real positive number. Then at a later time t

1 2 X n
|(t)i = |0 , ti = e 2 |0 | 0 |n, ti
n=0 n!

1 2 X n En
= e 2 |0 | 0 ei ~ t |ni
n=0 n!

1 2 X n
= e 2 |0 | eit/2 0 eint |ni
n=0 n!
= eit/2 |(t)i (14.198)

where (t) = 0 eit = ei(t) .

Finally, we have
r
~ X 1 2k
b20 ei0 t + ei0 t


h, t| x |, ti = ||
2m0 k!
k
X ||2k
r r
~ 2 ~
=2 || cos(0 t )(b0 ) , x0 =
2m0 k! 2m0
k
r
2~
= x0 cos(0 t ) , x0 = (14.199)
m0
and
h, t| p |, ti = p0 sin(0 t ) , p0 = 2m~ (14.200)
In addition, we have (for  1)

x 1 p 1
= << 1 , = << 1 (14.201)
x0 2 p0 2
This says that the relative uncertainties in the position and momentum of the
oscillator are quite accurately defined at any time. Hence the name quasi-
classical state.

1125
Let us look at some numbers. We consider a pendulum of length 1 meter and
of mass 1 gram and assume that the state of this pendulum can be described
by a quasi-classical state. At time t = 0 we assume that the pendulum is
at hx(0)i = 1 micron from its classical equilibrium position, with zero mean
velocity.

An appropriate choice is hx(0)i = x0 , hp(0)i = 0 = 0. We also have

r
g
= 2 = = 3.13 s1 (0) = 3.9 109 (14.202)
`

The relative uncertainty in the position is

x 1 1
= = = 1.3 1010 (14.203)
x0 2 2(0)

We note that after 1/4 period of oscillation,

2
T = period = (14.204)

(T /4) = (0)eiT /4 = (0)ei/2 = i(0) = 3.9i 109 (14.205)

14.7.2 Construction of a Schrodinger-Cat State

Suppose that during the interval [0, T ] we add to the harmonic potential, the
coupling (interaction)

2
W = ~g a+ a = ~g N 2 (14.206)

We will assume that g  , T  1. Under these conditions, we can make the

approximation that, during the interval [0, T ], the Hamiltonian of the system is
simply W . Assume that at time t = 0, the system is in a quasi-classical state
|(0)i = |i.

The eigenvectors of W are {|ni} with W |ni = ~gn2 |ni. This implies that for
|(0)i = |i
n
1 2 X 2
|(T )i = e 2 || eign T |ni (14.207)
n=0 n!

Some special cases will be of interest later.

1 2 X n 2 1 2 X
n
|(T = 2/g)i = e 2 || ei2n |ni = e 2 || |ni = |i
n=0 n! n=0 n!

1 2 X n 2 1 2 X
n
|(T = /g)i = e 2 || ein |ni = e 2 || (1)n |ni = |i
n=0 n! n=0 n!

1126

12 ||2
X n 2
|(T = /2g)i = e ein /2 |ni
n=0 n!

1
X 2 n 1
= e 2 || [1 i + (1 + i)(1)n ] |ni
n=0 n! 2
n
2 X 1 h
1
i
= e 2 || ei/4 + ei/4 (1)n |ni
n=0 n! 2
1 h i
= ei/4 |i + ei/4 |i (14.208)
2
Now, suppose that is pure imaginary, that is, = i. In this case, in the state
|i, the oscillator has a zero mean position and a positive velocity.
r
~
hxi = ( + ) = 0
2m

r
m~
hpi = i ( ) = 2m~
2
Similarly, in the state |i, the oscillator also has a zero mean position, but a
negative velocity.

If ||  1, the states |i and |i are macroscopically different. The state

(14.208) is a quantum superposition of such states. It therefore constitutes a
(harmless) version of Schrodingers cat, where we represent dead and alive cats
by simple vectors in Hilbert space.

14.7.3 Quantum Superposition Versus Statistical Mixture

We now consider the properties of the state (14.208) in a macroscopic
situation
||  1. We will choose = i pure imaginary and we set p0 = 2m~.

The probability distributions for position and momentum are given by

 2
P r(X) ei/4 hX | i + ei/4 hX | i
 
 2 2 2
ei/4 e(Xi 2) /2 + ei/4 e(X+i 2) /2 
 

2
 
eX cos2 2X (14.209)
4

 2
P r(P ) ei/4 hP | i + ei/4 hP | i
 
 2 2 2
ei/4 e(P 2) /2 + ei/4 e(P + 2) /2 
 

2)2 2)2
e(P + e(P + (14.210)

1127
where in the last expression
used the fact that, for  1, the two
we have
Gaussians centered at 2 and 2 have a negligible overlap.

These probability distributions are plotted in Figure 14.10 Figure 14.11 below
for = 5i

Figure 14.10: X Probability Distribution

Figure 14.11: P Probability Distribution

Suppose that a physicist (Alice) prepares N independent systems all in the state
(14.208) and measures the momentum of each of these systems. Suppose the

1128
measuring apparatus has a resolution p such that:

m~ << p << p0 (14.211)

For N  1, the results of these measurements is that Alice (plotting a his-

togram) will find two peaks, each of which contains roughly half of the events,
centered respectively at p0 and p0 (resembling Figure 14.11).

The state (14.208) represents the quantum superposition of two states which
are macroscopically different, and therefore leads to the paradoxical situations
mentioned earlier.

Another physicist (Bob) claims that the measurements done by Alice have not
been performed on N quantum systems in the state (14.208), but that Alice is
actually dealing with a nonparadoxical statistical mixture, that is, half of the N
systems are in the state |i and the other half in the state |i.

Assuming that this is true, the statistical mixture of Bob leads (after N mo-
mentum measurements) to the same momentum distribution as that measured
by Alice: the N/2 oscillators in the state |i all lead to a mean momentum p0
and the N/2 oscillators in the state |i all lead to a mean momentum p0 .
Up to this point, there is therefore no difference and no paradoxical behavior
related to the quantum superposition (14.208).

In order to settle the matter, Alice now measures the position of each of the
N independent systems, all prepared in the state (14.208). Assuming that the
resolution x of the measuring apparatus is such that
r
1 ~ 1 1
x << X << = (14.212)
|| m ||

Alice has sufficient resolution to observe the oscillations of the function

 
cos2 2X (14.213)
4
in the distribution P r(X). The shape of the distribution for x will therefore
reproduce the probability law for X as drawn in Figure 14.10 above, that is, a
modulation of period
 2 1/2
~ /(2m2 ) (14.214)
with a Gaussian envelope.

We continue with the assumption that Bob is dealing with a statistical mixture.
If Bob performs a position measurement on the N/2 systems in the state |i,
he will find a Gaussian distribution corresponding to the probability law
2 2
P r(X) |hX | i| eX (14.215)

1129
He will find the same distribution for N/2 systems in the state |i. The sum
of his results will be a Gaussian distribution, which is quite different(see Figure
14.11) from the result expected by Alice.

The position measurement should, in principle, allow one to discriminate be-

tween the quantum superposition and the statistical mixture.

In our earlier discussion of numbers for a pendulum we found that = 3.9109 .

Therefore, the resolution x which is necessary in order to tell the difference be-
tween a set of N systems in a quantum superposition (14.208), and a statistical
mixture consisting of N/2 systems in the state |i and N/2 systems in the state
|i is given by
r
1 ~
x << 5 1026 m (14.216)
|| m
Clearly, it is impossible to attain such a resolution in practice!

14.7.4 The Fragility of a Quantum Superposition

In a realistic physical situation, one must take into account the coupling of the
oscillator with its environment, in order to estimate how long one can discrimi-
nate between the quantum superposition (14.208), that is, the Schrodinger cat
which is alive and dead, and a simple statistical mixture, that is, a set of cats
(systems), half of which are alive, the other half beginning dead; each cat being
either alive or dead.

If the oscillator is initially in the quasi-classical state |0 i and if the environ-

ment is in a state |e (0)i, the wave function of the total system is the product
of the individual wave functions, and the state vector of the total system can
be written as the (tensor) product of the state vectors of the two subsystems:

|(0)i = |0 i |e (0)i (14.217)

The coupling is responsible for the damping of the oscillators amplitude.

At a later time t, the state vector of the total system becomes

|(t)i = |1 i |e (t)i (14.218)

where 1 = (t)et . The number (t) corresponds to the quasi-classical state

one would find in the absence of damping (evaluated earlier as (t) = 0 eit )
and is a real positive number.

From earlier
   
2 2
E(t) = ~ |(t)| + 1/2) = ~ |0 | e2t + 1/2) (14.219)

1130
The energy decreases with time. After a time much longer than 1/, the os-
cillator is in its ground state. This dissipation model corresponds to a zero
temperature environment. The mean energy acquired by the environment is
2 2
E(0) E(t) = ~ |0 | (1 e2t ) 2~ |0 | t , 2t << 1 (14.220)

For initial states of the Schrodinger cat type for the oscillator, the state vector
of the total system, at t = 0,

1  
|(0)i = ei/4 |0 i + ei/4 |0 i |e (0)i (14.221)
2
and, at a later time t,

1   E  E
|(t)i = ei/4 |1 i (+)
e (t) + e
i/4
|1 i ()
e (t) (14.222)
 
2

still with 1 = (t)et .  We assume

E that
 t isE chosen such that 1 is pure
 (+)  ()
imaginary, |1 |  1, and e (t) and e (t) are two normalized states of
the environment that are a priori different (but not orthogonal).

The probability distribution of the oscillators position, measured independently

of the state of the environment, is then

1 2 1 2
P r(x) = |hx | 1 i| + |hx | 1 i|
2  2 D  E

+ Real i hx | 1 i hx | 1 i (+)
 ()
e (t)  e (t) (14.223)
D  E
(+)  ()
Let = e (t)  e (t) . We then have 0 1, real.

This says that the probability distribution of the position keeps its Gaussian
envelope, but the contrast of the oscillations (cross term) is reduced by a factor
.

The probability distribution for the momentum is given by

1 2 1 2

P r(p) = |hp | 1 i| + |hp | 1 i| + Real i hp | 1 i hp | 1 i (14.224)
2 2
Since the overlap of the two Gaussians hp | 1 i and hp | 1 i is negligible for
|1 |  1, the crossed term, which is proportional to does not contribute
significantly. We recover two peaks centered at |1 | 2m~. The distinction
between a quantum superposition and a statistical mixture can be made by
position measurements. The quantum superposition leads to a modulation of
spatial period
 2 1/2
~ /(2m2 ) (14.225)

1131
with a Gaussian envelope, whereas only the Gaussian is observed for a statistical
mixture.

In order to see this modulation, the parameter must not be too small, say
1/10.

In a very simple model, the environment is represented by a second oscillator, of

the same mass and frequency as the first one. We will assume that this second
oscillator is initially in its ground state |e (0)i = |0i. If the coupling between
the two oscillators is quadratic, we can take for granted that
 E  E
 ()  ()
1. the states e (t) are quasi-classical: e (t) = |i

2. and that, for short times (t  1): ||2 = 2t|0 |2

A simple calculation then gives
2 X n ()n 2 2 2
h | i = e|| = e|| e|| = e2|| (14.226)
n
n!

From earlier considerations we must have

2
= h | i = e2|| 1/10 || 1 (14.227)

For times shorter than 1/, the energy of the first oscillator is
2
E(t) = E(0) 2t |0 | ~ (14.228)

The energy of the second oscillator is

2 2
E 0 (t) = ~ |((t)| + 1/2) = ~/2 + 2t |0 | ~ (14.229)

The total energy is conserved: the energy transferred during the time t is
2 2
E(t) = 2t |0 | ~ = ~ |((t)| (14.230)

In order to distinguish between a quantum superposition and a statistical mix-

ture, we must have E ~. In other words, if a single energy quantum ~ is
transferred, it becomes problematic to tell the difference.

If we return to the numerical example of the pendulum we have the following

results: with 1/2 = 1 year = 3 107 s, the time it takes to reach || = 1 is
2
(2 |0 | )1 2 1012 s

Conclusion
Even for a system as well protected from the environment as we have assumed
for the pendulum, the quantum superpositions of macroscopic states are unob-
servable. After a very short time, all measurements one can make on a system

1132
initially prepared in such a state coincide with those made on a statistical mix-
ture. It is therefore not possible, at present, to observe the effects related to the
paradoxical character of a macroscopic quantum superposition. However, it is
quite possible to observe mesoscopic kittens, for systems which have a limited
number of degrees of freedom and are well isolated.

14.8 The Quantum Eraser

We now investigate a quantum process where the superposition of two probabil-
ity amplitudes leads to an interference phenomenon. The two amplitudes will
be associated with two quantum paths as in the double slit experiment. In the
investigation we will show that the interference disappears if an intermediate
measurement gives information about which path has actually been followed.
Then we will show how the interference can actually reappear if the path infor-
mation is erased by a quantum device.

We consider a beam of neutrons, which are particles with charge zero and spin
1/2, propagating along the xaxis with velocity v. We will treat the motion of
the neutrons classically as uniform linear motion. Only the evolution of their
spin states will be treated quantum mechanically.

14.8.1 Magnetic Resonance

The eigenstates of the z component of the neutron spin are denoted |n : i.
A constant uniform magnetic field B ~ 0 = B0 uz is applied along the zaxis.
The magnetic moment of the neutron is denoted by = n Sn where n is the
gyromagnetic ratio and Sn is the spin operator of the neutron.

~ 0 are
The magnetic energy levels of the neutron in the presence of the field B
E = n ~B0 /2 = ~0 /2 where 0 = n B0 .

The neutrons cross a cavity of length L between times t0 and t1 = t0 + L/v.

Inside the cavity, in addition to the constant field B ~ 0 , a rotating field B
~ 1 (t)
~ 1 (t) lies in the (x, y) plane and has a constant angular
is applied. The field B
frequency :
~ 1 (t) = B1 (cos tux + sin tuy )
B (14.231)
Let |n (t)i = + |n : +i + |n : i be the neutron spin state at time t and
consider a neutron entering the cavity at time t0 .

The Hamiltonian for the system is

   
H = n B ~0 + B
~ 1 (t) = n Sn B~0 + B
~ 1 (t)
  
= n B0 Sz + B1 cos tSx + sin tSy (14.232)

1133
which gives in the |n : i basis

1 eit
 
~ 0
H= (14.233)
2 1 eit 0

Therefore, the evolution equations are

1 eit
     
~ 0 1 0
H |n (t)i = + +
2 1 eit 0 0 1
   
d d+ 1 d 0
= i~ |n (t)i = i~ + i~ (14.234)
dt dt 0 dt 1
or
d+ ~ ~
i~ = 0 + + 1 eit (14.235)
dt 2 2
d ~ it ~
i~ = 1 e + 0 (14.236)
dt 2 2
or
d+ 1 1
i = 0 + + 1 eit (14.237)
dt 2 2
d 1 it 1
i = 1 e + 0 (14.238)
dt 2 2
Defining
(t) = (t)ei(tt0 )/2 (14.239)
we get

d+ 0 1 it0
i = + + e (14.240)
dt 2 2
d 0 1 it0
i = + e + (14.241)
dt 2 2
which has constant coefficients.

We assume that near resonance, | 0 |  1 , and that terms proportional to

0 may be neglected in these equations. The equations become

d+ 1 it0
i = e (14.242)
dt 2
d 1 it0
i = e + (14.243)
dt 2
whose solution is
1 (t t0 ) 1 (t t0 )
(t) = (t0 ) cos i (t0 )eit0 sin (14.244)
2 2

1134
Defining

1 (t1 t0 ) (t1 t0 ) (t1 + t0 )

= , = , = (14.245)
2 2 2
we get

+ (t1 ) = ei + (t1 ) = ei + (t0 ) cos i (t0 )eit0 sin

 
(14.246)
(t1 ) = ei (t1 ) = ei (t0 ) cos i+ (t0 )e+it0 sin
 
(14.247)
+ (t1 ) = + (t0 )ei cos i (t0 )ei sin
 
(14.248)
(t1 ) = (t0 )ei cos i+ (t0 )ei sin
 
(14.249)
   
+ (t1 ) + (t0 )
= U (t0 , t1 ) (14.250)
(t1 ) (t0 )

where
ei cos iei sin
 
U (t0 , t1 ) = (14.251)
iei sin ei cos
is the time evolution matrix.

14.8.2 Ramsey Fringes

The neutrons are initially in the spin state |n : i. They successively cross
two identical cavities of the type described above. This is called a Ramsey
configuration and it is shown in Figure 14.12 below.

Figure 14.12: Ramsey Configuration

The object A is a detecting atom described later. The same oscillating field
~ 1 (t), is applied in both cavities. The magnitude B1 of this field is applied so
B
~ 0 is applied throughout
as to satisfy the condition = /4. The constant field B

1135
the entire experimental setup. At the end of the setup, one measures the number
of outgoing neutrons which have flipped their spin and are in the final state
|n : +i. This is done for several values of near = 0 .

The initial state condition corresponds to

+ (t0 ) = 0 , (t0 ) = 1 (14.252)
At time t1 the state is
 i
iei iei
     
+ (t1 ) 1 e 0 1
= = (14.253)
(t1 ) 2 ie i
ei 1 2 ei
or
1
|(t1 )i = iei |n : +i + ei |n : i


(14.254)
2
This says that
i 1
+ (t1 ) = ei , (t1 ) = ei (14.255)
2 2
and the probability of finding it in the state |n : i is
1
P = (14.256)
2
The same neutron enters the second cavity at time t 00 = t1 + T , with T = D/v,
where D is the distance between the two cavities. Between the two cavities the
~ 0.
spin precesses freely about B

We then have
+ (t00 )
  i
iei ei
    
e 0 + (t1 ) 1
= = (14.257)
(t00 ) 0 ei (t1 ) 2 ei ei
where
0 T
= (14.258)
2
so that the spin state at t 00 is
+ (t00 ) iei ei0 T /2
   
1
= (14.259)
(t00 ) 2 ei ei0 T
Now let t 01 be the time when the neutron leaves the second cavity with t01 t00 =
t1 t0 = D/v. Now 0 = (t01 + t00 )/2 is given by
t00 = t1 + T , t01 = 2t1 t0 + T
0
= (2t1 t0 + T + t1 + T )/2 = (3t1 + 2T t0 )/2
so that (for the second cavity)
 i0 0
cos 0 iei sin 0

e
U 0 = U (t00 , t01 ) = i 0 0 (14.260)
ie sin 0 ei cos 0

1136
where
0 = = 1 (t1 t0 )/2 , 0 = = 1 (t1 t0 )/2 (14.261)
so only the parameter changes into 0 .

Thus the probability amplitude for detecting the neutron in state + after the
second cavity is obtained by
1. Applying U 0 to the vector
+ (t00 ) iei ei0 T /2
   
1
= (14.262)
(t00 ) 2 ei ei0 T

2. calculating the scalar product of the result with the + state.

We have
1 0

+ (t01 ) = iei ei/2 ei0 T /2 iei ei/2 ei0 T /2 (14.263)
2
Since

+ = t1 , 0 = (3t1 + 2T t0 t1 + t0 ) = (t1 + T )
2
we get
1 i(t1 +T )/2  i(0 )T /2 
+ (t01 ) = e e + ei(0 )T /2
2
(0 )T
= ei(t1 +T )/2 cos (14.264)
2
Therefore, the probability that the neutron spin has flipped in the two-cavity
system is
2 (0 )T
P+ = |+ (t01 )| = cos2 (14.265)
2
With the approximation | 0 |  1 , the probability for a spin flip in a single
cavity is independent of and equal to 1/2. In contrast, the two-cavity result
exhibits strong modulation of the spin flip probability between 1 ( = 0 ) and
0 ((0 )T = ). This modulation results from an interference process of the
two quantum paths corresponding to:
1. a spin flip in 1st cavity and no flip in 2nd cavity
2. no flip in 1st cavity and spin flip in 2nd cavity
In practice, the velocities of the neutrons have some dispersion around the mean
value v. This results in a dispersion in the time T to get from one cavity to the
other. A typical experimental result giving the intensity of the outgoing beam
in the + state as a function of the frequency f = /2 of the rotating field B1
is shown in Figure 14.13 below.

1137
 Figure 14.13: Experimental Results

Since cos2 /2 = (1 + cos )/2, the averaged probability distribution is

   
( 0 )T 1 1
cos2 = + cos(( 0 )T )
2 2 2
Z
1 1
= + P (T ) cos(( 0 )T )dT (14.266)
2 2

For
1 2 2
P (T ) = e(T T0 ) /2 (14.267)
2
we get
 
( 0 )T 1 1 (0 )2 2 /2
cos2 = + e cos(( 0 )T0 ) (14.268)
2 2 2
This form agrees with the observed variation with frequency in Figure 14.13
of the experimental signal. The central maximum which is located at /2 =
748.8 kHz corresponds to = 0 . For that value a constructive interference
appears whatever the neutron velocity. The lateral maxima and minima are less
peaked, however, since the position of the lateral peak is velocity dependent.
The first two lateral maxima correspond to ( 0 )T = 2. Their amplitude
is reduced compared to the central peak by the exponential factor.

This experiment can be compared to a Youngs double slit interference experi-

ment with polychromatic light.

Suppose that we insert between the two cavities a device which can measure the
z component of the neutron spin (how this works will be discussed shortly). We

1138
define P++ as the probability of detecting a neutron in the + state between the
two cavities and in the + state when it leaves the second cavity. The probability
P++ is the product of two probabilities, namely, the probability of finding the
neutron in the state + when leaving the first cavity (p = 1/2) and, knowing
that it is in the + state of finding it in the + state when it leaves the second
cavity (p = 1/2). This gives P++ = 1/4. Similarly, P+ = 1/4. The sum
P++ + P+ = 1/2 does not display any interference, since one has measured
in which cavity the neutron spin has flipped. This is very similar to the elec-
tron double slit interference experiment if one measures through which slit the
electron passes.

14.8.3 Detection of the Neutron Spin State

In order to measure the spin of the neutron, one lets it interact during a time
with a spin 1/2 atom at rest. The atoms spin operator is Sa . Let |a : zi
be the two eigenstates of the observable Saz . After the interaction between
the neutron and the atom, one measures the spin of the atom. Under certain
conditions (which we will derive shortly) one can deduce the spin state of the
neutron after the measurement.

Let |a : xi be the eigenstates of Sax and |a : yi those of Say . We can then

write
1
|a : xi = (|a : +zi |a : zi) (14.269)
2
1
|a : yi = (|a : +zi i |a : zi) (14.270)
2

and
1
|a : yi = ((1 i) |a : +xi + (1 i) |a : xi) (14.271)
2
We assume that the neutron-atom interaction does not affect the neutrons
trajectory. We represent the interaction between the neutron and the atom by
a very simple model. This interaction is assumed to last a finite time during
which the neutron-atom interaction Hamiltonian has the form
2A
V = Snz Sax (14.272)
~
~0
where A is a constant. We neglect the action of any external field, including B
during this time, i.e., we assume the atom-neutron interaction dominates for a
short period of time.

The operators Snz and Sax commute since they act on two different Hilbert
spaces. Therefore, h i
Snz , V = 0 (14.273)

1139
The common eigenvectors of Snz and V and the corresponding eigenvalues are

|n : +i |a : xi Snz = +~/2 V = A~/2 (14.274)

|n : i |a : xi Snz = ~/2 V = A~/2 (14.275)

The operators Snz and V form a complete set of commuting operators as far as
the spin variables are concerned.

From now on we assume that A = /2. Suppose that the initial state of the
system is
|(0)i = |n : +i |a : +yi (14.276)
Expanding in terms of energy eigenstates, we get

|(0)i = |n : +i |a : +yi
1
= |n : +i ((1 + i) |a : +xi + (1 i) |a : xi) (14.277)
2
and
1  
|(t)i = |n : +i (1 + i) |a : +xi eiAt/2 + (1 i) |a : xi e+iAt/2
2
so that
1  
|( )i = |n : +i (1 + i) |a : +xi eiA /2 + (1 i) |a : xi e+iA /2
2
which for A = /2 gives
1  
|( )i = |n : +i (1 + i) |a : +xi ei/4 + (1 i) |a : xi e+i/4
2  
1 1 1
= |n : +i (1 + i) |a : +xi (1 i) + (1 i) |a : xi (1 + i)
2 2 2
1
= |n : +i (|a : +xi + |a : xi) = |n : +i |a : +i (14.278)
2
Similarly, if |(0)i = |n : i |a : yi then |( )i = |n : i |a : i.

Physically, this means that the neutrons spin state does not change since it
is an eigenstate of V , while the atoms spin precesses around the xaxis with
angular frequency A. At time = /2A it lies along the zaxis.

We now suppose that the initial spin state is

|(0)i = (+ |n : +i + |n : i) |a : +yi (14.279)

After the neutron-atom interaction described above, one measures the zcomponent
Saz of the atoms spin. The state after the interaction is(using linearity)

|( )i = + |n : +i |a : +i + |n : i |a : i (14.280)

1140
The measurement of the z-component of the atoms spin gives +~/2 with prob-
ability |+ |2 and state |n : +i |a : +i after the measurement or ~/2 with
probability | |2 and state |n : i |a : i after the measurement. In both
cases, after measuring the zcomponent of the atoms spin, the neutron spin
state is known - it is the same as that of the measured atom. It is not necessary
to let the neutron interact with another measuring apparatus in order to know
the value of Snz .

14.8.4 The Actual Quantum Eraser

We have seen above that if one measures the spin state of the atom between
the two cavities, the interference signal disappears. We now want to show that
it is possible to recover an interference if the information left by neutron on the
detecting atom is erased by an appropriate measurement.

A neutron, initially in the spin state , is sent into the two-cavity system.
Immediately after the first cavity, there is a detecting atom of the type described
above, prepared in the spin state +y. By assumption, the spin state of the atom
evolves only during the time interval when it interacts with the neutron.

The successive states for the neutron are:

|(t1 )i = |after 1st cavity and before atom interactioni

1
= iei |n : +i |a : +yi + ei |n : i |a : +yi


2
|(t1 + )i = |just after atom interactioni
1
= iei |n : +i |a : +i + ei |n : i |a : i


2
0
|(t0 )i = |entering 2nd cavityi
1  
= iei(+0 T /2) |n : +i |a : +i + ei(+0 T /2) |n : i |a : i
2
|(t01 )i = |after 2nd cavityi
 
i(+0 T /2) i i 0
1 ie e |n : +i ie |n : i |a : +i
=  0

2 +ei(+0 T /2) iei |n : +i + ei |n : i |a : i

The probability of finding the neutron in the state + at time t01 (after the 2nd
cavity) is the sum of the probabilities for finding

1. the neutron in state + and the atom in state +, i.e., the square modulus
of the coefficient of |n : +i |a : +i, which = 1/4 in this case)

2. the neutron in state + and the atom in state , which = 1/4 in this case
also).

1141
We therefore get P+ = 1/4 + 1/4 = 1/2 - there is no interference since the
quantum path leading in the end to a spin flip of the neutron can be determined
from the state of the atom.

At time t01 , Bob measures the zcomponent of the neutron spin and Alice
measures the ycomponent of the atoms spin. Assume that both measurements
give +~/2.

We can write
 
i(+0 T /2) i i 0
1 ie e |n : +i ie |n : i (|a : +yi + |a : yi)
|(t01 )i =  0

2 2 + ei(+0 T /2) iei |n : +i + ei |n : i (|a : +yi |a : yi)

The probability amplitude that Bob finds +~/2 along the zaxis while Alice
finds +~/2 along the yaxis is the coefficient of the term |n : +i |a : +yi in
the above state. Equivalently, the probability amplitude is found by projecting
the state onto |n : +i |a : +yi and squaring. We get
  1 0
2
P Snz = +~/2, Say = +~/2 = iei(++0 T /2) iei( +0 T /2) 
 
8
1 ( 0 )T
= cos2 (14.281)
2 2
which clearly exhibits a modulation reflecting an interference phenomenon. Sim-
ilarly, one finds that
  1 ( 0 )T
P Snz = +~/2, Say = ~/2 = sin2 (14.282)
2 2
which is also modulated.

Let us now discuss the following three statements:

1. When Alice performs a measurement on the atom, Bob sees at once an

interference appear in the signal he is measuring on the neutron.

2. Knowing the result obtained by Alice on each event, Bob can select a
subsample of his own events which displays an interference phenomenon.

3. The experiment corresponds to an interference between two quantum

paths for the neutron spin. By restoring the initial state of the atom,
the measurement done by Alice erases the information concerning which
quantum path is followed by the neutron spin and thus allows interference
to reappear.

Statement (1) is clearly wrong. As seen earlier, if atom A is present, Bob no

longer sees oscillations (in 0 ) of the probability for detecting the neutron
in the state +. This probability is equal to 1/2 whatever Alice does. Notice

1142
that if the statement were correct, this would imply instantaneous transmission
of information from Alice to Bob. By seeing interference reappear, Bob would
know immediately that Alice is performing an experiment, even though she may
be far away.

Statement (2) is correct. If Alice and Bob put together all their results, and if
they select the subsample of events for which Alice finds +~/2, then the number
of events for which Bob also finds +~/2 varies like

( 0 )T
cos2 (14.283)
2
Thus, they recover interference for this subset of events. In the complementary
set where Alice found ~/2, the number of Bobs results giving +~/2 varies like

( 0 )T
sin2 (14.284)
2
This search for correlation between events occurring in different detectors is a
common procedure in particle physics for example.

Statement (3), although less precise but more picturesque than than statement
(2), is nevertheless acceptable. The

( 0 )T
cos2 (14.285)
2
signal found earlier can be interpreted as the interference of the amplitudes
corresponding to two quantum paths for the neutron spin which is initially in
the state ; either its spin flips in the 1st cavity, or it flips in the 2nd cavity. If
there exists a possibility to determine which quantum path is followed by the
system, interference cannot appear. It is necessary to erase this information,
which is carried by the atom, in order to observe some interference. After Alice
has measured the atoms spin along the yaxis, she has, in some sense restored
the initial state of the system, and this enables Bob to see some interference. It is
questionable to say that information has been erased - one may feel that, on the
contrary, extra information has been acquired. Notice that the statement does
not specify in which physical quantity the interference reappears. Notice also
that the order of the measurements made by Alice and Bob has no importance,
contrary to what this third statement seems to imply.

1143
1144
Chapter 15

States and Measurement

15.1 State Preparation/Determination

Density Matrix Revisited
In Chapter 6 we introduced the two basic concepts needed to discuss the formal
structure of quantum mechanics.

Postulate 1 defined observables as Hermitian operators and identified their

eigenvalues with the possible results of any measurement of the corresponding
dynamical variable.

Postulate 2 stated that every physical system has a state (or density ) operator
associated with it, or, alternatively, we might say that every physical system has
an associated state vector.

This means that we must have some reproducible preparation procedure, which
we identify with the term state, that determines a probability distribution for
each dynamical variable.

This presents us with two kinds of problems:

1. State Preparation - How do we actually prepare a state that is repre-

sented by a particular state operator or state vector?

2. State Determination - For a given system, how do we determine the

state operator or state vector?

15.1.1 State Preparation

Suppose that at t = t0 we have a known pure state represented by a state vec-
tor |0 i. It is always possible to mathematically(theoretically) construct a time
development operator U (t1 , t0 ) that will generate any desired pure state via

1145
the relation |1 i = U (t1 , t0 ) |0 i. It is not always possible, however, to realize
U (t1 , t0 ) in practice (in the laboratory).

Example: Consider an atom in its ground state. We can prepare an excited

state or a linear combination of excited states by using a pulse of electromag-
netic radiation from a laser.

If we know the initial state, the mathematical mapping is one-to-one (pure state
to pure state) and invertible and thus reversible.

The more fundamental problem, however, is how do we prepare a particular

state from an arbitrary unknown initial state? This mapping is not invertible
and, therefore, if we are able to carry it out in the laboratory, it must involve
some kind of irreversible process.

The fundamental laws of nature, however, are reversible, as far as we know.

This means that the effective irreversibility that arises must come about by a
coupling of the system to some kind of apparatus or to the environment so that
we have somewhere to transfer the entropy or information that is lost in the
irreversible process.

This says that even in the microscopic world where reversibility holds, we can
have, effectively, an irreversible transformation.

Now, we know that it is always possible to prepare the lowest energy state
of a system simply by waiting long enough for the system to decay into its
ground state. This follows because the decay of an atomic excited state by
spontaneous emission is due to the atom-electromagnetic field coupling and the
survival probability of an excited state, due to this interaction, decays to zero
(usually exponentially with time). This means that the probability of obtain-
ing the ground state can be made arbitrarily close to one just by waiting long
enough.

The method assumes that the energy of the excited state is radiated away to
infinity (it never returns to the system) and that the electromagnetic field is
initially in its lowest energy state since otherwise we have a nonvanishing prob-
ability for the atom to absorb energy from the field and become re-excited (no
longer guaranteed to be in the ground state).

In thermal equilibrium, the probability of obtaining an excited atomic state is

proportional to eEx /kB T where Ex = excitation energy of the atomic state,
kB = Boltzmann constant and T = the effective temperature of the radiation
field. This factor is usually very small. The presence of the cosmic background
radiation at a temperature of 3 K kB T 0.0002 eV gives a lower limit to
the probabilities unless special shielding and refrigeration techniques are used in
any experiment. This background increases in importance as system gets larger

1146
since Ex gets smaller.

Suppose that this just waiting method can be used successfully to produce the
ground state for some system. It then turns out that we can then prepare a
wide range of states for a spinless particle.

Suppose that we wish to prepare the state characterized by the state function

(~r) = R(~r)eiS(~r) (15.1)

where R(~r) and S(~r) are real functions.

Step #1
Construct a potential W1 (~r) such that

~2 2
0 + W1 0 = E0 (15.2)
2m
where 0 (~r) = R(~r) is the ground state wave function for the system. By
definition of the ground state, R(~r) is then a nodeless function. We can do this
by choosing
~2 2 R(~r)
W1 (~r) = E + (15.3)
2m R(~r)
Proof : Direct substitution gives

~2 2 ~2 2 ~2 2 R(~r)
0 + W 1 0 = 0 + E0 + 0 = E0
2m 2m 2m R(~r)
or
2 R 0 2 0 R = 0



(15.4)
which certainly has a solution 0 (~r) = R(~r).

Step #2
Wait until the probability that the system has decayed to its ground state is
sufficiently close to one.

Step #3
Apply a pulse potential

W2 (~r, t) = ~S(~r) (t) (15.5)

where (
1/ 0 < t <
(t) = (15.6)
0 otherwise

1147
During the short time interval 0 < t < we can approximate the Schrodinger
equation by
(~r, t)
i~ = W2 (~r, t)(~r, t) (15.7)
t
since W2 overwhelms any other interactions in the limit 0.

Step #4
Integrate the equation with the initial condition (~r, 0) = R(~r). We get

(~r, 0 + ) = R(~r)eiS(~r) (15.8)

which is the state we wanted to prepare.

In general, we can usually carry out this procedure since we are only limited by
our ability to produce the potentials W1 and W2 in the laboratory.

The Filtering Method

An example of this method is the Stern-Gerlach(SG) apparatus. Its mode of
operation is simple. The potential energy of a magnetic moment ~ in a magnetic
~ is E = ~
field B ~ If the magnetic field is spatially inhomogeneous, then a
B.
force arises given by  
F~ = E = ~
~ B (15.9)
~
~ S(spin).
The magnitude and sign of the force depends on the spin state since
Therefore, different spin states will be deflected by this force into sub-beams
propagating in different directions.

Thus, the SG apparatus physically separates spin states in space.

By blocking off and thus eliminating all but one of the sub-beams, which is an
irreversible process, we can select(or filter out) a particular spin state. We will
discuss this procedure in detail shortly.

No-Cloning Theorem
Why dont we just make exact replicas, or clones, of a prototype of the state
(assuming we can find one)? We do this all the time in the world of macro-
physics, i.e., duplicate keys, copying a computer file, etc.

Mathematically, to clone an arbitrary quantum state |i we would need a device

in some suitable state |i and a unitary time development operator U such that

U |i |i = |i |i | 0 i (15.10)

The dimension of the final device state vector | 0 i is smaller than that of the
initial device state vector |i because the overall dimension of the space is

1148
conserved under unitary transformations, i.e., if |i is a 1particle state and
|i is an N particle state, then | 0 i must be an (N 1)particle state.

We now show that such a device is impossible. We do it by assuming that it is

possible and then arriving at a contradiction.

So, we assume it is possible to find a U operator. Then, we assume that there

exists two arbitrary states |1 i and |2 i for which the cloning transformation
works, i.e.,

U |1 i |i = |1 i |1 i | 0 i
U |2 i |i = |2 i |2 i | 00 i

It is possible that the final device states | 0 i and | 00 i have a dependence on the
state we are attempting to clone.

Now U is a linear operator. This implies that for the superposition state
1
|S i = (|1 i + |2 i) (15.11)
2
we must have
1 1
U |S i |i = |1 i |1 i | 0 i + |2 i |2 i | 00 i (15.12)
2 2
But |S i is not any special state and thus, by our assumption, we should have
obtained
U |S i |i = |S i |S i | 000 i (15.13)
but we do not!

Thus, it is impossible to build a device to clone an arbitrary, unknown quantum

state.

Classical states are just special limiting cases of quantum states. Since we are
able to copy an unknown classical state, does this theorem make any sense?

We were able to prove the impossibility of cloning states because of the linearity
of quantum mechanical time evolution process.

We could not prove the impossibility if we only required cloning to work on a

discrete set of states. The ability in quantum mechanics to form superpositions
was essential to the proof.

In any case, the set of discrete states must also be orthogonal. This follows from
the fact that the inner product between state vectors is preserved by a unitary
transformation.

1149
This means we must have
  
h| h1 | U + U |2 i |i = h| h1 | I |2 i |i
= h1 | 2 i h | i
= (h 0 | h1 | h1 |) (|2 i |2 i | 00 i)
2
= |h1 | 2 i| h 0 | 00 i

But since h | i = 1 , |h 0 | 00 i| 1 and |h1 | 2 i| 1, the relation

2
h1 | 2 i h | i = |h1 | 2 i| h 0 | 00 i (15.14)

would only work mathematically if h1 | 2 i = 0 or the discrete state vectors

are orthogonal.

Now, states that are classically different will certainly be orthogonal, so the
no-cloning theorem that we proved for quantum states would not be in conflict
with the ability to copy classical states.

Our discussion so far involves only pure states.

If, however, we can prepare ensembles corresponding to several different pure

states |i i, then by combining them with weights we can prepare the mixed
state represented by the state operator
X
= wi |i i hi | (15.15)
i

In the laboratory, nature usually presents us with mixed states (states that are
not pure) and it is the preparation of pure states that is difficult.

15.1.2 State Determination

Suppose we have some apparatus that is designed (we think) to produce a cer-
tain state. We need to be able to test or calibrate it in order to determine what
state is actually being produced. In addition, some natural process can produce
an unknown state and we need to be able to determine its properties.

The procedure in either case must be repeatable(whether under the control of

the experimenter or occurring spontaneously in nature). This says that we must
be able to create an ensemble of systems and can carry out measurements on
each of them.

Because a measurement involves an interaction with the system, it will change

the values of the state parameters and hence change the state representing the
system. This means that any further measurements on the same system will
be useless as far as enhancing our understanding of the original state. We must

1150
submit a system to the preparation procedure and then carry out only a single
measurement on it. To obtain any more information, we must repeat the state
preparation procedures before another measurement is carried out. We can use
the same system each time in the preparation procedure, or another identical
system each time. The results will be the same.

What sort of measurements are sufficient to determine the state operator (or
state vector) associated with the preparation procedure?

Suppose we have a dynamical variable R whose associated operator R has a dis-

crete, non-degenerate eigenvalue spectrum, i.e., R |rn i = rn |rn i. Also suppose
that we do a single measurement and get the result R = r3 . This only says that
we could have any state |i where hr3 | i =
6 0 or any state operator where
hr3 | |r3 i =
6 0.

If we repeat the preparation-measurement sequence many times, however, we

can determine the relative frequency of the result R = r3 , i.e., we will be mea-
suring
P rob(R = rn |) = hrn | |rn i (15.16)

for the case of non-degenerate eigenvalues. This says that the measurement
of the probability distribution of the dynamical variable R gives the diagonal
matrix elements of the state operator in this representation.

To obtain information about the off-diagonal elements hrm | |rn i =

6 0 we need
to measure some other dynamical variable whose operator does not commute
with R.

We can formally construct a set of operators such that their probability distribu-
tions would determine all of the matrix elements of the state operator. Consider
the Hermitian operators:

|rm i hrn | + |rn i hrm | |rm i hrn | |rn i hrm |

Amn = and Bmn = (15.17)
2 2i
We then have
  X
T r Amn = hrp |Amn |rp i
p
X |rm i hrn | + |rn i hrm |
= hrp | |rp i
p
2
X1
= [pn hrp | |rm i + pm hrp | |rn i]
p
2
1
= [hrn | |rm i + hrm | |rn i] = Re (hrm | |rn i) (15.18)
2

1151
and similarly  
T r Bmn = Im (hrm | |rn i) (15.19)

Therefore, if Amn and Bmn represent observables, then the measurement of

their averages, i.e.,   D E
T r Q = Q (15.20)

determines the all the matrix elements of the state operator .

The remaining questions are

1. Are they observables?
2. How would we actually perform the necessary measurements?
These questions are not answerable within the formal approach. We must look
at special cases to see what measurements are actually required.

15.1.3 Spin State (S = 1/2)

What is the most general state operator for a spin = 1/2 system? Since this
is a world of 2 2 matrices and in that world the four matrices I, x , y , z
form a basis for all the 2 2 matrices or, equivalently, all the operators in the
spin = 1/2 world. Therefore, we can always write
1  
= I + ~a (15.21)
2
This has T r() = 1 as required. If we choose ax , ay and az real, then = +
so that it is Hermitian. We can determine ax , ay and az as follows:
1
hx i = T r (x ) = T r x + ax x2 + ay y x + az z x


2
1   1  
= T r x + ax I iay z + iaz y = ax T r I = ax (15.22)
2 2
and similarly hy i = ay and hz i = az or hi = T r () = ~a. Physically,
therefore, ~a is the polarization vector of the state.

The eigenvalues of are (1 |~a|)/2 0 0 |~a| 1. Pure states have

|~a| = 1(maximum polarization) and unpolarized states have |~a| = 0 , which says
that it is isotropic and the average of any spin component is zero. Thus, in
order to determine the state operator, we must determine the components of ~a.
We have
1 D E 1 1
Si = ~i Si = ~ T r (i ) = ~ai (15.23)
2 2 2
and it therefore sufficient to measure
D E D E D E
Sx , Sy and Sz (15.24)

1152
which can be done using an SG apparatus (see discussion later in this chapter).

For some unit vector n = (sin cos , sin sin , cos ) we have the operator

ei sin
 
cos
n = (15.25)
ei sin cos

Its eigenvalues are given by the characteristic equation

(cos )(cos + ) sin2 = 0

2 = 1 = 1

The eigenvectors of n are

 i/2
cos 2 ei/2 sin 2
  
e
(15.26)
ei/2 sin 2 ei/2 cos 2

Only the relative magnitudes and relative phases of the components of any state
vector have physical significance (the norm and overall phase are irrelevant).
Looking at the eigenvectors, it is clear that all possible values of the relative
magnitude and relative phase can be obtained by varying and , or, in other
words, the relative magnitude and relative phase of any 2component vector
uniquely determines and . This implies that any pure state vector of an
S = 1/2 system can always be associated with a spatial direction n for which it
is the +hbar/2 eigenvector for that component of spin, i.e., emphany pure state
always represents a definite spin in some direction in this case.

15.1.4 Spin State (S =1)

In this case, the state operator is a 3 3 Hermitian matrix(or operator). It
depends on 8 independent real numbers (parameters), i.e., since

= + 11 , 22 , 33 real
ij = ji i 6= j

the eight associated real numbers are

11 , 22 , Re12 , Im12 , Re13 , Im13 , Re23 , Im23

33 = 1 11 22 since T r = 1

We can determine three of these numbers by measuring

D E D E D E
Sx , Sy , Sz (15.27)

where D E
S = polarization, as before. (15.28)

1153
The five other parameters are obtained by measuring the average values of the
so-called quadrupole operators

S S + S S , = x, y, z (15.29)

Only five of the six possible different operators are independent, namely,

(, ) = (x, y), (z, x), (y, z), (x, x), (y, y) (15.30)

(, ) = (z, z) is not independent since

Sx2 + Sy2 + Sz2 = 2~2 (15.31)

Now for spin = 1 we have the matrix representation (Sz is diagonal)

0 1 0 0 i 0 1 0 0
~ ~ ~
Sx = 1 0 1 , Sy = i 0 i , Sz = 0 0 0
2 0 1 0 2 0 i 0 2 0 0 1

and we have
1 sin ei

cos 2
0
1 sin ei
n S = 2 0 1 sin ei (15.32)
2
0 1 sin ei cos
2

with eigenvalues/eigenvectors

+~ 0 ~
1 i
1
+ cos )ei sin e 1
cos )ei

2 (1 2 2 (1
1 sin cos 12 sin
2
1 i 1
sin e i 1 i
2 (1 cos )e 2 2 (1 + cos )e

Unlike the spin = 1/2 case, it is no longer true that every vector must be an
eigenvector of the component of spin in some direction. This is so because it
requires four real parameters to specify the relative magnitudes and the relative
phases of the components of a general 3component vector. The eigenvectors
above only contain two parameters and , however. Therefore, the pure states
of a spin = 1 system need not be associated with a spin eigenvalue in any spatial
direction. Let us define the quantities

1
a = T r(S ) = x, y, z
~
1
q = 2 T r(S2 ) = x, y
~
1 h i
q = 2 T r S S + S S = xy, yz, zx
~

1154
these are the eight real numbers we need defined in terms of the measurable
quantities:
D E
S = x, y, z ,
D E
S2 = x, y
D E
S S + S S = xy, yz, zx

Now, we can do direct evaluations of the type

1
ax = T r(Sx )
~
1 XX
= hsz | |s0z i hs0z | Sx |sz i
~ s 0
z sz

= 2(12 + 23 )

where |1i = |sz = +~i , |2i = |sz = 0i , |3i = |sz = ~i.

Writing down all such possible equations and solving them for the matrix el-
ements of the state operator we can express the most general state operator
as
1 + 12 (ax qxx qyy ) 1 1

(a + qxx i(ay + qyz ))
2 2 x 2 (qxx qyy iqxy )
1 (a + q + i(a + q )) 1 + qxx + qyy 1
(a qxx i(ay qyz ))
= 2 2 x xx y yz
2 2 x
1 1 1
2 (q xx qyy + iq xy )
2 2
(ax qxx + i(ay qyz )) 1 2 (ax + qxx + q yy )

With this parameterization of the state operator, we now ask how do we measure
the parameters?

An SG apparatus with the magnetic field gradient along the xdirection can be
used to perform measurements on an ensemble of particles that emerge from the
state preparation. We can determine the relative frequencies of the Sx values
+~, 0, ~ in this way. Then we can calculate
D E D E
Sx Sx2
ax = and qxx = (15.33)
~ ~2
and similarly for gradients in the y and zdirections. In this way we can
measure ax , ay , az , qxx , qyy , qzz which provides a check since we must have qzz =
2 qxx + qyy . What about qxy , qyz , qzx ? Here we take advantage of the time
evolution of the unknown state operator in a uniform magnetic field B. ~ We
have
H = ~ B ~ = S B
~ = Sz for B
~ = B z (15.34)
Now suppose that at t = 0 the state operator is and the magnetic field is
turned on for a time interval t, after which we measure Sx2 . By doing this many

1155
times for each of several t values, we can evaluate
d D 2 E
Sx  (15.35)
dt t=0

from the data. Now, for any observable

D E  
R = T r (t)R (15.36)
t

where
(t) ih i
= (t), H (15.37)
t ~
In the Schrodinger picture we then have
!
d D E d    R
R = T r (t)R = T r R +
dt t dt t t
!
ih i R
= Tr H, R +
~ t
!
i  R
= Tr H R H R +
~ t
!
i  R
= Tr RH H R +
~ t
!
i h i R
= Tr (t) H, R + (t)
~ t

From this equation of motion we then have

d D 2 E i n h io i n h io
Sx  = T r H, Sx2 = T r Sz , Sx2
dt t=0 ~ n h ~ io
= T r Sx Sy + Sy Sx = ~2 qxy

In a similar manner we can determine qyz and qzx . Thus, we have managed to
determine the spin = 1 state completely. This procedure can be generalized to
higher spin states.

Now let us turn to the problem of determining the orbital state for a spin-
less particle. The orbital state of a spinless particle can be described by the
coordinate representation of the state operator

h~r| |~r 0 i (15.38)

This is a function of two variable ~r and ~r 0 . It is called the density matrix. Its
diagonal elements
2
h~r| |~ri = |h~r | i| (15.39)

1156
yield the position probability density.

To determine the density matrix for an arbitrary state we will need the prob-
ability distributions for the position and one or more dynamical variables with
operators that do not commute with the position operator. In 1933, Pauli posed
this question: Are the position and momentum probability densities sufficient to
determine the state?

The answer turns out to be NO!. Here is a counterexample.

Consider a pure state that is an eigenfunction of the orbital angular momentum,

i.e.,
nlm (~r) = fnl (r)Yml (, ) (15.40)
Now the states nl,m (~r) both have the same position and momentum distri-
butions, but they are distinct states of the system.

A sufficient set of measurements to determine the orbital state of a particle does

not seem to be known.

15.1.5 Composite Systems

Suppose that we have a composite system made up of components, i.e., a mul-
tiparticle state is a good example. Can we define states for the components
separately or only for the composite system as a whole? Is the state of a com-
posite system determined by the state of its parts?

To answer the first question, we consider a 2component system (components

labeled 1 and 2). The basis set for the vector space is the set of direct product
vectors of the form
|am bn i = |am i |bn i (15.41)
where

{|am i} = set of basis vectors for component 1 alone

{|bn i} = set of basis vectors for component 2 alone

The average of any dynamical variable R is given by

D E   X
R = T r R = ham bn |R |am bn i
m,n
X X
= ham bn | |am0 bn0 i ham0 bn0 | R |am bn i (15.42)
m,n m0 ,n0

Now suppose that R(1) is an operator representing a dynamical variable which

applies exclusively to component 1, i.e.,

R(1) R(1) I(2) (15.43)

1157
Then we have
D E X X
R(1) = ham bn | |am0 bn0 i ham0 | R(1) |am i hbn0 | I(2) |bn i
m,n m0 ,n0
X X
= ham bn | |am0 bn0 i ham0 | R(1) |am i nn0
m,n m0 ,n0
X
= ham bn | |am0 bn i ham0 | R(1) |am i (15.44)
m,m0 ,n

If we define the operator

(1) = T r(2) operator in the space of component 1 alone (15.45)

where T r(2) means the trace over the space of component 2, then we have
X
ham | (1) |am0 i = ham bn | |am0 bn i (15.46)
n

and we can write

D E XX  
R(1) = ham | (1) |am0 i ham0 | R(1) |am i = T r (1) R(1) (15.47)
m m0

The operator (1) is called the partial (or reduced) state operator for component
1. Similar results hold for
D E  
R(2) I(1) R(2) R(2) = T r (2) R(2) (15.48)

Let us now prove that (1) and (2) are in fact state operators. We must have

T r(1) (1) = 1 , (1) = (1)+ , hu| (1) |ui 0 for all |ui (15.49)

Now we have
X X
T r(1) (1) = ham | (1) |am i = ham bn | |am bn i = T r =1
m m,n

  X
ham | (1) (1)+ |am0 i = ham bn | + |am0 bn i


n
X
= ham bn | (0) |am0 bn i = 0 (1) = (1)+
n

To prove the last condition we assume hu| (1) |ui < 0 and look for a contradic-
tion. In the space of component 1 we use the orthonormal basis {|um i} where
|u1 i = |ui instead of using the basis {|am i}. We also use the product states
|um bn i = |um i |bn i. Then, our assumption implies that
X
0 > hu1 | (1) |u1 i = hu1 bn | |u1 bn i (15.50)
n

1158
but this is impossible since is non-negative. Thus, (1) and (2) are state
operators. (1) is sufficient to calculate the average of any dynamical variable
that belongs exclusively to component 1 and similarly for (2) .

They are not sufficient, in general, for determining the state of a composite sys-
tem. The reason is that they provide no information about correlations between
components 1 and 2.

If we find D E D ED E
R(1) R(2) = R(1) R(2) (15.51)

for all R(1) and R(2) , then the composite state is said to be an uncorrelated
state. In this case, we have
D E  h i
R(1) R(2) = T r R(1) R(2)
   
= T r (1) R(1) T r (2) R(2) = (1) (2) (15.52)

This is the only case where the total state operator is determined by the partial
state operators of the components.

Pure State Factor Theorem

If is a state operator and if (1) = T r(2) and (2) = T r(1) and if (1) describes
a pure state, then
= (1) (2) (15.53)

i.e., a pure partial state operator must be a factor of the total state operator.

Proof : We start with a representation of that guarantees nonnegativeness,

namely, the spectral representation in terms of eigenvectors and eigenvalues of

X
= k |k i hk | (15.54)
k

This form is nonnegative as long as the eigenvalues k > 0. Now let us expand
the eigenvectors in terms of the product vectors
X
k
|k i = Cmn |am bn i (15.55)
m,n

This implies
X X X
k k
= k Cmn Cm 0 n0 |am bn i ham0 bn0 | (15.56)
k m,n m0 ,n0

1159
so that

X X X X
(1) = T r(2) = hbr | k k
Cmn k
Cm 0 n0 |am bn i ham0 bn0 |
|br i
r k m,n m0 ,n0
XX X
k k
= k Cmn Cm 0 n0 hbr | am bn i ham0 bn0 | br i

r,k m,n m0 ,n0

XX X
k k
= Cmn Cm0 n0 |am i ham0 | rn rn0

r,k m,n m0 ,n0

X X X
k k
= k Cmn Cm0 n |am i ham0 | (15.57)
k m,m0 n

Now, since this is a pure state, we must have

(1) = |i h| (15.58)

Since the original basis {|am i} is arbitrary, we can choose |a1 i = |i which
implies that
X X X
(1) = |am i ham0 | k k
Cmn Cmk
0 n = |a1 i ha1 | (15.59)
m,m0 k n

This says that

X X XX
k k k k 0
k Cmn Cm0n = k Cmn Cm0 n = 0 unless m = m = 1

k n n k

or that X  k 2
k Cmn  = 0 for m 6= 1 (15.60)
k
k
Therefore, for m 6= 1 and any k such that k 6= 0 we must have Cmn = 0 since
0. Thus
X XX
k k
= k C1n C1n 0 |a1 bn i ha1 bn0 |

k n n0
!
X XX
k k
= |a1 i ha1 | k C1n C1n 0 |bn i hbn0 | (15.61)
k n n0

This is the form required by the theorem. The first factor is

(1) = |i h| = |a1 i ha1 | (15.62)

and the second factor is identified with

X XX
(2) = k k
C1n k
C1n0 |bn i hbn0 | (15.63)
k n n0

1160
Summarizing, we have shown that partial states for the components of a system
can be defined, but the states of the components do not suffice for determining
the state of a whole composite system. The relation between the states of the
components and the composite state is very complex. The theorem helps when
one component is pure, which implies that factors as specified.

Now a factorization of the form = (1) (2) implies that there is no correla-
tions between the components 1 and 2. Therefore, a component described by a
pure state cannot have any correlations with the rest of the system. Does this
make sense?

Consider a many-particle spin system described by the state vector

|i = |i |i |i ................... (15.64)

i.e., all spins are up in the zdirection. This seems like a high degree of correla-
tion. Yet, we must interpret the product form of the state vector as an absence
of correlations among the particles.

We resolve this dilemma by noting that the correlation existing here is gener-
ated from the quantum mechanical probability distributions. Since |i is an
eigenvector of the z-components of the spins, there are no fluctuations in these
dynamical variables and no variability (no fluctuations) implies the degree of
correlation is undefined.

If, instead, we consider the components of the spin in any direction other than
z, they will be subject to fluctuation and those fluctuations will indeed be cor-
related in the state |i.

Correlated states of multi-component systems are called entangled states. Many

modern quantum mechanical experiments involving quantum reality use entan-
gled states, as does quantum computing.

Consider the apparatus as shown in Figure 15.1 below.

Figure 15.1: Experimental Setup

The source (inside box) emits pairs of particles in variable directions, but always

1161
with opposite momentum
~kb = ~ka , ~kb0 = ~ka0 (15.65)

The two output ports on each side of the source box restrict each particle to
two possible directions. This says that the state of the emerging pairs is
1  E  E  E  E
|12 i = ~ka ~kb + ~ka0 ~kb0 (15.66)
2
The momenta of the particles are correlated in this state. This means that if
particle 1 on the right has momentum ~~ka , then particle 2 on the left must have
momentum ~~kb and if 1 has ~~ka0 , then 2 must have ~~kb0 .

Now by inserting mirrors in the proper places we can combine the beams a and
a0 on the right and combine beams b and b0 on the left. Looking at only one
side of the apparatus, it would appear that the amplitudes from paths a and
a0 should produce an interference pattern (like a double slit experiment) and
similarly for paths b and b0 on the left.

This, in fact, would be observed experimentally if the particles were not corre-
lated and the state was of the form

|12 i = |1 i |2 i (15.67)

Correlations between the particles, in fact, leads to a qualitative difference. The

2-particle configuration space wave function will have the form
~ ~ ~ ~
12 (~r1 , ~r2 ) ei(ka ~r1 +kb ~r2 ) + ei(ka0 ~r1 +kb0 ~r2 ) (15.68)

and the position probability density is

2 1 h i(~ka ~ka0 )~r1 i(~kb ~kb0 )~r2 ~ ~ ~ ~

i
|12 (~r1 , ~r2 )| 1 + e e + ei(ka ka0 )~r1 ei(kb kb0 )~r2
2 h i
1 + cos ~ka ~ka0 ) ~r1 + (~kb ~kb0 ) ~r2 (15.69)

This form holds only in the regions where the beams overlap. The wave function
is zero outside the beams.

If you ignore particles on the left and place a screen to detect particles on the
right, then the detection probability for particle 1 is given by
Z
2
|12 (~r1 , ~r2 )| d~r2 (15.70)

This is featureless, i.e., no interference pattern exists in the single particle prob-
ability density.

Only in the correlations between particles can interference be observed,i.e., if

1162
we select only those particles on the right that are detected in coincidence with
particles on the left in a small volume near ~r1 , then their spatial density is
proportional to
Z h i
2
|12 (~r1 , ~r2 )| (~r2 )d~r2 = 1 + cos ~ka ~ka0 ) ~r1 (15.71)

and we do see an interference pattern.

15.2 Measurement and Interpretation of States

All experiments involve state preparation followed by a measurement. Mea-
surement and states are so closely linked in quantum mechanics that a detailed
analysis of the measurement process will allow us to devise a better interpreta-
tion of the state vector.

15.2.1 Example of Spin Measurement

We now return to a discussion of the measurement of a spin component using a
Stern-Gerlach(SG) apparatus.

As we discussed earlier, its mode of operation is simple. The potential energy

of a magnetic moment ~ is E = ~
~ in a magnetic field B ~ If the magnetic
B.
field is spatially inhomogeneous, then a force arises given by
 
F~ = E = ~
~ B (15.72)

The magnitude and sign of the force depends on the spin state since ~
~ S(spin).
Therefore, different spin states (different spin components) will be deflected by
this force into sub-beams propagating in different directions. The SG apparatus
physically separates spin states in space. By blocking off and thus eliminating
all but one of the sub-beams, which is an irreversible process, we can select(or
filter out) a particular spin state. In addition, the value of the spin component
can be inferred from the location of the sub-beam or equivalently, the deflection
of the beam. A schematic of an SG apparatus is shown in Figure 15.2 below:

Figure 15.2: Stern-Gerlach Setup

1163
The velocity of the incident beam is in the ydirection and the magnetic field
is in the xzplane, which is transverse to the beam. Some idealizations are
necessary to simplify our analysis:

1. the magnetic field vanishes outside of the gap between the poles

2. only the zcomponent of the field is significant

3. within the gap, the field has a constant gradient

This means, that relative to a coordinate origin located some distance below
the magnet, the components of the magnetic field can be written as

Bx = By = 0 , Bz = zB 0 , B 0 = field gradient (15.73)

In this case, the magnetic force is in the zdirection and the y-component of
the velocity will be constant. We can, therefore, choose a frame of reference
moving uniformly in the ydirection (with the beam velocity). In this frame
the incident particle is at rest and it experiences a time-dependent magnetic
field that is only nonzero during the time interval T that it takes for the beam
to pass through the magnet field region. The spin Hamiltonian, H = B, ~
can then be written as:

0
t<0
H(t) = czz 0 < t < T (15.74)

0 T <t

Now, we must have B ~ = 0 for any magnetic field. The above magnetic field
does not, in fact, satisfy this relation. We would need to have, at least, that

Bx = xB 0 , Bz = B0 + zB 0 (15.75)

If, however, Bx  B0 , which is true in practice, then any component of the

magnetic moment in the xyplane will precess rapidly about the zaxis and
the force due to Bx will average to zero. Therefore, we can use our approximate
Hamiltonian.

Now suppose that the state vector for the particles is given by

|0 i = a |+i + b |i t0 (15.76)
  2
where a2  + |b| = 1 and |i = spin up/down eigenvectors of z .

The time evolution of this state (equation of motion) then gives for t 0
i i i
|(t)i = e ~ Ht |0 i = ae ~ Ht |+i + be ~ Ht |i
= aeiczt/~ |+i + beiczt/~ |i (15.77)

1164
so that for t T

|1 i = |(T )i = aeiczT /~ |+i + beiczT /~ |i (15.78)

This says that the effect of the interaction is to create a correlation between the
spin and the momentum of the particle. The state vector above says that

if z = +1 then Pz = +T c
if z = 1 then Pz = T c

or that the trajectory of the particle will be deflected either up or down accord-
ing to whether z is positive of negative. Thus, by observing the deflection of
the particle we can infer the spin value.

This demonstrates that the essential feature of any measurement procedure is

the establishment of a correlation between the dynamical variable to be mea-
sured (spin component in this case) and some macroscopic indicator that can
be directly observed (the beam deflection in this case).

15.2.2 A General Theorem of Measurement Theory

What are the essential ingredients of a measurement?

(I) an object and (II) an apparatus together

with an interaction that produces a correlation
between some dynamical variable of (I) and an
appropriate indicator variable of (II)

Suppose that we want to measure the dynamical variable R (assume it has a

discrete spectrum for now) that belongs to the object (I). We assume that the
corresponding operator R possesses a complete set of eigenvectors, i.e.,

R |riI = r |riI (15.79)

We assume that the apparatus (II) has an indicator variable A, and a corre-
sponding operator A that also possesses a complete set of eigenvectors, i.e.,

A |, miII = |, miII (15.80)

where is the indicator eigenvalue and m labels all the other quantum numbers
needed to specify a unique eigenvector of the apparatus.

We assume that the apparatus is prepared in an initial state prior to measure-

ment given by |0, miII , i.e., the indicator eigenvalue is set to zero.

Now introduce some interaction between (I) and (II) that produces a unique
correspondence between the value r of the dynamical variable R of (I) and the

1165
indicator of (II). Since the interaction causes a time development of the state
vectors, the time development operator U implicitly specifies the interaction.
We can study this process further via the time development operator U without
specifying any more details of the interaction.

Suppose that the initial state of (I) is an eigenstate of the R, say |riI . Then the
initial state of the combined system (I) and (II) is given by

|riI |0, miII (15.81)

If we require that the measurement should not change the value of the quantity
being measured, then we must have

U |riI |0, miII = |riI |r , m0 iII (15.82)

Clearly r is unchanged by the interaction. However, the value of m might change

and the value of changes to the value that corresponds to r, i.e., the indicator
variable now tells us (or allows us to infer) that the measured value is r. Since
we only care about the indicator variable, we need not put any restrictions on
the changes in m.

An assumption of this form is often made in the theory of measurement. It can

be significantly relaxed without affecting any of the arguments.

First, there is no reason why the state of the object (I) should not change during
the interaction (in practice it usually does). It is also not necessary for the state
of the apparatus to remain an eigenvector corresponding to a unique value of
m. We can assume a much more general result of the form
X 0 0
U |riI |0, miII = urr,m
,m
|r0 iI |r , m0 iII = |r ; (r, m)i (15.83)
r 0 ,m0

The labels (r, m) in the last vector are not eigenvalues but just indicate the
initial state from which the vector evolved.

The only restrictions here are that the final state vector is related to the initial
state vector by a unitary transformation, and that the particular value of r in
the initial state vector should correspond to a unique value of the indicator r in
the final state vector. This last condition is required if the apparatus is actually
to be able to carry out a measurement.

The values of r that correspond to different values of r should be clearly

distinguishable to the observer, i.e., they should be macroscopically distinct
values.

In the SG example the dynamical variable being measured is the spin component
z and the indicator variable is the momentum Pz . In this case, therefore, the

1166
indicator variable is not physically separate from the object of the measurement.
It only needs to be kinematically separate. Also in this case, we look at the
deflected beams and use the position coordinates of the point where the beam
strikes a screen as the indicator variable.

Now consider a general initial state for object (I) of the form
X
|iI = cr |riI (15.84)
r

This is not an eigenstate of the dynamical variable R that we are measuring. It

follows, however, from the linearity of the time development operator that the
final state of the system will be
!
X
U |iI |0, miII = U cr |riI |0, miII
r
X  f
= cr |r ; (r, m)i = m (15.85)
r

Thus, the final state vector is a coherent superposition of macroscopically distinct

indicator eigenvectors, which is the general theorem.

Finally, the probability, in the final state, that the indicator variable A of the
apparatus (II) has the value r is |cr |2 , which is just the same as the probability
in the initial state that the dynamical variable R of the object (I) had the value
r. This result is required if we are to have a faithful mapping from the initial
value of r to the final value r .

15.2.3 The Interpretation of the State Vector

We have just seen that if the initial state is not an eigenvector of the dynamical
variable being measured, then the final state vector for the whole system (object
of measurement + apparatus) must be a coherent superposition of macroscopi-
cally distinct indicator eigenvectors.

This enables us to get a better handle on how to interpret the quantum state.

Consider two standard interpretations:

(A) A pure state |i gives a complete description of an individual system. A

dynamical variable represented by the operator Q has a value q in the
state |i if and only if
Q |i = q |i

(B) A pure state describes the statistical properties of an ensemble of similarly

prepared systems.

1167
(A) is the more common interpretation in the literature, although it is not al-
ways made explicit. It assumes that because the state vector plays such an
important role in the mathematical formalism of quantum mechanics, it must
have an equally important role in the interpretation. It makes a strong corre-
spondence between the properties of the world and the properties of the state
vector.

Is it consistent with the measurement theorem, however?

 f
Because the final state vector m of the measurement process is not an eigen-
vector of the indicator variable, (A) says that the indicator cannot have a definite
value, i.e., we are in a superposition of macroscopically distinct indicator states.
In other words, this superposition state implies that we have a macroscopic un-
certainty.

Suppose that the indicator variable is the position of a needle on a meter or

a mark on chart recorder. Suppose, also, that for two adjacent values of the
measured variable, r and r 0 , the separation between the two corresponding in-
dicator variables r and r 0 is several centimeters. It would obvious to any
observer that the indicator position would be well defined to within a fraction
of a centimeter, but that the state vector involves a superposition of terms cor-
responding to values of r that differ by many centimeters.

Thus, it seems that an interpretation of the final state as a description of an

individual system cannot be reconciled with observation.

There is no such difficulty with (B) where the state vector is simply some ab-
stract quantity that characterizes the probability distributions of the dynamical
variables of an ensemble of similarly prepared systems (each member of the en-
semble consists of an object and a measuring apparatus).

The first idea for the measurement theorem came from Schrodinger in 1935 in
his now famous Cat experiment. We consider a box containing a cat, a flask
of poison gas, a radioactive atom, and an automatic device which releases the
poison gas when the atom decays.

If the atom were isolated, then after a time equal to one half-life, its state vector
would be

1
|atom i = (|undecayedi + |decayedi)
2
1
= (|ui + |di) (15.86)
2

1168
Now the atom is coupled to the cat via the apparatus. Therefore, the state of
the system after one half-life is
1
|system i = (|uiatom |aliveicat + |diatom |deadicat ) (15.87)
2
This is an entangled or correlated state. It is also a superposition of macroscopi-
cally distinct states (alive cat and dead cat). This is typical of any measurement
process.

Schrodinger argued the plausible interpretation that an individual systems prop-

erties are smeared over a range of values contained in the state vector. The
problem with this interpretation is that it implies a macroscopic smearing for
classical objects such as the unfortunate cat.

The physicists that believe interpretation (A) are now forced to introduce a new
postulate, called reduction of the state vector or collapse of the state vector.
 f
A new process arises during a measurement so that a transition m |r0 ; (r0 , m)i
occurs.

The process is called reduction or collapse and the final state is called the re-
duced state. Here the new final state vector is now an eigenstate of the indicator
variable A with eigenvalue r0 corresponding to the actual observed indicator
position.

The big question is - what is this new mechanism?

Proposed mechanisms
1. The reduction process is caused by an unpredictable and uncontrollable
disturbance of the object by the measuring apparatus. Any interaction
between the object (I) and the apparatus (II) that might act as the cause
of this disturbance must already be included in the Hamiltonian that we
use to construct the time development operator.

If the interaction satisfied the minimal condition that we derived earlier

(15.85)
!
X
U |iI |0, miII = U cr |riI |0, miII
r
X  f
= cr |r ; (r, m)i = m (15.88)
r

for it to be a measurement interaction, then it must, as above, lead to

a superposition and not to a reduced state. Therefore, it is difficult to
understand how the disturbance theory can make sense.

1169
 2. The observer causes the reduction process when she reads the result of the
measurement from the apparatus.

This is a variation of (1) with the observer instead of the apparatus causing
the disturbance. It also makes no sense. It also leads some physicists
to speculation about whether quantum mechanics can be applied to the
consciousness of the observer, which is an arena we do not need to enter.

3. The reduction is caused by the environment, where the environment means

the rest of the universe other than (I) and (II). The phenomenon is called
decoherence.

Here the proponents never quite make clear which part of the environment
is essential. If we include within (II) all these essential parts , then it is just
another disturbance model. We will have more to say about decoherence
later.

4. In proving the measurement theorem, the initial state of the apparatus

was assumed to be a definite pure state |0, miI . But, in fact, is an abbre-
viation for a set of many microscopic quantum numbers, which are never
determined in any experiment. It is very improbable that they will have
the same values on every repetition of the state preparation. Therefore,
the initial state should not be described as a pure state, but as a mixed
state involving a distribution of values. Proponents of this view assumed
that this would be the way to circumvent the measurement theorem.

We can defeat the last view by re-deriving the theorem using general state
operators.

15.2.4 The Measurement Theorem for General States

 i 
Instead of the pure vector m = |iI |0, miI |0, miI we now assume that the
initial state for the system is a more general state for the system ((I) and (II))
represented by X
i =
 i 
i 
wm m m  (15.89)
m

where the wm are the probabilities associated with each of the microscopic states
labeled by m, where m represents the quantum numbers(large number) of the
apparatus other than the indicator . The final state must be a mixture of
indicator eigenvalues, say of the form
X 2
X
d = |cr | vm |r ; (r, m)i hr ; (r, m)| (15.90)
r m

which is diagonal with respect to r .

If it were not so, then we would have a coherent superposition of macroscopically

1170
distinct indicator position eigenvectors, which is not allowed by interpretation
A. The final state is more plausible than the reduced state, which would have
prescribed a unique measurement result r0 . The new conjecture is consistent
with the prediction that the result of the measurement being r has probability
|cr |2 .

This conjecture is not correct, however. The actual final state of the measure-
ment process is given by
X
f = U i U + =
 f 
f 
wm m m  (15.91)
m
 f  i 
where m = U m . We then get
XX X
f = cr1 cr2 wm |r1 ; (r1 , m)i hr2 ; (r2 , m)| (15.92)
r1 r2 m

The terms with r1 6= r2 correspond to a coherent superposition of macroscop-

ically distinct indicator eigenvectors. These non-diagonal terms do not vanish
and thus the state will not reduce the diagonal form above. The measurement
theorem seems to be valid for these general states as well as the pure states.

It seems that in all cases where the initial state is not an eigenstate of the dynam-
ical variable being measured, the final state will involve coherent superpositions
of macroscopically distinct indicator eigenvectors. This makes interpretation
(A) untenable.

Thus, if we attempt to maintain the idea that the statistical quantum theory
is, in principle, able to produce a complete description of an individual physical
quantum system, then we seem to always end up in a theoretical box full of im-
plausible results. If one, however, views the quantum mechanical description as
the description of an ensemble of systems, it seems possible that the theoretical
difficulties will vanish.

15.2.5 Which Wave Function?

Why then do so many physicists consider an individual particle to have its own
wave function?

It turns out that taking this view seldom leads to any serious errors. This
is because the predictions of quantum mechanics that are derived from a wave
function consist of probabilities, and the operational significance of a probability
corresponds to a relative frequency. This means that one usually has to invoke
an ensemble of similar systems when one makes a comparison with experiment
independent of how one interprets the wave function.

Since so many results do not seem to depend in any critical manner on which

1171
interpretation one makes, should we dismiss the subject of interpretation as
irrelevant?

I do not think so!

Let us consider this interesting case. Electrons are emitted from a hot cathode
and then accelerated to form a beam to be used in interference experiments.

Using interpretation (A) we can account for the energy spread in the beam via
two different assumptions:

1. Each electron is emitted in an energy eigenstate (a plane wave), but the

particular energy varies from one electron to the next.

2. Each electron is emitted as a wave packet which has an energy spread

equal to the energy spread of the beam.

Do these two assumptions lead to quantitatively different predictions about the

resulting interference and thus allow us to distinguish between them experimen-
tally?

Let us analyze this system such that the electron beam is moving along the x-
axis (a one dimensional problem). Then, assumption (1) says that each electron
has a wave function of the form

k (x, t) = ei(kxt) (15.93)

and has an energy

~2 k 2
E = ~ = (15.94)
2M
The observed energy distribution (spread) of the beam enables us to calculate
the probability density (we call it W ()). The state operator corresponding to
this process will then be
Z
= |k i hk | W ()d (15.95)

In the coordinate representation we get

Z
(x, x0 ) hx| |x0 i = k (x, t)k (x0 , t)W ()d
Z
0
= eik(xx ) W ()d (15.96)

Note that all of the time dependence cancels out, so that we have a steady state!

All observable quantities, including the interference pattern, can be calculated

from the state function (x, x0 ).

1172
Now, assumption (2) says that if an individual electron is emitted at t0 in a
wave packet state
Z
t0 (x, t) = A()ei(kx(tt0 )) d (15.97)

2
then the energy distribution is given by |A()| = W (). The state function for
the emission process is obtained by averaging over the emission time (assumed
to be uniformly distributed):
T /2
Z
0 1
hx| |x i = lim t0 (x, t)t0 (x, t)dt0
T T
T /2
T /2 Z
Z Z
1 0 0 0
= lim A()ei(kx(tt0 )) d A ( 0 )ei(k x (tt0 )) d 0
T T
T /2

Now, integrating over t0 and then taking the limit, the integral is zero unless
= 0 (k = k 0 ). We thus get
Z
2 0
(x, x0 ) = |A()| eik(xx ) d (15.98)

This is the same result as obtained from assumption (1). Assumptions (1)
and (2) do not lead to any observable differences and any controversy over the
supposed wave functions of individual electrons seems to be pointless.

Now let us adopt interpretation (B), which was that

A pure state describes the statistical properties

of an ensemble of similarly prepared systems

We now regard the state operator as the fundamental description of the state
generated by the thermal emission process, which now yields an ensemble of
systems each of which is a single electron. In this case, we can obtain (x, x0 )
without ever speculating about individual wave functions.

Since this is a steady state process we must have

d
=0 (15.99)
dt
(in the Schrodinger picture). This implies that
h i
H, = 0 (15.100)

and thus H and possess a complete set of common eigenvectors. This set is the
free particle states | i, which have the coordinate representation hx | i =

1173
eikx and satisfy the eigenvalue equation H | i = ~ | i. In this case the state
operator has the form Z
= | i h |W ()d (15.101)

W () again describes the energy distribution in the beam. In the coordinate

representation this becomes
Z
0
(x, x0 ) = eik(xx ) W ()d (15.102)

Of course we get the same result as earlier, since all of these clever approaches
must agree with the same set of experimental predictions in the end!!

This last approach seems, however, to be superior since it avoids any (probably
pointless) speculations about the form of the supposed wave function of an
individual electron.

15.2.6 Spin Recombination Experiment

Is there any evidence that the state vector retains its form and does not undergo
any reduction process?

Let us consider a spin recombination experiment involving a single crystal neu-

tron interferometer. The crystal is cut into the shape shown inn Figure 15.3
below from a perfect (no dislocations or grain boundaries) crystal of silicon
about 10 cm long.

Figure 15.3: Single Crystal Neutron Interferometer

The diffraction beams are shown in Figure 15.4 below:

Figure 15.4: Diffraction Beams

1174
The incident beam at A is divided into a transmitted beam AC and a diffracted
(Bragg-reflected) beam AB. Similar divisions occur at B and C, with the trans-
mitted beams leaving the apparatus (no further role in the experiment). The
diffracted beams from B and C recombine coherently at D, where a further
Bragg reflection takes place. The interference of the amplitudes of the two
beams is observable by means of two detectors, D1 and D2 .

The amplitude at D1 is the sum of the transmitted part of CD plus the diffracted
part of BD, and similarly, the amplitude at D2 is the sum of the transmitted
part of BD plus the diffracted part of CD.

We assume for simplicity that the transmission and reflection coefficients are
the same at each of the vertices A,B,C and D and that free propagation of the
plane waves takes place between these vertices.

Only two distinct propagation directions are involved in this experiment. At

each diffraction vertex the amplitudes are redistributed between these two modes.
Figure 15.5 below shows what happens at a general diffraction vertex.

Figure 15.5: Diffraction Vertex

The time evolution and propagation processes are generated by linear opera-
tors. This implies that the relation between the amplitudes of the outgoing and
incoming waves is of the form
 0     
a1 a1 t r
= U where U = (15.103)
a02 a2 s u
or

a01 = ta1 + ra2

a02 = sa1 + ua2

U is a unitary matrix. The elements t an u are transmission coefficients and the

elements r and s are reflection coefficients.

Because U is unitary we have these relations:

2 2 2 2
U U + = I |t| + |u| = 1 and |r| + |s| = 1 (15.104)

1175
 det U = 1 |tu rs| = 1 (15.105)
 
1 u r
U 1 = U +
tu rs s t

s

t
= |u| = |t| and |s| = |r| (15.106)
r u

Now complex numbers can be regarded as 2-dimensional vectors, to which the

triangle inequality
|a + b| |a| + |b| (15.107)
applies. Using the inequality we have

|tu rs| = 1 |tu| + |rs| 1 (15.108)

Now since |u| = |t| and |s| = |r| we must have |tu| + |rs| = 1. This result is
compatible with |tu rs| = 1 only if tu and rs have the same complex phase,
and thus rs/tu must be real and negative.

If the amplitude at A is A , then the amplitudes at B and C will be

B = A reiAB and C = A teiAC (15.109)

where AB is the phase change due to the propagation through the empty space
between A and B, and similarly for AC . The amplitude that emerges toward
the detector D1 is the sum of the amplitudes from paths ABDD1 and ACDD1
(and similarly for detector D2 ):

D1 = A (a02 , a1 )eiAB (a01 , a2 )eiBD (a02 , a1 ) + (a01 , a1 )eiAC (a02 , a1 )eiCD (a02 , a2 )

= A seiAB reiBD s + teiAC seiCD u

= A s rseiAB eiBD + tueiAC eiCD

(15.110)
D2 = A (a02 , a1 )eiAB (a01 , a2 )eiBD (a01 , a1 ) + (a01 , a1 )eiAC (a02 , a1 )eiCD (a01 , a2 )

= A seiAB reiBD t + teiAC seiCD r

= A trs eiAB eiBD + eiAC eiCD

(15.111)

Any perturbation that has an unequal effect on the phases associated with the
two paths will influence the intensities of the beams reaching the detectors D1
and D2 . Since rs/tu is negative, it follows that if the interference between the
two terms in D2 is constructive, then the interference between the two terms
in D1 will be destructive and vice versa. The best way to detect such a per-
turbation is to monitor the difference between the counting rates in D1 and D2 .

1176
An experiment of this sort in 1975 detected a quantum interference due to
gravity. The interferometer was rotated about a horizontal axis parallel to the
incident beam causing a difference in the gravitational potential on paths AC
and AB and thus a phase shift in the interference pattern. In the spin recombi-
nation experiment, a beams of neutrons with spin polarized in the +zdirection
is incident from the left as shown in Figure 15.6 below:

Figure 15.6: Spin Recombination Experiment

At point A the beam is split into a transmitted beam AC and a Bragg-reflected

beam AB of equal intensity. Similar splitting occur at B and at C. The trans-
mitted beams at B and C exit the apparatus. We now insert a spin-flipper into
the beam CD as shown. The dots indicate spin up beams and the x indicates
a spin down beam. The spin up and spin down beams are then recombined at
D. The detectors then determine the spin state of the beams that emerge from
the apparatus after point D.

We let the vectors |+i and |i denote the spin-up and spin-down eigenvectors
of z . The neutrons at point B have the spin state |+i and the neutrons after
the spin-flipper have the spin state |i. We then ask this question.

What will be the spin state when the beams recombine at D?

At one time, when the beams are at B and C, they are separated by several
centimeters. This means that their spatial wave functions do not overlap. One
might suppose, in this case, that all coherence is lost and that no interference
is possible, i.e., that the spin state should be an incoherent mixture of spin-up
and spin-down states of the form

1
inc = (|+i h+| + |i h|) (15.112)
2

This state would also result if we applied the reduction hypothesis.

If, on the other hand, the coherence is maintained somehow, then the spin state
will be of the form

1
coh = |ui hu| where |ui = ei |+i + ei |i


(15.113)
2

1177
Both of these state operators predict that hz i = 0, i.e.
inc
= T r inc z = h+| inc z |+i + h| inc z |i


hz i
= h+| inc |+i h| inc |i
1 1
= h+| ( (|+i h+| + |i h|)) |+i h| ( (|+i h+| + |i h|)) |i
2 2
1 1
= =0
2 2
coh
= T r coh z = h+| coh z |+i + h| coh z |i


hz i
= h+| coh |+i h| coh |i
1 1
= h+| (|ui hu|) |+i h| (|ui hu|) |i = =0
2 2
or that the zcomponent of spin is equally likely to be positive or negative.

However, inc actually predicts zero polarization in any direction, while coh
predicts that the spin is polarized in some direction in the xyplane. We can
see this by computing
coh
= T r coh x = h+| coh x |+i + h| coh x |i


hx i
= h+| coh |i h| coh |+i = h+| (|ui hu|) |i h| (|ui hu|) |+i
1
= h+| ei |+i + ei |i ei h+| + ei h| |i



2
1
h| ei |+i + ei |i ei h+| + ei h| |+i



2
1 i i 1 i i
= e e e e = cos( ) (15.114)
2 2
Even though the phases and are not necessarily known in advance, their
difference can be systematically varied by placing known phase-shifters in one
of the beams. Such an experiment was done in 1982 and it found a periodic
dependence of hx i on the phase shift and no such dependence for hz i. This
confirms that the coherent superposition is the correct state.

Let us look at the state operators in more detail.

If we account for both the position and spin variables, the state function(vector)
should be
|i = + (~r) |+i + (~r) |i (15.115)
where the wave functions + (~r) and (~r) vanish outside the beams. The spin
state operator is given by
2 2
= |i h| = |+ | |+i h+| + + |+i h| + + |i h+| + | | |i h|
(15.116)

1178
or    2

h+| |+i h+| |i |+ | +
= = 2 (15.117)
h| |+i h| |i + | |
Along AB, AC, and from B to the left of D, (~r) = 0 and from the right of
the spin-flipper to the left of D, + (~r) = 0. Both components are nonzero to
the right of D.

At the point D the off-diagonal terms are nonzero, which means that we have
a coherent superposition. In this experiment, the preservation of the coherence
over a distance of several centimeters is possible because the scatterer is cut
from a single crystal of silicon and the relative separation of the components
are stable to within the interatomic separation distance.

Suppose that the spectrometer were not such a high precision device and that
the relative separations of points A, B, C, and D are subject to random fluctu-
ations that are larger than the spatial spread of the neutron wave function.

This gives rise to random fluctuations in the phases and and hence, in
the phases of the off-diagonal terms. Different neutrons passing through the
spectrometer at different times would experience different configurations of the
spectrometer, and to determine the observed statistical distributions we must
average over these fluctuations.

If we regard the noise fluctuations as a part of the state preparation procedure,

then should be averaged over the noise. If the phase difference ( ) fluc-
tuates by more than 2, then the off-diagonal terms will average to zero, will
be diagonal and will reduce to inc which corresponds to the state operator for
the reduced state.

The reduced state can therefore be significant under certain conditions. It does
not seem, however, to be a fundamental object, but , instead, arises only due
to an effect on the system (neutron + spectrometer) from its environment(the
cause of the noise fluctuations). The separation of the system and the environ-
ment is, however, artificial. If the reduction takes place in this manner, then it
is not a new fundamental process, and it would not have anything to do with
measurement.

Let us now include the environment not as an external effect on the system,
but as an integral part of the system. The neutrons that follow path ABD
will interact differently with the environment than those that follow path ACD.
These interactions will affect the state of the environment and therefore the
final state must now be

|i = + (~r) |+i |e1 i + (~r) |i |e2 i (15.118)

where |e1 i is the state of the environment if the neutron followed path ABD

1179
and |e2 i is the state of the environment if the neutron followed path ACD. If
|e1 i = |e2 i, then this inclusion of the environment has no effect. If we recalculate
the state operator we now get
 2

|+ | + he2 | e1 i
= 2 (15.119)
+ he1 | e2 i | |
If the difference between the effects of taking paths ABD and ACD on the en-
vironment is so great that |e1 i and |e2 i are orthogonal, then the state operator
reduced once again to inc .

We have thus seen two possible methods of handling the influences, if any, of
the environment on the experiment. One method treats the effect of the envi-
ronment as an outside perturbation, which introduces random phases and we
lose coherence if the random phase fluctuations are large enough. In the second
method, the environment is included in the state vector of the system. It is then
the effect of the apparatus on the environment, rather than the environment on
the apparatus, that is important. Clearly, these two approaches are equivalent.

15.2.7 Joint and Conditional Probabilities

In all of our discussions, an experiment consists of a state preparation followed
by the measurement of a single quantity. Suppose, instead, of a single mea-
surement, it involves a sequence of measurements of two or more dynamical
variables. In this case, we will not only need the probability distributions for
the individual quantities to be measured, but also will need to consider correla-
tions between values of the quantities to be measured.

We can deal with this by computing the joint probability distribution for the
results of two or more measurements, or the probability for one measurement
conditional on both state preparation and the result of another measurement.

We discussed these joint and conditional probabilities earlier in Chapter 5. Ax-

iom 4 related them by
Prob(A B|C) = Prob(A|C)Prob(B|A C) (15.120)
We let event C be the state preparation that corresponds to the state operator
so we will replace C with . The events A and B will be the results of two
measurements following that state preparation. Let R and S be two dynamical
variables represented by the Hermitian operators R and S such that
R |rn i = rn |rn i and S |sm i = sm |sm i (15.121)
We now define two projection operators.

Let be some interval (in eigenvalue space). Then the operators

X X
MR () = |rn i hrn | , MS () = |sm i hsm | (15.122)
rn sm

1180
project onto the subspace spanned by those eigenvectors whose eigenvalues lie
in the interval .

Let A denote the event of R taking a value in the range a (R a ). Let B

denote the event of S taking a value in the range b (S b ).

We suppose that the first of these events takes place at time ta and the second at
time tb . We use the Heisenberg representation and assume that the specification
of ta is implicit in the operators R and MR () and that the specification of tb
is implicit in the operators S and MS ().

Review and Digression on Probability Distributions

Let us now extend some of our earlier discussions (Chapter 5). In general we
have D E  
R = T r R (15.123)
Now if g(r)dr is the probability that the observable R lies between r and r + dr,
then by definition
D E Z  
F (R) = F (r0 )g(r0 )dr0 = T r F (R) (15.124)

Let us now extract the probability density g(r) using these results.

Discrete Spectrum
Let R have a purely discrete spectrum. We can then write
X
R = rn |rn i hrn | (15.125)
n

Now consider the function

(
1 R<r
F (R) = (r R) = (15.126)
0 R>r
We then have
D E Zr
(r R) = g(r0 )dr0 = Pr ob(R < r|)

 
= T r (r R)
" !#
X
= T r (r rn ) |rn i hrn |
n
X
= (r rn ) hrn | |rn i (15.127)
n

1181
Therefore,

g(r) = Pr ob(R < r|)
r
X
= (r rn ) hrn | |rn i
r n
X
= (r rn ) hrn | |rn i (15.128)
n

This makes it clear that g(r) = 0 if r is not an eigenvalue.

The probability that R will have the discrete value r in the virtual ensemble
characterized by is

P rob(R = r|) = lim [P rob(R < r + |) P rob(R < r |)]

0
X
= r,rn hrn | |rn i (15.129)
n

If we define the projection operator

X
P (r) = |rn i hrn | r, rn (15.130)
n

which projects onto the subspace spanned by all the degenerate eigenvectors
with eigenvalue r = rn . We then have
 
P rob(R = r|) = T r P (r) (15.131)

In the special case of a pure state where = |i h| and a non-degenerate

eigenvalue, this reduces to
2
Prob(R = rn |) = |hrn | i| (15.132)

A particular dynamical variable will, in general, have a non-vanishing statisti-

cal dispersion in most states. In the case of a discrete variable, however, it is
possible for all of the probability to be concentrated on a single value. If the dy-
namical variable R takes on the unique value r0 (assumed to be a non-degenerate
eigenvalue) with probability = 1 in some state, then from
X
P rob(R = r|) = r,rn hrn | |rn i (15.133)
n

we must have
hr0 | |r0 i = 1 (15.134)
But, any state operator must satisfy T r2 1 which implies
X X 2
hrn | |rm i hrm | |rn i = |hrn | |rm i| 1 (15.135)
m,n m,n

1182
Since one term in the sum already accounts for an amount = 1, all of the other
diagonal and non-diagonal matrix elements of must vanish.

Therefore the only state for which R takes on the non-degenerate eigenvalue r0
with probability = 1 is the pure state = |r0 i hr0 |. This is what we mean by
an eigenstate.

Continuous Spectrum
Let Q be a Hermitian operator having a purely continuous spectrum such that

Q |q 0 i = q 0 |q 0 i (15.136)

Then we can write Z

Q = q 0 |q 0 i hq 0 | dq 0 (15.137)

The eigenvectors are normalized by the relation

hq 0 | q 00 i = (q 0 q 00 ) (15.138)

Now let g(q)dq be the probability that the corresponding observable Q lies
between q and q + dq. We then get

D E Zq
(q Q) = g(q 0 )dq 0 = Pr ob(Q < q|)


  Z
= T r (q Q) = T r (q q 0 ) |q 0 i hq 0 | dq 0

Zq
= hq 0 | |q 0 i dq 0 (15.139)

or

g(q) = Pr ob(Q < q|) = hq| |qi (15.140)
q
Once again in the pure state we get
2
g(q) = |hq | i| (15.141)

The expressions for the probability and the probability density always consist
of a relation between two factors: one characterizing the state (the state func-
tion) and one characterizing a portion of the spectrum being observed(the filter
function).

In the expression  
P rob(R = r|) = T r P (r) (15.142)

1183
they are and P (r) respectively.

2 2
In the expressions |hrn | i| and |hq | i| they are the state vector and an
eigenvector belonging to the observable.

The two objects have very distinct natures,i.e., the state vector must be nor-
malized and therefore belongs to Hilbert space. The filter function does not
necessarily belong to Hilbert space, but rather to an extended or rigged Hilbert
space.

Returning to our discussion, we have the following situation. Associated with

any dynamical variable R and its corresponding Hermitian operator R is a fam-
ily of projection operators MR () which are related to the eigenvectors and
eigenvalues of R by X
MR () = |rn i hrn | (15.143)
rn

The sum is over all eigenvectors(even degenerate) whose eigenvalues lie in the
subset . For a continuous spectrum the sum becomes an integral. Then the
probability that the value of R will lie within is given by

P rob(R |) = T r (MR ()) (15.144)

If the region contains only one eigenvalue, then this reduces the earlier result
for P rob(R = r|) and in the continuous spectrum case it is equal to the integral
of the probability density over the region . This result satisfies all of the 4
probability axioms in Chapter 5.

We then have

P rob(A|C) = P rob(R a |) = T r (MR (a )) (15.145)

The joint probability P rob(A B|C) can be evaluated only if we can find(from
quantum mechanics) a projection operator for the compound event A B. This
is possible if the projection operators MR (a ) and MS (b ) commute. In that
case the product MR (a )MS (b ) is also a projection operator that projects
onto the subspace spanned by those common eigenvectors of R and S with
eigenvalues in the ranges a and b , respectively. We then have

P rob(A B|C) = P rob ((R a ) (S b )|)

= T r (MR (a )MS (b )) (15.146)

This is just the joint probability that both events A and B occur on the condition
C, or, in other words, it is the probability that the result of the measurement
of R at time ta is in the range a and the result of the measurement of S at
time tb is in the range b , following the state preparation corresponding to .
If R and S commute this calculation is possible for arbitrary ranges.

1184
The last term P rob(B|A C) is

P rob(B|A C) = P rob ((S b )|(R a ) ) (15.147)

It is the probability for a result of the S measurement, conditional on the state

preparation and a certain result of the R measurement. We do not know how
to deal with these expressions directly.

Two paths are now open to us.

1. We can regard the preparation of the state and the following measure-
ment of R as a composite operation that corresponds to the preparation
of a new state 0 .

2. We can define
P rob(A B|C)
P rob(B|A C) = (15.148)
P rob(A|C)
since we know how to calculate the right-hand-side.

Filtering-Type Measurements
If we want to regard the initial state preparation followed by the R measure-
ment as a composite operation that results in a new state 0 , then we require a
detailed description of the R measurement apparatus and a dynamical analysis
of its operation. This is only possible for particular cases and no general treat-
ment can be done.

Are there any types of measurements that we can treat without too much diffi-
culty?

Let us consider a measurement of the filtering type where the ensemble of sys-
tems generated by the state preparation is separated into sub-ensembles
according to the value of the dynamical variable R (SG apparatus is an exam-
ple of this type of measurement).

If we consider the result of a subsequent S measurement on only that sub-

ensemble for which R a , and ignore the rest, then we are determining the
conditional probability

P rob(B|A&C) = P rob ((S b )|(R a ) ) (15.149)

This filtering process, which has the effect of removing all values of R except
those for which R a , can be regarded as preparing a new state that is
represented by
MR (a )MR (a )
0 = (15.150)
T r [MR (a )MR (a )]

1185
so that we have
P rob(B|A ) = P rob ((S b ) | (R a ) )
= P rob ((S b ) |0 ) = T r (0 MS (b )) (15.151)
Finally, we can calculate a joint probability for two filtering-type measurements
as:
P rob(A B|C) = Pr ob(A|C) Pr ob(B|A C)
= Pr ob(A|) Pr ob(B|A )
= T r (MR (a )) T r (0 MS (b ))
 
MR (a )MR (a )
= T r (MR (a )) T r MS (b )
T r [MR (a )MR (a )]
T r (MR (a ))
= T r (MR (a )MR (a )MS (b ))
T r [MR (a )MR (a )]
T r (MR (a ))
= T r (MR (a )MS (b )MR (a ))
T r [MR (a )MR (a )]
T r (MR (a ))
= T r (MR (a )MS (b )MR (a ))
T r [MR (a )]
= T r (MR (a )MS (b )MR (a )) (15.152)
In the case when MR (a ) and MS (b ) commute, this reduces to the earlier
expression
P rob(A B|C) = T r (MR (a )MS (b )) (15.153)
The derivation of this last relation required that R and S commute, but no such
restriction was required to derive
P rob(B|A ) = T r (0 MS (b )) (15.154)
The latter, however, is restricted to filtering-type measurements.

We have just seen that both results are consistent with

P rob(A B|C) = P rob(A|C)P rob(B|A C) (15.155)
when all the various conditions are satisfied together. It does seem strange
however that the conditions for evaluating the two sides of the equation should
be different. This is not really a puzzle, however.

The derivation of P rob(AB|C) was implicitly based on the assumption that the
measurements of R and S were equivalent to, or at least compatible with, a joint
filtering according to the eigenvalues of R and S, i.e., a product of projection
operators. This is only possible if R and S commute. In this case the time order
of measurements is irrelevant, as is clear from the symmetry with respect to the
two projection operators, i.e.,
P rob(A B|C) = T r (MR (a )MS (b )) (15.156)

1186
Remember these are time-dependent operators in the Heisenberg picture.

If the operators R and S do not commute, the above relation is not true, as
the definition of the joint probability. In this case we must carefully observe
the time orderings, because it is the R measurement that serves as (part of the)
state preparation for the S measurement and not vice versa. This is clear in the
lack of symmetry in the last result

P rob(A B|C) = T r (MR (a )MS (b )MR (a )) (15.157)

Application to Spin Measurements

Let us now illustrate these ideas in a spin = 1/2 system.

Assume a state represented by |i has been prepared. It is then subjected to

three successive measurements of the filtering type:

a measurement of z at time t1
a measurement of u at time t2
a measurement of x at time t3

where the udirection is in the zxplane, making an angle with respect to

the zaxis. These filtering measurements will split the initial beam first into
two, then into four, and finally into eight separated subbeams as in Figure 15.7
below:

Figure 15.7: Filtering Splitting Beam

Seven SG machines are required to carry out this experiment. We assume that
the spin vector is a constant of the motion between the measurements.

Each of the eight final outcomes of this experiment corresponds to a particular

combination of results (+1 or 1) for the three (z , u , x ) measurements,
and the probability of these various outcomes is, in fact, the joint probability
for the results of the three measurements.

The full notation for this joint probability should be

P rob(z = a, u = b, x = c| X) (15.158)

1187
with a = 1, b = 1 and c = 1. As indicated above, the probability is condi-
tional on the state preparation (denoted by ) and the configuration of the SG
machines (denoted by X). We abbreviate it P rob(a, b, c| X) with the time
ordering assumed.

It is important to note that this is the joint probability for the results of three ac-
tual measurements, and not the joint distribution for hypothetical simultaneous
values of three noncommuting observables. Moreover, the various subbeams in
this experiment are all separated in space and no attempt is made to recombine
them. Thus, questions of relative phase and coherence are irrelevant.

We write the initial state vector as

|i = |z+i + |zi (15.159)
in the basis of the z eigenvectors. The amplitudes are divided at each filter-
ing operation and the division is calculated using projection operators. The
absolute squares of these amplitudes then give the probabilities of the various
measurements.

After the measurement of z at t1 we have

Pz (a|&X) = h| Mz (a) |i
2
= h| (|ai ha|) |i = |ha | i|
(
||2 for a = +1 or |ai = |z+i
= (15.160)
||2 for a = 1 or |ai = |zi
Similarly, after the measurement of u at t2 we have
   
Pzu (a, b| X) = h| Mz (a) Mu (b) Mz (a) |i
= h| Mz (a)Mu (b)Mz (a) |i (15.161)
where
Mu (1) = |ui hu| (15.162)
and

|u+i = cos |z+i + sin |zi , |ui = sin |z+i + cos |zi (15.163)
2 2 2 2
Finally after the measurement of x at t3 we have
   
Pzu (a, b, c| X) = h| Mz (a)Mu (b) Mx (c) Mu (b)Mz (a) |i
= h| Mz (a)Mu (b)Mx (c)Mu (b)Mz (a) |i (15.164)
Now we can always write
X
Pzu (a, b| X) = Pzu (a, b, c| X) (15.165)
c=1

1188
i.e., the probability for the z and u measurements independent of what happens
at the x measurement (the later x measurement should not affect them!).

We then get
X
Pzu (a, b| X) = Pzux (a, b, c| X)
c=1
X
= h| Mz (a)Mu (b)Mx (c)Mu (b)Mz (a) |i
c=1

= h| Mz (a)Mu (b)(Mx (c = +1) + Mx (c = 1))Mu (b)Mz (a) |i

= h| Mz (a)Mu (b)IMu (b)Mz (a) |i
= h| Mz (a)Mu (b)Mz (a) |i (15.166)

as expected.

So the presence of the x filter has no effect on the earlier measurements. Sim-
ilarly, we have
X X
Pz (a| X) = Pzux (a, b, c| X)
b=1 c=1

= h| Mz (a) |i (15.167)

since the absence of the x and u filters has no effect on the measurement of
z .

Now several interesting conditional probabilities can be calculated from these

joint probability distributions using the formula

P rob(A B|C)
P rob(B|A C) = (15.168)
P rob(A|C)

(1) Conditioning on a Prior Measurement

Let C be the preparation of the state , A be the result z = +1 for the
first measurement, and B be to result of the measurement of u . Then the
probability that the second measurement will yield u = +1 conditional on
both the state preparation and the result z = +1 in the first measurement is

Pzu (+1, +1|)

P rob(u = +1|(z = +1) ) =
Pz (+1|)
h| Mz (+1)Mu (+1)Mz (+1) |i
=
h| Mz (+1) |i
 cos 2
 

= 2
2
= cos2 (15.169)
|| 2

1189
This is the same as the probability of obtaining u = +1 conditional on the new
state being | 0 i = |z+i, i.e., as if the first measurement collapsed or reduced
the state to correspond to the measurement. We have not, however, assumed
any reduction process. This result follows from existing quantum mechanical
rules. It does not, however, say that any collapse occurred!

(2) Probability Distribution for x regardless of z and u

The probability of the result x = +1 in the final measurement regardless of
the results of the prior measurements is
X X
Px (+1| X) = P (a, b, +1| X)
a=1 b=1
X X
= h| Mz (a)Mu (b)Mx (+1)Mu (b)Mz (a) |i (15.170)
a=1 b=1

Now

P (a, b, c| X) = h| Mz (a)Mu (b)Mx (c)Mu (b)Mz (a) |i (15.171)

or
1 2
P (+1, +1, +1| X) = || cos2 [1 + sin ] (15.172)
2 2
1 2
P (1, +1, +1| X) = || sin2 [1 + sin ] (15.173)
2 2
1 2
P (+1, 1, +1| X) = || sin2 [1 sin ] (15.174)
2 2
1 2
P (1, 1, +1| X) = || cos2 [1 sin ] (15.175)
2 2
and therefore carrying out the sum we get
1h 2 2
i
Px (+1| X) = 1 + (|| || ) sin cos (15.176)
2
This is the probability of obtaining the result x = +1 with the z and u filters
in place, but ignoring the results of the z and u measurements. It is not equal
to the probability of obtaining x = +1 with the z and u filters absent, which
is
1 2
Px (+1|) = h| Mx (+1) |i | + | (15.177)
2
These two results differ because the particle must pass through the z and u
filters before reaching the x filter and clearly the presence of the other filters
is relevant!

Thus, we must always explicitly take the dynamical action of the apparatus into
account when developing a theory of measurement.

1190
(3) Conditioning on Both Earlier and Later Measurements
We now calculate the probability for a particular result of the intermediate u
measurement, conditional on specified results for the preceding z measurement
and the following x measurement.

The later measurement cannot have any causal effect on the outcome of the ear-
lier measurement, but it can give relevant information because of the statistical
correlations between the results of successive measurements!

We use
P rob(A B|C)
P rob(B|A C) = (15.178)
P rob(A|C)
with C = X, A = (z (t1 ) = +1) (x (t3 ) = +1) and B = (u (t2 ) = +1).
We get

P rob((u (t2 ) = +1)|(z (t1 ) = +1) (x (t3 ) = +1) X)

P (+1, +1, +1| X)
= (15.179)
Pzx (+1, +1| X)
From earlier we have
1 2
P (+1, +1, +1| X) = || cos2 [1 + sin ] (15.180)
2 2
and

Pzx (+1, +1|&X)

= P (+1, +1, +1| X) + P (+1, 1, +1| X)
1 2 1 2
= || cos2 [1 + sin ] + || sin2 [1 sin ]
2 2 2 2
1 2
= || (1 + sin cos ) (15.181)
2
Putting it all together we get

P rob((u (t2 ) = +1)|(z (t1 ) = +1) (x (t3 ) = +1) X)

 
2 1 + sin
= cos (15.182)
2 1 + sin cos
Thus, the probability distribution for u is well-defined for all . There is ,
however, no quantum state 0 such that
 
0 2 1 + sin
P rob(u = +1| ) = cos (15.183)
2 1 + sin cos
This is clear from the fact that

P rob((u (t2 ) = +1)|(z (t1 ) = +1) (x (t3 ) = +1) X) = 1 (15.184)

1191
i.e., the probability is 1 for both = 0 and = /2, which is very different than
case (1).

How do we resolve this paradox?

We must remember that a quantum state is characterized by a well-defined

state preparation procedure that can yield a statistically reproducible ensemble
of systems, and not merely by the specification of abstract information.

This is why the probabilities in these examples have been conditional on the
apparatus configuration X.

Now the angle specifies the direction of the u filter and it must be included
in X, i.e., we should write X . By conditioning on the final result x = +1 we
select a subensemble discarding those cases in which the result is 1. However,
a part of the specification of this subensemble is that its members have passed
through the u filter. Thus, the conditions that define the subensemble include
the value of the angle , which therefore may not be changed.

In the usual situation, as in case (1), all of the specifications correspond to

operations performed before the measurement of interest. Hence, they define
an ensemble whose composition does not depend upon what measurement we
may choose to perform. We then have a well-defined state, which yields a well-
defined probability distribution for any subsequent measurement that we may
choose to perform.

But this is not possible if we specify conditional information both before and
after the measurement of interest, as in the last example.

Thus, the paradox is resolved and makes us realize that we must pay very careful
attention to the state preparation concepts.

Comparing Approaches

Let us now look at all of this theory in a more standard manner(as is done in
many textbooks) and then compare the two approaches.

The standard axioms of quantum theory are:

1. The state is described by a vector |i in a linear space.

2. The observables are represented by Hermitian operators and a function of

an observable f (observable) is represented by f (operator).

3. The expectation value of an observable is h| operator |i.

1192
 4. The dynamics are given by the Schrodinger equation

i~ |i = H |i where H = Hamiltonian or energy operator
t

5. The only possible results of a measurement of an observable are the eigen-

values of the corresponding operator.
6. In a measurement of the observable A with the result a, the original state
changes into state |ai where

A |ai = a |ai

In this text, we made a different set of assumptions:

Postulate 1

For each dynamical variable or observable,

which is a physical concept, there corresponds
a Hermitian, linear operator, which is a
mathematical object.

The possible values of any measurement of

the observable are restricted to the eigenvalues
of the corresponding operator.

Postulate 1 is equivalent to axioms (2) and (5).

Postulate 2

(a) A density operator exists for every real physical system.

(b) The expectation value of an operator B is given by
D E
B = T r(W B)

Postulate 2 is equivalent to axioms (1) and (3).

We also assumed that we were in a linear vector space and hence could prove
Stones theorem which is equivalent to axiom (4) above.

We did not assume axiom (6).

In the discussion just completed we saw that axiom (6) seems to be incorporated
in our theory but is not generally true.

It is at the heart of all the controversy in quantum mechanics.

1193
Let us see how the standard approach proceeds.

First, both approaches imply that if a system is in the state (linear combination)
X
|i = cn |an i where A |an i = an |an i (15.185)
n

then the probability during a measurement of A that we measure the eigenvalue

am is given by
2 2
|ham | i| = |cn | (15.186)
The conceptual difficulties that we will elucidate within the structure of the
quantum mechanical state do not imply a weakness in quantum theory, whose
experimental validity has never encountered any limitations, but only a weakness
of our imagination!!

The Density Matrix in the Standard Approach

Pure and Mixed Ensembles
If a system is in a state |i, then the observable A has the average or expectation
value D E
A = h| A |i (15.187)

If we define a new operator called the density operator given by

= |i h| (15.188)

then we have these properties

X X
T r(A) = hn|A |ni = hn | i h| A |ni
n n
X
= h| A |ni hn | i = h| AI |i = h| A |i (15.189)
n

= h| I |i = h | i = 1
T r() = T r(I) (15.190)
2
= (|i h|)(|i h|) = |i h | i h| = |i h| = (15.191)
If we let {|ni} and {|mi} be two different basis systems, then we have
X X
T r(X) = hn|X |ni = hn | mi hm| X |ni
n n,m
X X
= hm| X |ni hn | mi = hm|X |mi
n,m m

so that the trace is independent of the basis.

If the systems or objects under investigation are all in one and the same state
|i, we call this a pure ensemble or we say that the system is in a pure state.

1194
In order to verify the probability predictions contained in the state vector |i ex-
perimentally , we must , in fact, investigate an ensemble of identically prepared
objects. If X
|i = cn |ni (15.192)
n

then the eigenvalue an will be the A measurement result Nn times for an en-
semble of N objects. The larger N , the more precisely Nn /N approaches the
probability |cn |2 , i.e.,
2 Nn
|cn | = lim (15.193)
N N
and the expectation value correspondingly becomes
D E X
2 1 X
A = |cn | an = lim Nn an (15.194)
N N
n n

In addition to the inherent statistical character residing in the states themselves,

an ensemble can also contain a statistical distribution of states.

If an ensemble with distinct states is present, we call this a mixed ensemble,

a mixture, or we say a mixed state. Suppose that of the N representatives of
the ensemble, N1 are in the state |1 i, N2 are in the state |2 i, etc. Then the
probability that an arbitrarily chosen element of the ensemble is in the state
|i i is given by
Ni X X Ni 1 X
pi = pi = = Ni = 1 (15.195)
N i i
N N i

The expectation value of A is then

D E X
A = pi hi | A |i i (15.196)
n

This expectation value can also be represented by a density matrix of the form
X
= pi |i i hi | (15.197)
i

with the properties

  X X
T r A = hn|A |ni = pi hn | i i hi | A |ni
n n,i
X X D E
= pi hi | A |ni hn | i i = pi hi | A |i i = A (15.198)
n,i i

T r () = 1 (15.199)
X
2 = pi pj |i i hi | j i hj | =
6 (15.200)
i,j

1195
For each |i
X 2
h| |i = pi |h | i i| 0 non negative (15.201)
i

If X
|mi = Pm |mi = Pm |mi hm| (15.202)
m

then X 2
h| |i = Pm |hm | i| 0 Pm 0 (15.203)
m

and
!
X X X X
Pm = hm| |mi = hm| pi |i i hi | |mi
m m m i
X X 2
= pi |hm | i i| = 1
i m

We also have X X
2 = 2
Pm |mi hm| T r2 = 2
Pm <1 (15.204)
m m

Thus, the criterion for a pure state is T r2 = 1 and for a mixed state T r2 < 1.

Now the expectation value of the projection operator |ni hn| (discrete spectrum)
is given by
X 2
X  2
T r (|ni hn| ) = pi |hn | i i| = pi c(i)
n  (15.205)
i i

which is equal to the probability of obtaining the state |ni as a result of a

measurement.

The expectation value of the projection operator |xi hx| (continuous spectrum)
is given by
X 2
X 2
T r (|xi hx| ) = pi |hx | i i| = pi |i (x)| (15.206)
i i

which is equal to the probability of obtaining the state |xi as a result of a mea-
surement.

Now let us consider a system consisting of two subsystems 1 and 2 with or-
thonormal basis states {|1ni} and {|2ni}, respectively. A general pure state in
the direct product space is then
X X 2
|i = cnm |1ni |2mi where |cnm | = 1 (15.207)
m,n n,m

1196
The corresponding density operator is
X X
= |i h| = cnm cn0 m0 |1ni |2mi h1n0 | h2m0 | (15.208)
n,m n0 .m0

If we carry out measurements concerning only subsystem 1, that is, if the op-
erators corresponding to the observables being measured only act on the states
|1ni, then
D E   X
A = T r A = h2m00 | h1n00 |A |1n00 i |2m00 i
m00 ,n00
!
X X X
00 00
= h2m | h1n | cnm cn0 m0 0 0
|1ni |2mi h1n | h2m | A |1n00 i |2m00 i
m00 ,n00 n,m n0 .m0
X X X
= cnm cn0 m0 (h2m00 | h1n00 |)(|1ni |2mi) h1n0 | A |1n00 i h2m0 | 2m00 i
m00 ,n00 n,m n0 .m0
X X X
= cnm cn0 m0 m,m00 m0 ,m00 n,n00 h1n0 | A |1n00 i
m00 ,n00 n,m n0 .m0
X
= cnm cn0 m h1n0 | A |1ni
n,n0 ,m

Now we can write

X X X
cnm cn0 m h1n0 | A |1ni = h1n00 | cnm cn0 m |1ni h1n0 | A |1n00 i
n,n0 ,m n00 n,n0 ,m

X  
= T r1 cnm cn0 m |1ni h1n0 | A = T r1 A
n,n0 ,m

where
X
= cnm cn0 m |1ni h1n0 | = T r2 appropriate density operator (15.209)
n,n0 ,m

so that finally D E h i
A = T r1 (T r2 ) A (15.210)
where T rk means trace over subsystem k. Now
X X
2 = cnm cn0 m cn1 m1 cn01 m1 |1ni h1n0 | 1n1 i h1n01 |
n,n0 ,m n1 ,n01 ,m1
X X
= cnm cn0 m cn1 m1 cn01 m1 |1ni n0 n1 h1n01 |
n,n0 ,m n1 ,n01 ,m1
! !
X X X
= cnm cn0 m cn0 m1 cn01 m1 |1ni h1n01 |
n,n0 ,n01 m m1

1197
which say that, in general
2 6= (15.211)
If, however, the cnm take the form
X 2
X 2
cnm = bn dm with |bn | = 1 = |dm | (15.212)
n m

then we have
! !
X X X
2 = cnm cn0 m cn0 m1 cn01 m1 |1ni h1n01 |
n,n0 ,n01 m m1
! !
X X 2
X 2
= bn bn0 |dm | bn0 bn01 |dm1 | |1ni h1n01 |
n,n0 ,n01 m m1
X   X X 2
= (bn bn0 ) bn0 bn01 |1ni h1n01 | = bn bn01 |1ni h1n01 | |bn0 |
n,n0 ,n01 n,n0 ,n01 n0
X
= bn bn01 |1ni h1n01 | =
n,n0 ,n01

In this case we have,

! !
X X
|i = bn |1ni dm |2mi (15.213)
n m

i.e., it is the direct product of two pure states of the subspaces 1 and 2.

It is also true, in this case, that

! !
X X X
T r1 2 = cnm cn0 m cn1 m1 cn01 m1 =1 (15.214)
n,n0 n m1

Except for this special case, represents the density operator of a mixed en-
semble. If the information of a subspace is ignored, then the pure state becomes
a mixed state.

Although the total system is in a pure state, the density operator , which
yields all expectation values pertaining only to subsystem 1, represents a mixed
ensemble in general.

Projection Operators
The projection operator in subspace 2, |2mi h2m| , projects onto the state |2mi.
Therefore,
X X
(|2mi h2m|) |i = |i = (|2mi h2m|) cnm |1ni |2m0 i = cnm |1ni |2mi
m0 ,n n

1198
and the state operator becomes

|2mi h2m| |2mi h2m|

(15.215)
T r1 h2m| |2mi

This implies that if represents a pure ensemble, the so does the projected state
operator.

In general, filters can be represented mathematically by the application of pro-

jection operators onto the state operator as above.

The von Neumann Equation

We now re-derive the equation of motion for the state operator. The Schrodinger
equation

i~ |i i = H |i i i~ hi | = hi | H (15.216)
t t

gives

" #
X X   


= pi |i i hi | = pi |i i hi | + |i i hi |
t t i i
t t
" #
1 X   1 X
= pi H |i i hi | |i i hi | H = H, pi |i i hi |
i~ i i~ i
ih i
= H, (15.217)
~

which is the von Neumann equation. This holds for time-dependent Hamiltoni-
ans also. It describes the time evolution of the state operator in the Schrodinger
picture.

Now, if we start in the state |(t0 )i at time t0 , then we have the formal solution
of the Schrodinger equation

|(t)i = U (t, t0 ) |(t0 )i (15.218)

where

i~ U (t, t0 ) = H(t)U (t, t0 ) (15.219)
t

1199
We then get
" #
(t) ih i i X
= H, = H, pi |i i hi |
t ~ ~ i
" #
i X
+
= H, pi U (t, t0 ) |(t0 )i h(t0 )| U (t, t0 )
~ i
iX
= pi H U (t, t0 ) |(t0 )i h(t0 )| U + (t, t0 )
~ i
iX
+ pi U (t, t0 ) |(t0 )i h(t0 )| U + (t, t0 )H
~ i
X
= pi U (t, t0 ) |(t0 )i h(t0 )| U + (t, t0 )
i
t
X
+ pi U (t, t0 ) |(t0 )i h(t0 )| U + (t, t0 )
i
t
 
= U (t, t0 ) |(t0 )i h(t0 )| U + (t, t0 )
t
 
= U (t, t0 )(t0 )U + (t, t0 )
t
or
(t) = U (t, t0 )(t0 )U + (t, t0 ) (15.220)
2
Theorem: The quantity T r is time independent. Hence, a pure(mixed) state
remains a pure(mixed) state.
Proof :
T r2 (t) = T rU (t0 )U + U (t0 )U + = T rU (t0 )I(t0 )U +
= T rU (t0 )(t0 )U + = T r(t0 )(t0 )U + U = T r2 (t0 )
Therefore, if T r2 (t0 ) = 1, then T r2 (t) = 1 and the state remains pure and
similarly for the mixed state. The expectation value in the two pictures is given
by D E    
A = T r (t)A = T r (t0 )AH (t) (15.221)
or the time dependence comes from the state operator in the Schrodinger picture
and from the operator in the Heisenberg picture.

Spin = 1/2 Systems

How things actually work is particularly transparent in state spaces of low di-
mension. We consider spin = 1/2 (a 2dimensional state space). The states
(in Dirac and spinor notation)
   
1 0
|i + = and |i = (15.222)
0 1

1200
are eigenstates of z , i.e., z = . Now a rotation in spin space through
an angle about an axis n is represented by the unitary operator
i
U = e 2 n = cos I + in sin (15.223)
2 2
and
   
i
2i n
e 2 n m e = cos I + in sin (m ) cos I in sin
2 2 2 2

= cos2 (m ) + i [n , m ] sin cos + (n ) (m ) (n ) sin2
2 2 2 2
Now
(n )(m ) = n mI + i (n m) (15.224)
or
[(n ), (m )] = 2i (n m) (15.225)
Alternatively,
[(n ), (m )] = ni mj [i , j ] = 2iijk ni mj k = 2ikij ni mj k
= 2ik (n m)k = 2i (n m) = 2im (n ) (15.226)
Therefore,
i i
e 2 n m e 2 n = cos2 (m ) + i [2i (n m)] sin cos
2 2 2

+ (n ) (n m + i (n m)) sin2
2

= (m n) (n ) sin2 + (m (n )) sin + (m ) cos2
2 2

+ i (n ) ((n m) ) sin2
2
2
= (m n) (n ) sin + (m (n )) sin + (m ) cos2
2 2
2
+ i (n (n m) + i (n (n m)) sin
2
2
= (m n) (n ) sin + (m (n )) sin + (m ) cos2
2 2
2
(n m) (n ) sin
2
2
= (m n) (n ) sin + (m (n )) sin + (m ) cos2
2 2

(m (m n) (n )) sin2
2

= (m n) (n ) sin2 + (m (n )) sin ((m (n (n ))
2

+ (m n) (n )) cos2 + (m (n (n ))) sin2
2 2
= (m n) (n ) (m (n (n ))) cos + (m (n )) sin (15.227)

1201
or
i i
e 2 n e 2 n = n (n ) (n (n )) cos + (n ) sin (15.228)

For rotations about the xaxis these reduce to (using n = x)

 
i
x 0 1
Ux = e 2 = cos I + ix sin = cos I + i sin (15.229)
2 2 2 1 0 2

i i
e 2 x t e 2 x = (t x)(x ) t (x (x )) cos + t (x ) sin
= tx x t (x (y z z y)) cos + t (y z z y) sin
= tx x t (y y z z) cos + t (y z z y) sin
= tx x + (y ty + z tz ) cos + (y tz z ty ) sin

1 0 0 x
= (tx , ty , tz ) 0 cos sin y (15.230)
0 sin cos z

The eigenstates of t , i.e., the eigenstates of spin = 1/2 in the t direction are
given by the equation
i i
Ux (t )Ux+ = e 2 x (t )e 2 x = (15.231)

or
(t )(Ux+ ) = (Ux+ ) (15.232)
The two eigenfunctions are

cos 2 i sin 2
   
Ux+ + = , Ux+ = (15.233)
i sin 2 cos 2

Some Special Cases

For = /2, we get the eigenfunctions of y :
   
y 1 1 1 y 1 i 1
+ = = (+ + i ) , = = (i+ + )
2 i 2 2 1 2
(15.234)
and similarly for the eigenfunctions of x :
   
1 1 1 1 1 1
x+ = = (+ + ) , x = = (+ )
2 1 2 2 1 2
(15.235)
A rotation through 2 gives U = 1 and . In a rotation through
360 the spinor changes its sign. It takes a rotation through 4 to produce
U = +1 and +. This property is related to the spinor nature of and
the transformation relation for t is related to the vector nature of .

1202
Let us now discuss the spin part of the density matrix. Imagine we are dealing
with electron beams. An electron beam of spin has the density operator

= |i h| (15.236)

while an electron beam of spin has the density operator

= |i h| (15.237)

If one mixes the two beams in a 50:50 ratio, the density operator is
1
M = (|i h| + |i h|) (15.238)
2
This state has unknown relative phases.

It is not a pure state since

1
2M = M 6= M (15.239)
2
In contrast, the pure state (a superposition instead of a mixture) has known
relative phases, i.e.
1
|i = |i + ei |i


(15.240)
2
and has the density operator
1
|i h| + |i h| + ei |i h| + ei |i h|


= (15.241)
2
In the pure state case, we clearly have interference terms. We can also see the
differences in the matrix representations, which are given by

nm = hn| |mi (15.242)

In these two cases we get

ei
   
1 1 0 1 1
M = , = (15.243)
2 0 1 2 ei 1
The mixed state density operator is diagonal.

The difference between the two density operators shows up in expectation values
and hence has measurable consequences. For the pure state,

1 ei
   
D E   1 h1| A |1i h1| A |2i
A = T r A = T r
2 ei 1 h2| A |1i h2| A |2i
i
h2| A |1i h1| A |2i + ei h2| A |2i
 
1 h1| A |1i + e
= Tr
2 ei h1| A |1i + h2| A |1i ei h1| A |2i + h2| A |2i
1   
= h1| A |1i + h2| A |2i + 2Re ei h1| A |2i
2

1203
while for the mixed state we have
   
D E   1 1 0 h1| A |1i h1| A |2i
A = T r M A = T r
M 2 0 1 h2| A |1i h2| A |2i
 
1 h1| A |1i h1| A |2i
= Tr
2 h2| A |1i h2| A |2i
1 
= h1| A |1i + h2| A |2i
2
Again, we see no interference (quantum) effects from the mixed state. It is clear
from these expressions that the mixed-state expectation value arises from the
pure-state expectation value by averaging over the relative phase value, i.e.,
Z2
1
M = d (15.244)
2
0

Polarization of Spin = 1/2 Particles

The most general density operator in spin space is given by
1  ~ 
= I + b ~ (15.245)
2
1 , 2 , 3 is a basis for all 2 2 matrices. The
since the set of four matrices I,
property T r = 1 works because T rI = 2 and T ri = 0. If we choose ~b to point
in the zdirection, then we have
1   1 1+b 0

= I + bz = (15.246)
2 2 0 1b

This implies that (1 b)/2 are probabilities and thus we must have |~b| 1. If
we have a pure state, then we must also have 2 = , which implies that
1   1 
2 = I + 2~b ~ + (~b ~ )(~b ~ ) = I + 2~b ~ + b2 I
4 4
1 1 + b2 ~

1  ~  2
= I + b ~ = = I + b ~ =
2 2 2

or |~b| = 1 in the pure state case. Now a straightforward calculation gives

  
1 ~ 
hi i = T ri = T r I + b ~ i
2
     
1 1~ 1
= T r i + T r (b ~ )i = T r bi I = bi
2 2 2
This implies that the degree of polarization of the spin state is characterized by
b = |~b|. The case b = 0 represents a completely polarized beam.

1204
(4) The Measurement Process
We have a beams of atoms or electrons moving through the inhomogeneous field
of a magnet as shown in Figure 15.8 below.

Figure 15.8: Stern-Gerlach Setup

The force is given by

Bz
mz (15.247)
z
which splits the beams into many beams each with a different value of mz .
We assume that the total angular momentum is spin only and S = 1/2. The
motion in the ydirection is force-free and can be separated out. We then have
the Hamiltonian
p2z p2
H = + B(z)B z z + (B + B 0 z + ....) B z (15.248)
2m 2m
describing the motion in the zdirection. We have Taylor expanded the mag-
netic field. The time-development of the state vector is given by the equation
   
u+ u+
i~ = i~ = H = H (15.249)
t t u u
 2 
u pz 0
i~ = BB B B z u (15.250)
t 2m

If we let the spatial wavefunction be a wave packet f (z) concentrated about the
zaxis before entering the magnetic field at time t = 0, then at time t, it is
then approximately
u (z, t) = f (z Ct2 )eit (15.251)
where the constant
B 0 B
C= (15.252)
2m
is the acceleration during the period the particle is in the magnetic field region
and
BB
= (15.253)
~

1205
This says that the particles with spin and are deflected downwards and
upwards, respectively.

The Stern-Gerlach Experiment as a Model for an Idealized Measure-

ment
In this experiment the object being measured is the spin of the particle and the
measuring apparatus is the position of the particle after traversing the magnetic
field (location on some screen). After passing through the magnetic field we
know that
for , z < 0 and for , z > 0 (15.254)
This apparatus, whose readout (position of some pointer) is the zcoordinate
of the particle, is thus appropriate for distinguishing from and, by the actual
size of the deflection, for determining the size of the magnetic moment and hence
the spin.

The requirement that the zcoordinate of the particle serve as the pointer of
a measuring instrument implies that the deflections must be macroscopically
distinguishable. Formally, this implies that the overlap of the two wave packets
f (z + Ct2 ) and f (z Ct2 ) must be negligible. We also assume that we have
calibrated the device.

What happens to general states?

We consider the initial state(a superposition)

1
(z, 0) = (+ + ) f (z) (15.255)
2
After traversing the field this becomes
1
(z, t) = + f (z + Ct2 )eit + f (z Ct2 )eit


(15.256)
2

The polarization and the pointer (z-coordinate) are coupled or entangled.

There is a unique correlation between the state of the spin and the state of
the pointer. Neither the spin nor the pointer are in an eigenstate (neither have
definite values).

In the basis of the states |zi |i, the density operator is

zz0 = h| hz| (|(t)i h(t)|) |z 0 i |0 i

f (z + Ct2 )f (z 0 + Ct2 ) f (z + Ct2 )f (z 0 Ct2 ) e2it
 
1
=
2 f (z Ct2 )f (z 0 + Ct2 ) e2it f (z Ct2 )f (z 0 Ct2 )
(15.257)

1206
Measurement of spin observables
After the spin 1/2 particle has passed through the Stern-Gerlach apparatus,
suppose that its spin is measured. The measurement can take place in two
ways:
(i) ignoring the pointer position z
(ii) for a particular pointer position z
In case (i), if one ignores the pointer position, then is equivalent to
Z  
1 1 0
= dz hz| |zi = (15.258)
2 0 1
This result holds because there is no overlap of the pointer wave functions
f (z + Ct2 ) and f (z Ct2 ). For an observable F (~ ), depending only on spin
operators, we then have
T rz, (F (~ )) = T r (F (~ )) (15.259)
where the density operator corresponds to a mixed ensemble. Thus, the pure
ensemble is replaced by the mixed ensemble .

In case (ii), filtering out the pointer position: in this case, we consider only the
particles with pointer position zpositive, i.e., we construct
Z  
1 0 0
T rz>0, (F (~ )) = dszz = (15.260)
2 0 1
Because of normalization, the density is operator is actually
 
0 0
= (15.261)
0 1
The particles deflected up have the spin wave function |i.

For a measurement with the result zpositive(spin negative), the state goes
over to |i. This is consistent with the quantum mechanical postulate that
measurements are repeatable.

The fact that the particles which have been filtered off at a particular pointer
position are in that eigenstate corresponding to the eigenvalue measured is called
the collapse or reduction of the state vector.

Going over from to with respect to all observables related to spin can also
be regarded as a reduction of the state vector.

The density matrix describes an ensemble composed of 50% spin-up and 50%
spin-down states. If N particles are subjected to this Stern-Gerlach experiment,
then as far as their spins are concerned, they are completely equivalent to N/2
particles in the state |i and N/2 in the state |i.

1207
15.2.8 A General Experiment and Coupling to the Envi-
ronment
We now consider a general experiment. Let O be the object and A the apparatus
including the readout. At the time t = 0, let the state of the whole system O +A
be X
|(0)i = cn |O, ni |Ai (15.262)
n
where the |O, ni are object states and |Ai is the (metastable) initial state of
the apparatus. At a later time t, after the interaction of the object with the
measuring apparatus, the state becomes
X
|(t)i = cn |O, ni |A(n)i (15.263)
n

where the final states of the apparatus |A(n)i . n = 1, 2, 3, ..... must be macro-
scopically distinguishable.

As before the density operator for this pure states is

(t) = |(t)i h(t)| (15.264)
If we read off the result of the measurement, the density operator for observables
relevant to the object O is
X 2
= T rA(n) (t) = |cn | |O, ni hO, n| (15.265)
n

We needed to make use of the fact that the final states of the apparatus are
macroscopically distinguishable, and thus do not overlap, i.e., that hA(n) | A(m)i =
mn .

Thus, if we do not read off the result of the measurement, a mixture occurs with
respect to O. If, on the other hand, we read off a particular value, e.g., A(m),
the density operator is then
|O, mi hO, m| (15.266)
2
The probability of measuring the value A(m) is clearly |cm | . The fact that in
a measurement with the result A(m) the density operator changes from
|(t)i h(t)| |O, mi hO, m| (15.267)
is known as the reduction of the wave function or state vector.

We now take into account the fact that the object and the apparatus are never
completely isolated from the environment, and take Z to be an additional vari-
able representing all further macroscopic consequences which couple to the state
A of the apparatus. The initial state is then
X
|(0)i = cn |O, ni |Ai |Zi (15.268)
n

1208
and after passage through the apparatus the state evolves into
X
|(t)i = cn |O, ni |A(n)i |Z(n)i (15.269)
n

If we do not read off Z, which always happens in practice, since we cannot keep
track of all the macroscopic consequences, the density operator of the (object
+ apparatus) is the mixture
X 2
= |cn | |A(n)i |O, ni hO, n| hA(n)| (15.270)
n

The subsystem (object + apparatus) is thus a mixed state. N such subsystems

behave like
2
N |c1 | subsystems in the state |O, 1i |A(1)i
2
N |c2 | subsystems in the state |O, 2i |A(2)i
....................
2
N |cn | subsystems in the state |O, ni |A(n)i
....................

This is a completely different density operator than that of the pure state, i.e.,
|(t)i h(t)|.

(i) If we do not read off A(n) either, then is equivalent to

X 2
= |cn | |O, ni hO, n| (15.271)
n

(ii) If we read off A(n), then the probability of obtaining the particular reading
A(m) is  
2
T rO,A |A(m)i hA(m)| = |cm | (15.272)

and in this case, the density operator is then

|A(m)i |O, mi hO, m| hA(m)| (15.273)

From that point onwards, it does not matter if we disregard A. Taking the trace
over A(n) yields, for the observable O, the density operator

|O, mi hO, m| (15.274)

The key problem in the theory of measurement is this reduction of the wave
function and in particular the question of when it takes place. This problem is
illustrated quite drastically by phenomena of Schrodingers Cat discussed in
Chapter 14.

1209
Influence of an Observation on Time Evolution
In order to further analyze the measurement process and its impact, we return
to the Stern-Gerlach experiment and consider this additional setup.

After the atomic beams have traversed the Stern-Gerlach apparatus, we recom-
bine them by means of a complicated field configuration in such a way that all
of the deformation and spreading of the wave function is carefully undone, i.e.,
the state
f (z) c1 ei+ + + c2 ei


(15.275)
is formed. This is the initial wave function once again, essentially. The phase
factors are inserted to characterize any path length differences.

Now, in the region where the beams + and are macroscopically separated,
we can set up a real measuring device whose pointer Z which reacts to z by the
interaction U (z Z), so that positive (negative) z leads to positive(negative)
Z. We then have for the initial state

|a i = f (z) (c1 + + c2 ) |Z = 0i (15.276)

for the intermediate state

|c i = c1 + f (z + Ct2 ) |Z = 1i + c2 f (z Ct2 ) |Z = +1i (15.277)

and for the state after traversing the entire setup

|e i = f (z)(c1 + ei+ |Z = 1i + c2 ei+ |Z = +1i) (15.278)

The pointer positions are described by |Z = 0i and |Z = 1i.

We now compare these two situations:

I. We turn on the coupling to the measuring device Z and obtain the final
state

|e i = f (z)(c1 + ei+ |Z = 1i + c2 ei+ |Z = +1i) (15.279)

II. We turn off the coupling to the measuring device Z and obtain the final
state
|e i = f (z) c1 ei+ + + c2 ei |Z = 0i


(15.280)

The resulting density operators are quite different. Although the state

|e i = f (z)(c1 + ei+ |Z = 1i + c2 ei+ |Z = +1i) (15.281)

is a pure state, it is equivalent to a mixture, as far as the statements relating to

the particle are concerned. This is due to the fact that the macroscopic states
|Z = 1i do not overlap (they are orthogonal).

1210
In situation II, both the total state and the state of the atom are pure states, In
situation I, the final state of the total system (atom + pointer) (characterized
by spin z and Z), is mixed, unless c1 or c2 vanish.

Evidently, the physical situation differs according to whether the interaction

U (z Z) between the system and the pointer is turned on or not.

Even if we do not read the result, we still influence the atomic system.

In order to further illustrate the back-reaction of the experiment on the object,

we now consider the following thought experiment due to Heisenberg. We con-
nect two Stern-Gerlach devices in series as shown in Figure 15.9 below

Figure 15.9: Back Reaction Experiment

Let the initial state of the atoms in the atomic beam be |ai, let {|ei} be a basis
for the final states, and let {|ci} be a basis for the intermediate state. We now
determine the probability of transition to the final state |ei, by representing this
in terms of the transition amplitudes of |ai to |ci to |ei.

For an isolated system the transition probability is

 2
X 
I (1) (2) 
Pae = Uac Uce  (15.282)

 
c

because X
(1) (2)
Uac Uce = Uae (15.283)
c

holds for an isolated system.

On the other hand, one could also say that the transition probability is the
product of the probabilities
 (1) 2  (2) 2
   
Uac  Uce  (15.284)

1211
summed over all intermediate states c :
X  2  2
II (1)   (2) 
Pae = Uac  Uce  (15.285)
c

These two probability expressions correspond to different experiments.

II
For Pae there is a measurement in the intermediate region, and this introduces
unknown phase factors eic which must be averaged over.

Experiment 1: Between SG1 and SG2 , the atoms remain unperturbed. There
I
is no coupling to the external world, and the transition probability is Pae .

Experiment 2: Between SG1 and SG2 , there is an influence on the atoms,

making possible a determination of the stationary state. However, the result
of the measurement is not recorded, and a mixture is formed. The transition
I
probability is Pae I.

Experiment 3: Between SG1 and SG2 , an influence on the atoms occurs

making possible a determination of the stationary state. Let us assume that c
 (2) 2

is found. The probability for state |ei behind SG2 is then given by Uce  .

15.3 Kochen-Specker Theorem

Some, perhaps many, users of quantum theory may find a pragmatic approach
(or interpretation), where a sharp distinction is made between the system and
the observer or observing equipment(or even a full instrumentalist interpre-
tation, where the primary emphasis is placed on the act of measurement with
quantum theory being viewed as a scheme for predicting the probabilistic spread
of the results obtained) satisfactory, but it is hard to believe that either gives
the ultimate view of physical reality. Indeed, with their reluctance to grant
objective status to the properties of individual systems, it is arguable whether
either gives a picture of reality at all.

However, attempts to move towards a more realist philosophy by ascribing latent

properties to individual systems most often tend to be too vague to do much
more than stimulate a search for new conceptual categories. In particular, there
is the unresolved issue of how the transition from potentiality to actuality is
made in realistic interpretations.

It is not surprising, therefore, that many physicists have sought an interpreta-

tion of quantum theory in which the probabilistic results would have the same
status as those in classical statistical physics where almost all of these difficult
philosophical problems do not arise.

Most discussions of this type involve the epistemic interpretation of probabil-

1212
ity, in which probabilistic assignments refer to our knowledge of an objectively
existing state of affairs, for example, an expression like A (the dispersion)
would be a measure of our ignorance of A which, it is supposed, does have an
actual value. Such a position is consistent with a completely realist view of the
world, in which individual objects and properties have an unequivocal existence
that is independent of any act of observation or measurement.

Interpretations of this type are naturally coupled with the assumptions that

1. quantum states refer to individual systems directly, not just to the out-
come of repeated measurements, or to any associated ensemble of systems

2. a (perfect) measurement of a physical quantity reveals the value that it

possessed immediately beforehand

Most problems discussed by quantum theorists would not arise in a scheme

where properties are possessed in this way. For example, state-vector reduc-
tion would merely be an acquisition of further knowledge about the system
by making a measurement - a clear analogue of the classical idea of conditional
probability. True, this raises the question of whether the measurement itself can
be described in quantum-mechanical terms, but the significance of this problem,
in this case, is no greater than it is in classical statistical physics.

Since an epistemic interpretation of quantum theory promises to remove most of

the conceptual problems, one might wonder why the theory is not automatically
presented in this way from the outset.

The reason is that the concept of an individual system possessing a value for all
its physical quantities is difficult to reconcile with the actual formalism of quan-
tum theory. As we have seen, one obstacle to this view are the Bell inequalities,
as discussed earlier in Chapter 8 and will discuss extensively in Chapter 16.

Another obstacle is the Kochen-Specker theorem, which we discuss now. We

are concerned here with the existence of a value function V|i (A) that is to be
interpreted as the value of the physical quantity A when the quantum state (of
an individual system) is |i. There is no difficulty with this concept in clas-
sical physics, where, to each physical quantity A there corresponds a function
fA : S < (where S is the classical state space) such that the value of A in
the state s S is just the value of fA at s:

Vs (A) = fA (s) (15.286)

The situation in quantum mechanics is very different since we have no prima

facie idea how to specify the value of V|i (A) for any particular pair (A, |i). If
|i is an eigenvector |ai of the self-adjoint operator A that represents A, then
we might be inclined to say that V|ai (A) = a, where a is the corresponding
eigenvalue, but it is unclear how to go beyond this special case.

1213
As a consequence, nothing useful can be said about the existence of quantum
value-functions without postulating further properties for them.

We note that there are plenty of functions that satisfy the eigenvector require-
ment. For example, the function
V|i (A) = h| A |i (15.287)
does so (assuming that |i is normalized). However, as we know, this gives
the expected value of A, so it is not a good choice for a value function in any
situation in which the dispersion A is non-zero. Therefore, to get a sensible
notion of a value function it is clearly necessary to go beyond the eigenvector
requirement.

The most natural condition to impose is that, for any function F : < <,
V|i (F (A)) = F (V|i (A)) [condition A] (15.288)
Thus, in any quantum state |i, the value of a function of a physical quantity
is equal to the function evaluated on the value of the quantity; for example, the
value of L2x is the square of the value of the angular momentum Lx .

Earlier we discussed the meaning of the expression F (A).

If we are quantizing a given classical system with state space S then F (A) can
stand for F (fA )(s), where fA : S <, and where the map F (fA ) : S <
is defined by F (fA )(s) := F (fA (s)) for all s S. In this case, it is viable to
regard a value function V|i as being defined on real-valued functions f on the
classical state space S.

However, if we do not wish to assume any such classical background, then

something different is needed.

One possibility is to define F (A) as that physical quantity which is associated

with the operator F (A). Of course, this is only meaningful if the quantization
map from physical quantity to self-adjoint operator is
(1) onto so that F (A) is the quantum representative of some physical quan-
tity.
(2) one-to-one so that the quantity thus defined is unique.
Under these circumstances, nothing is lost by thinking of V|i as a function of
self-adjoint operators, rather than physical quantities.

Another way of looking at this whole argument is to start with a value-function

0
V|i that is a function of self-adjoint operators, rather than physical quantities.
The analogue of the condition A above is then the relation
0 0
V|i (F (A)) = F (V|i (A)) (15.289)

1214
which deals only with mathematical entities. A value-function V|i of physical
quantities can then be defined by
0
V|i (A) := V|i (A) (15.290)

If the quantization map is one-to-one and onto, then F (A) is well-defined as the
inverse image of F (A), and the V|i function constructed in this way satisfies
condition A.

Note that we are driven towards such a strategy because, on the one hand(and
unlike in the pragmatic approach), an operational definition in terms of mea-
surements is not appropriate in the context of the more realist interpretation
being sought. But, on the other hand, we must avoid defining F (A) to be the
physical quantity that satisfies condition A for all |i H, since, then condi-
tion A would be a tautology. The potential trap is that this is precisely how
a function of a physical quantity is defined in classical physics. This discus-
sion impinges also on the basic function-preserving requirement F[ (A) = F (A),
which does become a tautology if F (A) is defined to be the physical quantity
that is associated with the operator F (A).

The intuitively plausible requirement of condition A has the following important

implications (all can be proved).
h i
1. If A, B = 0, the value function is additive in the sense that, for all
|i H
V|i (A + B) = V|i (A) + V|i (B) (15.291)
where A + B denotes the physical quantity associated with the self-adjoint
operator A + B.
h i
2. If A, B = 0, the value function is multiplicative in the sense that, for all
|i H
V|i (AB) = V|i (A)V|i (B) (15.292)
where AB denotes the physical quantity associated with the self-adjoint
operator AB.

3. Let I denote the physical quantity that corresponds to the identity oper-
Then using the result in (2) above with A := I, we see that, for
ator I.
each state |i
V|i (B) = V|i (I)V|i (B) (15.293)
for all physical quantities B. Thus, for each |i H, we get

V|i (I) = 1 (15.294)

provided that there is at least one quantity B for which V|i (B) 6= 0.

1215
The multiplicative property of (2) above has an important implication for any
physical quantity P whose representative is a projection operator P . In this
case, the property P 2 = P implies at once that

2
V|i (P ) = V|i (P 2 ) = V|i (P ) (15.295)
and hence
V|i (P ) = 0 or 1 (15.296)
Thus, thinking of P as a proposition, we see that a quantum value function V|i
gives it a false or true assignment in the state |i.
n o
Now consider a collection P1 , P2 , ....., Pn of projection operators that forms
a resolution of the identity, that is,
Pi Pj = 0 if i 6= j , P1 + P2 + ..... + Pn = I (15.297)
For example, these could be the spectral projectors of a self-adjoint operator
A with discrete eigenvalues, so that Pi = PA=ai . In a realist interpretation,
the associated proposition Pi asserts thatnA has a particular o value ai , which
implies that the collection of propositions P1 , P2 , ....., Pn should be mutually
exclusive - only one can be true at any given time - and exhaustive - at least one
of them must be true at any given time. This is in accord with the commonsense
view of the nature of truth and falsity, that is, in a set of mutually exclusive
and exhaustive propositions one and only one can be true and the rest are false.
This expectation is borne out by the formalism. Indeed, the results above imply
that
V|i (A + B) = V|i (A) + V|i (B) (15.298)
n
X n
X
V|i ( Pi ) = V|i (Pi ) = 1 (15.299)
i=1 i=1

for all states |i.

However, since each V|i (Pi ) has the value 0 or 1, this sum can equal 1 only if
one of the propositions is given the value 1 and the rest are given the value 0.

A special case is when Pi := |ei i hei | where {|e1 i , |e2 i , ......} is an orthonormal
basis for the Hilbert space H. Then, according to the result just obtained
above, for any state |i a value function V|i must assign the number 1 to one
of the projectors/one-dimensional subspaces (more precisely, to the associated
proposition) and 0 to the rest. This is not a trivial requirement, since any
given vector will belong to many different orthonormal basis sets, and the value
given to the corresponding one-dimensional subspace must be independent of
the choice of such a set.

In fact, this requirement is so difficult that it cannot be satisfied.

1216
This will ultimately defeat any idea of a realist interpretation.

As stated above, this argument must apply to all possible sets of orthogonal
projectors which provide a resolution of the identity in the Hilbert space. Since
the projectors are in one-to-one correspondence with the rays in the Hilbert
space, the above constraint means that, for every complete orthogonal basis of
unit vectors in the Hilbert space, we must be able to associate the number 1
with one vector and the number 0 with all the other vectors in the basis in a
consistent manner.

15.3.1 Theorem (due to Kochen and Specker)

There is no such function V|i if the
Hilbert space H is such that dim(H) > 2.

It is convenient(for carrying out a proof) to translate this problem into a coloring

problem on the surface of a unit hypersphere in Hilbert space.

Translation into Coloring Problem

Can we color the hypersphere with two colors, red and blue, in such a way that
the following conditions are satisfied?

1. Every point (unit vector) is colored red or blue.

2. For every complete orthogonal set of unit vectors only one is colored red.

3. Unit vectors belonging to the same ray have the same color.

The two-color statement of the Kochen-Specker theorem is:

If the dimension of the Hilbert space is greater than

two, the coloring of the unit hypersphere in the way
described above is not possible.

Let us denote the two-color theorem (not possible case) in a real(complex)

real(complex)
Hilbert space of finite dimension N by TN and in an infinite-dimensional
real(complex)
separable Hilbert space by T .

We notice at once that

TNreal TNreal
+1 (15.300)
This follows by supposing the coloring is possible in the (N + 1)dimensional
case (that is (TNreal
+1 ) is true - the possible case) and considering the N dimensional
subspace orthogonal to any direction colored blue. This will now itself induce a
coloring of the unit hypersphere in the subspace in accordance with the speci-
fication of the theorem. So (TNreal real
+1 ) (TN ) or contrapositively, we have

1217
proved TNreal TNreal
+1 .

Repeated applications of this result show that if we can prove TNreal for any
real
given N , then the theorem is true for any greater value of N and, indeed, T
will follow by a similar argument, in which, assuming the coloring is possible
in the infinite-dimensional case, we show that it would be possible in a finite-
dimensional subspace that includes a direction colored red.

Further, for any dimension, finite or infinite,

TNreal TNcomplex (15.301)

since, if we could color the unit hypersphere in the Hilbert space defined over the
complex field, we could show the coloring to be possible for a real Hilbert space
of the same dimension by considering some particular complete orthogonal set
of vectors in the complex case, and generating a structure isomorphic to a real
Hilbert space by considering all the orthogonal sets obtained by real orthogonal
transformations from the initial set of orthonormal vectors.

With these results in mind, all we need to do is examine TNreal for low values
of N . We notice that T2real is false(it is possible to satisfy conditions (1)-(3)).
This corresponds to coloring the unit circle in a real Euclidean plane. The
construction for doing such a coloring is shown in Figure 15.10 below.

Figure 15.10: 2-Dimensional Construction

Alternate open-closed quadrants are colored blue and red as indicated. It is

then easily checked that, for any orthogonal pair of directions, one is indeed
colored red and the other blue, with opposite directions having the same color.

We now show that T3real holds. First, we give a plausibility argument due to
Belinfante.

Consider the unit sphere in Euclidean 3-space. Suppose that the coloring had
been carried out. Then we would expect 1/3 of the surface to be colored red
and 2/3 to be colored blue, since for every orthogonal triad of directions one
is colored red and the remaining two blue. But every time we color a point P ,
say, red, then we must color the whole equator, with P as the pole, blue, since
any direction orthogonal to P gets colored blue. This is shown in Figure 15.11
below.

1218
 Figure 15.11: 3-Dimensional Construction

P QR is one orthogonal triad of directions, so if P is red, then Q and R must be

blue. Rotating this orthogonal triad about OP sweeps out the complete blue
equator as shown. So it looks like we can never end up with enough red points,
if every red point is associated with an infinite number of blue points on the
corresponding equator.

This argument , of course, is quite unrigorous, since there is a lot of double

counting if one remembers that each point on a blue equator lies on infinitely
many other equators!

Kochen and Specker gave a rigorous proof which goes as follows.

First, we prove a lemma:

There is a finite angular distance between

any two points with opposite color.

More specifically, we shall show that if 1 and 2 are any two points on the surface
of a sphere with center at O, and if we denote by 12 the angle between the unit

vectors O1 and O2, then if
 
1
0 12 sin1 (15.302)
3
the points 1 and 2 cannot be assigned the opposite color.

Proof : We first introduce the representation of the points and orthogonality

relations on the sphere by means of a so-called Kochen-Specker diagram, in
which points on the sphere are represented by vertices in the diagram and if two
points are in orthogonal directions, the corresponding vertices of the diagram
are joined by a straight line. We show that the Kochen-Specker diagram shown
in Figure 15.12 below is constructible if
 
1
0 12 sin1
3

1219
 Figure 15.12: Kochen-Specker Diagram

This Kochen-Specker diagram has ten points on the unit sphere in 3dimensional
Euclidean space with  
1
0 12 sin1
3
as we shall prove.

Suppose first that 12 is any acute angle. Since 3 is orthogonal to 1 and 2, and

4 is orthogonal to 3, O4 must be in the plane defined by O1 and O2. Since

O4 is orthogonal to O2, we may choose 4 so that 14 is also acute, and clearly,
14 = /2 12 as shown in Figure 15.13 below.

Figure 15.13: Angle Diagram

Now write O5 = i, O6 = k and take a unit vector j orthogonal to i and k so as

to complete a set of orthonormal vectors {i.j, k}. Then O7, being orthogonal
to i, may be written as

O7 = (j + xk)(1 + x2 )1/2 (15.303)
Similarly,

O8 = (i + y j)(1 + y 2 )1/2 (15.304)

1220
for some numbers x and y.

The orthogonality relations in the diagram force

O9 = (xj k)(1 + x2 )1/2 (15.305)

and

O10 = (y i j)(1 + y 2 )1/2 (15.306)

But O1 is orthogonal to O7 and O8, so we must have

O1 = (xy i xj + k)(1 + x2 + x2 y 2 )1/2 (15.307)

Also, O4 is orthogonal to O9 and O10, so we must have

O4 = (i + y j + xy k)(1 + y 2 + x2 y 2 )1/2 (15.308)

But then, taking the inner product of O1 and O4, we have at once
xy
cos 14 = 1/2
((1 + + x2 x2 y 2 )(1
+ y 2 + x2 y 2 ))
= cos (/2 12 ) = sin 12 (15.309)

This expression is easily seen to achieve a maximum of 1/3 for x = y = 1. So

the Kochen-Specker diagram is constructible if
 
1
0 12 sin1
3

Now consider the color map V : S {red, blue}, whose domain is the surface
S of the unit sphere, and suppose that V (1) = red and V (2) = blue, then from
the Kochen-Specker diagram above, V (1) = red V (7) = V (8) = V (3) = blue.
But V (2) = blue implies that V (4) = red. Hence, V (9) = V (10) = blue. But
V (7) = V (9) = blue, which implies that V (5) = red and V (8) = V (10) = blue
implies that V (6) = red. But V (5) = V (6) = red is not possible since 5 and 6
are orthogonal directions.

Hence, we conclude there is a minimum angular distance between points of

opposite color which is certainly greater than sin1 31 , otherwise the above

contradiction would be derivable.

We now use the lemma to prove the impossibility of the coloring considered in
T3real

Suppose that we take 12 = 18 < sin1 (1/3), then we know that V (1) = red
implies that V (2) = red. But now introduce four additional points, labelled

1221
3,4,5,and 6, lying at 18 intervals along the equator through the points 1 and
2. Then, repeating the above argument

V (1) = red V (2) = red V (3) = red V (4) = red V (5) = red V (6) = red

But 16 = 5 18 = 90 .

So, if any point on the sphere is colored red, we have demonstrated two orthog-
onal red points, which contradicts the specification of the coloring.

But, considering any three orthogonal points on the sphere, one of them must
be colored red.

Hence, the coloring is not possible, that is, we have demonstrated T3real .

This argument can be put in the form of a Kochen-Specker diagram with a finite
set of vertices which cannot be colored in the specified way.

All we have to do is to choose three arbitrary points p0 , q0 , r0 and insert the ap-
propriate fivefold repetition of the Kochen-Specker diagram above, with 12 =
18 , between each pair of vertices of the triangle p0 , q0 , r0 . The resulting beau-
tiful example of an inconsistent Kochen-Specker diagram, that is, one which
cannot be colored, is shown in Figure 15.14 below.

Figure 15.14: An Inconsistent Kochen-Specker Diagram

1222
Notice that, since a is orthogonal to r0 and q0 , as also is p0 , then we can choose
a to coincide with p0 . Similarly, we can identify b and q0 and c and r0 . So the
total number of vertices in the diagram is made up of eight vertices in each of the
fifteen hexagons with three pairs of vertices identified, leaving (8 15) 3 = 117
distinct vertices.

In other words, what is demonstrated in Figure 15.14 is a collection of 117 points

on the unit sphere with the orthogonality relationships indicated, which can be
constructed, but which cannot be colored as specified. So we have proved T3real .

We have shown that if a value function exists for a Hilbert space of a particular
dimension N , then it necessarily does so for any space of dimension less than
N . Similarly, if a value function exists for a complex Hilbert space, then one
exists on any real Hilbert space of the same dimension.

Then, to prove the theorem, it was clear that it suffices to show that no such
function can exist in real, 3-dimensional euclidean space.

Reducing the problem in this manner had the advantage that it could be stud-
ied in a manifestly geometrical way, in which case it looks like a certain type of
map-coloring problem as shown above.

In their original proof, Kochen and Specker found a counter example to the ex-
istence of a value function, in the form of a set of 117 vectors in the vector space
<3 that could not be consistently assigned the numbers 0 or 1 in the desired
way as we have shown above.

This remarkable result shows that a value function V|i cannot exist without
violating one of the assumptions implicit in the statement of the theorem. The
obvious candidates are:
1. The requirement that V|i (F (A)) = F (V|i (A))

2. The assumption that the quantization map A 7 A is one-to-one.

The natural situation in which the first requirement might, perhaps, be expected
to fail is when we deal with a self-adjoint operator A that can be written as
functions A = f1 (A1 ) and A = f2 (A2 ) of a pair of self-adjoint operators A1 and
A2 with [A1 , A2 ] 6= 0.

The value condition implies, in this case, that

V|i (F (A)) = f1 (V|i (A1 )) = f2 (V|i (A2 )) (15.310)

and, if A1 and A2 do not commute, the wisdom of this assumption could be

questioned. For example, if we suppose that the value possessed by a quantity
can be found by making an appropriate measurement (which is an assumption,

1223
albeit not an unreasonable one), then the last equation implies that measuring
A1 and applying f1 to the result, will yield the same number as measuring
A2 and applying f2 to the result. However, there seems no good reason for
supposing this would be so if [A1 , A2 ] 6= 0.

On the other hand, dropping the last equation has a very peculiar effect. For
example, let A1 and A2 be a pair of non-commuting, self-adjoint operators whose
spectral resolutions contain a common projector P .

Then P can be written as one function of A1 , and as another function of A2 ,

that is, in general, if a self-adjoint operator A has a discrete spectrum with a
spectral resolution
M
X
A = ai Pi (15.311)
i=1

then the projector Pi can be written as a function of A

Pi = ai (A) (15.312)

where r (t) := 1 if t = r and 0 otherwise. Hence, the failure of the value

condition P (or equivalently, the truth or falsity of the associated proposition)
depends on the context in which P is taken: if viewed as belonging to the
spectral representation of A1 , it will be given one value; if viewed as belonging
to that of A2 , it will be given another.

For example, if P is thought of as belonging to the spectral representation of A1 ,

the remaining projectors in the spectral set are a natural choice for commuting
partners and, analogously, if P is viewed as part of the spectral representation
of A2 . More generally, this means that the value of any physical quantity A
whose representing operator A has degenerate eigenvalues may depend on the
context in which it is considered. In particular, it may depend on the choice
of other physical quantities whose associated operators commute with A, and
which therefore help to label the degeneracy subspaces of A.

Yet another way of expressing these ideas is to say that the operator A does not
represent a unique physical quantity. It has a different meaning depending on
what other compatible quantities are considered at the same time. This links
up with the second possible way of saving value functions, that is, the idea that
the quantization map A 7 A may be many-to-one.

The Kochen-Specker theorem is not directly relevant to the pragmatic approach

to quantum theory in so far as no attempt is made in that approach to think
of an individual system as possessing values for its observables. However, if
this approach is augmented with the assumptions (1) physical quantities do
have values in any individual systems and (2) these values can be revealed by
suitable measurements, then the theorem implies that the results obtained will

1224
depend on the context in which an observable is studied. In particular, if A, B
and C are three operators with
h i h i h i
A, B = 0 = A, C , B, C 6= 0 (15.313)

then the result of an individual measurement of A will depend on whether we

choose to measure B or C (or neither) at the same time.

A very important example of this situation arises naturally in the EPR context
of a pair of entangled systems that have become spatially separated, and on
which measurements are then made in such a way that the measurement events
are space-like separated. Remember that two events are space-like separated if
no signal can be sent from one to another without exceeding the speed of light
and for any such pair of events there is always some inertial frame of reference
with respect to which the two events are simultaneous.

If A is associated with the first system and B and C with the second, then we
expect to be able to measure A simultaneously with either B or C and hence
h i h i
A, B = 0 = A, C (15.314)

On the other hand, it is easy to arrange

h thati the operators B and C associated
with the second system are such that B, C 6= 0.

Note that the probabilistic predictions of the formalism concerning the results
of measuring A do not depend on what else is measured at the same time, that
is, the probability of getting result am when the state is |i is always just

h| Pm |i (15.315)

In addition, statements about the hypothetical values of quantities for individual

systems are not amenable to experimental tests. In consequence, the deductions
above concerning contextuality will not trouble a pragmatist. Indeed, she may
even support them, by citing Bohrs ideas about a property of a quantum system
having a meaning only within the context of a specific measurement situation.

However, within a realist interpretation, a contextual assignment of values is

in sharp variance with what is normally meant by saying that a property is
possessed.

In normal discourse, implicit in the statement the quantity A has a value a is

an understanding that this is independent of what else might be asserted at the
same time. But, according to the Kochen-Specker theorem, in the quantum case
we are obliged to talk instead of pairs (A, B) or (A, C) having certain values
(a, b) and (a0 , c) as pairs, and the implication of contextuality is that a may not
equal a0 .

1225
Note that the analogue of a value-function in a hidden variable theory would
involve the assumption that the value of a physical quantity A in a quantum
state |i depends on these additional variables. However, this does not change
the force of these arguments above.

In particular, it is only in the contextual sense that properties in a hidden

variable theory can be said to be possessed.

Let make a digression now to consider quantum logic.

15.4 The Logic of Quantum Propositions

15.4.1 The Meaning of True
Several times we have mentioned the idea that a projection operator P can be
regarded as the quantum representative of a proposition P , with the eigenvalues
1 and 0 being associated in some way with P being true or false respectively.
Equivalently, the proposition is represented by the subspace
n o
HP := |i H|P |i = |i (15.316)

of H. Furthermore, the discussion above showed that, if it existed, a value

function V|i satisfying

V|i (F (A)) = F (V|i (A)) (15.317)

would associate a number 0 or 1 with each projector. The obvious interpretation

is that V|i (P ) = 1 and V|i (P ) = 0 correspond, respectively, to the proposition
being true or false when the quantum state is |i.

Binary-valued functions of this type (called valuations) play an important role

in classical propositional logic, which suggests that it may be profitable to ex-
plore quantum theory further from this perspective, particularly the logic of
propositions represented by projectors.

A central ingredient will presumably be the fact that each (normalized) state
|i gives rise to a probability assignment

P rob(P | |i) := h| P |i = h| P 2 |i
   2
= h| P P |i = P |i (15.318)

which will be interpreted in some way as the probability that the state of affairs
represented by P is realized if the state is |i.

Since all propositions in physics can be reduced ultimately to statements about

1226
values(possessed, measured or otherwise) of physical quantities, nothing is lost
by focussing on this case.

In what follows, we shall only discuss propositions of the type A , which

asserts that the value of a physical quantity (or easily extended to a finite set
of compatible quantities) A lies in some subset < (or will be found to do
so if a suitable measurement is made). Assertions of this type refer to the state
of affairs at some specific time.

Propositions of the type A are what are normally assumed to be represented

by spectral projectors. If the spectrum of A is discrete, the operator PA is
just the sum of the projectors PA=ai where the eigenvalues ai belong to the
subset of real numbers. Of course, a special case is when = {aj } for some
specific eigenvalue aj , in which case PA = PA=aj .

The probabilistic predictions of quantum theory are that if the state is |i, then
2
Prob (A | |i) = PA |i (15.319)

In particular, P rob (A | |i) = 1 if and only if PA |i = |i, that is, |i

lies in the eigenspace HA (= HPA ) generated by all eigenvectors of A whose
eigenvalues are in .

Note that a given projection operator may represent more than one proposition
of this type. For example, two operators A and B may share a common spectral
projector so that PA = PB0 for some subsets and 0 of <. In this case,
we shall say that the propositions A and hB i0 are physically equivalent.
There are implications for such a definition if A, B 6= 0.

Important issues concerning this proposition-projector association include:

(1) the exact nature of the propositions represented by projection operators

(2) the way in which the concepts of truth and falsity apply to such
propositions

(3) the quantum analogue of the way in which the set of propositions in classi-
cal physics acquires the structure of a Boolean algebra from the underlying
mathematical representation of states and physical quantities

The nature of the proposition represented by a projection operator depends on

the interpretation of quantum theory that is adopted.

A realist might want to say that the projection operator PA represents the
physical quantity whose value is 1 if the value of A lies in and is 0 otherwise.
Hence, in this case, the proposition A is to be read as

1227
 The realist version of A :
A has a value and this value lies in the subset .

Of course, this leaves open the question of when such an assertion is justified.
The situation in classical physics is clear: the proposition is true in a state s S
if, and only if, fA : S < is such that fA (s) . However, as we saw in the
Kochen-Specker discussion, any attempt to assign possessed values for all quan-
tities in quantum theory can only be maintained at the expense of considering
these properties to be contextual. This means that it is not meaningful to say
that any particular proposition is either true or false without specifying what
other compatible propositions are to be considered at the same time.

One possibility is to adopt the minimal attitude that a physical quantity A

can only be said to have a value a if (1) a is an eigenvalue of A and (2) the
state |i lies in the associated eigenspace HA=a . One might then argue that,
since the first part of the realist version of A is A has a value.....,
the truth of A requires that |i belong to one of the eigenspaces of
A. Thus, |i must be an eigenvector of A. The second part .......... and
this value lies in the subset , then means that the corresponding eigen-
value must belong to the set = {a1 , a2 , ....., ad }. Note that the first re-
quirement excludes any |i that is the sum of elements in more than one of
the eigenspaces HA=a1 , HA=a2 , ......., HA=ad , even though the formalism gives
P rob(A | |i) = 1 for all such states. In these circumstances, it is clearly
inappropriate to regard the spectral projector PA as representing the propo-
sition A . Therefore, if one wanted to maintain the proposition-projector
link, it would be natural to adopt the weaker condition that A is true in a
state |i if PA |i = |i, that is, if |i is any linear combination of eigenvec-
tors of A whose corresponding eigenvalues lie in .

The situation is less ambiguous in the instrumentalist approaches to quantum

theory. The projector PA now represents a physical observable that is defined
operationally by specifying how it is to be measured, that is, measure A, and
then, if the result lies in , assign the number 1 to the observable, otherwise
assign the number 0. The proposition A should now be read as asserting
that a measurement of A will yield a result that lies in the subset .

On the face to it, the statement if a measurement of A is made, then the result
will lie in is applicable to a single system, as is the realist claim that A
has a value that lies in . However, the instrumentalist form of the proposi-
tion is not a statement about how things are, but rather a claim about what
would happen if a certain operation was performed. This seems to cry out
for a positivist-type verification, that is, make the measurement and see what
happens - if the result does not lie in , then the proposition is certainly false.

But what if the result does lie in ? Does this mean this proposition is true?

1228
Certainly not if, as is arguably the case, a more precise rendering of the propo-
sition is:

The modal version of A :

if a measurement of A is made, then, necessarily,

the result will lie in the subset .

Put in the modal form, the proposition is manifestly counterfactual and cannot
be verified by any single, or finite set, of measurements. Indeed, it could be
argued that, in this form, the proposition does not apply at all to an individual
system but only to a collection of such on which repeated measurements are to
be made. Thus, in this reading, it is consistent to say that states and proposi-
tions apply only to such collections.

The discussion above shows why, in the relative-frequency, anti-realist interpre-

tations of quantum theory, we are lead naturally to read the proposition A
as:

The probability-one version of A :

if a measurement of A is made, then, with,

probability one, the result will lie in the subset .

in which case it is clear when it would be asserted as true, namely,

if the quantum state is |i, a proposition T represented

by a projection operator PT is true if P rob(T | |i) = 1

Some Comments on these Ideas

1. Since
2
P rob(T | |i) = PT |i

the condition P rob(T | |i) = 1 implies that PT |i = |i, that is, |i is,
in fact, an eigenvector of PT with eigenvalue 1.

2. This result implies that the proposition A is true in any state |i that
can be written as a linear superposition of the eigenvectors associated with
the eigenvalues in . Of course, this does not mean that the probability
of getting any specific eigenvalue is equal to 1. Indeed, this number will
be strictly less than 1 if |i has non-zero expansion coefficients for more
than one of the eigenspaces. Note that, unlike in the realist case, this
causes no problems, since the modal form of A does not say that a
measurement of A necessarily gives any one particular element of .

1229
 3. Care must be taken when handling propositions in this modal form. For
example, if the proposition A = am is interpreted as if A is measured, then
the result will necessarily be am , then the value of P rob(A = am | |i)
cannot be read as the probability that the proposition is true when the
state is |i. Rather, P rob(A = am | |i) refers to the relative frequency
with which result am will be found on a given collection of systems.

For example, the statement there is a 70% chance that the proposition if A
is measured, then the result will necessarily be am is true, is quite different
from if A is measured, then there is a 70% chance that the result will be am .
It is the latter that is intended in the quantum formalism by the statement
P rob(A = am | |i) = 0.7. Note that, if desired, one might introduce an extended
class of modal propositions of the form

if a measurement of A is made, then the probability

that the result will lie in IS r.

where the real number r satisfies 0 r 1. The special value r = 1 then

corresponds to the probability-one proposition above.

15.4.2 Is False the Same as Not True?

The next issue to consider is what it means to say that a proposition is false
and under what circumstances in quantum theory such an assertion would be
justified.

In classical physics or logic, this causes no difficulties:

a proposition is false if and only if it is not true

If we adopt this idea in quantum theory then, in the relative-frequency interpre-

tations, A is false means that, in a long series of repeated measurements
of A, one or more numbers not lying in will occur with a non-zero relative
frequency. In general, this suggests the definition that a proposition T is false
in a state |i if P rob(T | |i) < 1.

On the other hand, a realist may balk at the idea that the proposition A
could be said to be false if there is still a non-zero probability of finding values
in . When confronted with the existing formalism of quantum theory - in
particular the idea that propositions are to be associated with projection oper-
ators - a realist may feel that the most natural assignment of truth and falsity
in a state |i is that a proposition A is (1) true in state |i if P rob(A
| |i) = 1 (that is, |i HA ) and (2) false if P rob(A | |i) = 0 (that

is, |i HA ).

However, any vector |i H can be written as a sum of components in HA

1230

and HA , and, according to these assignments, the proposition A cannot
be said to be either true or false if both components are non-zero. This suggests
that some sort of multi-valued logic should be used. The simplest such attempt,
due to Reichenbach is to use a three-valued logic in which a proposition can be
either true, false, or indeterminate. The typical quantum mechanical situation
in which a non-trivial superposition of eigenvectors leads to a proposition being
neither true nor false would then be assigned to the indeterminate category.

A more complex possibility is to use a continuous spectrum of truth values lying

between 0 and 1, with the truth value of the proposition T in a state |i being
set equal to P rob(T | |i). This is the idea that lies behind the concept of fuzzy
logic.

But, in either case, it is not really clear what the meaning is of a proposition
that is deemed to apply to a single system but which is then said to be neither
true nor false. Philosophers have debated this for some time without reaching
any definite conclusions.

In addition, serious worries have been expressed about using the idea of using a
multi-valued logic within a mathematical framework that is, itself, based firmly
on classical two-valued logic.

Another major problem with multi-valued logics is to decide on the appropriate

analogues of the usual logical connectives and, or, negation, and logical impli-
cation that are usually introduced in the context of classical physics. Many
mathematical attempts have been tried, but no clear consensus has emerged.

The difficulties with logical connectives can be alleviated by adopting the strat-
egy in which a proposition A is understood to be false in state |i if it is
not true, that is, if P rob(A | |i) < 1. This approach is quite natural in
instrumentalist interpretations of quantum theory, and results in a two-valued
propositional structure, albeit at the expense of introducing a certain asymme-
try between the concepts of true and false. It is still necessary to define the
logical connectives, however.

The Logical Connectives

The construction of a quantum logic on the set of two-valued (true/false) propo-
sitions originally started by Birkhoff and von Neumann. They started with the
set (H) of all closed subspaces of a Hilbert space H (or, equivalently, with the
set of all projection operators) and suggested certain analogues of the classical
logical operations. The basic definitions are as follows.

1. A proposition T implies another U , written T  U , if for all states |i

such that P rob(T | |i) = 1, we have P rob(U | |i) = 1, that is, U is true
in all states in which T is true. This is equivalent to HT being a closed

1231
 subspace of HU .

The analogous classical relation is ST SU .

If PT and PU are the associated projection operators, the relation T  U ,

is equivalent to
PT PU = PU PT = PT (15.320)
Note that if T implies U and U implies T , it is natural to say that
the propositions T and U are equivalent. Indeed, The former means
PT PU = PT and PT PU = PU , that is, PT = PU , which is the defini-
tion of physical equivalence suggested earlier. As is the case in classical
physics, the appropriate objects to study are the equivalence classes of
propositions which, in the quantum case, are the projection operators.
This means that it is more appropriate to view the logical operations as
defined on projection operators (or, equivalently, topologically closed sub-
spaces), than on propositions, per se.
2. The operator that corresponds to the identically false proposition is the
that maps every vector in H to the null vector, and the
null operator
operator that is the identically true proposition is the unit or identity
Thus, for any projection operator P , we have
operator I.
 P  I.

In classical physics, the analogue of these two propositions are the empty
set and the whole space S respectively, and for all subsets ST of S we
have ST S.
3. If T is a proposition, the proposition T (not T ) is defined by requiring
that, for all states |i, P rob(T | |i) = 1 if, and only if, P rob(T | |i) = 0.
This is equivalent to the relation HT = (HT ) . This relation determines
a unique projection operator PT that represents T :

PT = I PT (15.321)

which actually implies the (superficially stronger) relation between the

probabilities of T and T

P rob(T | |i) + P rob(T | |i) = 1 (15.322)

The classical analogue is ST = S ST .

Unlike the classical case, the quantum definition has the peculiar prop-
erty that T can be false in a state |i (that is, T is not true, so that
P rob(T | |i) < 1) without implying that T is true (which requires that
P rob(T | |i) = 0). But, as expected, T true does imply that T is false!
This curious asymmetry stems directly from (1) the existence of linear
superpositions of eigenstates of the operator PT and (2) our desire to have
a binary-valued, rather than trinary-valued, logic.

1232
4. Let T and U be a pair of propositions with associated closed subspaces
HT and HU , and projection operators PT and PU , respectively. Then
the proposition T U (to be interpreted as some type of conjunction,
T AN D U ) is defined by requiring that P rob(T U | |i) = 1 in a state
|i if and only if P rob(T | |i) = 1 and P rob(U | |i) = 1, that is, T U is
true in the state |i if and only if both T AN D U are true. This means

HT U = HT HU (15.323)

which is reminiscent of the classical result ST U = ST SU .

Expressed in terms of subspaces, the quantum definition seems rather

natural. However, there is no simple way of translating it into the language
of projection operators since the operator PT PU := PT U that projects
onto HT HU is not a simple function of the projection operators PT and
PU . If [PT , PU ] = 0 it can be shown that the projection operator PT PU
that represents the proposition T U is

PT PU = PT PU (15.324)

5. In classical physics, the set that represents the disjunction T U (T OR U )

of a pair of propositions T and U , is the union ST SU of the sets ST
and SU that represent them, that is, it is the smallest subset of the state
space S that contains both ST and SU .

The obvious analogue in quantum theory is the set HT U equal to the

union of HT and HU . But this cannot be correct since HT HU is not a
linear subspace of H, that is, it is not closed under the operation of taking
sums of vectors. This suggests that the more appropriate definition would
be HT U = HT + HU , which is defined to be the linear sum of all vectors
in HT and HU . However, even this is not correct if H has an infinite
dimension since then HT + HU may not be topologically closed. This
defect is remedied by the final definition

HT U = HT + HU (15.325)

which is the smallest, topologically closed linear subspace of that contains

both HT and HU .

As in the case of the conjunction, there is no easy way to write this

last relation in terms of the associated projection operators. However,
if [PT , PU ] = 0, then the projection operator PT PU := PT U that rep-
resents the proposition T U is

PT PU = PT + PU PT PU (15.326)

In particular, if PT PU = 0 then PT PU = PT + PU (this is the classical

result).

1233
Some Comments

1. The closed subspaces HT and HU of H satisfy the equation

HT + HU = (HT HU

) (15.327)
This is the analogue of the DeMorgan rule of classical set theory
ST SU = S ((S ST ) (S SU )) (15.328)
which corresponds to the logical identity T U = (T U ).

This fact provides further justification for the consistency of the above
assignments of logical connectives.
2. Care must be taken in interpreting the assignment of logical connectives.
This is particularly true of the idea of T U . One may interpret as
and in the normal sense only if the two operators commute since, as
mentioned earlier, only then is it unequivocally meaningful to assert that
states of the system exist in which both properties are possessed at the
same time. In the non-commuting case, the best that can be said (instead
of
PT PU = PT PU (15.329)
is that
PT PU = lim (PT PU )k (15.330)
k
which is sometimes read as saying that T U refers to the results of an
infinite sequence of measurements of T and U .
3. The fact that most pairs of operators do not commute is reflected in
the crucial property of the set of quantum propositions that the algebraic
structure associated with the definitions of and fails to be distributive.

This can be seen easily with the aid of a simple example. Let H|1 i and
H|2 i be one-dimensional subspaces of a two-dimensional Hilbert space
H, spanned by (that is, all complex multiples of) the linearly-independent
vectors |1 i and |2 i respectively, and let |i be any non-trivial linear
combination of |1 i and |2 i. Then we have


H|i H|1 i + H|2 i = H|i H = H|i (15.331)
but if we use the standard distributive property we have


H|i H|1 i + H|2 i = H|i H|1 i + H|i H|2 i = H + H = H
(15.332)
We thus have the contradiction H|i = H . This result contrasts sharply
with the distributive property of classical propositions, which is direct con-
sequence of the Boolean nature of the algebra if intersections and unions
of subsets of a classical state space.

1234
Most of the strange, non-classical features of quantum theory can be traced back
to this non-distributive property of the logical structure of quantum proposi-
tions.

It is useful to think about Gleasons theorem in light of these discussions.

Gleasons Theorem

The probability rules for density matrices imply that for any proposition P ,
with associated projector P (H), the probability associated with P in the
state is
 
P rob(P |) = T r P (15.333)

An interesting and important question arises at this point. Are there any other
ways of assigning probabilities to the elements of (H), or, is the standard
quantum formalism unique? More precisely, can we find a new probabilistic
theory by starting with some Hilbert space H, with its non-distributive algebra
(H) of subspaces/projection operators, and then construct a probability map
P rob : (H) < that is not of the type given above? In particular, what is
the space of states in such a theory?

To tackle this question we must first decide what requirements should be satis-
fied by a probability map P rob : (H) <. The analogous question in classical
statistical physics is answered by defining a general probability measure to be a
real-valued function of subsets of S that satisfies the three conditions

0 (W ) 1 for all such subsets W

() = 0 , (S) = 1
(W1 W2 ....) = (W1 ) + (W2 ) + ......

In the quantum case, one obvious condition is 0 P rob(P ) 1 for all proposi-
tions P (H). Two other, essentially trivial, requirements are P rob() = 0
and P rob(I) = 1, where and I correspond respectively to the propositions
that are identically false and true.

They key question is what should be required of the probabilities P rob(T U )

and P rob(T U ) for a pair of propositions T and U .

It is helpful to start by considering the analogous question in classical physics,

where propositions correspond to subspaces of the state space S. In this case,
bearing in mind that ST U = ST SU and ST U = ST SU , it is easy to see
with the aid of a Venn diagram that one required condition is

P rob(T U ) = P rob(T ) + P rob(U ) (15.334)

1235
which, if {P1 , P2 , ...., PM } is any finite set of propositions that are pairwise
exclusive, generalizes at once to
M
X
P rob(P1 P2 ........ PM ) = P rob(Pi ) (15.335)
i=1

An important technical requirement in classical probability theory is countable

additivity, that is, this relation should continue to hold in the limit M .

The quantum analogue of two mutually-exclusive propositions is a pair of pro-

jection operators PT , PU that are orthogonal, that is, PT PU = 0 = PU PT . In
this case, problems of incompatibility do not apply, and it is natural to suppose
that the last relation passes across in the form
M
X
P1 P2 ........ PM = Pi (15.336)
i=1

for any collection of pairwise orthogonal projection operators.

Thus, a minimal set of requirements on any quantum probability function P rob :

(H) < is
0 P rob(P ) 1 for all P (H)
P rob(0) = 0 , P rob(I) =1

!
X X
P rob Pi = P rob(Pi )
i=1 i=1

for any finite or countably infinite set {P1 , P2 , ....} of projection operators that
are pairwise mutually orthogonal.

Typically, such sets arise as the spectral projectors of a self-adjoint operator.

One way of satisfying these relations is to define

 
P rob(P ) = T r P (15.337)

where is any density operator.

A remarkable theorem due to Gleason shows that, provided that dim(H) > 2,
the converse is true - the only way of satisfying
0 P rob(P ) 1 for all P (H)
P rob(0) = 0 , P rob(I) =1

!
X X
P rob Pi = P rob(Pi )
i=1 i=1

1236
is with the aid of a density operator and the relation
 
P rob(P ) = T r P (15.338)

Gleasons theorem places strong constraints on any attempts to modify the

standard quantum formalism as, for example, might be desired in the construc-
tion of hidden variable theories. It also gives a rather deep reason why density
operators play such an important role in quantum theory.

15.5 Quantum Mechanical Picture of the World

Let us look at Kochen-Speker and Gleason in another way to help our under-
standing. Quantum mechanics provides us with the most detailed picture of
the natural world. The mention of quantum mechanics conjures up ideas of
uncertainty, indeterminacy, and probability. This section aims to explain these
ideas in a way based on the knowledge we have gained in this text.

If you look at most papers on this subject it seems that only the experts can
know the details, because only they have the knowledge of the esoteric mathe-
matics required. Hopefully we can simplify the arguments using only elementary
mathematics(especially geometric arguments) in this presentation.

We first build a mathematical model of the subatomic world based on the prop-
erties revealed by the SternGerlach experiment. We then analyze the model and
prove a theorem whose physical implications are immediately obvious. One must
follow the reasoning carefully and analyze the figures to understand the proof of
the theorem. We will find that the subatomic world cannot be completely de-
termined and that its properties cannot be definite before an experiment reveals
these properties to us.

15.5.1 The Model

To understand atoms or subatomic particles, the tools of classical physics, as
we have seen, are not sufficient. We need new kinds of concepts and laws, which
are contained in quantum mechanics. Because matter and radiation consists
of subatomic particles, the macroscopic world must be ultimately explained
according to quantum principles.

As we saw earlier, one of the first surprising results on the behavior of small
particles was observed by Stern and Gerlach in 1922. In that experiment, a
beam of silver atoms was sent through a magnet as shown in Fig. 15.15 below.

1237
 Figure 15.15: Stern-Gerlach Setup

According to classical mechanics, the silver atoms should be deflected by a cer-

tain angle. There should be a large range of scattering angles and, therefore, a
band of impacts on a screen placed behind the magnet. Surprisingly, only two
narrow strips were observed, one above the initial direction and one below. The
fact that only two strips were obtained was totally unexpected in the context
of classical physics; the only explanation of this fact was that a property of
the silver atomscalled intrinsic angular momentum or spincould take on only
two possible values, ~/2. More precisely, the projection of the spin along the
direction of the magnetic field can have only two values in this case.

It was subsequently discovered that for atoms other than silver, the beam split
into three, four, or more branches, depending on the kind of atom. If the beam
of particles splits into two branches, we say the particle has spin 1/2; if it splits
into three, we say the particle has spin 1; if it splits into four, we say the particle
has spin 3/2, and so forth.

The discovery of angular momentum quantization was a key moment in the

creation of quantum mechanics. We are interested only in a particular aspect of
the SternGerlach experiment. In fact, we will not even talk directly about the
angular momentum, although it will constantly enhance the argument. Let us
consider spin 1 particles. If we shoot a beam of those particles at a SternGer-
lach device positioned according to Fig. 15.16 (oriented in the zdirection
represented by the unit vector z), some of the particles will deflect up (in state
|1z i = |z i), some of them down(in state |1z i = |z i), and a third group will
not deflect up or down(in state |0z i = |z i).

1238
 Figure 15.16: Stern-Gerlach oriented in zdirection

If the magnet is turned to face in any other direction, say the xdirection
represented by the unit vector x, as in Fig. 15.17 below

Figure 15.17: Stern-Gerlach oriented in xdirection

the result will be the same: a beam goes to the right, another one goes to the left,
and a third beam is not deflected. The corresponding states are |1x i = |x i),
|1x i = |x i), and |0x i = |x i as before.

From this experiment, we come to the important conclusion that the state of
a system is characterized by a state vector that is labeled by a number or an
arrow representing a direction in space. This type of characterization, as we
have seen, turns out to be a basic principle of quantum mechanics.

We can use a SternGerlach device to answer the following question: Is the sys-
tem in the |0x i = |x i state? The answer will be yes for the beam that does
not deflect when the magnetic field points in the xdirection. The beams that
are deflected will give us negative answer to this question. This kind of ques-
tion, which is answered with only two alternatives: yes or no, 1 or 0, is called
a projector. Thus, the question, Is the system in the |0x i = |x i state? is
called the x or 0x projector (or projection operator) and is represented by the
operator symbol Px or P0x . In the same way, we can apply a 0y projector P0y
which corresponds to the question, Is the system in the |0y i = |y i state? In

1239
general, we can define the projector P0n corresponding to an arbitrary direction
represented by the unit vector n.

But quantum mechanics, as we have seen, is very peculiar. Suppose that we

apply the projector P0x with an affirmative response, so that we are sure that
the system (the atom) is in the |0x i state. If we now introduce a second Stern-
Gerlach magnet, pointing in any other direction n, then the projector P0n will
be answered sometimes affirmatively and sometimes negatively. To be precise,
the probability of obtaining yes is equal to the square of the projection (proba-
bility amplitude) of n on x, i.e., n x as shown in Fig. 15.18 below.

Figure 15.18: Projection = Probability Amplitude

This, as we know, is the same as the square of the scalar product (probability
amplitude) of the states representing the atom in each case, i.e., h0n | 0x i, which
is just the Born Rule. If the direction n is orthogonal to x, then the projection
is zero. Therefore, the probability that the system is in the |0n i is zero. In this
case, the answer to the question Pn will be no if Px had been affirmative.

Gleason, in his remarkable theorem published in 1957 (A. Gleason, Measures

on the closed subspaces of a Hilbert space, J. Math. Mech. 6(6), 885893
(1957)), proved that a probability measure defined over a spherical sur-
face [or a linear space of dimension 3 if for any vector a and any number ,
(a) = (a) (a and a are the same physical state)] has to be of the form:
prob(P = 1|Pz = 1) = cos2 . (Note: P = 1 is a convenient way of saying that
P takes on the value 1. Other authors write v(P=1 ) instead. Gleason showed
the equivalent result for any dimension of Hilbert space.

Questions referring to perpendicular directions are mutually exclusive: if one of

them is affirmative, the other will be negative. Because these questions or pro-
jectors do not interfere with each other, we say that they are compatible, in the
sense that they can be measured together. (If the directions are not perpendic-
ular, they can interfere: we can obtain two affirmative results). Mathematically

1240
this looks like:

1 1 0
1 1
|0x i = 0 |0y i = 0 |0z i = 1
2 1 2 1 0

Thus,
h0x | 0y i = h0x | 0z i = h0y | 0z i = 0
In three-dimensional space, there are at most three compatible projectors. For
example, the projectors, P0x , P0y , and P0z are compatible, because they do not
interfere with each other and can be implemented at the same time, meaning
that the response will be yes for one of the three and no for the other two. If
the system is in the |0x i state, for instance, it cannot be in the |0y i state or the
|0z i state.

In general, a set of three directions that are mutually perpendicular will be

called an orthogonal triple. In an orthogonal triple, one of the projectors will
yield yes (1) and the other two will yield no (0). The maximum number of
compatible projectors is three because we are considering a spin-1 particle. The
spin of other kinds of particles as well as other quantum observable quantities
will require a different number of dimensions. But the important point is that
any quantum observable quantity can be split into projectors so that our analysis
can be extended to every quantum observable quantity.

15.5.2 The Question

With these ideas in mind, we turn our attention to the central idea of this sec-
tion: What picture of the world does quantum mechanics provide? We know
quantum mechanics is a probabilistic theory. In each experiment we have a
certain probability of obtaining one result or another. But when we throw a
die, we also obtain a result with a certain probability. Is it the same kind of
randomness? Do the probabilities have the same nature in both cases?

When we throw the die, we could in principle know what number we would see if
we knew the precise value of the force applied to the die. But there are so many
elements we do not control that we accept the result as a random event, even
though in theory, we could predict the result if we analyzed the details of the
enormous variety of variables that are relevant. We use probabilities because
analyzing all those details would be too laborious. It is a question of practical
difficulty; not just of principle.

What would happen in a quantum experiment? Would there be any way of de-
termining the result of an experiment through a complete understanding of the
variables? We know that when we repeat a quantum experiment we can obtain
different results, even with the same initial conditions. That does not happen in
the case of the die: if the forces were the same, the result would be the same as

1241
well. In a quantum experiment we may need to have more profound knowledge.
Moreover, it is possible that todays technology is insufficient to discover the
subtle variables that cause the same experiment to produce different results. Is
it not plausible to think that there should be something within the system that
explains why we obtain one result and not another? Einstein, among others,
thought this way. For him, quantum mechanicsdespite its achievementsseemed
to be an incomplete theory because it could not predict with certainty the re-
sult of an experiment. Even though our insufficient knowledge prevents us from
identifying it, could each object have a hidden variable that could explain its
apparently random behavior in a particular experiment? Or do we have to ad-
mit that quantum uncertainty is more radical? Is matter really intrinsically
random so that there is no possibility, even in principle, of predicting the result
of an experiment?

An alternative lies between these two pictures of the world. If the hidden vari-
ables explaining the apparently random results exist, we would have a totally
deterministic world. It is a world where there is a cause for everything that
happens. In such a world of hidden variables, the properties of a subatomic
particle, or any other quantum system, would be clearly defined. If we re-
turn to the set of projectors we have constructed, then each projector would
have a plain answer: yes (1) or no (0). As we have said the projectors take
on only two values, 0 or 1, with a certain probability. When we say that the
projectors are defined, we mean that they take on the value 1 (or 0) with
probability 1. Therefore the expectation value of any projector in that case is
hP i = prob(P = 1) = 0 or 1. In brief definite values means hP i = 0 or 1.
If there were a definite world out there, we could ascribe the values 0 or 1 to all
the projectors, with the condition that if three projectors are mutually orthog-
onal, only one of them will be ascribed the value 1. Note that the assumption
hP i = 0 or 1 is by itself contradictory to quantum mechanics because in the
latter prob(P = 1|Pz = 1) = cos2 . We do not have to wait for the conclusion
of the theorem. It is a matter of principle: a theory with definite values clashes
with quantum mechanics. So for that matter we do not need the KSB theorem.
The merit of thePtheoremPis to show that any theory with definite values and
satisfying prob( Pi ) = prob(Pi ) (Pi mutually orthogonal) is contradictory
by itself.

But this proposition is a mathematical one and can be analyzed as such. Thus,
mathematics will tell us which is the correct picture of the world. (This power
of mathematics is possible only because we have previously constructed the pro-
jectors, a mathematical structure, superimposed on the physical world).

In 1932 the problem of hidden variables was discussed by Von Neumann. He

presented it like this: Suppose that the correct picture of the world is the first
one; that is, let us suppose that apparently identical states have a hidden vari-
able that makes them different. Then we could select a set of systems that were
exactly in the same pure state, that is, with the same value for the hidden vari-

1242
able. In this case, our set would be a true dispersionless state (dispersion is the
deviation from the mean). On the contrary, if there were no hidden variables,
then apparently identical systems (systems that are in the same state) would
have slight differences which would correspond to their different behavior in the
experiments; thus, they do show dispersion. Von Neumann proved that there
are no dispersionless states; therefore there cannot be hidden variables. But his
demonstration was not conclusive. The matter was not resolved until 19661967.

It was Bell, who used a result obtained previously by Gleason and Kochen and
Specker, who proved that there cannot be pure dispersionless states, that is, it
is impossible to ascribe values 0, 1 to the projectors consistently. We present
here a simplified version, proposed by Gill and Keanes (and based on the work
of Piron).

As we have said, the question is whether it is possible to maintain a world pic-

ture in which the quantum observable quantities, the properties of subatomic
particles, exist before we measure or look at them (looking at things is a way
of measuring them). In that world we could assign values to every projector.

Because the projectors point in any direction of space, we can talk about projec-
tors or directions alike. Besides, let us remember that, if we consider a sphere of
unit radius, each direction has one corresponding point on it: the one where that
direction goes through the sphere. Thus we can equally well speak of projec-
tors, directions, or points on a sphere. Consequently, the existence of an outside
world, determinate and deterministic, would mean that we can ascribe values
0 or 1 to all the points on a sphere. But remember that we have a restriction:
in each orthogonal triple, we ascribe the value 1 to only one point and the 0 to
the other two.

To see whether it is possible to assign values 0 or 1 to all the points on a spherical

surface with that restriction, we need a geometrical lemma.

15.5.3 The Geometrical Lemma

Let us now look at the orthogonal triple A, AE , AP as shown in Fig.15.19 below.
We have rotated the equatorial plane so that the great circle has A as its most
northerly point, and crosses the equator at the point AE . Call AP the point of
the northern hemisphere orthogonal to that great circle. Of these three points,
one must have the value 1; the other two, 0.

1243
 Figure 15.19: Orthogonal Triple

Proposition - If the value of AP , a direction perpendicular to a great circle, is

1, then all the directions in the great circle have to take on the value 0.

To see that this statement is true, it is sufficient to note that any direction B in
the great circle can make up an orthogonal triple with AP and another direction
B 0 in the circle, perpendicular to B as shown in Fig.15.20 below.

Figure 15.20: If value of AP is 1, B and B 0 take on the value 0

Because an orthogonal triple can have only one 1, B and B 0 must take on the
value 0.

We next prove a lemma which will be fundamental to our argument.

Proposition - Starting from one point A of the northern hemisphere, we can

reach any more southerly point than A by a finite sequence of great circle de-
scents.

The great circle descent from a point A is the great circle which has A as its
most northerly point. (From now on, the great circle descent from A will be
called the great circle from A). Beginning at any point A (not at the north
pole or on the equator), we can descend along the great circle from A to an-
other point B; then we can descend along the great circle from B to another

1244
point C; we can once more descend along the great circle from C; and so forth.
The lemma states that we can go along great circle descents from one point to
another further south. That means that we can travel from Madrid to Manila
going along a sequence of great circle descents.

The proof of this lemma is straightforward if we project the points of the north-
ern hemisphere from the center of the sphere onto the plane tangent to the
sphere at the north pole. In Fig.15.21 below we see that the projection of a
parallel is a circumference (parallels are lines of constant latitude). We also
see that a great circle descent projects onto a straight line, the tangent to the
circumference projected by the parallel that contains the summit of the great
circle. Besides, the meridians give rise to straight lines passing through the
north pole.

Figure 15.21: Parallel and great circle that are tangent at S project from sphere
center onto tangent plane at North Pole, giving rise, respectively, to a circum-
ference and a straight line tangent to it at point s. Point C of great circle from
S projects onto point c of straight line. Meridian through S gives rise to straight
line P S, perpendicular to straight line sc.

These considerations imply that the great circle descent from S projects onto a
straight line, tangent to the circumference at the point s of the tangent plane
and perpendicular at that point to the line projected by the meridian. To prove
the lemma, we begin with the simplest case: How can we go along great cir-
cle descents from the point S to the point T , both on the same meridian (see
Fig.15.22 below)?

1245
 Figure 15.22: S and T are on the same meridian

First, let us project the points S and T onto the plane tangent to the sphere at
the north pole (see Fig.15.23 below)

Figure 15.23: The projection of the great circle from S (DS ) is the straight line
ds, perpendicular to the projection of its meridian (the straight line P st)

We will obtain the points s and t. We also project the great circle from S (DS )
and obtain the straight line ds , perpendicular to the projection of the meridian
P st. Remember that we want to find a point R such that we can descend to R
along the great circle from S, and then to T along the great circle from R. We
are going to search for R using the tangent plane, because R will be the summit
of its great circle if the projections of its meridian and the great circle from it
are perpendicular. Thus we should search for a point r in ds such that the angle
P rt is 90 ; in this way RT will be an arc of the great circle from R, because
its projection rt will be perpendicular at r to the projection P r of the meridian
through R. To find the point r it is sufficient to recall a statement of elementary
geometry: An inscribed angle spanning a semicircumference is 90 . Therefore
we draw the semicircumference whose diameter is the segment P t (see Fig.15.24
below); the point where it intersects ds is the point r for which we were looking.

1246
Figure 15.24: r is intersection of semicircumference(diameter = segment P t)
with line ds

Next we find the point R in the sphere surface whose projection is r. In Fig.
15.25 below we see the spatial construction that enables us to find the point R,
and the arcs of the great circle descents DS and DR that we travel on to reach
T from S.

Figure 15.25: Join r with center of sphere and find R on spherical surface

We have already looked at the case in which the two points that we want to
join using great circle descents are on the same meridian. Next, we examine the
procedure used when the two points are on different meridians. In particular,
we will examine the case when the two points are almost antipodespoints whose
geographic longitudes differ by 180 .

Let us suppose that we intend to descend from the point S to another one on
a lower latitude, which is on the other side (that is, 180 away) on the same

1247
meridian. First we project the point S onto the plane tangent to the sphere at
the north pole (see Fig. 15.26 below).

Figure 15.26: Project point S onto plane tangent to sphere at North pole

Next, we divide the angle of 180 that separates the points of interest in the
projection plane into, for example, eight parts. We start at s and make the
construction in Fig. 15.27 below.

Figure 15.27: Going to the other side of the meridian

We call T the point on the sphere surface whose projection onto the plane is
exactly the point t. The points S and T become linked by the arcs of great circle
descents corresponding to the projections perpendicular to the lines dividing
the 180 angle. (If we divide the 180 into smaller equal angles and make the
previous construction, we could get from S to latitudes lower but almost equal
to that of S). To find the spherical surface points we join each of the dividing
points with the spheres center (see Fig.15.28 below).

1248
 Figure 15.28: Project back to find the summits of the great circle descents

The points S and T are thus linked by the arcs of the succession of great circle
descents shown in Fig. 15.29 below.

Figure 15.29: Find the arcs of the great circle descents

As we have mentioned, if we increase the divisions of the 180 angle, we will

reach latitudes as close as we wish to that of the starting point. The connection
with southerly points could be achieved either by varying one of the angles
or following the method we have used before to join two points on the same
meridian. In any case it is clear that we can always reach, starting from one
point A of the northern hemisphere, any more southerly point than A (even if
it is on the other side on the same meridian) by a finite sequence of great circle
descents.

15.5.4 The Theorem

Using this lemma, we can determine if it is possible to assign values 0 or 1 to the
points of the spherical surface so that we have exactly one 1 in each orthogonal

1249
triple. First let us see which distribution of zeros and ones the lemma implies.
Let us begin with an orthogonal triple and assign 1 to one of the points and 0
to the other two. Let us place the 1 at the North Pole to facilitate the picture;
all the points on the equator will thus take on the value 0 (see the proposition).
Let us now consider the triple in Fig.15.30, where the great circles plane forms
an angle of more than 45 with the equator.

Figure 15.30: Great circle A forms an angle of more than 45 with the equator

If A has value 0, because AE is on the equator and is therefore 0-valued, the

point AP would take on the value 1; consequently, all the points of the great
circle descent from A would be 0-valued as well. But the latitude of AP is less
than 45 , which means that AP is more southerly than A; thus we can reach
AP along a sequence of great circle descents from A. Because we can reason
similarly about each one of these great circle descents, we ultimately conclude
that AP is 0-valued. (If the most northerly point of a great circle, say B, takes
on the value 0, because BE is on the equator and there- fore is 0-valued, BP
would take on the value 1. Therefore all the points of the great circle descent
from B would also take on the value 0). We thus arrive at a contradiction and
conclude that in a 45 dome around the north pole, all the points are 1-valued.

If all the points on the spherical surface within 45 of a 1-valued point have
to take the value 1, then the points on the great circles perpendicular to those
projectorswhich occupy the central zone of the spherical surfacehave to take
on the value 0. In short, from the hypothesis that the projectors have well-
defined values, we have reached the conclusion that the spherical surface must
be distributed according to Fig.15.31 below: If we choose as the North Pole the
direction of the original 1-valued projector, the points in the 45 domes around
the North Pole and around the South Pole are 1-valued. The remaining points
of the spherical surface are 0-valued.

1250
 Figure 15.31: Distribution of ones and zeros on the spherical surface

But the spherical surfaces division into 1s and 0s that we have concluded is con-
tradictory (in addition to clashing with quantum mechanics, that is, for quan-
tum mechanics, and contrary to the result we have obtained, there is a nonzero
probability that a projector within the upper skullcap takes on the value zero).
Indeed, from the hypothesis that projectors have well-defined values, we have
proven that if a projector is 1-valued, it has to be surrounded by a 45 dome
made up of 1s. But let us look at a projector that forms, for example, 40 with
the North Pole (P40 in Fig. 15.31). It also has the value 1; therefore all the
projectors until P85 (a projector that forms 85 with the North Pole) should be
1-valued as well. But we had concluded that P85 was 0-valued. Consequently,
we arrive at a contradiction. Thus it is impossible to distribute 1s and 0s on
a spherical surface with the required condition. (In terms of color we see that
if 1 means red and 0 means green and in each orthogonal triple there must be
only one red point, it is impossible to color the spherical surface with these two
colors). We have proven the KSB no-go theorem: It is impossible to ascribe
values 0 or 1 to each point on a spherical surface so that we have exactly one 1
in each orthogonal triple.

We conclude that we cannot maintain a picture of a completely determinate

world. We cannot think that reality exists out there, without observing it. Ein-
stein was not happy with this state of affairs. He thought that there should be an
objective world having definite properties whether or not they were measured.
It was Pais who said that during a walk with him Einstein suddenly stopped,
turned to Pais, and asked him whether he really believed that the moon existed
only when one looked at it. The theorem we have just proved shows that the
properties of microscopic systems are not completely defined before we look at
them. The value of the projectors remains indefinite until we measure them. We
need to accept a radically random world. The microscopic universe is blurred
until we make an experiment and force it to choose one option. For this reason

1251
Peres has said that unperformed experiments have no results, preventing us from
using counterfactual arguments.

Therefore, we have to admit that there are some commeasurable or co-decidable

(simultaneously decidable) projectors and others that are not. Hence, there are
some quantities that we can measure at the same time, but whose measuring
excludes measuring other quantities that are not compatible with them. And
we cannot say anything about the latter. The crux of the KSB theorem is that
we cannot assign any value at all to them.

One of the authors of the theorem, Specker, related this problem to a ques-
tion about the different kinds of Gods knowledge of reality. In particular, the
KSB theorem is related to the future contingencies of the philosopher and Je-
suit Pedro Fonseca (1528-1599): those future events whose occurrence depends
on an action taken by a creature exercising free will. According to Fonseca,
God knows the outcome of unperformed experiments, or as he puts it, he knows
what would happen if He put the will of the creatures in circumstances different
from those in which He put them. Father Fonseca calls this knowledge middle
science. But the KSB theorem proves that God does not have middle science.
He cannot know what would happen if something different from what actually
happens were to happen, that is, if a different projector had been measured.

15.5.5 The Discrete Version

Kochen and Speckers original paper did not deal with a continuous set of pro-
jectors. They searched for a finite number of projectors or directions leading
to a contradiction like the one we have found. They arrived at a contradiction
with a set of 117 directions. In 1990 Peres reduced that number to only 33. In
1991 Penrose heard Peres lecture on this work, and realized that Peres direc-
tions matched those of a geometrical figure he had seen in a lithograph, The
Waterfall, of the Dutch painter Escher (see the cap on the left tower of Fig.
15.32 below).

Figure 15.32: The left-hand tower has the quantum polyhedron

1252
The figure is a polyhedron which is composed of three interpenetrating cubes,
each obtained from the others by 90 rotations about lines joining the mid-
points of its opposite edges (see Fig. 15.33 below).

Figure 15.33: The quantum polyhedron is made o three interpenetrating cubes

The 33 directions of Peres are the lines through opposite vertices, opposite edge
midpoints, and opposite face midpoints of each cube.

In Figs.15.34-15.39, each letter corresponds to the straight line going through

the center of the cubes and the point closer to that letter. The straight lines
going through the center of a face are designated as AS (upper), AL (lateral),
and AF (frontal) for the cube A, BS , BL , and BF for the cube B,and CS ,
CL , and CF for the cube C. The polyhedrons symmetry allows the following
operations.

Figure 15.34: The directions on the cube A

1253
Figure 15.35: The directions on the two interpenetrating cubes A and B

Figure 15.36: The interpenetrating cubes A and B in perspective

Figure 15.37: The directions on the two interpenetrating cubes A and C

1254
 Figure 15.38: The interpenetrating cubes A and C in perspective

Figure 15.39: The directions on the cube C

In the orthogonal triple AS XY , we choose as the AS direction the one with the
value 1 so that the directions X and Y will take on the value 0. The directions
AL and AF are 0-valued too, because they are perpendicular to AS . In the
triple BS BF Y , the direction Y is 0-valued. From the two others, we choose BS
as the one with the value 1. In the triple CS CL X, the X is 0-valued. From the
two others, we choose CS as the one with the value 1. In the triple DEAF ,
AF is 0-valued. We can still choose the value of the two others. We choose D
as the one with the value 1. The directions F and G are perpendicular to D as
well; so they will take on the value 0.

In the triple HF Y , H has the value 1, because F and Y are 0-valued. J takes
on the value 0 because it is perpendicular to H. (It is clear that H and F are
perpendicular because F becomes H when we turn the cube A 90 around Y ,
to generate the cube C).

In the triple KCL J, K takes on the value 1 because it is orthogonal to CL and

J, which are 0-valued. L takes on the value 0 because it is perpendicular to K.
In the triple M Y L, M takes on the value 1 because it is orthogonal to Y and L.
N takes on the value 0 because it is perpendicular to M . In the triple OAL N , O
takes on the value 1 because it is orthogonal to AL and N . P takes on the value
0 because it is perpendicular to O. In QXP , Q takes on the value 1 because it
is orthogonal to X and P . R takes on the value 0 because it is perpendicular

1255
to Q. In SBF R, S takes on the value 1 because it is orthogonal to BF and
R, which are 0-valued. T takes on the value 0 because it is orthogonal to S.
Finally, in XGT , X takes on the value 1 because it is orthogonal to G and T ,
which had the value 0. But X was 0-valued! Thus we arrive at a contradiction.

By using only a finite number of directions, we have shown again that it is not
possible to assign 0,1 values to every direction if the condition that there is only
one 1-valued direction for each orthogonal triple of directions is fulfilled. So we
have again proved the KSB theorem.

How far does this theorem lead us? If, as shown, projectors and quantum
observable quantities are not completely determined, we have a picture of mi-
croscopic reality with indefinite and ambiguous properties. Physical properties
do not have, in general, objective existence independent of the act of observa-
tion. We should accept what Jordan had already said in 1934: Observations
not only disturb what has to be measured, they produce it! In a measurement of
position, for example, the electron is forced to a decision. We compel it to as-
sume a definite position; previously it was, in general, neither here nor there; it
had not yet made its decision for a definite position.... And Jordan concluded:
We ourselves produce the results of measurement. The KSB theorem proves
that we do not have an alternative picture of the subatomic world. That is its
importance.

15.6 Problems
15.6.1 Measurements in a Stern-Gerlach Apparatus
(a) A spin1/2 particle in the state |Sz +i goes through a Stern-Gerlach an-
alyzer having orientation n = cos z sin x (see figure below).

Figure 15.40: Tilted Stern-Gerlach Setup

1256
 What is the probability of finding the outgoing particle in the state?
(b) Now consider a Stern-Gerlach device of variable orientation as in the figure
below.

Figure 15.41: Variable Orientation Stern-Gerlach Setup

More specifically, assume that it can have the three different directions

n1 = n = cos z sin x
n2 = cos + 23 z sin + 23 x

n3 = cos + 34 z sin + 43 x

with equal probability 1/3. If a particle in the state |Sz +i enters the
analyzer, what is the probability that it will come out with spin eigenvalue
+~/2?
(c) Calculate the same probability as above but now for a Stern-Gerlach an-
alyzer that can have any orientation with equal probability.
(d) A pair of particles is emitted with the particles in opposite directions in
a singlet state |0, 0i. Each particle goes through a Stern-Gerlach analyzer
of the type introduced in (c); see figure below. Calculate the probability
of finding the exiting particles with opposite spin eigenvalues.

Figure 15.42: EPR Stern-Gerlach Setup

15.6.2 Measurement in 2-Particle State

A pair of particles moving in one dimension is in a state characterized by the
wave function
   
1 1
(x1 , x2 ) = N exp (x1 x2 + a)2 exp (x1 + x2 )2
2 2

1257
 (a) Discuss the behavior of (x1 , x2 ) in the limit a 0.
(b) Calculate the momentum space wave function and discuss its properties
in the above limit.
(c) Consider a simultaneous measurement of the positions x1 and x2 of the
two particles when the system is in the above state. What are the ex-
pected position values? What are the values resulting from simultaneous
measurement of the momenta p1 and p2 of the two particles?

15.6.3 Measurements on a 2 Spin-1/2 System

(a) Consider a system of two spin-1/2 particles in the singlet state
1  
|1, 0i = |i(1) |i(2) |i(1) |i(2)
2
(1)
and perform a measurement of Sz . Comment on the fact that a simulta-
(2)
neous measurement of Sz gives an outcome that can always be predicted
form the first-mentioned measurement. Show that this property, entan-
glement, is not shared by states that are tensor products. Is this state
1 
|i = |1, 1i + 2 |1, 0i + |1, 1i
2
entangled, i.e., is it a tensor product?
(b) Consider now the set of four states |ai, a = 0, 1, 2, 3:
1
|0i = (|1, 1i + i |1, 1i)
2
1
|1i = (|1, 1i + i |1, 1i)
2
1  i/4 
|2i = e |1, 0i ei/4 |0, 0i
2
1  i/4 
|3i = e |1, 0i + ei/4 |0, 0i
2
Show that these states are entangled and find the unitary matrix Ua such
that
|ai = Ua |i
where {|i} = |1, 1i , |1, 1i , |1, 0i , |0, 0i.
(c) Consider a one-particle state |i = C+ |i+C |i and one of the entangled
states considered in (b), for example |0i. Show that the product state can
be written as
1
|i |0i = (|0i |i + |1i | 0 i + |2i | 00 i + |3i | 000 i)
2
where the states | 0 i, | 00 i, | 000 i, are related to |i through a unitary
transformation

1258
15.6.4 Measurement of a Spin-1/2 Particle
A spin1/2 electron is sent through a solenoid with a uniform magnetic field
in the y direction and then measured with a Stern-Gerlach apparatus with field
gradient in the x direction as shown below:

Figure 15.43: EPR Stern-Gerlach Setup

The time spent inside the solenoid is such that t = , where = 2B B/~ is
the Larmor precession frequency.
(a) Suppose the input state is the pure state |z i. Show that the probability
for detector DA to fire as a function of is
1 2 1
PDA = (cos(/2) + sin(/2)) = (1 + sin )
2 2
Repeat for the state |z i and show that
1 2 1
PDA = (cos(/2) sin(/2)) = (1 sin )
2 2

(b) Now suppose the input is a pure coherent superposition of these two states,
1
|x i = (|z i + |z i)
2
Find and sketch the probability for detector DA to fire as a function of .

(c) Now suppose the input state is the completely mixed state
1
= (|z i hz | + |z i hz |)
2
Find and sketch the probability for detector DA to fire as a function of .
Comment on the result.

15.6.5 Mixed States vs. Pure States and Interference

A spin-interferometer is shown below.

1259
 Figure 15.44: Spin-Interferometer Setup

Spin1/2 electrons prepared in a given state (pure or mixed) are separated

into two paths by a Stern-Gerlach apparatus( gradient field along z). In one
path, the particle passes through a solenoid, with a uniform magnetic field along
the xaxis. The two paths are then recombined, sent through another Stern-
Gerlach apparatus with field gradient along x, and the particles are counted in
detectors in the two emerging ports.

The strength of the magnetic field is chosen so that t = , for some phase ,
where = 2B B/~ is the Larmor frequency and t is the time spent inside the
solenoid.
(a) Derive the probability of electrons arriving at detector DA as a function
of for the following pure state inputs:
(i) |z i , (ii) |z i , (iii) |x i , (iv) |x i
Comment on your results.
(b) Remember that for a mixed state we have
X
= pi |i i hi |
i

where pi is the probability of |i i.

This is a statistical mixture of the states {|i i}, not a coherent superpo-
sition of states. We should think of it classically, i.e., we have one of the
set {|i i}, we just do not know which one.

Prove that X 2
PDA = T r [|x i hx | ] = pi |hx | i i|
i
2
where |hx | i i| = the probability of detector B firing for the given input
state (we figured these out in part (a)). Repeat part (a) for the following
mixed state inputs:
1 1 1 1
(i) = |z i hz | + |z i hz | , (ii) = |x i hx | + |x i hx | ,
2 2 2 2
1 2
(iii) = |z i hz | + |z i hz |
3 3

1260
15.6.6 Which-path information, Entanglement, and Deco-
herence
If we can determine which path a particle takes in an interferometer, then we
do not observe quantum interference fringes. But how does this arise?

Consider the interferometer shown below:

Figure 15.45: Spin-Interferometer Setup

Into one arm of the interferometer we place a which-way detector in the form
of another spin1/2 particle prepared in the state |z iW . If the electron which
travels through the interferometer, and is ultimately detected (denoted by sub-
script D), interacts with the which-way detector, the which-way electron flips
the spin |z iW |z iW , i.e, the which-way detector works such that
If |iD = |z iD nothing happens to |z iW
If |iD = |z iD , then |z iW |z iW (a spin flip)
Thus, as a composite system

|z iD |z iW |z iD |z iW |z iD |z iW |z iD |z iW (15.339)

(a) The electron D is initially prepared in the state |x iD = (|z iD + |z iD ) / 2.
Show that before detection, the two electrons D and W are in an entangled
state
1
|DW i = (|z iD |z iW + |z iD |z iW )
2
(b) Only the electron D is detected. Show that its marginal state, ignoring
the electron W, is the completely mixed state,
1 1
D = |z iD D hz | + |z iD D hz |
2 2
This can be done by calculating
X 2
Prob(mD ) = |hmD , mW | DW i|
mW

for some observable.

1261
 This state shows no interference between |z iD and |z iD . Thus, the
emphwhich-way detector removes the coherence between states that ex-
isted in the input.
(c) Suppose now the which-way detector does not function perfectly and does
not completely flip the spin, but rotates it by an angle about x so that
| iW = cos (/2) |z iW + sin (/2) |z iW
Show that in this case the marginal state is
1 1
D = |z iD D hz | + |z iD D hz |
2 2
+ cos (/2) |z iD D hz | + sin (/2) |z iD D hz |
Comment on the limits 0 and .

15.6.7 Livio and Oivil

Two scientists (they happen to be twins, named Oivil and Livio, but never
mind .....) decide to do the following experiment. They set up a light source,
which emits two photons at a time, back-to-back in the laboratory frame. The
ensemble is given by
1
= (|LLi hLL| + |RRi hRR|)
2
where L refers to left-handed polarization and R refers to right-handed polar-
ization. Thus, |LRi would refer to the state in which photon number 1 (defined
as the photon which is aimed at Oivil, say) is left-handed and photon number
2 (the photon aimed at scientist Livio) is right-handed.

These scientists(one of whom has a diabolical bent) decide to play a game with
Nature: Oivil (of course) stays in the lab, while Livio treks to a point a light-
year away. The light source is turned on and emits two photons, one directed
toward each scientist. Oivil soon measures the polarization of his photon; it is
left-handed. He quickly makes a note that his sister is going to see a left-handed
photon, about a year from that time.

The year has passed and finally Livio sees her photon, and measures its po-
larization. She sends a message back to her brother Oivil, who learns in yet
another year what he know all along; Livios photon was left-handed.

Oivil then has a sneaky idea. He secretly changes the apparatus, without telling
his forlorn sister. Now the ensemble is
1
= (|LLi + |RRi) (hLL| + hRR|)
2
He causes another pair of photons to be emitted with this new apparatus and
repeats the experiment. The result is identical to the first experiment.

1262
 (a) Was Oivil lucky, or will he get the right answer every time, for each ap-
paratus? Demonstrate the answer explicitly using the density matrix for-
malism.

(b) What is the probability that Livio will observe a left-handed photon, or a
right-handed photon, for each apparatus? Is there a problem with causal-
ity here? How can Oivil know what Livio is going to see, long before
she sees it? Discuss what is happening here. Feel free to modify the
experiment to illustrate any points you wish to make.

15.6.8 Measurements on Qubits

You are given one of two quantum states of a single qubit(physical system
representing a 2-valued state): either |i = |0i or |i = cos |0i + sin |1i.
You want to make a single measurement that best distinguishes between these
two states, i.e., you want to find the best basis for making a measurement to
distinguish the  states. So let us measure the qubit in the basis {|i =
 two
|0i + |1i ,  }, where , are to be determined for optimal success. For
outcome |i we guess that the qubit was in state |0i; for outcome  we guess


that the qubit was in state |i. Determine the optimal measurement basis
given this procedure. You can take and to be real numbers, in which case
2 2
the normalization || + || = 1 implies that you can write and as, e.g.
= sin and = cos . HINT: you will need to first construct the probability
of a correct measurement in this situation. You should convince yourselves that
this is given by

P r[qubit was |0i]P r[|i |qubit was |0i]+P r[qubit was |i]P r[ |qubit was |i]
 

where, e.g.
P r[|i |qubit was |0i] = | h | 0i |2
If the state you are presented with is either |i or |i with 50% probability each,
what is the probability that your measurement correctly identifies the state?

15.6.9 To be entangled....
Let HA = span{|0A i , |1A i} and HB = span{|0B i , |1B i} be two-dimensional
Hilbert spaces and let |AB i be a factorizable state in the joint space HA HB .
Specify necessary and sufficient conditions on |AB i such that UAB |AB i is an
entangled state where

UAB = |0A i h0A | |0B i h0B | |1A i h1A | |1B i h1B |

15.6.10 Alice, Bob and Charlie

Let Alice, Bob and Charlie be in possession of quantum systems whose states live
in HA = span{|0A i , |1A i}, HB = span{|0B i , |1B i} and HC = span{|0C i , |1C i},

1263
respectively. Suppose that a joint state of these systems has been prepared as
the (three-way) entangled state
1
|ABC i = (|0A 0B 0C i + |1A 1B 1C i)
2
(a) What is the reduced density operator on HA HB if we take a partial
trace over HC ?
(b) Suppose Charlie performs a measurement specified by the partial projec-
tors 1A 1B |0C i h0C | and 1A 1B |1C i h1C |. Compute the probabilities
of the possible outcomes, as well as the corresponding post-measurement
states. Show that this ensemble is consistent with your answer to part (a)
(c) Suppose Charlie performs a measurement specified by the partial projec-
tors 1A 1B |xC i hxC | and 1A 1B |yC i hyC |, where
1 1
|xC i = (|0C i + |1C i) , |yC i = (|0C i |1C i)
2 2
Again, compute the probabilities of the possible outcomes and the corre-
sponding post-measurement states and show that this ensemble is consis-
tent with your answer from part (a).
(d) Suppose Alice and Bob know that Charlie has performed one of the two
measurements from parts (b) and (c), but they do not know which (as-
sume equal probabilities) measurement he performed nor do they know
the outcome. Write down the quantum ensemble that you think Alice
and Bob should use to describe the post-measurement state on HA HB .
Is this consistent with the reduced density operator from part (a)? How
should Alice and Bob change their description of the post-measurement
state if Charlie subsequently tells them which measurement he performed
and what the outcome was?

1264
Chapter 16

The EPR Argument and Bell Inequality

16.1 Hidden variables and Bells Inequalities-

1st Try
As we have seen, when a quantum system possesses more than one degree of
freedom, the associated Hilbert space is a tensor product of the spaces associ-
ated with each degree of freedom. This structure leads to specific properties
of quantum mechanics, whose paradoxical nature has been discussed in earlier
chapters. In this section we will study an example of such a situation by con-
sidering entangled states for the spins of two particles.

The system under consideration is a hydrogen atom which is dissociated into

an electron and a proton. We consider the spin states of these two particles
when they have left the dissociation region and are located in spatially distinct
regions, e.g. a few meters from one another. They are then considered to be
free particles whose spin states do not evolve.

16.1.1 The Electron Spin

Consider a unit vector u n the (z, x) plane given by u = cos uz + sin ux
where ux and uz are unit vectors in the x and z directions. We note that
Se = Se u is the component of the electron spin in the u direction.

In the eigenbasis |e : i of Sez , the matrix representing Se is

   
h+| Se |+i h+| Se |i ~ cos sin
= (16.1)
h| Se |+i h| Se |i 2 sin cos

which has eigenvalues ~/2 (true for any direction).

1265
The corresponding eigenvectors are

|e : +i = cos |e : +i + sin |e : i (16.2)
2 2

|e : i = sin |e : +i + cos |e : i (16.3)
2 2
If the electron is emitted in the state |e : +i, the probability P+ () of finding
the electron in the state |e : +i is given by
2
P+ () = |he : + | e : +i|
  2
 
=  cos he : +| + sin he : |
 cos |e : +i + sin |e : i 
2 2 2 2
 2

= cos2

=  cos cos + sin sin (16.4)
2 2 2 2  2

and similarly,
2
P () = |he : | e : +i|
  2
 
=  sin he : +| + cos he : | cos |e : +i + sin |e : i 
2 2 2 2
 2

= sin2

=  sin cos + cos sin (16.5)
2 2 2 2  2

Using these results, the expectation value of Se in the |e : +i state is then

D E    
~ ~
Se = + P+ () + P ()
2 2
 
~ 2 2 ~
= cos sin = cos( ) (16.6)
2 2 2 2

16.1.2 Correlations Between the Two Spins

We assume that after the dissociation, the electron-proton system is in the
factorized spin state |e : +i |p : i. Now if |u1 i E and |u2 i E and
|v1 i F and |v2 i F , then |ui|vi G = EF and if A and B act respectively
in E and F , then C = A B acts in G. Then one has hu2 | hv2 | C |u1 i |v1 i =
hu2 | A |u1 i hv2 | B |v1 i.

We now determine the probability P+ () of finding +~/2 when measuring the

component Se of the electron spin in this state.

The projector on the eigenstate |e : +i, corresponding to the measured value,

is |e : +i he : +| Ip , where Ip is the identity operator on the proton states.

1266
Therefore,

P+ () = hp : | he : +| (|e : +i he : +| Ip ) |e : +i |p : i
= hp : | Ip |p : i he : + | e : +i he : + | e : +i
2
= |he : + | e : +i| = cos2 (16.7)
2
and the state after the measurement is |e : +i |p : i. The proton spin is
not affected, because the initial state is factorized (and all probability laws are
factorized).

For u = cos uz + sin ux and u = cos uz + sin ux we can calculate the

expectation values (as earlier)
D E ~ D E ~
Se = cos( ) , Sp = cos( ) (16.8)
2 2
The correlation coefficient between the two spins E(, ) is defined by
D E D ED E
Se Sp Se Sp
E(, ) = D ED E1/2 (16.9)
2
Se 2
Sp

Now
2 ~2 2 ~2
Se = Ie , Sp = Ip (16.10)
4 4
and
D E
Se Sp = he : +| Se |e : +i hp : +| Sp |p : +i
~2
= cos( ) cos( ) (16.11)
4
Thus,
2
~2
~4 cos( ) cos( ) + 4 cos( ) cos( )
E(, ) = ~2
=0
4

This just reflects the fact that in a factorized state, the two spin variables are
independent.

Correlations in the Singlet State

Now assume that, after the dissociation, the two particles are in the singlet spin
state
1
|s i = (|e : +i |p : i |e : i |p : +i) (16.12)
2

1267
If we measure the component Se of the electron spin along the direction u =
cos uz + sin ux , we find the following results and corresponding probabilities:
there are two possible values
~
+ projector |e : +i he : +| Ip
2
~
projector |e : i he : | Ip
2
with probabilities
  1
2 2
P+ () = |he : + | e : +i| + |he : + | e : i| =
2
  1
2 2
P () = |he : | e : +i| + |he : | e : i| =
2
This result is a consequence of the rotational invariance of the singlet state.

Now suppose the result of this measurement is +~/2 and then later on, one
measures the component Sp of the proton spin along the direction u =
cos uz + sin ux .

Since the electron spin is measured to be +~/2, the state after that measurement
is

he : + | s i |e : +i
1
= (he : + | e : +i |e : +i |p : i he : + | e : i |e : +i |p : +i)
2

cos |e : +i |p : i sin |e : +i |p : +i (16.13)
2 2
The probabilities for the two possible results of measurement of the proton spin,
~/2, are

P+ () = sin2 , P () = cos2 (16.14)
2 2
What would have happened if we had measured the proton spin first?

If the proton had been measured first then we would have

1 1
P+ () = , P () = (16.15)
2 2
as we found for the electron when the electron was measured first.

The fact that the measurement on the electron affects the probabilities of the
results of a measurement on the proton, although the two particles are spatially
separated, is in contradiction to Einsteins assertion or belief that the real states
of two spatially separated objects must be independent of one another. This is
the starting point of the EPR paradox. Quantum mechanics is not a local theory

1268
as far as measurements are concerned.

Note, however, that this non-locality does not allow the instantaneous trans-
mission of information, From a measurement of the proton spin, one can not
determine whether the electron spin has been previously measured. It is only
when, for a series of experiments, the results of the measurements on the electron
and the proton are later compared, that one can find this non-local character
of quantum mechanics.
D E D E
We now recalculate the expectations values Se and Sp in the singlet state.
D E D E
We get (using the same process as above) Se = 0 = Sp . This is so be-
cause one does not worry about the other variable.

Finally, we can calculate the correlation coefficient in the singlet state. We have,
since the spins are correlated now, that
D E ~2 

~2
Se Sp = sin2 cos2 = cos( ) (16.16)
4 2 2 4
and therefore
2
~4 cos( ) + 0
E(, ) = ~2
= cos( ) (16.17)
4

in the singlet state.

16.1.3 A Simple Hidden Variable Model

For Einstein and several other physicists, the solution to the paradox uncovered
above comes from the fact that the states of quantum mechanics, in particu-
lar the singlet state above, provide and incomplete description of reality. A
complete theory (for predicting spin measurements, in the present case) should
incorporate additional variables or parameters, whose knowledge would ren-
der measurements independent for two spatially separated objects. However,
present experiments cannot determine the values of these parameters, which
are therefore called hidden variables. The experimental result should then con-
sist of some averaging over these unknown parameters.

In the case of interest, a very simplified example of such a theory is the follow-
ing. We assume that, after each dissociation, the system is in a factorized state
|e : +i |p : i, but that the direction varies from one event to another. In
this case is the hidden variable. We assume that all directions of are equally
probable, i.e., the probability density that the decay occurs with direction is
uniform and equal to 1/2.

Since we are ignorant of the value of , the expectation value of an observable

1269
A is now defined to be

Z2
D E 1
A = he : +| hp : | A |e : +i |p : id (16.18)
2
0

Let us now use this new definition of the expectation value to investigate the
correlation coefficient. We have from earlier
D E D ED E
Se Sp Se Sp
E(, ) = D ED E1/2
2
Se 2
Sp

Using our earlier results we have

D E ~2 Z d
Se = cos( ) =0 (16.19)
4 2
D E
and similarly Sp = 0. We also have

~2 ~2
Z
D E d
Se Sp = cos( ) cos( ) = cos( ) (16.20)
4 2 8
Therefore, in this simple hidden variable model
1
E(, ) = cos( ) (16.21)
2
In such a model, one finds a non-vanishing correlation coefficient, which is an
interesting observation. Even more interesting is that the correlation is smaller
than the prediction of quantum mechanics by a factor of 2.

The first precise experimental tests of hidden variable descriptions versus quan-
tum mechanics have been performed on correlated pairs of photons emitted in
an atomic cascade. Although, we are not dealing with spin1/2 particles in this
case (see discussion later in this chapter), the physical content is basically the
same as in this case. As an example Figure 16.1 below presents the experimental
results of Aspect, et al,

Figure 16.1: Data from Aspect, etal

1270
It gives the variation of E(, ) as a function of the difference , which is
found experimentally to be the only relevant quantity, i.e., the results do not
depend in any way on or separately! The circles indicate the size of exper-
imental errors.

The experimental points agree with the predictions of quantum mechanics and
clearly disagree, therefore, with the predictions of this particular hidden vari-
ables theory.

16.2 Bells Theorem and Experimental Results

As proved by Bell in 1965, the disagreement between the predictions of quantum
mechanics and those of hidden variable theories is actually very general when
one considers correlation measurements on entangled states.

We can, however, show that the correlation results for hidden variable theories
are constrained by what is known as Bells inequality, which is violated by quan-
tum mechanics.

Consider a hidden variable theory, whose results consists of two functions A(, u )
and B(, u ) giving respectively the results of the electron and proton spin mea-
surements. Each of these two functions takes only two values +~/2 and ~/2.
It depends on the value of the hidden variable for the considered electron-
proton pair. The nature of the hidden variable need not be further specified
for this discussion. The result A of course depends on the axis u chosen for
the measurement of the electron spin, but it does not depend on the axis u .
Similarly B does not depend on u . This locality hypothesis is essential for the
following discussion.

Note that we assume here that the hidden variable theory reproduces the one
operator averages found for the singlet state:
D E Z
Se = P ()A(, u )d = 0 (16.22)
D E Z
Sp = P ()B(, u )d = 0 (16.23)

If this was not the case, such a hidden variable theory should clearly be rejected
since it would not reproduce a well-established experimental result.

Let us now consider the quantity

A(, u )B(, u ) + A(, u )B(, u0 ) + A(, u0 )B(, u0 ) A(, u0 )B(, u )
(16.24)
for any set u , u , u0 , u0 . We can rewrite this as

A(, u )(B(, u ) + B(, u0 )) + A(, u0 )(B(, u0 ) B(, u )) (16.25)

1271
Now the two quantities B(, u ) and B(, u0 ) can take on only two values ~/2.
Therefore, one has either

B(, u ) + B(, u0 ) = ~ , B(, u0 ) B(, u ) = 0 (16.26)

or
B(, u ) + B(, u0 ) = 0 , B(, u0 ) B(, u ) = ~ (16.27)
Therefore, since |A(, u )| = |A(, u0 )| = ~/2, we have the result

A(, u )(B(, u )+B(, u0 ))+A(, u0 )(B(, u0 )B(, u )) = ~2 /2 (16.28)

We then define the quantity S as

S = E(, ) + E(, 0 ) + E(0 , 0 ) E(0 , ) (16.29)

and we get

A(, u )B(, u ) + A(, u )B(, u0 )

Z  
P ()d
+A(, u0 )B(, u0 ) A(, u0 )B(, u )
~2
= [E(, ) + E(, 0 ) + E(0 , 0 ) E(0 , )]
4
~2 ~2
Z
= P ()d =
2 2

or |S| 2, which is Bells inequality.

Now let us consider a special case = 0 = 0 0 = /4. The quantum

mechanical result for S is

SQ = cos( ) cos( 0 ) cos(0 0 ) + cos(0 ) (16.30)

If we set 1 = , 2 = 0 , 3 = 0 0 , we can look for the extrema of

f (1 , 2 , 3 ) = cos(1 + 2 + 3 ) cos 1 cos 2 cos 3 (16.31)

The extrema correspond to 1 = 2 = 3 and sin 1 = sin 31 whose solutions

between 0 and are 1 = 0, /4, 3/4, .

Defining the function g(1 ) = 3 cos 1 + cos 31 we have

g(0) = 2 , g(/4) = 2 2 , g(3/4) = 2 2 , g() = 2 (16.32)

The plot in Figure 16.2 below shows g():

1272
 Figure 16.2: Bell Function

The shaded areas correspond to results which cannot be explained by hidden

variable theories. This system therefore constitutes a test of the predictions of
quantum mechanics versus any local hidden variable theory.

16.3 The EPR(Einstein-Podolsky-Rosen) Argument-

Quick Overview
The nondeterministic character of quantum mechanics is very disturbing for
the classical physicist. Hence, there were repeated attempts to replace quan-
tum theory by a statistical theory.

According to these theories, there exist hidden variables whose values prescribe
the values of all observables for any particular object, except that the hidden
variables are unknown to the experimenter, thus yielding the probabilistic char-
acter of the theory. The probabilistic character of quantum mechanics would
then be quite analogous to that of classical statistical mechanics, where one can
imagine that the motion of all the particles is, in principle, known.

For example, let us consider a particle of spin = 1/2 in an eigenstate of Sx

with eigenvalue ~/2. According to quantum mechanics, the zcomponent is
not fixed. If one measures it for a very large number of such particles, one finds
~/2 50% of the time and ~/2 50% of the time. According to the idea of hidden
variables, for each particle, parameters unknown to us would determine whether
~/2 or ~/2 results. These hidden variables would prescribe ~/2 each 50% of
the time.

By means of a number of thought experiments, Einstein attempted to demon-

strate the incompleteness of the quantum mechanical description and to get
around the indeterminism and the uncertainty relation. Each of these argu-
ments was refuted, in turn, by Bohr.

1273
An argument - sometimes referred to as a paradox - due to Einstein, Podolsky
and Rosen (EPR) , played a pivotal role in the discussion of indeterminism and
the existence of hidden variables; we consider the argument as reformulated by
David Bohm.

Let two spin = 1/2 particles in the singlet state

1
|0, 0i = (|i |i |i |i) (16.33)
2
be emitted from a source an then move apart in space. Even if the two particles
are separated by an arbitrarily large distance and can no longer communicate
with one another, one finds the following correlations in this state during a
measurement of the one particle spin states:

If one measures the zcomponent of the spin and finds

particle 1 spin up, particle 2 is spin down. If one
finds particle 1 spin down, particle 2 has spin up.

If, instead, one measures Sx , then +~/2 for

particle 1 implies ~/2 for particle 2, etc.

This expresses the nonlocality of quantum theory.

The experiment on particle 1 influences the result of the experiment on particle

2, although they are widely separated. The nonlocality is a consequence of the
existence of correlated many-particle states such as the direct product

|i |i (16.34)

and the fact that one can linear superimpose such states.

The nonlocality of quantum mechanics does not lead to contradictions with

relativity theory. Although a measurement of a spin component of particle 1
immediately reveals the value of that component for particle 2, no information
can be transmitted in this way. Since particle 1 takes values ~/2 each 50% of
the time, this remains true for particle 2, even after the measurement of particle
1. Only by a subsequent(slow) comparison of the results is it possible to verify
the correlation.

Einstein, Podolsky and Rosen gave the following argument in favor of hidden
parameters in conjunction with the EPR thought experiment.

By the measurement of Sz or Sx of particle 1, the values of Sz or Sx of particle 2

are known. Because of the separation of the particles, there was no influence on
particle 2, and therefore the values of Sz , Sx etc, must have been fixed before the
experiment. Thus, there must be a more complete theory with hidden variables.

1274
In the EPR argument, the predictions of the quantum states

1
|0, 0i = (|i |i |i |i)
2

are used, but the inherent nonlocality is denied.

In the remainder of our discussion, we will consider local hidden variables. These
would predetermine which value each of the components of S ~ of particle 1 has
and likewise for particle 2. Each of the particles would carry this information
independently of the other.

16.3.1 The Bell Inequality again

We now show again that such local hidden variables lead to predictions differ-
ent from those of quantum mechanics. We then compare the predictions with
experiment.

We consider a correlation experiment in which a particle of total spin = 0 decays

into two particles each with spin = 1/2. At a sufficiently large distance from the
source, a rotatable polarizer and a detector are set up for each particle as shown
below, so that the particles can be detected and we can investigate whether any
correlation in the spin orientations exists. The setup is shown in Figure 16.3
below.

Figure 16.3: Bell-EPR Experiment

Polarizer 1 with angular setting only lets particle 1 through if its spin in the
direction n has the value +~/2 and polarizer 2 with angular setting only
lets particle 2 through if its spin in the direction n has the value +~/2. The
particles are counted by detectors 1 and 2. If they respond, then the spin is
positive, otherwise it is negative.

We consider the correlation between various angular settings of the polarization

experiment.

A measure of the correlation is N (; ), defined as the relative number of exper-

iments resulting in particle 1 at angle being positive and particle 2 at angle
being positive.

1275
Using the spin projection operator
1
P = (1 + ~ n ) (16.35)
2
quantum mechanics gives
1 1
N (; ) = h0, 0| (1 + ~1 n ) (1 + ~2 n ) |0, 0i
2 2
1 1
= h0, 0| (1 + ~1 n ) (1 ~1 n ) |0, 0i
2 2
1
= (1 n n ) (16.36)
4
since h0, 0| ~1 |0, 0i = 0 in the singlet state. For coplanar detectors this reduces
to
1
N (; ) = sin2 (16.37)
2 2
If hidden variables were really present, we could represent N (; ) by the fol-
lowing sum
N (; ) = N (; ) + N (; ) (16.38)
Here, N (; ) is the relative number of particle pairs in which particle 1 has
positive spin at angles and and negative spin at , while N (; ) is the
relative number of particle pairs in which particle 1 has negative spin at |gamma
instead. In theories with hidden variables, all of these quantities are assumed
to be known.

Now one has N (; ) N (; ) since N (; ) = N (; )+N (; ) and both

quantities on the right-hand side of the equation are nonnegative. Similarly,
N (; ) N (; ). Thus,
N (; ) N (; ) + N (; ) (16.39)
This is a simple version of the Bell inequality.

Remarks
1. In experiments one often works with the correlation defined by
P (; ) = h0, 0| (~1 n ) (~2 n ) |0, 0i = 4N (; ) 1 (16.40)
instead of N (; ) itself. Using
1
N (; ) = sin2 (16.41)
2 2
we get
P ( ) P (; ) = cos( ) (16.42)
and Bells inequality becomes
P (; ) 1 P (; ) + P (; ) (16.43)

1276
 2. The limit prescribed by the Bell inequality can be determined as follows.
In
N (; ) N (; ) + N (; ) (16.44)
we substitute for , , the values 0, , /2 respectively to obtain

N (0; ) N (0; /2) + N (/2; ) (16.45)

In the singlet state,

1
N (0; ) = , N (0; /2) = N (/2; ) (16.46)
2
so that
1
N (0; /2) (16.47)
4
Other values can be obtained by different combinations of angles.

Finally, we contrast the consequences of the Bell inequality with quantum me-
chanics and compare with experiments.

To this end, we compute N (; ), N (; ) and N (; ) for the three angles

= 0 , = 45 and = 90 using
1
N (; ) = sin2 (16.48)
2 2
to get
1
sin2 45 2 sin2 22.5 (16.49)
2
or
0.5 0.29 (16.50)
which is clearly not true! Therefore, quantum mechanics and hidden variables
are incompatible.

The comparison of quantum mechanics and the Bell inequality is shown in Fig-
ure 16.4 below, which gives the correlation P () = P (; 0) according to quantum
mechanics and the Bell inequality.

1277
 Figure 16.4: Bell-EPR Experiment Theoretical Predictions

The experimental demonstration of the violation of the Bell inequality by Lamehl-

Rachti for protons and Aspect for photons is shown in Figure 16.5 below which
gives the experimental results on the spin correlation of proton pairs

Figure 16.5: Bell-EPR Experiment - Data

Clearly, quantum mechanics is correct. This means that any theory that has
the same probabilistic predictions as quantum mechanics must be nonlocal.

16.4 EPR and Bell - The Details

Let us first rethink some quantum mechanical ideas in a context needed for this
discussion. This review will hopefully reinforce the ideas you have learned so
far.

1278
16.4.1 Single-Photon Interference
All good discussions on quantum mechanics present a long an interesting anal-
ysis of the double slit experiment. The crux of the discussion comes when the
light intensity is reduced sufficiently for photons to be considered as presenting
themselves at the entry slit one by one. For a long time this point was very con-
tentious, because correlations between two successive photons cannot be ruled
out a priori.

Since 1985, however, the situation has changed. An experiment was performed
by Grangier, Roger and Aspect. It was an interference experiment with only a
single photon. They used a light source devised for an EPR experiment which
guarantees that photons arrive at the entry slit singly.

The experiment is difficult to do in practice, but is very simple in principle and

it provides an excellent experimental introduction to the concepts of quantum
mechanics.

The light source is a beam of calcium atoms, excited by two focused laser beams
having wavelengths 0 = 406 nm and 00 = 581 nm respectively. Two-photon
excitation produces a state having the quantum number J = 0. When it
decays, this state emits two monochromatic photons having the wavelengths
1 = 551.3 nm and 2 = 422.7 nm respectively, in a cascade of two electronic
transitions from the initial J = 0 level to the final J = 0 state, passing through
an intermediate J = 1 state, as shown in Figure 16.6 below

Figure 16.6: Calcium based light source - Energy Levels

The mean lifetime of the intermediate state is 4.7 ns. To simplify the terminol-
ogy, we shall call the 1 = 551.3 nm light green, and the 2 = 422.7 nm light
violet.

1279
Next we describe the experiment, exhibiting its three stages which reveal the
complications of the apparatus in progressively greater detail (Figures 16.7-16.9
below).

1. The first stage is a trivial check that the apparatus is working properly;
nevertheless it is already very instructive (Figure 16.7 below).

Figure 16.7: Single Photon Interference Experiment - Stage 1

Figure 16.7 shows interference with a single photon (first stage). In the
sketch, solid lines are optical paths and dashed lines are electrical connec-
tions.

On either side of the source S one positions two photomultiplier tubes

P MO and P MA . These are very sensitive, and can detect the arrival
of a single photon. Detection proceeds through photoelectric absorption,
followed by amplification which produces an electric signal proportional
to the energy of the incident photon. The associated electronic logic cir-
cuits can identify the photons absorbed by each detector: the channel
P MO responds only to green light, and the channel P MA responds only
to violet light. The electronic gate is opened (for 9 ns - this is twice the
mean lifetime and corresponds to an 85% probability that the photon has
been emitted) when green light is detected by P MO . If, while the gate is
open, violet light is emitted by the same atom towards(not all of the vio-
let photons go towards the source) P MA , then P MA detects this photon,
producing a signal that passes through the gate and is counted in NA .
The counter NO registers the number of green photons detected by P MO .
It turns out that NA  NO . As the observation period becomes very
long(approximately 5 hours), the ratio NA /NO tends to a limit that is
characteristic of the apparatus. It represents the probability of detecting
a violet photon in P MA during the 9 ns following the detection of a green
photon by P MO .

1280
 The purpose of this arrangement is to use a green photon in order to open
a 9 ns time window, in which to detect a violet photon emitted by the
same atom. As we shall see, there is only an extremely small probability
of detecting through the same window another violet photon emitted by
a different atom.

We will assume that a second observer is in the lab. This observer always
feels compelled to present what he thinks are simple-minded truths using
ordinary words. We will called this second observer Albert. Albert, as we
shall see, has a tendency to use, one after another, the three phrases, I ob-
serve, I conclude, and I envisage. Consulted about the above experiment,
Albert states, with much confidence,

I observe that the photomultiplier P MA detects violet light when the source
S is on, and that it ceases to detect anything when the source is off. I con-
clude that the violet light is emitted by S, and that it travelled from S to
P MA .

I observe that energy is transferred between the light and the photomulti-
plier P MA always in the same amount, which I will call a quantum.

I envisage the quanta as particles, emitted by the source, propagating freely

from S to P MA , and absorbed by the detector. I shall call this quanta pho-
tons.

Albert stops talking at this point.

2. The second stage of the experiment introduces the concept of individual

photons in Figure 16.8 below which is interference with a single photon
(second stage).

Figure 16.8: Single Photon Interference Experiment - Stage 2

1281
Across the path of the violet light one places a half-silvered mirror LS ,
which splits the primary beam into two secondary beams(equal intensity),
one transmitted and detected by P MA , the other reflected and detected
by P MB . As in the first stage, the gate is opened for 9 ns, by P MO . While
it is open, one registers detection by either P MA (counted as NA ) or by
P MB (counted as NB ) or by both, which we call a coincidence (counted
as NC ). The experiment runs for 5 hours again and yields the following
results:
(a) The counts NA and NB are both of the order of 105 . By contrast,
NC is much smaller, being equal to 9.
(b) The sequence of counts from P MA is random in time, as is the se-
quence of counts from P MB .
(c) The very low value of NC shows that counts in P MA and P MB are
mutually exclusive (do not occur at same time).
The experimenters analyze the value of NC in depth; their reasoning can
be outlined as follows:
(a) Suppose two different atoms each emit a violet photon, one being
transmitted to P MA and the other reflected to P MB with both ar-
riving during the 9 ns opening of the gate; then the circuitry records
a coincidence. In the regime under study, and for a run of 5 hours,
quantum theory predicts that the number of coincidences should be
NC = 9. The fact that this number is so small means that, in prac-
tice, any given single photon is either transmitted or reflected.
(b) If light is considered as a wave, split into two by LS and condensed
into quanta on reaching P MA and P MB , then one would expect the
photon counts to be correlated in time, which would entail NC  9.
Classically speaking this would mean that we cannot have a trans-
mitted wave without a reflected wave.
(c) Experiment yields NC = 9; this quantum result differs from the
classical value by 13 standard deviations; hence the discrepancy is
very firmly established, and allows us to assert that we are indeed
dealing with a source of individual photons.
Albert leaves such logical thinking to professionals. Once he notes that
NC is very small, he is quite prepared to treat it as if it were zero. He
therefore says I observe that light travels from the source to P MA or to
P MB , because detection ceases when the source is switched off.

I observe the counts NA and NB correspond to a game of heads or tails,

in that the two possibilities are mutually exclusive, and that the counts are
random.

I observe that the optical paths 1 and 2 are distinguishable, because the

1282
experiment allows me to ascertain, for each quantum, whether it has trav-
elled path 1 (detection by P MA ) or path 2 (detection by P MB ).

I envisage that, on arrival at the half-silvered mirror, each photon from

the source is directed at random either along path 1 or along path 2; and
I assert that it is the nature of photons to play heads or tails..

Digression: The Mach-Zender Interferometer and Quantum In-

terference

(a) The next experiment uses a Mach-Zender interferometer so let us see

how it operates.

Background information: Consider a single photon incident on

a 50-50 beam splitter (that is, a partially transmitting, partially re-
flecting mirror, with equal coefficients). Whereas classical electro-
magnetic energy divides equally, the photon is indivisible. That is,
if a photon-counting detector is placed at each of the output ports
(see Figure 16.9 below), only one of them clicks. Which one clicks is
completely random (that is, we have no better guess for one over the
other).

Figure 16.9: 50-50 Beam Splitter

The input-output transformation of the waves incident on 50-50 beam

splitters and perfectly reflecting mirrors are shown in Figure 16.10
below.

Figure 16.10: Input/output Transformations

We can easily show that with these rules, there is a 50-50 chance
of either of the detectors shown in the first figure above to click.

1283
 According to the rules given in the figure
1 1
1,out = in , 2,out = in (16.51)
2 2
since nothing enters port #2.

By our probability postulate the probability to find a photon at po-

sition 1 or 2 is
2 2
P1,out = |1,out | dx = 12 |in | dx = 21
R R 
2 2 50 50% chance
P2,out = |2,out | dx = 12 |in | dx = 21
R R

Note: As we see from the experimental discussion below, the photon

is found at one detector or the other, never both. The photon is
indivisible. This contrasts with classical waves where half of the
intensity goes one way and half the other; an antenna would also
receive energy. We interpret this as the mean value of a large number
of photons.
(b) Now we set up a Mach-Zender interferometer(shown in Figure 16.11
below):

Figure 16.11: Mach-Zehnder Interferometer

?The wave is split at beam-splitter b1, where it travels either path

b1-m1-b2(call it the green path) or the path b1-m2-b2 (call it the blue
path). Mirrors are then used to recombine the beams on a second
beam splitter, b2. Detectors D1 and D2 are placed at the two output
ports of b2.

Assuming the paths are perfectly balanced (that is equal length), we

can show that the probability for detector D1 to click is 100% - no
randomness!

To find the wavefunctions impinging on detectors D1 and D2 let us

apply the transformation rules sequentially.

1284
 (1) Beamsplitter #1

1 1
1,out = in , 2,out = in
2 2

(2) Propagation a distance L/2 along each path mean that the phase
of the wavefunction changes by eikL/2 so that the wavefunctions
are
1 1
1,atmirror = eikL/2 in , 2,atmirror = eikL/2 in
2 2

(3) Reflection off mirrors means wavefunctions become

1
1,af termirror = eikL/2 in (16.52)
2
1 ikL/2
2,af termirror = e in
2

(4) Propagation a distance L/2 along each path mean that the phase
of the wavefunction changes by eikL/2 so that the wavefunctions
are
1 1
1,atb2 = eikL in , 2,atb2 = eikL in
2 2

(5) After beamsplitter #2

1,atb2 + 2,atb2
out,1 = = eikL in
2
1,atb2 2,atb2
out,2 = =0
2

Therefore,
Z Z
2 2
P1,out = |out,1 | dx = |in | dx = 1
Z
2
P2,out = |out,2 | dx = 0

Thus, we have a 100% chance of detector D1 firing and a 0% chance

of detector D2 firing. There is no randomness.
(c) Classical logical reasoning would predict a probability for D1 to click
given by

PD1 = P (transmission at b1|green path)P (green path)

+ P (ref lection at b2|blue path)P (blue path)

1285
Now we know that there is a 50-50 probability for the photon to take
the blue or green path which implies that

P (green) = P (blue) = 1/2

Also with the particle incident at b2 along the green path there is
a 50% chance of transmission and similarly for reflection of the blue
path.

Therefore,

P (transmission at b2|green) = P (ref lection at b2|blue) = 1/2

and
1 1 1 1 1
PD1 = + =
2 2 2 2 2
so that classical reasoning implies a 50-50 chance of D1 firing, that
is, it is completely random!

The quantum case is different because the two paths which lead to
detector D1 interfere. For the two paths leading to D1 we have

1 eikL in + 1 eikL in
total = 2 2
2
Z
2 1 1 1 1 1 1 1 1
PD1 = |total | dx = PD1 = + + + = 1
2 2 2 2 2 2 2 2

where the last two terms are the so-called interference terms. Thus,
PD1 = 1. The paths that lead to detector D2 destructively interfere
so that PD2 = 0.

We now ask how would you set up the interferometer so that detector
D2 clicked with 100% probability? How about making them click at
random?

Leave the basic geometry the same, that is, do not change the direc-
tion of the beam splitters or the direction of the incident light.

We now want constructive interference for the paths leading to D2

and destructive interference for D1.

We can achieve this by changing the relative phase of the two paths
by moving the mirror so that the path lengths are not the same.

Suppose we move the mirror on the green path (at an angle of 45 ) so

that the path lengths in the green path are both changed to L + L.

1286
 We then have
1 eik(L+L) + 1 eikL
2 2
out,1 = = in
2
1 eik(L+L) 1 eikL
2 2
out,2 = = in
2
and
Z Z
2 1 2 2  2
PD1 = dx |2,out | = dx |in | eikL  eikL + 1
4
1 ikL
1 ikL
+ 1 eikL + 1 = + 1 eikL + 1




= e e
4 4  
1 ikL ikL 1 + cos(kL) 2 kL
= (2 + e +e )= = cos
4 2 2

Similarly we have
Z Z
2 1 2 2  2
PD2 = dx |1,out | = dx |in | eikL  eikL 1
4
1 ikL
1 ikL
1 eikL 1

ikL


= e 1 e 1 = e
4 4  
1 ikL ikL 1 cos(kL) 2 kL
= (2 e e )= = sin
4 2 2

Therefore, to achieve PD1 = 0 and PD2 = 1 we choose

kL = m (m odd) L = m =m
k 2
We can make the both random if
   
2 kL 2 kL 1 kL
cos = sin = =p (p odd)
2 2 2 2 4

L = p =p
2k 4
Returning to our discussion.....
3. The third stage then consists of an interference experiment as shown in
Figure 16.12 below, which is the interference with a single photon (third
stage).

1287
 Figure 16.12: Mach-Zehnder Interferometer Inserted

A so-called Mach-Zehnder interferometer is used, allowing one to obtain

two interference profiles. The beam of violet light from the source S
is split into two by the mirror LS . After reflection from two different
mirrors, these secondary beams meet on a second half-silvered mirror LS .
Here, each secondary beam is further split into two; thus one establishes
two interference regions, region (1,2) where one places P MA , and region
(1,2) where one places P MB .

A very high precision piezoelectric system allows one of the mirrors to

be displaced so as to vary the path difference between the two arms of
the interferometer. In this way one can shift the pattern of interference
fringes by regular steps, without moving the detectors P MA and P MB ; the
standard step corresponds to a change of /50 in the difference between
the two optical paths.

A sweep, taking 15 sec for each standard step, yields two interference plots
corresponding, respectively, to the paths (1,2) and (1,2); the fringes
have good contrast(difference in intensity between maxima and minima),
and their visibility

(NA,max NA,min ) / (NA,max + NA,min ) (16.53)

was measured as 98% as shown in Figure 16.13 below which gives the two
interference plots obtained with the Mach-Zehnder interferometer. Note
that the maximum counting rates in P MA correspond to minima in P MB ,
indicating a relative displacement of /2 between the two interference pat-
terns

1288
 Figure 16.13: Interference Results

If we recall that we are reasoning in terms of photons, and that the photons
are being processed individually, then we must admit that the interference
does not stem from any interaction between successive photons, but that
each photon interferes with itself.

What would Albert have to say? He seems exasperated but is still polite.
His statements are brief:

I observe that the optical paths differ in length between LS and LS , and
are then coincident over (1,2) and over (1,2).

In P MA I observe a process that seems perfectly natural to me, namely

light + light light
In P MB I observe a process that I find astounding, namely
light + light darkness
Such superposition phenomena with light I shall call interference; con-
structive in P MA and destructive in P MB .

In the situation considered before, I envisaged light as consisting of par-

ticles called photons, which travelled either along path 1 or along path 2.

1289
In the present situation I want to know for each individual photon which
path it has travelled; to this end I should like to ask you to close off path
2, since this will ensure that the photons travel by path 1..

Clearly Albert is perturbed. He awaits the new experimental results with

some anxiety.

On closing either path, whether 1 or 2, one observes that all interference

phenomena disappear. For instance, instead of a very high count NA and
a very low count NB , we now obtain essentially equal counts from P MA
and P MB .

Albert is visibly displeased and now very wary. He then continues with
his analysis of the experiment:

I observe that in order to produce interference phenomena it is necessary

to have two optical paths of different lengths, both open.

Whenever a photon is detected, I note my inability to ascertain whether

the light has travelled by path 1 or by path 2, because I have no means for
distinguishing between the two cases.

If I were to suppose that photons travel only along 1, then this would imply
that path 2 is irrelevant, which is contrary to what I have observed. Simi-
larly, if I were to suppose that photons travel only along 2, then this would
imply that path 1 is irrelevant, which is also contrary to my observations.

If I envisage the source S as emitting particles, then I am forced to con-

clude that each individual photon travels simultaneously along both paths
1 and 2; but this result contradicts the results of the previous experiment
(second stage), which compelled me to envisage that every photon chooses,
at random, either path 1 or path 2.

I conclude that the notion of particles is unsuited to explaining interfer-

ence phenomena.

I shall suppose instead that the source emits a wave; this wave splits into
two at LS , and the two secondary waves travel one along path 1 and
the other along path 2. They produce interference by mutual superposition
on LS constructively in (1,2) and destructively in (1,2). At the far
end of (1,2) or of (1,2) I envisage each of the waves condensing into
particles, which are then detected by the photomultipliers (essentially by
P MA since the contrast is 98% means only very few photons are detected
by P MB

It seems to me that I am beginning to understand the situation. I en-

1290
visage light as having two complementary forms: depending on the kind
of experiment that is being done, it can manifest itself either as a wave,
or as a particle, but never as both simultaneously and in the same place.
Thus, in the experiment where the path followed by the light cannot be
ascertained (third stage), light behaves first like a wave, producing inter-
ference phenomena; but it behaves like a particle when, afterwards, it is de-
tected through the photoelectric effect. I conclude that light behaves rather
strangely, but nevertheless I have the impression that its behavior can be
fully described once one has come to terms with the idea of wave-particle
duality..

Albert leaves the room slowly, hesitantly, even reluctantly. He might be

impressed by all the completeness of all that he has just described or
maybe he is worried that more needs to be said.

In fact, something does remain to be said, since the problem of causality

remains open. Let us look carefully at the experimental layouts in the
second and third stages: we see that they have LS in common, and that
they differ only beyond some boundary (indicated by the dashed circle
downstream from LS ). We have stated that light behaves like a particle
or like a wave depending on whether or not one can ascertain the path
it takes through the apparatus; but in the two experiments under con-
sideration, the choice between the alternatives must be decided on LS ,
before the light has crossed the crucial boundary, that is, at a stage where
nothing can as yet distinguish between the two kinds of apparatus, since
they differ only beyond the point of decision. It is as if the light chose
whether to behave like a wave or like a particle before knowing whether
the apparatus it will pass through will elicit interference phenomena or
the photoelectric effect. Hence the question of causality is indeed opened
up with vengeance.

Albert comes back abruptly. He is disconcerted and wearily says:

Originally I supposed that light would behave like a wave or like a particle,
depending on the kind of experiment to which it was being subjected.

I observe that the choice must be made on the half-silvered mirror LS ,

before the light reaches that part of the apparatus where the choice is ac-
tually implemented; this would imply that the effect precedes the cause.

I know that both waves and particles obey the principle of causality, that
is, that cause precedes effect.

I conclude that light is neither wave nor particle; it behaves neither like
waves on the sea, nor like projectiles fired from a gun, nor like any other
kind of object that I am familiar with.

1291
I must ask you to forget everything I have said about this experiment,
which seems to me to be thoroughly mysterious.

Albert leaves, but quickly returns with a contented smile, and his final
statement is not without a touch of malice. I observe in all cases that the
photomultipliers register quanta when I switch on the light source.

I conclude that something has travelled from the source to the detector.
This something is a quantum object, and I shall continue to call it a
photon, even though I know that it is neither a wave nor a particle.

I observe that the photon gives rise to interference when one cannot as-
certain which path it follows; and that interference disappears when it is
possible to ascertain the path.

For each detector, I observe that the quanta it detects are randomly dis-
tributed in time.

If I repeat the experiment several times under identical conditions, then I

observe that the photon counts registered by each photomultiplier are re-
producible in a statistical sense. For example, suppose that in the first and
in the second experiments P MA registers NA0 and NA00 respectively; then
one can predict that NA00 has a probability of 0.68 of being in the interval
NA0 (NA0 )1/2

Thus, these counts enable me to determine experimentally, for any kind of

apparatus, the probability that a given detector will detect a quantum, and
it is precisely such probabilities that constitute the results of experiments.

I assert that the function of a physical theory is to predict the results of

experiments.

What I expect from theoretical physicists is a theory that will enable me

to predict, through calculation, the probability that a given detector will
detect a photon. This theory will have to take into account the random
behavior of the photon, and the absence or presence of interference phe-
nomena depending on whether the paths followed by the light can or cannot
be ascertained..

Albert leaves, wishing the physicists well in their future endeavors.

Physicist have indeed worked hard and the much desired theory has in-
deed come to light, namely, quantum mechanics, as we have seen in our
discussions. As we have seen, it applies perfectly not only to photons, but
equally well to electrons, protons, neutrons, etc; in fact, it applies to all
the particles of microscopic physics. For the last 75 years it has worked

1292
 to the general satisfaction of physicists.
Meanwhile, it has produced two very interesting problems of a philosophical
nature.
1. Chance as encountered in quantum mechanics lies in the very nature of the
coupling between the quantum object and the experimental apparatus. No
longer is it chance as a matter of ignorance or incompetence: it is chance
quintessential and unavoidable.
2. Quantum objects behave quite differently from the familiar objects of
our everyday experience: whenever, for pedagogical reasons, one allows
an analogy with macroscopic models like waves or particles, one always
fails sooner or later, because the analogy is never more than partially
valid. Accordingly, the first duty of a physicist is to force her grey cells,
that is her concepts and her language, into unreserved compliance with
quantum mechanics (as we have been attempting to do); eventually this
will lead her to view the actual behavior of microsystems as perfectly
normal. As a teacher of physics, our duties are if anything more onerous
still, because we must convince the younger generations that quantum
mechanics is not a branch of mathematics, but an expression of our best
present understanding of physics on the smallest scale; and that, like all
physical theories, it is predictive.
In this context, let us review the basic formalism of quantum mechanics.

16.4.2 Basic Formalism

We will introduce the elements of quantum mechanics as axioms. Physicists
have devised a new mathematical tool. The transition amplitude from initial
to final state, and it is this amplitude that enables one to calculate the needed
probabilities.
1. For the experiment where the photon travels from the source S to the de-
tector P MA (see Figure 16.14(a) below), we write the transition amplitude
from S to P MA as

hphoton arriving at P MA | photon leaving Si

which we symbolize simply as

hf1 | ii , hf2 | ii

In this case there are two probabilities:

2 2
|hf1 | ii| , |hf2 | ii|

The total probability is their sum:

2
|hf1 | ii| + |hf2 | ii|

1293
 More generally, we would write
2
X 2
|hf | ii| = |hfk | ii|
k

Figure 16.14 below shows three arrangements sufficient to determine the

transition amplitude: (a) a single optical path; (b) two paths,allowing us
to ascertain which path has actually been taken; (c) two paths, not allow-
ing us to ascertain which path has actually been taken.

Figure 16.14: Three Arrangements

2. If a photon is emitted by the source S can take either of two paths, but it
is impossible to ascertain which path it does take (Figure 16.14(c) above),
then there are again two transition amplitudes:
hphoton arriving at P MA | photon leaving Sialong path 1
hphotonarrivingatP MB | photon leaving Sialong path 2
which we symbolize simply as
hf | ii1 , hf | ii2
To allow for interference, we assert that in this case it is the amplitudes
that must be added; the total amplitude reads
hf | ii = hf | ii1 + hf | ii2

1294
 The total probability is then:
2
|hf | ii1 + hf | ii2 |

More generally, we would write

X
total amplitude: hf | ii = hf | iik
k
 2
X 
2
total probability: |hf | ii| =  hf | iik 
 
 
k

where the sums are over all possible paths.

3. If one wants to analyze the propagation of the light more closely, one
can take into account its passage through the half-silvered mirror LS ,
considering this as an intermediate state (Figure 16.14(b) above). The
total amplitude for path 1 is

hphoton arriving at P MA | photon leaving Si

However, it results from two successive intermediate amplitudes:

hphoton arriving at LS | photon leaving Si

hphoton arriving at P MA | photon leaving LS i

Here we consider the total amplitude as the product of the successive

intermediate amplitudes; symbolically, labelling the intermediate state as
v, we have
hf | ii = hf | i h | ii
Finally, consider a system of two mutually independent photons. If photon
1 undergoes a transition from a state i1 to a state f1 , and photon 2 from
a state i2 to a state f2 , then

hf1 f2 | i1 i2 i = hf1 | i1 i hf2 | i2 i

The four rules just given suffice to calculate the detection probability in any
possible experimental situation. They assume their present form as a result of
a long theoretical evolution; but they are best justified a posteriori, because
in 75 years they have never been found to be wrong. Accordingly, we may
consider them as the basic principles governing the observable behavior of all
microscopic objects, that is, objects whose action on each other are of order
(Plancks constant). From these principles (they are equivalent to our earlier
postulates - just look different because we are using the amplitude instead of
the state vector as the fundamental mathematical object in the theory) one can
derive all the requisite formalism, that is, all of quantum mechanics.

1295
Quantum mechanics as we have described it earlier and also above, works splen-
didly, like a well-oiled machine. It, and its basic principles, might therefore be
expected to command the assent of every physicist; yet it has evoked, and on
occasion continues to evoke, reservations both explicit and implicit. For this
there are two reasons:

1. Quantum mechanics introduces unavoidable chance, meaning that its char-

acteristic randomness is inherent in the microscopic phenomena them-
selves.

2. It attributes to microscopic objects properties so unprecedented that we

cannot represent them through any macroscopic analogs or models.

Both features are revolutionary, and it is natural that they should have provoked
debate. On the opposite sides of this debate we find two great physicists, Neils
Bohr and Albert Einstein, and we will now discuss how the debate evolved from
its beginnings in 1927 to its conclusion in 1983 (that is 56 years!).

16.5 Inseparable Photons (the EPR Paradox)

including some history
Though ornithologists have known about inseparable parrots for a long time,
to physicists the existence of inseparable photons has been brought home only
during the last two decades, through a beautiful series of experiments by Alain
Aspect and his research group at Orsay Laboratory in Paris. The experiments
are exemplary, in virtue both of the difficulties they had to overcome and the re-
sults achieved, which are exceptionally clear-cut. In fact, the significance of the
experiments extends beyond the strict confines of physics, because they provide
the touchstone for settling a philosophical debate that has divided physicist for
75 years. The division dates back to the appearance of two mutually contradic-
tory interpretations of quantum mechanics at the Como conference in 1927. To
sketch the debate, we start with a brief summary of the philosophy of physics.

16.5.1 The Philosophical Stakes in the Debate

Our summary is best presented diagrammatically as shown in Figure 16.15 be-
low where we present the philosophical elements in a debate between physicists.

Figure 16.15: Philosophical Ideas

1296
 1. For the physicist who is a realist, a physical theory reflects the behavior
of real objects, whose existence is not brought into question.
2. For the physicist who is a positivist, the purpose of a physical theory is to
describe the relations between measurable quantities. The theory does not
tell one whether anything characterized by these quantities really exists,
nor even whether the question makes sense.
3. For the physicist who is a determinist, exact knowledge of the initial con-
ditions and of the interactions allows the future to be predicted exactly.
Determinism is held to be a universal characteristic of natural phenom-
ena, even about those which we know, as yet, little or nothing. In this
framework, any recourse to chance merely reflects our own ignorance.
4. For the physicist who is a probabilist, chance is inherent in the very nature
of microscopic phenomena. To her, determinism is a consequence, on
the macroscopic level, if the laws of chance operating on the microscopic
level; it is appropriate to measurements of mean values of quantities whose
relative fluctuations are very weak.
From these four poles, realism, positivism, determinism, and chance, the physi-
cist chooses two, one on each axis. Though sometimes the choice is made in full
awareness of what it entails, most often it is made subconsciously. In our de-
scription of quantum mechanics, we might adopt without reservations, the point
of view of the elementary particle physicist. For a start, she believes firmly in
the existence of particles, since she spends her time in accelerating, deflecting,
focusing, and detecting them. Even though she has never seen or touched them,
to her their objective existence is not in any doubt. Next she observes that they
impinge on the detectors quite erratically, whence she has no doubts, either,
that their behavior is random. Accordingly, the elementary particle experimen-
talist has chosen realism and chance, most often without realizing that she has
made choices at all.

There are other philosophical options that can be adopted with eyes fully open:
realism and determinism are the choices of Albert Einstein; positivism and
chance are those of Neils Bohr. They are well acquainted and each thinks very
highly of the other: which is no bar to their views being incompatible, nor to
the two men representing opposite poles of the debate.

16.5.2 From Como to Brussels (1927-30)

On September 26,1927, in Como, Niels Bohr delivered a memorable lecture. His
stance is that of an enthusiastic champion of the new quantum mechanics. He
puts special weight on the inequalities proved by Heisenberg the year before:
1 1
xpx ~ , tE ~ (16.54)
2 2

1297
They imply that it is impossible to define exact initial conditions for a mi-
croscopic object, which automatically makes it impossible to construct, on the
microscopic scale, a deterministic theory patterned on classical mechanics. Only
a probabilistic theory is possible, and that theory is quantum mechanics.

Einstein disagrees with this point of view, and his opposition to Bohrs the-
ses becomes public at the Brussels conference in 1930: he adopts the role of a
dissenter who knows precisely how to press home the most difficult questions.
Deeply shocked by the retreat from determinism, he tries to show via his thought
(gedanken) experiments he can contravene the Heisenberg inequalities.

At the cost of several sleepless nights devoted to analyzing the objections of

his adversary, Bohr refutes all of Einsteins criticisms, and emerges from the
conference as the undoubted winner.

16.5.3 From Brussels to the EPR Paradox (1930-35)

Having lost the argument at Brussels, Einstein tries to define his objections
with ever greater precision. Believing as he does that position and momentum
exist emphobjectively and simultaneously, he considers quantum mechanics to
be incomplete and merely provisional. The points of view of the two antagonists
at this stage of the debate can be spelled out as follows.

For Einstein, a physical theory must be a deterministic and a complete repre-

sentation of the objective reality underlying the phenomena. It features known
variables that are observable, and others, unknown as yet, called hidden vari-
ables. Because of our provisional ignorance of the hidden variables, matter at
the microscopic level appears to us to behave arbitrarily, and we describe it
by means of a theory that is incomplete and probabilistic, namely by quantum
mechanics.

For Bohr, a physical theory makes sense only as a set of relations between
observable quantities. Quantum mechanics supplies a correct and complete de-
scription of the behavior of objects at the microscopic level, which means that
the theory itself is likewise complete. The observed behavior is probabilistic,
implying that chance is inherent in the nature of the phenomena.

Between chance as a matter of ignorance, as advocated by Einstein, and chance

unavoidable, as advocated by Bohr, the debate does not remain merely philo-
sophical. Quite naturally it returns to the plane of physics with the thought
experiment proposed by Einstein, Podolsky and Rosen in 1935, which in their
view proves that quantum mechanics is indeed incomplete. Their thought ex-
periment is published as a paper in the Physical Review, but it is so important
that it reverberates as far as the New York Times. Physicists call the proposal
the EPR paradox, after its proponents. It will take fifty years to untangle the
question, first in theory and then by experiment. We will not, of course, follow

1298
these fifty years blow by blow; instead, we confine attention to three decisive
stages reached respectively in 1952, 1964, and 1983. But we start with an
illustration that helps one see what the EPR paradox actually is.

16.5.4 Elementary Introduction to the EPR Paradox

Consider two playing cards, one red(diamond) and one black(spade) as shown
in Figure 16.16 below where we use two playing cards help us understand the
stakes in the EPR paradox.

Figure 16.16: EPR Setup

An experimenter in Lyons puts them into separate envelopes which she then
seals. She is thus provided with two envelopes looking exactly alike, and she
puts both into a container. She shakes the container so as the shuffle the pack,
and the system is ready for the experiment.

At 8:00 two travellers, one from Paris and one from Nice, come to the container
(in Lyons), take one envelope each, and then return to Paris and Nice, respec-
tively. At 14:00 they are back at their starting points; each opens her envelope,
looks at the card, and telephones to Lyons reporting the color. The experiment
is repeated every day for a year, and the observer in Lyons keeps a careful record
of the results. At the end of the year the record stands as follows:
1. The reports from Paris are red or black, and the sequence of these reports
is random. The situation is exactly the same as in a game of heads or
tails, and probability of each outcome is 1/2.
2. The reports from Nice are red or black, and the sequence of these reports
is random. Here too probability of each outcome is 1/2.
3. When Paris reports red, Nice reports black ; when Paris reports black, Nice
reports red. One sees that there is perfect(anti) correlation between the

1299
 report from Paris and the report from Nice.
Accordingly, the experiment we have described displays two features:
1. It is unpredictable and thereby random at the level of individual observa-
tions in Paris and Nice.
2. It is predictable, by virtue of the correlation, at the level where one observes
the Paris and the Nice results simultaneously.
Einstein and Bohr might have interpreted the correlation as follows.

According to Einstein
The future of the system is decided at 8:00 when the envelopes are chosen,
because he believes that the contents of the two envelopes differ. Suppose, for
instance, that Paris has (without knowing it) drawn a red card, and Nice the
black. The colors so chosen exist in reality, even though we do not know them.
The two cards are moved, separately, by the travellers between 8:00 and 14:00,
during which time they do not influence each other in any way. The results on
opening the envelopes read red in Paris and black in Nice. Since the choice at
8:00 was made blind, the opposite outcome is equally possible, but the results
at 14:00 are always correlated (either red/black or black/red). This correlation
at 14:00 is determined by the separation of the colors at 8:00, and we say the
theory proposed by Einstein is realist, deterministic, and separable(or local), by
virtue of a hidden variable, namely, the color.

According to Bohr
There is a crucial preliminary factor, inherent in the preparation of the system.
On shaking the container with the two envelopes, one loses information regard-
ing the colors. Afterwards, one only knows that each envelope contains either a
red card (probability 1/2) or a black card (probability 1/2). We will therefore
say that a given envelope is in a brown state, which is a superposition of a red
state and of a black state having equal probabilities. At 8:00 the two envelopes
are identical: both are in a brown state, and the future of the system is still
undecided. There is no solution until the envelopes are opened at 14:00, since it
is only the action of opening them that makes the colors observable. The result
is probabilistic. There is a probability 1/2 that in Paris the envelope will be
observed to go from the brown state to the red, while the envelope in Nice is
observed to go from the brown state to the black; there is the same probability
1/2 of observing the opposite. But the results of the observations on the two
envelopes are always correlated, which means that there is a mutual influence
between them, in particular at 14:00; in fact it is better to say that, jointly,
they constitute a single and non-separable system, even though one is in Paris
and the other is in Nice. Accordingly, the theory proposed by Bohr is positivist,

1300
probabilistic(non-deterministic) and non-separable(non-local), interrelating as
it does the colors that are actually observed.

Einsteins view appears to be common sense, while it must be admitted that

Bohrs is very startling; however, the point of this macroscopic example is, pre-
cisely, to stress how different the quantum view is from the classical.

Proceeding with impeccable logic but from different premises, both theories
predict the same experimental results. Can we decide between them? At the
level considered here it seems we cannot: for even if the envelopes were opened
prematurely while still in Lyons, one would merely obtain the same results at a
different time, and without affecting the validity of either interpretation. The
solution to the problem must be looked for at the atomic level, by studying the
true EPR set-up itself.

16.5.5 The EPR Paradox (1935-52)

Albert Einstein, Boris Podolsky, and Nathan Rosen meant to look for an experi-
ment that could measure, indirectly but simultaneously, two mutually exclusive
quantities like position and momentum. Such results would contravene the pre-
dictions of quantum mechanics, which allows the measurement of only one such
quantity at any one time; that is why the thought experiment is called the EPR
paradox.

In 1952, David Bohm showed that the paradox could be set up not only with con-
tinuously varying quantities like position and momentum, but also with discrete
quantities like spin. This was the first step towards any realistically conceiv-
able experiment. Meanwhile, objectives have evolved, and nowadays it is more
usual to talk of the EPR scenario, meaning some sensible experiment capable
of discriminating between quantum theory and hidden-variable theories. Such
a set-up is sketched in Figure 16.17 below where we present the simplest EPR
scenario.

Figure 16.17: EPR Scenario

A particle with spin 0 decays, at S, into two particles of spin 1/2, which di-
verge from S in opposite directions. Two Stern-Gerlach type detectors A and
B measure the xcomponents of the spins. Two types of response are possible:

1301
 1. spin up at A, spin down at B, a result denoted by (+1, 1)

2. spin down at A, spin up at B, a result denoted by (1, +1)

Thus far everyone is agreed, but the interpretation is yet to come.

Einstein reasons that if pairs of particles produced at S elicit different responses

(+1, 1) and (1, +1) from the detector system A,B, then the pairs must have
differed already at S, immediately after the decay.

It must be possible to represent this difference by a hidden variable , which

has an objective meaning, and which governs the future of the system. After
the decay the two particles separate without influencing each other any further,
and eventually they trigger the detectors A and B.

Bohr reasons that all the pairs produced at S are identical. Each pair con-
stitutes a non-separable system right up to the time when the photons reach
the detectors A and B. At that time we observe the response of the detectors,
which is probabilistic, admitting two outcomes (+1, 1) and (1, +1). To sum
up, Einstein restricts the operation of chance to the instant of decay (at S),
whose details we ignore, but which we believe creates pairs whose hidden vari-
ables are different.

By contrast, Bohr believes that chance operates at the instant of detection, and
that it is inherent in the very nature of the detection process: this chance is
unavoidable.

We are still in the realms of thought, and stay there up to 1964.

In 1964, the landscape changes: John Bell, a theorist at CERN, shows that it
is possible to distinguish between the two interpretations experimentally.

The test applies to the EPR scenario; it is refined by Clauser, Horne, Shimony,
and Holt, whence it is called the BCHSH inequality after its five originators.

16.5.6 The BCHSH Inequality (1964)

To set up an EPR scenario, one first needs a source that emits particle pairs.
Various experimental possibilities have been explored:

1. atoms emitting two photons in cascade

2. electron-positron annihilation emitting two high-energy photons

3. elastic proton-proton scattering

It is solution (1) that has eventually proved the most convenient; it has been
exploited by Alain Aspect at the Institute for Optics in Paris, in particular.

1302
Next one needs detectors whose response can assume one of two values, repre-
sented conventionally by +1 and 1. Such a detector might be

1. for spin 1/2 particles, a Stern-Gerlach apparatus responding to spin up or

spin down

2. for photons, a polarizer responding to parallel polarization or perpendicular

polarization

Our sketch of the EPR scenario can now be completed as in Figure 16.18 below
where we present the most general EPR scenario.

Figure 16.18: BCHSH Setup

Figure 16.18(a) views the apparatus perpendicularly to axis, showing the two
~ and B.
detectors A and B, with their polarizing directions denoted as A ~

Figure 16.18(b) views the apparatus along its axis, and shows that the analyzing
directions of the two detectors are not parallel, but inclined to each other at an
angle .

In Figure 16.18((c) we also a view along the axis of the apparatus, and shows the
actual settings chosen by Aspect: two orientations are allowed for each detector,
A~ 1 or A
~ 2 for one, and B
~ 1 or B
~ 2 for the other.

We adopt the following conventions:

~
(1) = 1 is the response of detector A when oriented along A

~
(2) = 1 is the response of detector B when oriented along B

1303
Since each detector has two possible orientations, called 1 and 2, we shall denote
their responses as 1 , 2 and 1 , 2 respectively. Now consider the quantity hi
defined by
hi = h1 1 i + h1 2 i + h2 1 i h2 2 i (16.55)
where the symbol h...i denotes the mean value over very many measured events.
We call hi the correlation function of the system.

The BCHSH inequality reads 2 hi 2. Its authors have proved that it

must be satisfied if mechanics at the microscopic level constitutes a theory that
is realist, deterministic, and separable: or in other words if the theory contains
a hidden variable. A sketch of the a proof is shown below.

A Proof of Bells Inequality

A theory that is deterministic and separable
Suppose that the pair a, b emerging from S can be characterized by a hidden
~ ) and (B,
variable . The responses of the detectors A, B are (A, ~ ) respec-
tively as shown in Figure 16.19 below.

Figure 16.19: Bell Inequality Setup

The theory is deterministic and separable:

1. deterministic, because the results are determined by the hidden variables
~ and B
plus the settings A ~

2. separable, because the response of A is independent of the response of B,

and vice versa
Since the value of is unknown and different for each pair, the responses of A and
B seem random. Lacking information about , we characterize it by choosing a
statistical distribution (), which then allows us to derive the distribution of
the responses (A, ~ ) and (B,
~ ), which can be compared with experiment.

Bells inequalities have the great virtue that they apply to any hidden variable
theory, irrespective of the choice of ().

Theorem 1: Consider the four numbers 1 , 2 and 1 , 2 , each of which can

assume only the values 1 or 1. Then the combination
= 1 1 + 1 2 + 2 1 2 2 (16.56)

1304
can assume only the values 2 and 2.

To prove the theorem, one constructs a truth table for all 16 possibilities, which
shows that 2 and 2 are indeed the only possible values of .

1 2 1 1
1 1 1 1 2
1 1 1 -1 2
1 1 -1 1 -2
1 1 -1 -1 -2
1 -1 1 1 2
1 -1 1 -1 2
1 -1 -1 1 -2
1 -1 -1 -1 -2
-1 1 1 1 -2
-1 1 1 -1 2
-1 1 -1 1 -2
-1 1 -1 -1 2
-1 -1 1 1 -2
-1 -1 1 -1 -2
-1 -1 -1 1 2
-1 -1 -1 -1 2

Table 16.1: Values

Theorem 2: Consider very many sets of four numbers 1 , 2 and 1 , 2 . The

mean value of lies in the range [2, 2]. In other words,

2 hi 2 (16.57)

This is obvious, because every value of lies in this range, and so therefore must
the mean. The endpoints are included in order to allow for limiting cases. Note
that both theorems are purely mathematical, neither involves any assumptions
about physics.

16.5.7 BCHSH Inequality(Bells inequality in real world)

Within the framework of a theory that is realist, deterministic, and separable,
we can describe the photon pair in detail. Realism leads us to believe that
polarization is an objective property of each member of the pair, independent
of any measurements that may be made later. Determinism leads us to believe
that the polarizations are uniquely determined by the decay cascade, and that
they are fully specified by the hidden variable , which governs the correlation
of the polarizations in A and B. Finally, separability leads us to believe that the
measurements in A and B do not influence each other, which means in particular

1305
that the response of detector A is independent of the orientation of detector B.

Now consider a pair of photons a, b, characterized by a hidden variable . The

response of the apparatus in its four settings would be as follows:

~1, B
1 and 1 in the orientation (A ~ 1)
~2, B
2 and 2 in the orientation (A ~ 2)
~1, B
10 and 20 in the orientation (A ~ 2)
~2, B
20 and 10 in the orientation (A ~ 1)

Recall that the variables and can only take on the values 1 and 1. It is
impossible in practice to make four measurements on one and the same pair
of photons, because each photon is absorbed in the first measurement made
on it; that is why we have spoken conditionally, that is, of what results would
be(a COUNTERFACTUAL statement). But if we believe that the photon
correlations are governed by a theory that is realist, deterministic, and sepa-
rable, then we are entitled to assume that the responses, of type or type ,
depend on properties that the photons possess before the measurement, so that
the responses correspond to some objective reality. In such a framework we can
appeal to the principle of separability, which implies, for instance, that detec-
tor A would give the same response to the orientations (A ~1, B
~ 1 ) and (A
~1, B
~ 2 ),
because the response of A is independent of the orientation of B.

Mathematically, this is expressed by the relation 1 = 10 .

Similarly one finds 2 = 20 , 1 = 10 , 2 = 20 .

Thus, we have shown that, for a given pair of photons, all possible responses
of the apparatus in its four chosen settings can be specified by means of only
four two-valued variables 1 , 2 and 1 , 2 . This reduction from eight to four
variables depends on the principle of separability. In this way, we are led to a
situation covered by Theorem 2, and therefore 2 hi 2.

By making many measurements for each of the four settings we can determine
the four mean values h1 1 i , h1 2 i , h2 1 i , h2 2 i and thus the mean value
of the correlation

hi = h1 1 i + h1 2 i + h2 1 i h2 2 i

According to quantum mechanics (which is positivist, probabilistic, and non-

separable), there are cases where the BCHSH inequality is violated. In par-
ticular, one can show that for photons in the configuration chosen by Aspect
quantum mechanics yields(we will derive this shortly)

hi = 3 cos 2 cos 6 (16.58)

1306

This leads to values well outsidethe interval [2, 2]; for example to hi = 2 2
when = 22.5 and to hi = 2 2 when = 67.5 .

Proof : The laboratory reference frame Oxyz serves to specify the orientations
of detectors and polarizers as shown in Figure 16.20 below:

Figure 16.20: Detector Orientations

Before any measurements have been made, the photon pair a, b forms a non-
separable entity, represented by the vector
1
|i = (|xA , xB i + |yA , yB i) (16.59)
2
The act of measurement corresponds to passage to the -basis. Hence, we
require the transition amplitudes from the two states |xA , xB i , |yA , yB i to the
four states
|A , B i , |A , B + /2i , |A + /2, B i , |A + /2, B + /2i (16.60)
In the -basis we have
1
|i = [cos(B A ) |A , B i
2
sin(B A ) |A , B + /2i
+ sin(B A ) |A + /2, B i
+ cos(B A ) |A + /2, B + /2i] (16.61)
The square of each amplitude featured here represents the detection probability.
For example, the probability of simultaneously detecting photon a polarized at
the angle A and the photon b polarized at the angle B is
 2
1 1
cos(B A ) = cos2 (B A ) (16.62)
2 2

1307
By convention, we write the responses of detector A to a photon in state |A i
(respectively |A + /2i) as = 1 and similarly with for detector B.

Let us analyze the four possible responses:

1. |A , B i gives = 1, = 1 so = 1; the probability is

1
P++ = cos2 (B A ) (16.63)
2

2. |A , B + /2i gives = 1, = 1 so = 1; the probability is

1
P+ = sin2 (B A ) (16.64)
2

3. |A + /2, B i gives = 1, = 1 so = 1; the probability is

1
P+ = sin2 (B A ) (16.65)
2

4. |A + /2, B + /2i gives = 1, = 1 so = 1; the probability is

1
P = cos2 (B A ) (16.66)
2

The mean value of hiAB follows immediately as

hiAB = P++ P+ P+ + P = cos 2(B A ) (16.67)

The settings chosen by Aspect are as shown in Figure 16.20 above. Correspond-
ing to it we have the four terms

h1 1 i = hiA1 B1 = cos 2(B1 A1 ) = cos 2 (16.68)

h1 2 i = hiA1 B2 = cos 2(B1 A2 ) = cos 2 (16.69)
h2 1 i = hiA2 B1 = cos 2(B2 A1 ) = cos 2 (16.70)
h2 2 i = hiA2 B2 = cos 2(B2 A2 ) = cos 6 (16.71)

For comparison with Bells inequality, we introduce the correlation function

hi = h1 1 i + h1 2 i + h2 1 i h2 2 i = 3 cos 2 cos 6 (16.72)

Thus, the BCHSH test turns the EPR scenario into an arena for rational con-
frontation between the two interpretations; it remains only to progress from
thought experiments to experiments conducted in the laboratory.

1308
16.5.8 The Beginnings of the Experiment at Orsay (1976)
Alain Aspects experiment studies the correlation between the polarizations
of the members of photon pairs emitted by calcium. The light source is a
beam of calcium atoms, excited by two focused laser beams having wavelengths
0 = 406 nm and 00 = 581 nm respectively. Two-photon excitation pro-
duces a state having the quantum number J = 0. When it decays, this state
emits two monochromatic photons having the wavelengths 1 = 551.3 nm and
2 = 422.7 nm respectively, in a cascade of two electronic transitions from the
initial J = 0 level to the final J = 0 state, passing through an intermediate
J = 1 state, as shown in Figure 16.21 below which shows the excitation and
decay of the calcium atom.

Figure 16.21: Calcium based light source - Energy Levels

The mean lifetime of the intermediate state is 4.7 ns. To simplify the terminol-
ogy, we shall call the 1 = 551.3 nm light green, and the 2 = 422.7 nm light
violet.

The polarizer, which works like a Wollaston prism is shown in Figure 16.22 be-
low where we can see the two-valued response of a Wollaston prism.

Figure 16.22: Wollason Prism - Polarizer

1309
The Wollaston prism is made of quartz or of calcite. It splits an incident beam
of natural (unpolarized) light into two beams of equal intensity, polarized at 90
to each other. If only a single unpolarized photon is incident, it emerges either
in the state |xi, with probability 1/2, or in the state |yi , with probability 1/2.
Thus, the response of the system is two-valued.

The photon is detected by the photomultiplier tubes (PM) downstream from

the prism. Every electric pulse from these detectors corresponds to the passage
of a photon, allowing the photons to be counted. The experimental layout is
sketched in Figure 16.23 below, which shows a sketch of the first Orsay experi-
ment.

Figure 16.23: First Orsay Experiment

It uses a coincidence circuit which registers an event whenever two photons are
detected in cascade. In this way four separate counts are recorded simultane-
ously, over some given period of time. In the EPR scenario envisaged by Bohm,
where = 0, the only possible responses are (+1, 1) or (1, +1) In the situ-
ation realized by Aspect, the angle is non-zero, and four different responses
are possible.

1. N++ the number of coincidences corresponding to = 1 and = 1, that

is, to = 1

2. N+ the number of coincidences corresponding to = 1 and = 1,

that is, to = 1

3. N+ the number of coincidences corresponding to = 1 and = 1,

that is, to = 1

1310
 4. N the number of coincidences corresponding to = 1 and = 1,
that is, to = 1
The resolving time of the coincidence circuit is 10 ns, meaning that it reckons
two photons as coincident if the they are separated in time by no more than
10 ns. The mean life of the intermediate state of the calcium atom is 4.7 ns.
Therefore, after a lapse of 10 ns, that is more than twice the mean lifetime,
almost all the atoms have decayed (actually 88%). In other words, the efficiency
of the coincidence counter is very high.

The experiment consists in counting, over some given time interval, the four
kinds of coincidence: N++ , N+ , N+ and N . The total number of events is
N = N++ + N+ + N+ + N .

Accordingly, the different kinds of coincidence have probabilities

P++ = N++ /N corresponding to = 1

P+ = N+ /N corresponding to = 1
P+ = N+ /N corresponding to = 1
P = N /N corresponding to = 1

and the measured average of is

N++ N+ N+ + N
hi = (16.73)
N
Each set of four coincidence counts corresponds to one particular setting of
~ B),
(A, ~ and yields a mean value hi. But in order to determine the correlation
function hi used in the BCHSH inequality, we need four mean values hi.
Therefore, we choose, in succession four different settings as shown in Figure
16.18(c); four counting runs then yield the four mean values h1 1 i , h1 2 i , h2 1 i , h2 2 i
which then determine the value of hi via

hi = h1 1 i + h1 2 i + h2 1 i h2 2 i (16.74)

The Results of the First Experiment at Orsay

The results of the first Orsay experiment are shown in Figure 16.24 below. The
angle which specifies the setting of the polarizers is plotted horizontally, and
the mean value hi vertically.

1311
 Figure 16.24: Results from First Orsay Experiment

From earlier, the correlation function predicted by quantum mechanics reads

hi = 3 cos 2 cos 6 (16.75)
It is drawn as the solid curve on the graph(the curve has been corrected for
instrumental effects, which explains why its ends are not precisely at 2 and 2).
According to the BCHSH inequality
2 hi 2 (16.76)
so that hidden-variable theories exclude the cross-hatched regions of the plane,
which correspond to hi > 2 or hi < 2.

The experimental results from 17 different values of are indicated on the figure
by squares, where the vertical size of the square gives plus or minus one standard
deviation (a measure of the experimental error).

Clearly, there can be no doubt that the BCHSH inequality is violated; many of
the experimental points fall outside the interval [2, 2]. At the point where the
violation is maximal ( = 22.5 ), one finds
hi = 2.70 0.015 (16.77)
which represents a departure of over 40 standard deviations from the extreme
value of 2. What is even more convincing is the precision with which the exper-
imental points lie on the curve predicted by quantum mechanics.

Quite evidently, for the EPR scenario one must conclude not only that hidden-
variable theories fail, but also that quantum mechanics is positively the right
theory for describing the observations.

The Relativistic Test

The EPR experiment just described shows that the measurements in A and B
are correlated. What is the origin of the correlations?

1312
According to quantum theory, before the measurement each particle pair consti-
tutes a single system extending from A to B, whose two parts are non-separable
and correlated. This interpretation corresponds to a violation of Bells inequal-
ity and agreement with experiment.

According to hidden-variables theories, the particle pair is characterized, at the

instant of decay, by its hidden variable , which determines the correlation be-
tween the polarizations measured in A and B. This interpretation satisfies Bells
inequality but disagrees with experiment.

Accordingly, the Orsay experiment supports the quantum interpretation (in

terms of the correlation between two parts A and B of a single system).

However, to clinch this conclusion, one must ensure that no influence is exerted
in the ordinary classical sense through some interaction propagated between the
two detectors A and B, that is, no influence which might take effect after the
decay at S, and which might be responsible for the correlation actually observed.

Let us therefore examine the Orsay apparatus in more detail as in Figure 16.25
below where we attempt to test Einsteinian non-separability.

Figure 16.25: Orsay Experiment - Details

When the detectors at A and B record a coincidence, this means that both have
been triggered within a time interval of at most 10 ns, the resolving time of the
circuit. Could it happen that, within this interval, A sends to B a signal capable
of influencing the response of B? In the most favorable case, such a signal would
travel with the speed of light in vacuum, which according to relativity theory is
the upper limit on the propagation speed of information, and thereby of energy.
To cover the distance AB, which is 12 m in the figure, such a signal would need
40 ns. This is too long by at least 30 ns, and rules out any causal links between
A and B in the sense of classical physics. One says that the interval between A
and B is space-like.

One of the advantages of the Orsay experiment is that it uses a very strong
light-source, allowing sufficient distance between the detectors A and B while

1313
still preserving reasonable counting rates. By increasing the distance AB step
by step, Aspect could check that the correlation persists, even when the interval
between A and B becomes space-like. This is the check that guarantees that
the two-photon system is non-separable irrespective of the distance AB.

It has become the custom to speak of the principle of Einsteinian separability in

order to denote the absence of correlations between two events separated by a
space-like interval. This is the principle that the Orsay experiment invites us to
reconsider, even though our minds, used to the world at the macroscopic level,
find it difficult to conceive of two microscopic photons 12 m apart as a single
indivisible object.

The Final Stage of the Experiment at Orsay (1983)

Though the results of the first Orsay experiment are unarguable and clear-cut,
the conclusion they invite is so startling that one should not be surprised at
the appearance of a last-ditch objection, which as it happens gave the exper-
imenters a great deal of trouble. In the preceding section we discussed the
possible role of interactions between A and B operating after the decay at S,
and duly eliminated the objection. But one can also ask whether correlations
might be introduced through an interaction operating before the decay. We
could imagine that the decay itself is preconditioned by the setting of detectors
A and B, such influences taking effect through the exchange of signals between
the detectors and the source. No such mechanism is known a priori, but we
do know that, if there is one, then Einsteinian non-separability would cease to
be a problem, because the mechanism could come into action long before the
decay, removing any reason for expecting a minimum 30 ns delay. Though such
a scenario is very unlikely, the objection is a serious one and must be taken into
account; to get around it, the experimenter must be able to choose the orienta-
tion of the detectors A and B at random after the decay has happened at S. In
more picturesque language, we would say that the two photons must leave the
source without knowing the orientations of the polarizers A and B. Briefly put,
this means that it must be possible to change the detector orientations during
the 20 ns transits over SA and SB.

The solution adopted at Orsay employs periodic switching every 10 ns. These
changes are governed by two independent oscillators, one for channel A and one
for channel B. The oscillators are stabilized, but however good the stabilization
it cannot eliminate small random drifts that are different in the two channels,
seeing that the oscillators are independent. This ensures that the changes of
orientation are random even though the oscillations are periodic, provided the
experiment lasts long enough (1 to 3 hours).

The key element of the second Orsay experiment is the optical switch shown in
Figure 16.26 below.

1314
 Figure 16.26: Second Orsay Experiment - Optical Switch

In a water tank, a system of standing waves is produced by electro-acoustic

excitation at a frequency of 25 M Hz(corresponds to 10 ns between switchings).

The fluid keeps changing from a state of perfect rest to one of maximum agita-
tion and back again. In the state of rest, the light beam is simply transmitted,
In the state of maximum agitation, the fluid arranges itself into a structure of
parallel and equidistant plane layers, alternately stationary (nodal planes) or
agitated (antinodal planes). Thus, one sets up a lattice of net-like diffracting
planes; the diffracted intensity is maximum at the so-called Bragg angles, just as
in scattering from a crystal lattice. Here the light beam is deviated through 102
radians (the angles in the figure are exaggerated for effect). The two numerical
values, 25 M Hz and 102 radians, suffice to show the magnitude of the technical
achievement. With the acoustic power of 1 watt, the system functions as an
ideally efficient switch.

The second Orsay experiment (using optical switches) is sketched in Figure

16.27 below.

Figure 16.27: Second Orsay Experiment using Optical Switches

In this set-up, the photons a and b leave S without knowing whether they will
go, the first to A1 or A2 and the second to B1 or B2 .

1315
The second experiment is less precise than the first, because the light beams
must be very highly collimated in order to ensure efficient switching. Neverthe-
less, its results exhibit an unambiguous violation of Bells inequality, reaching
5 standard deviations at the peak; moreover the results are entirely compatible
with the predictions of quantum mechanics.

16.6 The Principle of Non-Separability

Experiment has spoken. Half a century after the Como conference, Bohrs in-
terpretation once again beats Einsteins, in a debate more subtle and also more
searching. There were two conflicting theories:

Einstein Bohr
hidden variables quantum mechanics
realist positivist
deterministic probabilistic
separable non-separable

Table 16.2: Two Conflicting Theories

The violation of the BCHSH inequality argues for Bohrs interpretation, all the
more so as the measured values of hi are in close agreement with the predic-
tions of quantum mechanics.

It remains to ask oneself just why hidden-variable theories do fail. Of the three
basic assumptions adopted by such theories, namely realism, determinism, and
separability, at least one must be abandoned. In the last resort, it is separa-
bility that seems to be the most vulnerable assumption. Indeed, one observes
experimentally that the violation of the BCHSH inequality is independent of the
distance between the two detectors A and B, even when this distance is 12 m
or more. There are still die-hard advocates of determinism, who try to explain
non-separability through non-local hidden variables. Such theories, awkward
and barely predictive, are typically ad hoc, and fit only a limited number of
phenomena. They are weakly placed to defend themselves against interpreta-
tions furnished by quantum mechanics, which have the virtues of simplicity,
elegance, efficiency, and generality, and which are invariably confirmed by ex-
periment.

The principle of Einsteinian separability asserts that there are no correlations

between two phenomena separated by a space-like interval. In other words, no
interaction can propagate faster than light in vacuum. In an EPR scenario
this principle must be abandoned, and replaced by a principle asserting non-
separability:

1316
 in a quantum system evolving free of external
perturbations,and from well-defined initial
conditions, all parts of thesystem remain
correlated, even when the interval between
them is space-like

This assertion reflects the properties of the state vector of a quantum system.
For an EPR system, the state vector after the decay of the source reads

1
|i = (|xA , xB i + |yA , yB i) (16.78)
2

This expression combines the elements A and B in a non-separable manner,

which is what explains the observed correlations. The truth is that all this
has been well known ever since the beginnings of quantum mechanics, with the
concept of the electron cloud as the most telling illustration. It is for instance
hard to imagine separability between the 92 electrons of a uranium atom. What
is new is that quantum mechanics, considered hitherto as a microscopic theory
applicable on the atomic scale, is now seen to apply to a two-particle system
macroscopically, on the scale of meters. The truly original achievement of As-
pects experiment is the demonstration of this fact.

Quantum objects have by no means exhausted their capacity to astonish us by

their difference from the properties of the macroscopic objects in our everyday
surroundings. In the preceding sections we saw that a photon can interfere with
itself and we have shown that two photons 12 m apart constitute but a single
object. Thus, it becomes ever more difficult to picture a photon through analo-
gies with rifle bullets, surface waves in water, clouds in the sky, or with any
other object of our familiar universe. Such partial analogies fail under attempts
to make them more complete, and through their failure we discover new prop-
erties pertaining to quantum objects. The only fruitful procedure is to follow
the advice of Niels Bohr, namely, to bend ones mind to the new quantum con-
cepts until they become habitual and thereby intuitive. Earlier generations of
physicists have had to face similar problems. They had to progress from Aristo-
tles mechanics to Newtons, and then from Newtons to Einsteins. The same
effort is now required of us, at a time favorable in that, by mastering the EPR
paradox, quantum mechanics has passed a particularly severe test with flying
colors.

From this point of view, the principle of non-separability seems as important as

the principle of special relativity , and Aspects experiment plays the same role
now that the Michelson-Morley experiment played then.

1317
16.7 An Example and a Solution - Bells Theo-
rem with Photons
Two photons fly apart from one another, and are in oppositely oriented circularly
polarized states. One strikes a polaroid film with axis parallel to the unit vector
a, the other a polaroid with axis parallel to the unit vector b. Let P++ (a, b) be
the joint probability that both photons are transmitted through their respective
polaroids. Similarly, P (a, b) is the probability that both photons are absorbed
by their respective polaroids, P+ (a, b) is the probability that the photon at
the a polaroid is transmitted and the other is absorbed, and finally, P+ (a, b)
is the probability that the photon at the a polaroid is absorbed and the other
is transmitted.

The classical realist assumption is that these probabilities can be separated:

Z
Pij (a, b) = d ()Pi (a, )Pj (b, ) (16.79)

where i and j take on the values + and , where signifies the so-called hidden
variables, and where () is a weight function. This equation is called the
separable form.

The correlation coefficient is defined by

C(a, b) = P++ (a, b) + P (a, b) P+ (a, b) P+ (a, b) (16.80)

and so we can write

Z
C(a, b) = d ()C(a, )C(b, ) (16.81)

where

C(a, ) = P+ (a, ) P (a, ) (16.82)

C(b, ) = P+ (b, ) P (b, ) (16.83)

It is required that
(a) () 0
R
(b) d () = 1

(c) 1 C(a, ) 1 , 1 C(b, ) 1

The Bell coefficient

B = C(a, b) + C(a, b0 ) + C(a0 , b) C(a0 , b0 ) (16.84)

combines four different combinations of the polaroid directions.

1318
 (1) Show that the above classical realist assumptions imply that |B| 2.
 2
(2) Show that quantum mechanics predicts that C(a, b) = 2 a b 1.

(3) Show that the maximum value of the Bell coefficient is 2 2, according to
quantum mechanics.
(4) Cast the quantum mechanical expression for C(a, b) into a separable form.
Which of the classical requirements, (a), (b), or (c) above is violated?

Solution
(1) With the separability assumption, we have (16.81)
Z
C(a, b) = d ()C(a, )C(b, )

It follows that the Bell coefficient can be written in the form

B = C(a, b) + C(a, b0 ) + C(a0 , b) C(a0 , b0 )
Z
= d()[C(a, )(C(b, ) + C(b0 , ))

+ C(a0 , )(C(b, ) C(b0 , ))] (16.85)

Since |C(a, )| 1, |C(a0 , )| 1 and () 0, we have
Z    
|B| d() C(b, ) + C(b0 , ) + C(b, ) C(b0 , ) (16.86)
   

Now suppose that for a given , CM () is the maximum and Cm () is the

minimum of C(b, ) and C(b0 , ), so that CM () Cm (). Then
Z
|B| d() (|CM () + Cm ()| + CM () Cm ()) (16.87)

There are two cases to consider.

For the case CM () 0, we have |CM () + Cm ()| = CM () + Cm () so that

Z
|B| d() (CM () + Cm () + CM () Cm ())
Z Z Z
= 2 d()CM () 2 d() |CM ()| 2 d() = 2

For the case CM () < 0, we have |CM () + Cm ()| = CM () Cm () so that

Z
|B| d() (CM () Cm () + CM () Cm ())
Z Z Z
= 2 d()(Cm ()) 2 d() |Cm ()| 2 d() = 2

1319
Thus, in all cases |B| 2.

(2) A photon, traveling in the ydirection, might have right- or left-handed

circular polarization. The corresponding quantum states are written |Ri and
|Li respectively. These circular polarization states can be expressed as coherent
superpositions of linearly polarized states in the z and x directions:

1 1
|Ri = (|zi + i |xi) , |Li = (|zi i |xi) (16.88)
2 2

Under a rotation of the coordinate axes by an angle about the y direction,

|Ri ei |Ri and |Li ei |Li or equivalently

|z 0 i
      
|zi cos sin |zi
= (16.89)
|xi |x0 i sin cos |xi

If each photon is in a state of right-handed circular polarization, we write the

corresponding state vector as |R1 i |R2 i. However, since the photons are moving
in opposite directions, one along the positive, and the other along the negative
y axis, it follows that the actual directions in which the electric fields rotate, in
time, in the vicinity of the two photons, are opposed to one another. The same
holds for the state |L1 i |L2 i, corresponding to each photon being in a state of
left-handed circular polarization.

The linear combination of these two states,

1
|EP Ri = (|R1 i |R2 i + |L1 i |L2 i) (16.90)
2

corresponds to the more general situation in which the photons are in oppositely
oriented states of circular polarization, where the sense of this polarization is
not specified. We can write this entangled or Einstein-Podolsky-Rosen state in
the form
1
|EP Ri = (|z1 i |z2 i |x1 i |x2 i) (16.91)
2
which is a superposition of states of linear polarization.

Suppose now that a measurement of linear polarization is made on photon 1 in

the z direction, and of photon 2 in the z direction, that is, the z direction after
a rotation of the axes about the y axis. The probability amplitude associated
with this measurement on the EPR state is
1
hEP R | z1 z20 i = (hz1 | hz2 | hx1 | hx2 |) (|z1 i (cos |z2 i sin |x2 i))
2
1
= cos (16.92)
2

1320
where we have used hz1 | x1 i = 0. The probability that photon 1 is found to
have linear polarization in the direction z, and photon 2 in the direction z is
2 1
P++ (a, b) = |hEP R | z1 z20 i| = cos2 (16.93)
2
where we have assumed that a is in the z direction and b is in the z direction.

Suppose next that the linear polarization of the linear polarization of photon 1
were measured in the x direction, and that of photon 2 again in the z direction.
The probability amplitude is
1
hEP R | x1 z20 i = (hz1 | hz2 | hx1 | hx2 |) (|x1 i (cos |z2 i sin |x2 i))
2
1
= sin (16.94)
2
If photon 1 has polarization in the x direction, then it will not be transmitted
by a polarizer in the z direction - it will be absorbed. Hence,
2 1
P+ (a, b) = |hEP R | x1 z20 i| = sin2 (16.95)
2
Similarly,
2 1
P+ (a, b) = |hEP R | z1 x02 i| = sin2 (16.96)
2
2 1
P (a, b) = |hEP R | x1 x02 i| = cos2 (16.97)
2
The correlation coefficient is then
C(a, b) = P++ (a, b) + P (a, b) P+ (a, b) P+ (a, b)
= cos2 sin2 = 2 cos2 1 = cos 2 (16.98)
Since the unit vectors a and b are at an angle with respect to one another, it
follows that a b = cos and therefore
C(a, b) = 2 cos2 1 = 2(a b)2 1 (16.99)
0
(3) Suppose that the angle between the vectors a and a is x/2, between a and
b is y/2 and between b and b0 is z/2. Then the angle between a0 and b0 is
(x + y + z)/2 and according to quantum mechanics, the Bell coefficient has the
form
B = cos x + cos y + cos z cos(x + y + z) (16.100)
This function has extrema when
B
= sin x + sin(x + y + z) = 0
x
B
= sin y + sin(x + y + z) = 0
y
B
= sin z + sin(x + y + z) = 0
z

1321
or
sin x = sin y = sin z = sin(x + y + z) (16.101)
This has the solution

x = y = z and 3x = x x = /4 (16.102)

For this extremum

3 3 1
B = 3 cos cos = + =2 2 (16.103)
4 4 2 2
This is a maximum, since at this point

2B 2B 2B 3
= = = cos + cos = 2<0
x2 y 2 z 2 4 4

(4) Let the vector a be at an angle a with respect to some direction in the xz
plane, and let b be at an angle b with respect to the same direction. Then

C(a, b) = cos 2(a b )

= cos 2a cos 2b + sin 2a sin 2b
Z
= d ()C(a, )C(b, ) (16.104)

and with the assignments

() = ( + 1) + ( 1)
C(a, 1) = cos 2a , C(a, 1) = sin 2a
C(b, 1) = cos 2b , C(b, 1) = sin 2b

we then see that

() 0
1 C(a, ) , C(b, ) 1for = 1

but Z
d () = 1 + 1 = 2

so that the normalization condition (b) is violated.

16.8 Non-Locality, EPR and Bell - a last time

As we discussed earlier, the second major problem confronting hidden variables
and possessed properties was first understood in the context of the EPR para-
dox and then reinforced by the Bell inequalities. Let us look back at these ideas
in light of the above discussions.

1322
The original EPR analysis was rather complex in a technical sense and most dis-
cussions now use a simpler version due to Bohm. He considered a particle whose
decay produces two spin1/2 particles whose total spin angular momentum is
zero. These particles move away from each other in opposite directions, and the
components of their spins along various directions are subsequently measured
by two observers, N and L, say. The constraint on the total spin means that if
both observers agree to measure the spin along a certain direction n , and if N
measures +~/2, then L will necessarily get the result ~/2, and if N measures
~/2, then L will necessarily get the result +~/2.

There are no surprises if such correlations are analyzed in the context of clas-
sical physics. If one particle emerges from the decay with its internal angular
momentum vector pointing along some particular direction, then because of
conservation of angular momentum, the second particle is guaranteed to emerge
with its spin vector pointing in the opposite direction. Thus, the 100% anti-
correlations found in the measurements made by the two observers are simply
the result of the fact that both particles possess actual, and (anti-)correlated,
values of internal angular momentum and this is true from the time they emerge
from the decay to the time the measurements are made. There are no paradoxes
here, and everything is in accord with the simple realist view of classical physics.

The situation in quantum theory is radically different. Suppose first that the
measurements are made along the zaxes of the two observers. The spin part
of the state of the two particles can be written in terms of the associated eigen-
vectors as
1
|i = (|i |i |i |i) (16.105)
2
where, for example, |i |i is the state in which particles 1 and 2 have spin +~/2
and ~/2 respectively. Thus

Sz |i = +~/2 |i , Sz |i = ~/2 |i (16.106)

The pragmatic or instrumentalist interpretation of the entangled state |i is

straightforward. If, in a series of repeated measurements by N, a selection is
made of the pairs of particles for which the measurement of particle gave spin-
up, then - with probability one - a series of measurements by L on her particle
in these pairs will yield spin-down. Similarly, if N finds spin-down then, with
probability one, L will find spin-up. This correlation can be explained by say-
ing that the measurements by N(computed with the operator Sz I) cause a
reduction of the state vector from |i to |i |i or |i |i respectively according
to whether the spin-up or the spin-down result is selected. This new state is an
eigenstate of the operator I Sz associated with the second particle, and with
an eigenvalue that is the opposite of the result obtained by N.

This description is acceptable within the confines of the pragmatic approach

to quantum theory, but difficulties arise if one tries to enforce a more realist

1323
interpretation of the above entangled state.

The obvious question is how the information about each observers individual
results gets to the other particle to guarantee that the result obtained by the
second observer will be the correct one.

One might be tempted to invoke the reasoning of classical physics and argue
that both particles possess the appropriate value all the time.

However, the only way in standard quantum theory of guaranteeing that a cer-
tain result will be obtained is if the state is an eigenvector of the observable
concerned. But the state |i above is not of this type. In fact, it displays
the typical features of quantum entanglement - it is a superposition of states.
Any attempt to invoke a hidden variable resolution will have to cope with the
implications of the Kochen-Specker theorem.

There is also a question of whether this picture is compatible with special rela-
tivity. If the measurements by the two observers are space-like separated (which
can be easily arranged) then which of them makes the first measurement and
hence, in the standard interpretation, causes the state-vector reduction is clearly
reference-frame dependent.

The problem is compounded by considering what happens if the observers decide

to measure, say, the x-component of the spins, rather than the z-components.
The state above can now be written in terms of Sx eigenvectors as
1
|i = (|i |i |i |i) (16.107)
2
where |i and |i correspond to eigenvalues +~/2 and ~/2 respectively of
the operator Sx .

In one sense, this new entangled state is what might have been expected, and
confirms that there is the same type of 100% anti-correlation between Sx mea-
surements as that found for the observable Sz . Indeed, this argument can be
generalized to show that for any unit vector n, the entangled state can be
rewritten as a sum of two anti-correlated terms containing eigenvectors of the
projection n S of the spin operator along n. Thus, if one adopts the classical
type argument, one is obliged to conclude that both particles possessed exact
values of spin along any axis from the moment they left the decay. This might
not be easy to reconcile with the uncertainty relations associated with the an-
gular momentum commutators.

EPR considered these issues, and concluded that the difficulties could be re-
solved in one of only two ways:
1. When N makes her measurement, the result communicates itself at once

1324
 in some way to particle 2, and converts its state into the appropriate
eigenvector.

or

2. Quantum theory is incomplete and provides only a partial specification of

the actual state of the system.

In contemplating the first possibility it must be appreciated that the two par-
ticles may have moved a vast distance apart before the first measurement is
made and, therefore, any at once mode of communication would be in violent
contradiction with the spirit (if not the law) of special relativity. It is not
surprising that Einstein was not very keen on this alternative! An additional
objection involves the lack within quantum theory itself of any idea about how
this non-local effect is supposed to take place, so in this sense the theory would
be incomplete anyway.

EPR came to the conclusion that the theory is indeed incomplete, although they
left open the correct way in to complete it. One natural path is to suppose that
there exist hidden variables whose values are not accessible to measurement in
the normal way but which determine the actual values of what we normally
regard as observables - in the same way as do the microstates in classical sta-
tistical physics.

However, it is not a trivial matter to construct a hidden variable theory that

reproduces all of the empirical results of quantum mechanics (which are experi-
mentally correct!). In particular, such a theory would need to explain why it is
that certain observables are incompatible (those with non-vanishing commuta-
tor) in the sense that one cannot prepare a state of the system that violates the
predictions of the uncertainty relations. In addition, there is the need to come
to terms with the implications of the Kochen-Specker theorem.

Hidden variable theories capable of reproducing the results of conventional quan-

tum theory do in fact exist (Bohm for example) but they exhibit a non-locality
which is every bit as peculiar as that discussed above. One might think that this
is a deficiency of these particular theories and that others might exist without
this problem.

However, as we discussed earlier and will review here again, a very famous result
of John Bell shows that this is not possible, that is, any hidden-variable theory
that exactly replicates the results of quantum theory will necessarily possess
striking non-local features.

This result is of major importance in understanding and appreciating the con-

ceptual challenge posed by quantum theory.

1325
16.8.1 The Bell Inequalities
As with the Kochen-Specker result, the non-locality property we are about to
discuss is not just a feature of hidden variables theories. It applies to any realist
interpretation of quantum theory in which it is deemed meaningful to say that
an individual system possesses values for its physical quantities in a way that
is analogous to that in classical physics.

We will derive an inequality that is satisfied by certain correlation functions in

any such theory which is also local. We will then see that the predictions (which
are experimentally correct) of quantum theory violate this inequality.

The considerations of EPR were concerned with two observers who make mea-
surements along the same axis. Bell found his famous inequalities by asking
what happens if the observers measure the spin of the particles along different
axes. In particular, we consider a pair of unit vectors a and a0 for one observer
and another pair b and b0 for the other observer.

Now suppose a series of repeated measurements is made on a collection of sys-

tems whose quantum state is described by the entangled state vector |i. For
example, we could look at a series of decays, each of which produces a pair of
particles with zero total spin angular momentum. The central realist assump-
tion we are testing is that each particle has a definite value at all times for any
direction of spin. We let an denote 2/~ times the value of a S possessed by
particle 1 in the nth element of the collection. Thus an = 1 if a S = ~/2.

The key ingredient in the derivation of the Bell inequalities is the correlation be-
tween measurements made by the two observers along these different directions.
For directions a and b this is defined by

N
1 X
C(a, b) := lim an bn (16.108)
N N
n=1

and similarly for the other directions. Note that if the results are always totally
correlated(spins always in the same direction) then C(a, b) = +1, whereas if they
are totally anti-correlated(spins always in opposite directions) we get C(a, b) =
1.

Now look at the quantity

gn := an bn + an b0n + a0n bn + a0n b0n (16.109)

For any member n of the collection, each term in this sum will take on the value
+1 or 1 . Furthermore, the fourth term on the right hand side is equal to the
product of the first three (because (an )2 = 1 = (bn )2 ). Then thinking about the
various possibilities shows that gn can take on only the values 2. Therefore,

1326
the right hand side of the expression
 N
  N N N N

1 X  1 X 1 X 1 X 1 X 
gn  =  an bn + an b0n + a0n bn a0n b0n 
   
N  N N N N


n=1 n=1 n=1 n=1 n=1

representing the average value of gn must be less than or equal to 2. Thus in

the limit as N , we get
 
C(a, b) + C(a, b0 ) + C(a0 , b) C(a0 , b0 ) 2 (16.110)
 

which is one of the famous Bell inequalities.

It is important to emphasize the only assumptions that have gone into proving
this inequality are:
1. For each particle it is meaningful to talk about the actual values of the
projection of the spin along any direction.
2. There is locality in the sense that the value of any physical quantity is not
changed by altering the position of a remote piece of measuring equipment.
This means that both occurrences of an in the expression for the average value
of gn have the same value, that is, they do not depend on the direction (b or
b0 ) along which the other observer chooses to measure the spin of particle 2. In
particular, we are ruling out the type of context-dependent values that arose in
our discussion of the Kochen-Specker theorem.

We will now show that the predictions of quantum theory violate this inequality
over a range of directions for the spin measurements. The quantum mechanical
prediction for the correlation between the spin measurements along axes a or b
is  2
2
C(a, b) := h| a S(1) b S(2) |i (16.111)
~
where S(1) and S(2) are the spin operators for particles 1 and 2 respectively,
and the tensor product is as we defined earlier in this chapter. Since the total
angular momentum of the entangled vector |i is zero, it is invariant under the
unitary operators which generate rotations of the coordinate systems.

This means that C(a, b) is a function of a b = cos ab only and, hence, there is
no loss of generality in assuming that a points along the zaxis and that b lies
in the x z plane. Then the expression for C(a, b) becomes

C(a, b) = h| 1z (2z cos ab + 2x sin ab ) |i (16.112)

It is then straightforward to show that

C(a, b) = cos ab (16.113)

1327
Now we restrict our attention to the special case in which (1) the four vectors
a, a0 , b, b0 are coplanar and (2) a or b are parallel and (3) ab0 = a0 b = say.
Then the Bell inequality will be satisfied provided that

|1 + 2 cos cos 2| 2 (16.114)

This is violated for all values of between 0 and 90 . This means that if the
predictions of quantum theory are experimentally valid in this region then any
idea of systems possessing individual values for observables must necessarily in-
volve an essential non-locality. This applies in particular to any hidden variable
theory that is completely consistent with the results of quantum theory. Thus,
the important questions are:
1. Are the Bell inequalities empirically violated?
2. If so, are such violations in accord with the predictions of quantum theory?
In many experiments over the last two decades, the overwhelming conclusion
is that the predictions of quantum theory are vindicated and so we are obliged
either to stick with a pragmatic approach or a strict instrumentalist interpreta-
tion or else to accept the existence of a strange non-locality that seems hard to
reconcile with our normal concepts of spatial separation between independent
entities.

16.9 Bayesian Probability in QM

We turn to Bayesian probability arguments to deal with a realist.

16.9.1 Using Bayesian Ideas in Analysis of Experiments

In actual experimental tests, there are no infinite ensembles for accurate mea-
surements of mean values. Experimental physicists perform a finite number of
tests and then they state their results accompanied by a confidence level.

The real problem of theory versus experimental analysis is of a different nature

however.

I am a theorist and I believe that quantum mechanics gives a reliable description

of nature. I have a friend, however, who is a local realist.

We only have a finite number of trials of a Bell inequality experiment at our

disposal.

How many tests are needed to make my realist friend feel uncomfortable?

The problem is not whether the validity of a Bell inequality can be salvaged
by invoking clever loopholes, as some realists try to trick us into, but whether

1328
there can be any local realistic theory that reproduces the experimental results.

To simplify the discussion, I will assume that there are ideal detectors and that
the rate at which particles are produced by the apparatus is perfectly known.

Experimental Results Change Beliefs

First, we consider a yes-no test.

Quantum mechanics(QM) predicts that the probability of the yes result is q

and an alternative local realistic(LR) theory predicts a probability r.

An experimental test is performed n times and yields m yes results.

What can we infer about the likelihood of the two theories?

The answer is given by Bayes theorem

P (B|C)
P (B|A C) = P (A|B C) (16.115)
P (A|C)

Denote by p 0q = P (q|I) and p 0r = P (r|I) the prior probabilities that we assign

to the validity of the two theories. These are subjective probabilities expressing
our personal beliefs.

For example, if my friend is willing to bet 100 to 1 (for example) that LR is

correct and QM is wrong, then

p 0r
= 100 (16.116)
p 0q

The question is: how many experimental tests are needed to change my friends
opinion to
p 00r
= 0.01 (16.117)
p 00q
say, before he is driven to bankruptcy. This is a reversal (in belief) by a factor
of 104 .

In this case, P (r|{m, n} I) = p 00r is the new prior probability for my friend
after the experiments are finished and similarly we have P (q|{m, n} I) = p 00q .

If we define

Er = P ({m, n}|q I) , Eq = P ({m, n}|q I) (16.118)

which are just the probabilities of the experimentally found result (the actual
data - m successes in n trials) according to the two theories.

1329
These follow from the binomial theorem
n! n!
Er = rm (1 r)nm , Eq = q m (1 q)nm (16.119)
m!(n m)! m!(n m)!

It then follows from Bayes theorem that

P ({m, n}|r I)P (r|I) = P (r|{m, n} I)P ({m, n}|I) (16.120)

P ({m, n}|q I)P (q|I) = P (q|{m, n} I)P ({m, n}|I) (16.121)

or that

P ({m, n}|r I)P (r|I) P ({m, n}|q I)P (q|I)

= P ({m, n}|I) =
P (r|{m, n} I) P (q|{m, n} I)
or
P ({m, n}|r I)P (r|I) p 0 Er P (r|{m, n} I) p 00
= r0 = = r00 (16.122)
P ({m, n}|q I)P (q|I) p q Eq P (q|{m, n} I) pq

We define the ratio

P ({m, n}|q I) Eq  q m  1 q nm

= =D= (16.123)
P ({m, n}|r I) Er r 1r

as the confidence depressing factor for the hypothesis LR with respect to the
hypothesis QM.

16.9.2 Simple Example

Suppose that we flip coins and the yes-no question is: Did the coin come up
heads?

I, the theorist, will assume that the coin is unbiased and that therefore q = 0.5
and m = n/2 (assuming that I am correct). We then have
 n/2  n/2  n  n/2
1 1 1 1
D= = (16.124)
2r 2(1 r) 2 r(1 r)

Since we want 104 , we find

 n
5 5 4 4
= 104 n log = 4 n = 5 = 16 (16.125)
3 3 log 3 0.22

So that it would take only 16 coin flips to reverse my untrusting friends belief.

Now let us return to the Bell inequality.

1330
 Figure 16.28: Three Detector Experiment

16.9.3 Simple Ideas

Let us consider a device that has three widely separated detectors each of which
has two switch settings as shown in Figure 16.28 above.
When a detector is triggered it flashes either red or green. The detectors are
far apart from the source, there are no connections between the detectors and
no connections between the source and the detectors other than those mediated
by the group of three particles (as shown) that originate at the source and fly
away, one to each detector.

A run of the experiment consists of setting the switch on each detector to one of
its two positions (labeled 1 or 2), pressing a button at the source (to release a trio
of particles, one aimed at each detector), and recording the color subsequently
flashed at each detector.

There are eight possible switch settings:

111 112 121 122 211 212 221 222 (16.126)

We consider only the data acquired for four of the eight possible switch settings,
namely, those in which the number of detectors set to 1 is odd.

111 122 212 221 (16.127)

The other set

112 122 212 222 (16.128)

1331
will lead to similar results (1 2). As shown in Figure 16.28 we call the detec-
tors A, B, and C, and specify pertinent facts about them by listing three pieces
of information (switch settings or colors flashed) in that order.

If we run the experiment many times, then the observational results are the
following. If just one detector is set to 1 (and the others to 2), then an odd
number of red lights always flash, that is, either all three detectors flash red or
there is one red flash and two green ones.

If all three detectors are set to 1, then an odd number of red lights is never
observed to flash - either two of the three flash red or all three flash green.

This is summarized by the table:

Settings Result 1 Result 2 Result 3 Result 4

111 GGG GRR RGR RRG
122 RGG GRG GGR RRR
212 RGG GRG GGR RRR
221 RGG GRG GGR RRR

Table 16.3: Results

All four outcomes are equally likely in each case(this particular detail is not
important).

We will discuss a real, physical system that exhibits this behavior later.

Let us set aside, for the moment, the 111 case and consider the 122, 212, and
221 cases in which just one detector is set to 1. Because an odd number of red
lights always flash in any of these three cases, whenever the switches are so set
we can predict with certainty what one of the three detectors will do in a run,
merely by noting what happens to the other two. For should the other two flash
the same color (RR or GG), then the third will have to flash red, but should
the other two flash different colors (RG or GR), then the third will have to flash
green.

Now we follow the path set out by EPR to draw an inference that will seem
inescapable. Along the way we will use the so-called EPR reality criterion.

Since there are no direct connections between the detectors, their behavior can
only be coordinated due to the fact that all three are triggered by particles that
came from a common source. This fact and this fact alone must contain the
explanation for why we can learn in advance what color will flash at a given
detector, say A, from measurements made far away at B and C. Information
telling the detector at A what color to flash in order to maintain the observed

1332
consistency with the colors flashed at B and C must somehow be encoded in the
particle that triggers A. Since that particle could indeed have been coordinated
with the particles that triggered B and C when all three were back at their
common source, this explanation seems both inevitable and entirely reasonable.

We can apply this reasoning to any one of the three detectors (by moving it
farther from the source so that before it flashes we have had the opportunity
to observe what colors flash at the other two). We conclude that in each run
of the experiment each particle must be carrying to its detector instructions on
what color to flash, and that an odd number of the particles must specify red.

Thus, for a given choice of the switch settings (say 122) the particles heading for
detectors A, B, and C must respectively be carrying instructions RRR, RGG,
GRG, or GGR, but never GRR, RGR, RRG, or GGG.

Which of the four allowed groups of instructions they collectively carry is re-
vealed only when the lights flash. All of the above reasoning applies equally
well, of course, to 212 and 221 runs.

In the absence of connections between the detectors and the source, a particle
has no information about how the switch of its detector will be set until it ar-
rives there. Since in each run any detector might turn out to be the one set to
1 or one of the ones set to 2, to preserve the perfect record of always having
an odd number of red flashes in 122, 212, and 221 runs, it would seem to be
essential for each particle to be carrying instructions for how its detector should
flash for either of the two possible switch settings it might find upon arrival.

The existence of instructions of this sort is the EPR reality criterion.

The instructions carried by each particle can be symbolized by a pair of letters

1 R R G or G
2R G R G
would result in RRR if the switch settings were 122, GGR for 212, and GRG
for 221.

Since each of the three possible switch settings result in an odd number of red
flashes, this is indeed a legal set of instructions.

An example of an illegal set of instructions is

R R G
G R R
for this gives an even number of red flashes GRR for the switch setting 212,
which is never observed.

1333
Since there are eight ways the lights can flash, namely,

RRR RRG RGG RGR GRR GRG GGR GGG (16.129)

the total number of possible instruction sets is 8 8 = 64.

It is not hard to enumerate all the legal (odd number of red flashes) instruction
sets.

First note that three of the six positions in a legal instruction set corresponding
to any one of the three choices 122, 212, or 221 for the switch settings, must
contain an odd number of Rs, since that particular setting might be encountered
in any run, and since only odd numbers of red flashes are ever observed. Thus,
the only possible entries for the positions corresponding to the switch settings
122 are (leaving blank the entries not relevant to those three settings):

ABC ABC ABC ABC

1 R R G G
2 R R GG RG GR

so that 122 gives RRR, RGG, GRG, or GGR independent of the other entries.

We can next count the way to fill in the blanks in these four forms so as to
produce the correct data for switch settings 221. Since each of the four already
specifies the color flashed at detector B for setting 2, namely, R G R G, to
ensure that any 221 run produces an odd number of red flashes there are only
two choices available for the other two (A and C) unspecified 221 entries for
each of the four forms: RR or GG if the specified entry is R and RG or GR if
the specified entry is G so that we have

ABC ABC ABC ABC

1 RR RG GG GR
2 RRR RGG GRG GGR
ABC ABC ABC ABC
1 RR RR GR GG
2 GRR GGG RRG RGR

This raises the number of possible forms to eight, each of which leaves only the
entry for setting 1 at the detector B unspecified. But that entry is now entirely
determined by the entries at settings 2 for detectors A and C (having to be R,

1334
if the latter two entries are the same color and G, if they are different).
ABC ABC ABC ABC
1 RRR RGG GRG GGR
2 RRR RGG GRG GGR
ABC ABC ABC ABC
1 RGR RRR GGR GRG
2 RRR GGG RRG RGR
They are arranged in the same horizontal order as the forms in (1), with the
two possibilities for each form placed directly above one another. It is easy to
check explicitly that each instruction set (2) does indeed give an odd number of
red flashes when a single detector is set to 1.

122 gives RRR RGG GRG GGR

RRR RGG GRG GGR

221 gives RRR RGG GRG GGR

GRG GGR RRR RGG

212 gives RRR RGG GRG GGR

GGR GRG RGG RRR

Clearly, (2) represents the eight legal sets.

Now, finally, we consider the fourth type of run, in which all three detectors are
set to 1, and an odd number of red flashes is never observed.

The above instruction sets must determine the outcomes of these runs as well.
For who is to prevent somebody from flipping the two switches set to 2 over to
1, just before the particles arrive?

An inspection of the upper rows in (2) reveals that every one of the eight allowed
instruction sets results in an odd number of red flashes when all three switches
are set to 1.

If the instruction sets existed, then 111 runs would always have to produce an
odd number of red flashes. But they never do.

Thus, a single 111 run suffices all by itself to give data inconsistent with the
otherwise compelling inference of instruction sets.

Here the instruction sets(realistic theory) require an odd number of red flashes
in every 111 run, but quantum mechanics(experiment) prohibits an odd number
of red flashes in every 111 run.

1335
Something is wrong with the EPR idea of instruction sets or EPR reality.

The Quantum Mechanical Explanation

Here is how the device works. What emerges from the source are three spin1/2
particles (a, b, and c) in a spin state whose structure is given below. The parti-
cles fly apart to the detectors in the horizontal plane. We define the zdirection
for each particle to be along the line of flight. The detectors contain Stern-
Gerlach magnets which measure the vertical (x) component of the spin when
the switch is set to 1 and the horizontal component (y) perpendicular to the line
of flight when their switch is set to 2. They are set so that we get red flashes
for spin-up, and green flashes for spin-down.

Let us describe a spin state that produces the remarkable correlations (GHZ-
state) described earlier.

We measure angular momentum for each particle in units of ~/2 so that the spin
operators for each particle can be taken to be the Pauli matrices. Now consider
the three commuting Hermitian operators

xa yb yc , ya xb yc , ya yb xc (16.130)

They commute because all pairs of the six spin operators out of which they are
constructed commute, except for those associated with the x and y components
of the spin of a single particle, which anticommute. This does not cause any
trouble, however, because converting the product in one order to the product
in then other order always involves and even number of such anticommuting
exchanges.

Being commuting and Hermitian, the three operators above can be provided
with simultaneous eigenvectors. Since the square of each operator is the iden-
tity, the eigenvalues of each can only be 1.

The actual spin state that produces the remarkable correlations (the Greenberger-
Horne-Zeilinger or GHZ-state) is described by

1
|GHZi = (|1, 1, 1i |1, 1, 1i) (16.131)
2
where 1 specifies spin-up or spin-down along the appropriate zaxis.

For simplicity in the following argument, here we pick the state with all three
eigenvalues equal to +1, which preserves the symmetry among the particles. The
argument works for any such symmetric state and for any linear combination of
such states as in the above state.

Since the components of the spin vectors of different particles commute, we can

1336
simultaneously measure the x component for one particle and the y components
for the other two. Because the spin state is an eigenvector of all three of the
operators
xa yb yc , ya xb yc , ya yb xc
with eigenvalue +1, the product of the results of each of the three single spin
measurements has to be +1, regardless of which particle we pick for the x spin
measurement. Since +1 flashes red and 1 flashes green, there must indeed be
an even number of green flashes and thus an odd number of red flashes.

What about the result of three xspin measurements, declared earlier never to
result in an odd number of red flashes? Translating this into spin language tells
us that the product of the three results must always be 1. The Hermitian
operator corresponding to that product is

xa xb xc (16.132)

so for the declaration to be correct, it must be that the eigenvector of the first
three operators with eigenvalue +1 is also an eigenvector of the last opera-
tor(above) with eigenvalue 1.

This is easily confirmed. Indeed, the last operator is just minus the product of
the other three operators

xa xb xc = (xa yb yc )(ya xb yc )(ya yb xc ) (16.133)

Since we are in an eigenvector with eigenvalue +1 of each of the three operators

appearing on the right, we are indeed also in an eigenvector of xa xb xc with
eigenvalue 1.

The consequence of the EPR reality criterion specified earlier, if translated into
quantum theoretic terminology, would also assert that the state was an eigen-
vector of the operator xa xb xc , but with the wrong eigenvalue. In this sense, the
GHZ experiment provides the strongest possible contradiction between quan-
tum mechanics and the EPR reality criterion.

Alternatively, we can say it this way. We may measure, on each particle, either
x or y , without disturbing the other particles. The results of these measure-
ments will be called mx or my , respectively. From

xa yb yc |111i = |111i
ya xb yc |111i = |111i
ya yb xc |111i = |111i

and
xa xb xc |111i = |111i

1337
we can predict with certainty that, if the three x are measured, the results
satisfy
max mbx mcx = 1 (16.134)
Therefore, each of the operators xa , xb and xc corresponds to an EPR element
of reality, because its value can be predicted with certainty by performing mea-
surements on the two other, distant particles.

However, it follows from

xa yb yc |111i = |111i
ya xb yc |111i = |111i
ya yb xc |111i = |111i

that we can predict with certainty the value of xa by measuring yb and yc

rather than xb and xb . We then have

max mby mcy = +1 (16.135)

and likewise, by cyclic permutation,

may mbx mcy = +1 (16.136)

and
may mby mcx = +1 (16.137)
The product of the last four results gives

max mbx mcx max mby mcy may mbx mcy may mby mcx = 1 (16.138)
2 2 2 2 2 2
(max ) (mbx ) (mcx ) (mby ) (mcy ) (may ) = 1 (16.139)

But,
(mjx )2 = 1 (16.140)
so that we get a contradiction.

There is a tacit assumption in the above argument, that max in max mbx mcx =
1 is the same as max in max mby mcy = +1, in spite of the fact that these two
ways of obtaining max involve mutually exclusive experiments - measuring xb
and xc or measuring yb and yc .

This tacit assumption is of counterfactual nature, and cannot be experimentally

verified. It obviously adheres to the EPR reality criterion - but is simply wrong!

Saying it another way, the crucial minus sign in

xa xb xc = (xa yb yc )(ya xb yc )(ya yb xc )

1338
which is totally destructive of the possibility of these instruction sets, comes
from the fact that in working out the identity it is necessary to interchange the
anticommuting operators xb and yb in order to get rid of all the y components
(using (yi )2 = 1) and be left with a product of three x components. It is
only that one instance of uncompensated anticommutation that produces the
conclusion so devastating to the hypothesis of instruction sets.

This is extremely pleasing, for it is just the fact the x and y components of the
spin of a single particle do not commute, which leads the well-educated quantum
mechanician to reject from the start the inference instruction sets (which have
to specify the value of both of these non-commuting observables), making it
necessary for me to disguise what was going on earlier so that you would not
have dismissed this discussion as rubbish before reaching the interesting part.

There is no other Bellian refutation of EPR in which the mathematical details

of the refutation so closely reflect the broad interpretive doctrines of quantum
theory that EPR tried to challenge. The entries in the instruction sets are
precisely the conjectured cnumber values for all the xi and yi values that
appear to be the only explanation for the remarkable correlations. In addition,
the logic of the red and green lights in the simple model precisely parallels the
algebraic behavior of the four operators used here except for that one devastating
anticommutation.

Let us return now to using Bayesian ideas to convince our realist friends about
the validity of quantum mechanics within the context of the Bell inequalities.

16.9.4 More about the Greenberger-Horne-Zeilinger(GHZ)

State
We consider the GHZ state for a three-particle system
1
|GHZi = (|1, 1, 1i |1, 1, 1i) (16.141)
2
where 1 and +1 denote any two orthogonal states of each of the three particle
subsystems.

We have three distant observers examine the three subsystems. The first ob-
server has the choice of two tests. The first test can give two different results
that we label a = 1, and likewise the other test yields a0 = 1. Symbols b, b0 ,
c and c0 are similarly defined for the other two observers. Any possible values
of their results satisfy
a0 bc = ab0 c = abc0 = a0 b0 c0 = +1 (16.142)
Mermin has then shown that we have the inequality
2 ha0 bc + ab0 c + abc0 a0 b0 c0 i +2 (16.143)

1339
As we saw above quantum mechanics makes a very simple prediction for the
GHZ state: there are well chosen tests that give with certainty
a0 bc = ab0 c = abc0 = a0 b0 c0 = +1 (16.144)
It is important to remember that performing any such test can verify the value
1 for only one of these products (at a time) since each product corresponds to
a different experimental setup.

If, however, we take all these results together, they manifestly conflict with
a0 bc + ab0 c + abc0 a0 b0 c0 = 2.

Many physicists have erroneously, at this point, stated that a single experi-
ment is sufficient to invalidate local realism. This is sheer nonsense: a single
experiment can only verify one occurrence of one of the terms in
a0 bc = ab0 c = abc0 = a0 b0 c0 = +1 (16.145)
What does our realist friend think?

He believes that, in each experimental run, each term in the above result has a
definite value even if that term is not actually measured in that run.

We ask him to propose a rule giving the average values of the products in
a0 bc + ab0 c + abc0 a0 b0 c0 = 2 (16.146)
Suppose that he assumes
ha0 bci = hab0 ci = habc0 i = ha0 b0 c0 i = 0.5 (16.147)
This clearly attains the right hand side of (Mermins) inequality. This assump-
tion then leads to the prediction that if we measure a0 bc we shall find the result
1 (that is, yes) in 75% of the cases and the opposite result in 25% and like wise
for the other tests. This simply corresponds to the averages proposed above.

In our earlier discussion about confidence depressing factors, this corresponds

to
q = 1 and r = 0.75 (16.148)
If we assume that quantum mechanics is correct, then m = n (that is what
q = 1 means). Therefore, we have
 q m  1 q nm  1 n  4 n
D= = = (16.149)
r 1r r 3
Therefore it would take
 n
4 4
D= = 104 n = 32 (16.150)
3 log(1.33)
tests to undo the realists beliefs.

1340
16.10 Problems
16.10.1 Bell Inequality with Stern-Gerlach
A pair of spin1/2 particles is produced by a source. The spin state of each
particle can be measured using a Stern-Gerlach apparatus (see diagram below).

Figure 16.29: EPR Stern-Gerlach Setup

(a) Let n1 and n2 be the field directions(arrows in diagram) of the Stern-

Gerlach magnets. Consider the commuting observables

2 ~1 2 ~2
(1) = n1 S , (2) = n2 S
~ ~
corresponding to the spin component of each particle along the direction
of the Stern-Gerlach apparatus associated with it. What are the possible
values resulting from measurement of these observables and what are the
corresponding eigenstates?

(b) Consider the observable (12) = (1) (2) and write down its eigenvectors
and eigenvalues. Assume that the pair of particles is produced in the
singlet state

1  (1) (2) (1) (2)


|0, 0i = |Sz +i |Sz i |Sz i |Sz +i
2

What is the expectation value of (12) ?

(c) Make the assumption that the spin of a particle has a meaningful value
even when it is not being measured. Assume also that the only possible
results of the measurement of a spin component are ~/2. Then show that
the probability of finding the spins pointing in two given directions will be
proportional to the overlap of the hemispheres that these two directions
define. Quantify this criterion and calculate the expectation value of (12) .

1341
 (d) Assume the spin variables depend on a hidden variable . The expectation
value of the spin observable (12) is determined in terms of the normalized
distribution function f ():
Z
D E 4
(12) = 2 df ()Sz(1) ()S(2) ()
~
Prove Bells inequality
D E D E D E
() (12) (0 )  1 +  (12) ( 0 ) 
 (12)

  

(e) Consider Bells inequality for 0 = 2 and show that it is not true when
applied in the context of quantum mechanics.

16.10.2 Bells Theorem with Photons

Two photons fly apart from one another, and are in oppositely oriented circularly
polarized states. One strikes a polaroid film with axis parallel to the unit vector
a, the other a polaroid with axis parallel to the unit vector b. Let P++ (a, b) be
the joint probability that both photons are transmitted through their respective
polaroids. Similarly, P (a, b) is the probability that both photons are absorbed
by their respective polaroids, P+ (a, b) is the probability that the photon at
the a polaroid is transmitted and the other is absorbed, and finally, P+ (a, b)
is the probability that the photon at the a polaroid is absorbed and the other
is transmitted.

The classical realist assumption is that these probabilities can be separated:

Z
Pij (a, b) = d()Pi (a, )Pj (b, )

where i and j take on the values + and , where signifies the so-called hidden
variables, and where () is a weight function. This equation is called the
separable form.

The correlation coefficient is defined by

C(a, b) = P++ (a, b) + P (a, b) P+ (a, b) P+ (a, b)

and so we can write

Z
C(a, b) = d()C(a, )C(b, )

where

C(a, ) = P+ (a, ) P (a, ) , C(b, ) = P+ (b, ) P (b, )

It is required that

1342
 (a) () 0
R
(b) d() = 1

(c) 1 C(a, ) 1 , 1 C(b, ) 1

The Bell coefficient

B = C(a, b) + C(a, b0 ) + C(a0 , b) C(a0 , b0 )

combines four different combinations of the polaroid directions.

(1) Show that the above classical realist assumptions imply that |B| 2
 2
(2) Show that quantum mechanics predicts that C(a, b) = 2 a b 1

(3) Show that the maximum value of the Bell coefficient is 2 2 according to
quantum mechanics

(4) Cast the quantum mechanical expression for C(a, b) into a separable form.
Which of the classical requirements, (a), (b), or (c) above is violated?

16.10.3 Bells Theorem with Neutrons

Suppose that two neutrons are created in a singlet state. They fly apart; the
spin of one particle is measured in the direction a, the other in the direction b.

(a) Calculate the relative frequencies of the coincidencesR(up, up), R(up, down),
R(down, up) and R(down, down), as a function of , the angle between a
and b.

(b) Calculate the correlation coefficient

C(a, b) = R(up, up) R(up, down) R(down, up) + R(down, down)

(c) Given two possible directions, a and a0 , for one measurement, and two
possible directions, b and b0 , for the other, deduce the maximum possible
value of the Bell coefficient, defined by

B = C(a, b) + C(a0 , b) + C(a0 , b0 ) C(a, b0 )

(d) Show that this prediction of quantum mechanics is inconsistent with clas-
sical local realism.

1343
16.10.4 Greenberger-Horne-Zeilinger State
The Greenberger-Horne-Zeilinger (GHZ) state of three identical spin1/2 par-
ticles is defined by
1
|GHZi = (|za +i |zb +i |zc +i |za i |zb i |zc i)
2
where za + is the eigenvector of the zcomponent of the spin operator of particle
a belonging to eigenvalue +~/2 (zspin up), za is the eigenvector of the
zcomponent of the spin operator of particle a belonging to eigenvalue ~/2
(zspin down), and similarly for b and c. Show that, if spin measurements are
made on the three particle in the x or ydirections,
(a) the product of three spins in the xdirection is always ~3 /8
(b) the product of two spins in the ydirection and one spin in the xdirection
is always +~3 /8

(c) Consider a prize game for a team of three players, A, B, and C. The players
are told that they will be separated from one another and that each will
be asked one of two questions, say X or Y, to which each must give one of
two allowed answers, namely, +1 or 1. Moreover, either
(a) all players will be asked the same question X
or
(b) one of the three players will be asked X and the other two Y
After having been asked X or Y, no player may communicate with the
others until after all three players have given their answers, +1 or 1. To
win the game, the players must give answers such that, in case (a) the
product of the three answers is 1, whereas in case (b) the product of the
three answers is +1.
(a) Show that no classical strategy gives certainty of a win for the team
(b) Show that a quantum strategy, in which each player may take one of
the GHZ particles with her, exists for which a win is certain

1344
Chapter 17

Path Integral Methods

17.1 Historical Remarks

In this book we have developed the standard formulation of quantum mechan-
ics developed more or less concurrently by Schrodinger, Heisenberg, Dirac and
others in the 1920s and shown to be equivalent to one another soon thereafter.

In 1933, Dirac made the observation that the action plays a central role in
classical mechanics (he considered the Lagrangian formulation of classical me-
chanics to be more fundamental than the Hamiltonian formulation), but that
it seemed to have no important role in quantum mechanics as it was known at
the time. He speculated on how the situation might be rectified, and he arrived
at the conclusion that (in more modern language) the propagator in quantum
mechanics corresponds to  
S
exp i (17.1)
~
where S is the classical action evaluated along the classical path.

In 1948, Feynman developed Diracs suggestion, and succeeded in deriving a

new formulation of quantum mechanics, based on the fact that the propagator
can be written as a sum over all possible paths (not just the classical one)
between the initial and final points. Each path contributes exp (iS/~) to the
propagator. So while Dirac considered only the classical path, Feynman showed
that all paths contribute: in a sense, the quantum particle takes all paths, and
the amplitudes for each path add according to the usual quantum mechanical
rule for combining amplitudes.

17.2 Motivation
What do we learn from path integrals? Path integrals give us no dramatic new
results in the quantum mechanics of a single particle. In fact, most, if not all,

1345
calculations in quantum mechanics which can be done by path integrals can be
done with considerably greater ease using the standard formulation of quantum
mechanics. So why all the fuss?

It turns out that path integrals are considerably more useful in more complicated
situations, such as quantum field theory. Even if this was not case, however,
path integrals give a very worthwhile contribution to our understanding of quan-
tum mechanics.

First, path integrals provide a physically extremely appealing and intuitive way
of viewing quantum mechanics: anyone who can understand Youngs double slit
experiment in optics should be able to understand the underlying ideas behind
path integrals.

Second, the classical limit of quantum mechanics can be understood in a par-

ticularly clean way via path integrals.

17.3 Path Integrals in Quantum Mechanics

17.3.1 General ideas
Consider a particle moving in one dimension. The Hamiltonian has the usual
form:
p2
H= + V (q) (17.2)
2m
The fundamental question in the path integral (PI) formulation of quantum
mechanics is:

If the particle is at a position q at time t = 0, what

is the probability amplitude that it will be at some
other position q 0 at a later time t = T ?

It is easy to get a formal expression for this amplitude in the usual Schrodinger
formulation of quantum mechanics. Let us introduce the eigenstates of the
position operator q, which form a complete orthonormal set:
Z
0 0
q |qi = q |qi , hq | qi = (q q) , dq |qi hq| = 1 (17.3)

(when there is the possibility of an ambiguity, operators will be written with a

hat, otherwise the hat will be dropped). Then the initial state is
|(0)i = |qi (17.4)
Letting the state evolve in time and projecting on the state |q 0 i, we get the
amplitude A,
A = hq 0 | (T )i K(q 0 , T ; q, 0) = hq 0 | eiHT |qi (17.5)

1346
(except where otherwise noted, ~ will be set to 1). This object, for obvious
reasons, is known as the propagator from the initial space-time point (q, 0) to
the final point (q 0 , T ). Clearly, the propagator is independent of the origin of
time:

K(q 0 , T + t; q, t) = K(q 0 , T ; q, 0) (17.6)

We will derive an expression for this amplitude in the form of a summation(integral

really) over all possible paths between the initial and final points. In doing so,
we will derive the PI method from quantum mechanics.

Let us separate the time evolution in the above amplitude into two smaller time
evolutions, writing

eiHT = eiH(T t1 ) eiHt1 (17.7)

The amplitude becomes

A = hq 0 | eiH(T t1 ) eiHt1 |qi (17.8)

Inserting an identity operator gives

Z
A= dq1 hq 0 | eiH(T t1 ) |q1 i hq1 | eiHt1 |qi
Z
= dq1 K(q 0 , T ; q1 , t1 )K(q1 , t1 ; q, 0) (17.9)

This formula is none other than an expression of the quantum mechanical rule
for combining amplitudes. If a process can occur a number of ways, the am-
plitude for each of these ways add. A particle, in propagating from q to q 0 ,
must be somewhere at an intermediate time t1 . Labelling that intermediate
position q1 , we compute the amplitude for propagation via the point q1 (this
is the product of two propagators as in equation(17.9)) and integrate over all
possible intermediate positions. This result is reminiscent of Youngs double
slit experiment where the amplitudes for passing through each of the two slits
combine and interfere. We will look at this example in more detail later.

We can repeat the division of the time interval T . Let us divide it up into a
large number N of time intervals of duration = T /N . Then we can write the
propagator as

N
A = hq 0 | eiH |qi = hq 0 | e|iH eiH{z.........eiH} |qi (17.10)
N times

1347
We can again insert a complete set of states (identity operator) between each
exponential, which gives
Z Z
A = hq 0 | eiH dqN 1 |qN 1 i hqN 1 | eiH dqN 2 |qN 2 i hqN 2 | .........
Z Z
iH
..... dq2 |q2 i hq2 | e dq1 |q1 i hq1 | eiH |qi
Z
= dq1 ........dqN 1 hq 0 | eiH |qN 1 i hqN 1 | eiH |qN 2 i ........ hq1 | eiH |qi
Z
= dq1 ........dqN 1 KqN , qN 1 KqN 1 , qN 2 ........Kq2 , q1 Kq1 , q0

where we have defined q0 = q, qN = q 0 and Kqi , qj = hqj | eiH |qi i. Note

that the initial and final positions are not integrated over. This expression says
that the amplitude is the integral of the amplitude of all N legged paths, as
illustrated in Figure 17.1 below

Figure 17.1: Amplitude as a sum over N-legged paths

Apart from the mathematical details concerning the limit N , this is clearly
going to become a sum over all possible paths of the amplitude for each path:
X
A= Apath (17.11)
paths

where
X Z
= dq1 ........dqN 1 , Apath = KqN , qN 1 KqN 1 , qN 2 ........Kq2 , q1 Kq1 , q0
paths
(17.12)

1348
Let us look at this expression in detail.

The propagator for one sub-interval is

Kqj+1 , qj = hqj+1 | eiH |qj i (17.13)

We can expand the exponential, since is small:

 
1
Kqj+1 , qj = hqj+1 | eiH |qj i = hqj+1 | 1 iH H 2 2 + .... |qj i
2
= hqj+1 | qj i i hqj+1 | H |qj i + O( 2 ) (17.14)

The first term is a delta function, which we can write as

Z
dpj ipj (qj+1 qj )
hqj+1 | qj i = (qj+1 qj ) = e (17.15)
2

In the second term of equation(17.14), we can insert another identity operator

but this time in the form of an integral over momentum eigenstates between H
and |qj i. This gives

p2
  Z 
dpj
i hqj+1 | H |qj i = i hqj+1 | + V (q) |pj i hpj | |qj i
2m 2
Z  2 
dpj p
= i hqj+1 | + V (q) |pj i hpj | qj i
2 2m
!
p2j
Z
dpj
= i hqj+1 | + V (qj+1 ) |pj i hpj | qj i
2 2m
!
p2j
Z
dpj
= i + V (qj+1 ) hqj+1 | pj i hpj | qj i
2 2m
!
p2j
Z
dpj
= i + V (qj+1 ) eipj (qj+1 qj ) (17.16)
2 2m

where we used hq | pi = exp(ipq). Note that the operator p acted to the right
and the operator V (q) acted to the left.

The expression in equation(17.16) is asymmetric between qj+1 and qj . The

origin of this is our choice of putting the identity operator to the right of H
in the second term of equation(17.14). Had we put it to the left instead, we
would have obtained V (qj ) in equation(17.16). To not play favorites, we should
choose some sort of average of these two cases. In what follows, I will sim-
ply write V (qj ) where qj = 12 (qj+1 + qj ) (the exact choice does not matter in
the continuum limit, which we will take eventually; this choice is the common
choice). Combining equations (17.15) and (17.16), the sub-interval propagator

1349
is
! !
p2j
Z
dpj ipj (qj+1 qj ) 2
Kqj+1 , qj = e 1 i + V (qj ) + O( )
2 2m
Z
dpj ipj (qj+1 qj ) iH(pj , qj )
1 + O( 2 )


= e e (17.17)
2
There are N such factors in the amplitude. Combining them, and writing
qj = (qj+1 qj ) /, we get

Z NY 1 N 1
dpj X
Apath = exp i (pj qj H(pj , qj )) (17.18)
j=0
2 j=0

N
where we have neglected a multiplicative factor of the form 1 + O( 2 ) which
will tend toward one in the continuum limit. Then the propagator becomes
Z
K = dq1 ........dqN 1 Apath

Z NY1 Z NY 1 N 1
dpj X
= dqj exp i (pj qj H(pj , qj )) (17.19)
j=1 j=0
2 j=0

Note that there is one momentum integral for each interval (N total), while
there is one position integral for each intermediate position (N 1 total).

If N , this approximates an integral over all functions p(t), q(t). We adopt

the following notation:
T
Z Z
K Dp(t)Dq(t) exp i dt (pq H(p , q)) (17.20)
0

This result is known as the phase-space path integral. The integral is viewed as
over all functions p(t) and over all functions q(t) where q(0) = q, q(T ) = q 0 . But
to actually perform an explicit calculation, equation(17.20) should be viewed as
a shorthand notation for the expression equation(17.19) in the limit N .

If, as is often the case (and we have assumed in deriving the above expression),
the Hamiltonian is of the standard form, namely
p2
H= + V (q) (17.21)
2m
we can actually carry out the momentum integrals in equation(17.19). We can
rewrite this expression as
!
Z NY1 N 1 Z NY1 N 1
X dpj X p2j
K= dqj exp i V (qj ) exp i pj qj
j=1 j=0 j=0
2 j=0
2m
(17.22)

1350
The p integrals are all Gaussian, and they are uncoupled. One such integral is
Z r
dp i(pqp2 /2m) m imq2 /2
e = e (17.23)
2 2i
The careful reader may be worried about the convergence of this integral. If so,
a factor exp(p2 ) can be introduce and the limit 0 taken at the end (see
for example Chapter 1 - page 21).

The propagator becomes

1 1 1
Z NY N N
!!
imqj2
r
X Y m
K= dqj exp i V (qj ) exp
j=1 j=0 j=0
2i 2

1 1
!
 m N/2 Z NY N
X mqj2
= dqj exp i V (qj ) (17.24)
2i j=1 j=0
2

The argument of the exponential is a discrete approximation of the action of a

path passing through the points q0 = q, q1 , .., qN 1 , qN = q 0 . As above, we can
write this in the more compact form
Z
K = Dq(t) exp (iS(q(t))) (17.25)

This is our final result and is known as the configuration space path integral.
Again, equation(17.25) should be viewed as a notation for the more precise
expression equation(17.24), as N .

17.3.2 Examples
To solidify the above notions, let us consider a few explicit examples. As a
first example, we will compute the free particle propagator first using ordinary
quantum mechanics and then via PI. We will then mention some generalization
which can be done in a similar manner.

Free Particle
Let us compute the propagator K(q 0 , T ; q, 0) for a free particle, described by the
Hamiltonian
p2
H= (17.26)
2m
The propagator can be computed straightforwardly using ordinary quantum
mechanics. To this end, we write
Z
0 iHT 0 ip2 T /2m dp
K = hq | e |qi = hq | e |pi hp | qi
2
Z Z
dp ip2 T /2m 0 dp ip2 T /2m+i(q0 q)p
= e hq | pi hp | qi = e (17.27)
2 2

1351
The integral is a Gaussian. We obtain
 m 1/2 0 2
K= eim(q q) /2T (17.28)
2iT
Let us see how the same result can be obtained using PIs. The configuration
space PI (equation(17.25)) is

 m N/2 Z NY 1 N 1  2
m X q j+1 q j
K = lim dqj exp i
N 2i 2
j=1 j=0

 m N/2 Z NY1
m (qN qN 1 )2 + (qN 1 qN 2 )2
  
= lim dqj exp i 2 2
N 2i
j=1
2 +... (q2 q1 ) + (q1 q0 )
(17.29)

where q0 = q and qN = q 0 are the initial and final points. The integrals are
Gaussian and can be evaluated exactly, although the fact that they are coupled
complicates matters significantly. The result is
 m N/2 1  2i (N 1)/2 0 2
K = lim eim(q q) /2N
N 2i N m
 m 1/2 0 2
= lim eim(q q) /2N (17.30)
N 2iN

But N is the total time interval T , so that we get

 m 1/2 0 2
K= eim(q q) /2T (17.31)
2iT
in agreement with equation(17.28).

A couple of remarks are in order. First, we can write the argument of the
exponential as
 0 2
1 q q
T m (17.32)
2 T
which is just the action S(qclassical ) for a particle moving along the classical
path (a straight line in this case) between the initial and final points.

Secondly, we can restore the factors of ~ if we want, by ensuring the correct

dimensions. The argument of the exponential is the action, so in order to make
it a pure number we must divide by ~. Furthermore, the propagator has the
dimension of the inner product of two position eigenstates, which is inverse
length. In order that the coefficient have this dimension we must multiply by
~1/2 . The final result is
 m 1/2
K= eiS(qc )/~ (17.33)
2i~T

1352
This result typifies a couple of important features of calculations in this subject,
which we will see repeatedly as we continue the discussion. First, the propagator
separates into two factors, one which is the phase eiS(qc )/~ . Second, calculations
in the PI formalism are typically quite a bit more lengthy than using standard
techniques of quantum mechanics.

Harmonic Oscillator
As a second example of the computation of a PI, let us show how to compute
the propagator for the harmonic oscillator using this method.

Let us start with the somewhat-formal version of the configuration-space PI

(equation(17.25)):
Z
K(q 0 , T ; q, 0) = Dq(t) exp (iS(q(t))) (17.34)

For the harmonic oscillator

ZT  
1 2 1
S(q(t)) = dt mq m 2 q 2 (17.35)
2 2
0

The paths over which the integral is to be performed go from q(0) = q to

q(T ) = q 0 . To do this PI, suppose we know the solution of the classical problem,
qc (t):
qc + 2 qc2 = 0 , qc (0) = q , qc (T ) = q 0 (17.36)
We can write q(t) = qc (t) + y(t) and perform a change of variables in the PI to
y(t), since integrating over all deviations from the classical path is equivalent
to integrating over all possible paths. Since at each time q and y differ by a
constant, the Jacobian of the transformation is 1. Furthermore, since ? obeys
the correct boundary conditions, the paths y(t) over which we integrate go from
y(0) = 0 to y(T ) = 0. The action for the path qc (t) + y(t) can be written as a
power series in y:

ZT  
1 2 1
S(qc (t) + y(t)) = dt mq m 2 qc2
2 c 2
0
ZT  
1 1
+ (linear in y) + dt my 2 m 2 y 2 (17.37)
| {z } 2 2
=0 0

The term linear in y vanishes by construction - qc being the classical path, is

the path for which the action is stationary! So we may write

S(qc (t) + y(t)) = S(qc (t)) + S(y(t)) (17.38)

1353
We substitute this into equation(17.25) and obtain
Z
0 iS(qc (t))
K(q , T ; q, 0) = e Dq(t)eiS(y(t)) (17.39)

As mentioned above, the paths y(t) over which we integrate go from y(0) = 0
to y(T ) = 0. The only appearance of the initial and final positions is in the
classical path, i.e., in the classical action. Once again, the PI separates into
two factors. The first is written in terms of the action of the classical path and
the second is a PI over deviations from this classical path. The second factor is
independent of the initial and final points.

This separation into a factor depending on the action of the classical path and
a second one, a PI which is independent of the details of the classical path is a
recurring theme and an important one. Indeed, it is often the first factor which
contains most of the useful information contained in the propagator and it can
be deduced without even performing a PI. It can be said that much of the work
in the game of path integrals consists in avoiding having to compute one!

The evaluation of equation(17.39) is complicated and will be done at the end of

this chapter. The result is
 1/2
0 m
K(q , T ; q, 0) = eiS(qc (t)) (17.40)
2i sin (T )
where
m
q 02 + q 2 cos (T ) 2q 0 q



S(qc (t)) = (17.41)
2 sin (T )
We close this section with two remarks. First, the PI for any quadratic action
can be evaluated exactly (see derivation later), essentially since such a PI con-
sists of Gaussian integrals. The general result is given later.

Second, K(q 0 , T ; q, 0) is the amplitude to propagate from one point to another

in a given time interval. Now we ask this question: if a particle is initially
at position q, what is the wave function after the elapse of a time T ? If we
consider K as a function of the final position and time, it turns out to be the
wave function for a particle with a specific initial condition and, as such, the
propagator satisfies the Schrodinger equation at its final point.

17.4 The Classical Limit

17.4.1 Derivation of the Principle of Least Action
Since the example calculations performed above are somewhat dry and math-
ematical, it is worth backing up a bit and staring at the expression for the
configuration space PI, equation(17.25):
Z
K = Dq(t) exp (iS(q(t)))

1354
This innocent looking expression tells us something which is at first glance un-
believable and at second glance really unbelievable. The first glance observation
is that a particle, in going from one position to another, takes all possible paths
between these two positions. This is, if not actually unbelievable, at the very
least counter-intuitive. We could, however, argue away much of what makes us
feel uneasy if we could convince ourselves that while all paths contribute, the
classical path is the dominant one.

However, the second glance observation is not reassuring. If we compare the

contribution of the classical path (whose action is S(qc (t))), with that of some
other, arbitrarily wild, path (whose action is S(qw (t))), we find that the first
is exp(iS(qc (t))) while the second is exp(iS(qw (t))). They are both complex
numbers of unit magnitude - each path taken in isolation is equally important.
The classical path is no more important than any arbitrarily complicated path!

How are we to reconcile this really unbelievable conclusion with the fact that a
ball thrown in the air has a more-or-less parabolic motion?

The key, not surprisingly, is in how the different paths interfere with one an-
other. By considering the case where the rough scale of classical action is much
bigger than the quantum of action, ~, we will see the emergence of the Principle
of Least Action.

Consider two neighboring paths q(t) and q 0 (t) which contribute to the PI (see
Figure 17.2 below).

Figure 17.2: Two Neighboring Paths

Let q 0 (t) = q(t) + (t), with (t) small. Then we can write the action as a
functional Taylor expansion about the classical path.

If you are not familiar with the manipulation of functionals (functions of func-
tions) do not despair - the only rule needed beyond standard calculus is the
functional derivative
q(t)
= (t t0 ) (17.42)
q(t0 )

1355
where the last function is the Dirac delta function.

We have
Z
S(q(t))
S(q 0 ) = S(q + ) = S(q) + dt(t) + O( 2 ) (17.43)
q(t)

The two paths contribute exp(iS(q(t))) and exp(iS(q 0 (t))) to the PI. The com-
bined contribution is
  Z 
i S(q)
A ' eiS(q)/~ 1 + exp dt(t) (17.44)
~ q(t)

where we have neglected corrections of order 2 . We see that the difference in

phase between the two paths, which determines the interference between the
two contributions, is Z
1 S(q)
dt(t) (17.45)
~ q(t)
e see that the smaller the value of ~, the larger the phase difference between two
given paths. So even if the paths are very close together, so that the difference
in actions is extremely small, for sufficiently small ~ the phase difference will
still be large and on average destructive interference occurs.

This argument must be rethought, however, for one exceptional path - the path
which extremizes the action, i.e., the classical path qc (t). For this path

S(qc + ) = S(qc ) + O( 2 ) (17.46)

Thus, the classical path and a very close neighbor will have actions which differ
by much less than two randomly chosen but equally close paths (see Figure 17.3
below).

Figure 17.3: Paths near classical path interfere constructively

This means that for fixed closeness of two paths and for fixed ~, paths near the
classical path will on average interfere constructively (small phase difference)
whereas for random paths the interference will be on average destructive.

1356
Thus, heuristically, we conclude that is the problem is classical (action  ~),
the most important contribution to the PI comes from the region around the
path which extremizes the PI. In other words, the particles motion is governed
by the principle that the action is stationary. This, of course, is none other
than the Principle of Least Action from which the Euler-Lagrange equations of
classical mechanics are derived.

17.5 Topology and Path Integrals in Quantum

Mechanics
In path integrals, if the configuration space has holes in it such that two paths
between the same initial and final points are not deformable into one another,
interesting effects can arise. This property of the configuration space goes by the
following catchy name: non-simple-connectedness. We now study a situation of
this type.

17.5.1 Aharonov-Bohm Effect

The Aharonov-Bohm effect(already discussed in Chapter 8) is one of the most
dramatic illustrations of a purely quantum effect: the influence of the electro-
magnetic potential on particle motion even if the particle is perfectly shielded
from any electric and magnetic fields. While classically the effect of electric
and magnetic fields can be understood purely in terms of the forces these fields
create on the particles, Aharonov and Bohm devised an ingenious thought ex-
periment (which has since been carried out in the laboratory) showing that this
is no longer true in quantum mechanics. Their effect is best illustrated by a
refinement of Youngs double-slit experiment, where particles passing through
a barrier with two slits in it produce an interference pattern on a screen further
downstream as shown in Figure 17.4 below.

Figure 17.4: Aharonov-Bohm effect. Magnetic flux is confined within the shaded
area. particles are excluded from this area by a perfect shield

1357
Aharonov and Bohm proposed that such an experiment be performed with
charged particles. The setup had an added twist. A magnetic field from which
the particles are perfectly shielded exists in between the two slits. If we per-
form the experiment first with no magnetic flux and then with a nonzero and
arbitrary flux passing through the shielded region, the interference pattern will
change, in spite of the fact that the particles are perfectly shielded from the mag-
netic field and feel no electric or magnetic force whatsoever.

Classically, we say - no force, no effect. Not so in quantum mechanics. PIs

provide a excellent way of understanding this effect.

Consider first two representative paths q1 (t) and q2 (t) (in two dimensions) pass-
ing through slits 1 and 2 respectively, and which arrive at the same spot on the
screen as shown in Figure 17.5 below.

Figure 17.5: Two representative paths contributing to the amplitude for a given
point on the screen

Before turning on the magnetic field, let us suppose that the actions for these
paths are S(q1 ) and S(q2 ). Then the interference of the amplitudes is deter-
mined by
 
eiS(q1 )/~ + eiS(q2 )/~ = eiS(q1 )/~ 1 + ei(S(q2 )S(q1 ))/~ (17.47)

The relative phase is

12 (S(q2 ) S(q1 ))/~ (17.48)
Thus, the two paths interfere constructively if 12 = 2n and destructively if
12 = (2n + 1) and, in general, there is partial cancellation between the two
contributions.

~ We can describe this

How is this result affected if we add a magnetic field B?
~ ~
field by a vector potential, writing B = A. This affects the particles motion
by the following change in the Lagrangian (see Chapter 8):

L0 (~q,~q)
L(~q,~q) e ~v A(~
= L(~q,~q) ~ q) (17.49)
c

1358
Thus, the action changes by
Z Z Z
e ~ e d~q(t) ~ e ~ q)
dt ~v A(~q) = dt A(~q) = d~q(t) A(~
c c dt c
The integral is Z
~ q)
d~q(t) A(~ (17.50)

which is the line integral of A~ long the path taken by the particle. So including
the effect of the magnetic field, the action of the first path is
Z
0 e ~ q)
S (~q1 ) = S(~q1 ) d~q(t) A(~ (17.51)
c
q
~1

and similarly for the second path.

Let us now look at the interference between the two paths, including the mag-
netic field.
0 0 0
 0 0
 0
 0

eiS (~q1 )/~ + eiS (~q2 )/~ = eiS (~q1 )/~ 1 + ei(S (~q2 )S (~q1 ))/~ = eiS (~q1 )/~ 1 + ei12
(17.52)
where the new relative phase is

Z Z
e ~ q) ~ q )
012 = 12 d~q(t) A(~ d~q(t) A(~ (17.53)
~c
q
~2 q
~1

But the difference in the line integrals in equation(17.53) is a contour integral

Z Z I
~ q ) d~q(t) A(~
d~q(t) A(~ ~ q ) = d~q(t) A(~
~ q) = (17.54)
q
~2 q
~1

where is the flux inside the closed loop bounded by the two paths. So we can
write
e
012 = 12 (17.55)
~c
It is important to note that the change of relative phase due to the magnetic field
is independent of the details of the two paths, as long as each passes through
the corresponding slit. This means that the PI expression for the amplitude
for the particle to reach a given point on the screen is affected by the magnetic
field in a particularly clean way. Before the magnetic field is turned on, we may
write A = A1 + A2 , where
Z
0
A1 = D~q eiS (~q)/~ (17.56)
slit 1

1359
and similarly for A2 . Including the magnetic field,

i(S 0 (~
q ) ec ~ ~
Z R R
q A/~
d~ ie q A/~c
d~
A01 = D~q e 1 =e 1 A1 (17.57)
slit 1

where we have pulled the line integral out of the PI since it is the same for all
paths passing through slit 1 arriving at the point on the screen under consider-
ation. So the amplitude is
~ ~ ~
R R R
ie q A/~c
d~ ie q A/~c
d~ ie q A/~c
d~  H 
A=e 1 A1 + e A2 = e 2 1 A1 + eie q A/~c
d~
A2
~
R
q A/~c
ie d~  
=e 1 A1 + eie/~c A2 (17.58)

The overall phase is irrelevant and the interference pattern is influenced directly
by the phase e/~c. If we vary the phase continuously (by varying the mag-
netic flux), we can detect a shift in the interference pattern. For example, if
e/~c = , then a spot on the screen which formerly corresponded to construc-
tive interference will now be destructive and vice versa.

Since the interference is dependent only on the phase difference mod(2), as

we vary the flux we get a shift of the interference pattern which is periodic,
repeating itself when e/~c changes by an integer times 2.

17.6 Evaluation of Path Integrals - Details

We consider the case of particles moving in one dimension labelled by the posi-
tion coordinate x. The particles have associated with them a Lagrangian
1
L(x, x, t) = mx2 U (x) (17.59)
2
In order to define the path integral we assume a series of times tN > tN 1 >
tN 2 > ...... > t1 > t0 letting N go to infinity later. The spacings between the
times t + j + 1 and tj will all be identical, namely

tN t0
tj+1 tj = = N (17.60)
N
The discretization in time leads to a discretization of the paths x(t) which will
be represented through the series of space-time points

{(x0 , t0 ) , (x1 , t1 ) , (x2 , t2 ) ...... (xN 1 , tN 1 ) , (xN , tN )} (17.61)

Even though the time instances are fixed, we note that the xj the values are not.
They can be anywhere in the allowed volume which we will choose to be the
interval [, +]. In passing from one space-time instance (xj , tj ) to the next

1360
(xj+1 , tj+1 ) we assume that kinetic energy and potential energy are constant,
namely,
1 (xj+1 xj )2
m and U (xj ) (17.62)
2 2N
respectively. these assumptions lead then to the following Riemann form for the
action integral
N 1 
1 (xj+1 xj )2
X 
S[x(t)] = lim N m U (xj ) (17.63)
N
j=0
2 2N

where the form S[....] is the standard form for a functional.

The main idea is that one can replace the path integral now by a multiple
integral over x1 , x2 , etc. This allows us to write the propagator or evolution
operator as

K (xN , tN ; x0 , t0 ) =
Z Z Z

N 1  2

i X 1 (xj+1 xj )
lim CN dx1 dx2 ..... dxN 1 exp N m U (xj )
N ~ j=0
2 2N

(17.64)

Here, cN is a constant which depends on N and other constants in the exponent.

It needs to be chosen to make sure that the limit in equation(17.64) can be
properly taken. Its value is
 N/2
m
CN = (17.65)
2i~N

17.6.1 Propagator for a Free Particle

Rather then using the integration variables xj , it is more suitable to define new
integration variables yj , the origin of which coincides with the classical path of
the particle. To see the benefit of such an approach, we define a path y(t) as
follows
x(t) = xcl (t) + y(t) (17.66)
where xcl (t) is the classical path which connects the space-time points (x0 , t0 )
and (xN , tN ), namely,
xN x0
xcl (t) = x0 + (t t0 ) (17.67)
tN t0
It is essential for the following to note that, since x(t0 ) = xcl (t0 ) = x0 and
x(tN ) = xcl (tN ) = xN , it holds that

y(t0 ) = y(tN ) = 0 (17.68)

1361
Also we use the fact that the velocity of the classical path
xN x0
xcl = (17.69)
tN t0
is constant. The action integral S[x(t)|x(t0 ) = x0 , x(tN ) = xN ] for any path
x(t) can be expressed through an action integral over the path y(t) relative to
the classical path(note explicit new notation). We get
ZtN
1
dt m x2cl + 2xcl y + y 2


S[x(t)|x(t0 ) = x0 , x(tN ) = xN ] =
2
t0
ZtN ZtN ZtN
1 1
= m dt x2cl + mxcl dt y + m dt y 2 (17.70)
2 2
t0 t0 t0

The condition (17.68) implies for the second term on the RHS
ZtN
dt y = y(tN ) y(t0 ) = 0 (17.71)
t0

The first term on the RHS of (17.70) is using (17.67)

ZtN 2
1 1 (xN x0 )
m dt x2cl = m (17.72)
2 2 tN t0
t0

The third term can be written in the notation introduced above

ZtN
1
m dt y 2 = S[x(t)|x(t0 ) = 0, x(tN ) = 0] (17.73)
2
t0

i.e., due to (17.68), can be expressed through a path integral with endpoints
x(t0 ) = 0, x(tN ) = 0. The resulting expression for S[x(t)|x(t0 ) = x0 , x(tN ) =
xN ] is

S[x(t)|x(t0 ) = x0 , x(tN ) = xN ]
2
1 (xN x0 )
= m + S[x(t)|x(t0 ) = 0, x(tN ) = 0] (17.74)
2 tN t0
Inserting into the expression for the propagator we have
" # x(tZN )=0
2  
im (xN x0 ) i
K (xN , tN ; x0 , t0 ) = exp D[x(t)] exp S[x(t)]
2~ tN t0 ~
x(t0 )=0
(17.75)

1362
which can be written as
" #
2
im (xN x0 )
K (xN , tN ; x0 , t0 ) = exp K (0, tN ; 0, t0 ) (17.76)
2~ tN t0

Thus, we need to evaluate

K (0, tN ; 0, t0 ) =
N/2 Z Z Z

 N 1  2

m i X 1 (yj+1 yj )
lim dy1 dy2 ..... dyN 1 exp N m 2

N 2i~N ~ 2
j=0 N

(17.77)

The exponent E can be written, using

2y12 y1 y2 y2 y1 + 2y22 y2 y3 y3 y2
 
im
E=
2~N +2y32 ....... yN 2 yN 1 yN 1 yN 2 + 2yN
2
1
N
X 1
=i yj ajk yk (17.78)
j,k=1

where the ajk are the elements of the following symmetric (N 1) (N 1)

matrix
2 1 0 . . 0 0
1 2 1 . . 0 0

0 1 2 . . 0 0
m .

(ajk ) = . . . . . . (17.79)
2~N
.

. . . . . .
0 0 0 . . 2 1
0 0 0 . . 1 2
The following integral

Z Z Z

N
X 1
I= dy1 dy2 ..... dyN 1 exp i yj ajk yk (17.80)
j,k=1

must now be determined. We will exploit the fact that for any real, symmetric
matrix there exists a similarity transformation such that

a11 0 0 . . 0 0
0 a22 0 . . 0 0

0 0 a33 . . 0 0
S 1 aS =

. . . . . . . (17.81)
. . . . . . .

0 0 0 . . an1,n1 0
0 0 0 . . 0 ann

1363
where S can be chosen as an orthonormal transformation, i.e.,

S T S = I S T = S 1 (17.82)

The akk are the eigenvalues of the a matrix and are real. This property allows us
n
P
to simplify the bilinear form yj ajk yk by introducing new integration variables
j,k

n
X n
X
S 1


yj = y
jk k
, yk = Sjk yk (17.83)
k k

The bilinear form then reads in terms of yj

n
X n X
X n n X
X n n
X
y` S T


yj ajk yk = y` S`j ajk Skm ym = a S y =
`j jk km m
yj ajk ym
j,k j,k `,m j,k `,m j,k

where we have used (from equations (17.81) and (17.82))

n
X
ST


ajk = j`
a`m Smk (17.84)
`,m

For the determinant of a we have

n
Y
det(a) = ajj (17.85)
j=1

as well as
n
Y
det(a) = det(S 1 aS) = det(S 1 ) det(a) det(S) = det(a) = ajj (17.86)
j=1

We have assumed that det(a) 6= 0. Accordingly we have

n
Y
ajj 6= 0 (17.87)
j=1

so that none of the eigenvalues of a vanishes, i.e., ajj 6= 0 f or j = 1, 2, ...., n.

Substitution of the integration variables (17.83) in equation(17.80) gives
Z Z Z    n
!
 (y1 , ...., yn )  X
2
I= dy1 dy2 ..... dyN 1 det
 exp i akk yk
(y1 , ...., yn ) 
k
(17.88)
where we have introduced the Jacobian matrix
(y1 , ...., yn )
J= (17.89)
(y1 , ...., yn )

1364
with elements
yj
Jjs = (17.90)
ys
According to equation(17.84), J = S and hence det(J) = 1. We then have
Z Z Z n
X
!
I= dy1 dy2 ..... dyN 1 exp i akk yk2
k

Z Z
dy1 exp ia11 y12 ............ dyn exp iann yn2



=

n Z
Y
dyk exp iakk yk2


= (17.91)
k=1

which leaves us to determine integrals of the type

Z
2
dx eicx (17.92)

where c 6= 0. We first consider the case c > 0. One can relate the integral
(17.92) to the standard Gaussian integral
Z r
cx2
dx e = , c>0 (17.93)
c

by considering the contour integral

I
2
J= dz eicz (17.94)

along the path = 1 + 2 + 3 + 4 shown in Figure 17.6 below.

Figure 17.6: Contour path in the complex plane

1365
 2
The contour integral vanishes since eicz is an analytic function, i.e., the inte-
grand does not have any singularities anywhere inside the contour. The contour
integral (17.94) can be written as the sum of the following path integrals
I
2
J = J1 + J2 + J3 + J4 , Jk = dz eicz (17.95)
k

The contributions Jk can be expressed through integrals along a real coordinate

axis by realizing that the paths k can be parameterized by real coordinates x
Zp
2
1 : z = x J1 = dx eicx (17.96)
p
Zp
2
2 : z = ix + p J2 = i dx eic(ix+p) (17.97)
0

Z 2p Z 2p

ic( i x)2
2
3 : z = ix J3 = i dx e = i dx ecx (17.98)

2p 2p
Z0
2
4 : z = ix p J4 = i dx eic(ixp) (17.99)
p

for x, p <.

Substituting x for x into integral J4 we get

Z0 Zp
ic(ixp)2 2
J4 = (i) dx e = i dx eic(ix+p) = J2 (17.100)
p 0

We now show that the two integrals J2 and J4 vanish for p . This follows
from the calculation below.
 p   p 
Z  Z   
2 2 2
lim |J2 or 4 | = lim  i dx eic(ix+p)  lim  |i| dx eic(p x )  e2cxp 
     
p+ p+
  p+  
 0p  0

1 e2cp
Z 
lim |J2 or 4 | lim  dx e2cxp  = lim
  
=0
p+ p+
  p+ 2cp
0

Therefore, J2 and J4 do not contribute to the integral (17.95) for p . We

then have
Z Z
r
icx2 cx2 i
dx e = i dx e = (17.101)
c

1366
One can derive the same result for c < 0, if one chooses the same contour but
with a path that is reflected at the real axis. This gives
Z Z
2 2
J= dx eicx i dx ecx = 0 , c<0

Z s r
2 i i
dx eicx = = (17.102)
|c| c

Putting everything together we have

v s
n r
i u (i)n (i)n
Y u
I= =u Qn = (17.103)
akk det(a)
akk
t
k=1
k=1

This last result holds for a ddimensional real, symmetric matrix (ajk ) with
det(ajk ) 6= 0. In order to complete the evaluation of the propagator in equa-
tion(17.64) we split off the factor m/2~N in the definition (17.79) of (ajk )
defining a new matrix (Ajk ) through
m
ajk = Ajk (17.104)
2~N
Using
 N 1
m
det(ajk ) = det(Ajk ) (17.105)
2~N
which is a general property of determinants, we get
 N/2  (N 1)/2
m 2i~N 1
K(0, tN ; 0, t0 ) = lim p (17.106)
N 2i~N m det(Ajk )
In order to determine det(Ajk ) we consider the dimension n of (Ajk ), presently
N 1, as a variable and let n = 1, 2, ...... We seek to evaluate the determinant
of the n n matrix

2 1 0 . . 0 0
1 2 1 . . 0 0

0 1 2 . . 0 0

.
Dn = det . . . . . . (17.107)
. . . . . . .

0 0 0 . . 2 1
0 0 0 . . 1 2
For this purpose, we expand (17.107) in terms of subdeterminants along the
last column. One can readily verify that this procedure leads to the following
recursion equation for the determinants
Dn = 2Dn1 Dn2 (17.108)

1367
To solve this three term recursion relationship one needs two starting values.
Using  
2 1
D1 = det(2) = 2 , D2 = det =3 (17.109)
1 2
we get
Dn = n + 1 (17.110)
Therefore, we get
det(Ajk ) = N (17.111)
which gives
 1/2  1/2
m m
K(0, tN ; 0, t0 ) = lim = (17.112)
N 2i~N N 2i~(tN t0 )

and from (17.76)

1/2
im (x x0 )2
  
m
K(x, t; x0 , t0 ) = exp (17.113)
2i~(t t0 ) 2~ t t0

which agrees with our earlier result (17.33).

The propagator allows us to predict the time evolution of any state function
(x, t) of a free particle. The result can be generalized to three dimensions as
3/2
im (~r ~r0 )2
  
m
K(~r, t; ~r0 , t0 ) = exp (17.114)
2i~(t t0 ) 2~ t t0

17.6.2 Propagator for a Quadratic Lagrangian

We now need to develop equations to reproduce the result for a harmonic oscil-
lator in equations (17.40 and (17.41)).

We want to determine the propagator

x(tN
Z)=xN  
i
K(xN , tN ; x0 , t0 ) = D[x(t)] exp S[x(t)] (17.115)
~
x(t0 )=x0

for the quadratic Lagrangian

1 1
L(x, x, t) = mx2 c(t)x2 e(t)x (17.116)
2 2
We need to determine the action integral
ZtN
S[x(t)] = dt0 L(x, x, t0 ) (17.117)
t0

1368
for an arbitrary path x(t) with end points x(t0 ) = x0 and x(tN ) = xN . In order
to simplify this task we again define a new path y(t)

x(t) = xcl (t) + y(t) (17.118)

which describes the deviation from the classical path xcl (t) with end points
x(t0 ) = x0 and x(tN ) = xN so that

y(t0 ) = y(tN ) = 0 (17.119)

This gives

L(xcl + y, xcl + y, t) = L(xcl , xcl , t) + L0 (y, y(t), t) + L (17.120)

where
1 1
L(xcl , xcl , t) = mx2cl c(t)x2cl e(t)xcl (17.121)
2 2
1 1
L0 (y, y(t), t) = my 2 c(t)y 2 e(t)y (17.122)
2 2
L = mxcl y(t) c(t)xcl y e(t)y (17.123)

We now show that the contribution of L to the action integral (17.117) vanishes.
For this purpose we use

d
xcl y = (xcl y) xcl y (17.124)
dt
and get

ZtN ZtN
t
dtL = m (xcl y)|tN0
dt [mxcl (t) + c(t)xcl (t) + e(t)]y(t) (17.125)
t0 t0

According to (17.119) the first term on the RHS vanishes. Getting the Euler-
Lagrange equations from the Lagrangian (17.116) we have for the classical path

mxcl (t) + c(t)xcl (t) + e(t) = 0 (17.126)

and hence, the second term on the RHS also vanishes. Thus,

ZtN
dtL = 0 (17.127)
t0

We then have
 
i
K(xN , tN ; x0 , t0 ) = exp S[xcl (t)] K(0, tN ; 0, t0 ) (17.128)
~

1369
where
x(tN
Z)=xN ZtN

i
K(0, tN ; 0, t0 ) = D[y(t)] exp dt L0 (y, y, t) (17.129)
~
x(t0 )=x0 t0

For the quadratic Lagrangian, we have achieved a separation in terms of a

classical action integral and a propagator connecting the end points y(t0 ) = 0
and y(tN ) = 0 which is analogous to the result (17.76). A similar discretization
scheme gives in this case
K (0, tN ; 0, t0 )
N/2 Z Z Z
!
N 1 2 2
c y

m iN X m(yj+1 yj ) j j
= lim dy1 dy2 ..... dyN 1 exp
N 2i~N ~ j=0 22N 2

(17.130)
where cj = c(tj ), tj = t0 + N j. Again, we can write the exponent in (17.130)
using a quadratic form
NX1
E=i yj ajk yk (17.131)
j,k=1

where the ajk are the elements of the following symmetric (N 1) (N 1)

matrix

2 1 0 . . 0 0
1 2 1 . . 0 0

0 1 2 . . 0 0
m
(ajk ) = . . . . . . .
2~N

. . . . . . .

0 0 0 . . 2 1
0 0 0 . . 1 2

c1 0 0 . . 0 0
0 c2 0 . . 0 0

0 0 c3 . . 0 0
N

. . . . . . . (17.132)
2~

. . . . . . .

0 0 0 . . cN 2 0
0 0 0 . . 0 cN 1
In the case, det (ajk ) 6= 0, we can express the multiple integral in (17.130) using
(17.103) as
N 1 1/2
 N/2 !
m (i)
K (0, tN ; 0, t0 ) = lim
N 2i~N det(a)
" #1/2
m 1
= lim
N 1 (17.133)
N 2i~
N 2~N det(a)
m

1370
Therefore we need to evaluate the function
"  N 1 #
2~N
f (t0 , tN ) = lim N det(a) (17.134)
N m

According to (17.132) we have

 N 1
def 2~N
DN 1 = det(a) (17.135)
m
2

2 mN c1 1 0 . . 0 0
2
1 2 mN c2 1

. . 0 0
2

m
0 1 2 mN c3 . . 0 0

= . . . . . . .
2~N

. . . . . . .

2N
. 2 1

0 0 0 . m cN 2

2N
0 0 0 . . 1 2 m cN 1
In the following we will assume that the dimension n = N 1 of the matrix
is a variable (as in the earlier derivation). One can then derive a recursion
relationship for the Dn
2
 
Dn = 2 N cn Dn1 Dn2 (17.136)
m
using the method of expanding a determinant in terms of subdeterminants as
earlier. Using the starting values
2N
 
D0 = 1 ; D1 = 2 c1 (17.137)
m
this recursion relationship can be used to determine Dn1 . We can write
(17.136) as a 2nd-order difference equation
Dn+1 2Dn + Dn1 cN +1 Dn
= (17.138)
2N m
Since we are interested in the solution of this equation in the limit of vanish-
ing N we can interpret (17.138) as a 2nd-order differential equation in the
continuous variable t = nN + t0
d2 f (t0 , t) c(t)
= f (t0 , t) (17.139)
dt2 m
The boundary conditions at t = t0 (according to (17.137)) are

f (t0 , t0 ) = N D0 = 0 (17.140)
2N 2N

df (t0 , t)  D1 D0
= N = 2 c 1 1 = 1 c1 = 1 (17.141)
dt t=t0 N m m

1371
We finally get for the propagator (17.115)
 1/2  
m i
K(x, t; x0 , t0 ) = exp S[xcl (t)] (17.142)
2i~f (t0 , t) ~
where f (t0 , t) is the solution of ((17.139)-(17.141)) and where S[xcl (t)] is deter-
mined by first solving the Euler-Lagrange equations for the Lagrangian (17.116)
to obtain the classical path xcl (t) with end points xcl (t0 ) = x0 and xcl (tN ) = xN
and then evaluating (17.117) for this path. Note that the required solution
xcl (t) involves a solution of the Euler-Lagrange equations for boundary con-
dition which are different from those conventionally encountered in Classical
Mechanics where we usually find a solution for initial conditions xcl (t0 ) = x0
and xcl (t0 ) = v0 .

17.6.3 Propagator of a Harmonic Oscillator

In order to illustrate the evaluation of (17.142) we consider the case of a har-
monic oscillator (we just quoted the result of the calculation earlier). In this
case, the coefficients in the Lagrangian (17.116) are c(t) = m 2 and e(t) = 0,
i.e., the Lagrangian is
1 1
L(x, x, t) = mx2 m 2 x2 (17.143)
2 2
We first determine f (t0 , t). In this case we have

f = 2 f , f (t0 , t0 ) = 0 , f(t0 , t0 ) = 1 (17.144)

The solution which obeys these boundary conditions is

sin (t t0 )
f (t0 , t) = (17.145)

We now determine S[xcl (t)]. For this purpose we first find the path xcl (t) which
obeys xcl (t0 ) = x0 and xcl (t) = x and satisfies the Euler-Lagrange equation for
the harmonic oscillator

mxcl + m 2 x2cl = 0 xcl ( ) = A sin ( t0 ) + B cos ( t0 ) (17.146)

Imposing the boundary conditions gives

x x0 cos (t t0 )
B = x0 , A= (17.147)
sin (t t0 )
and the desired path is
x x0 c
xcl ( ) = sin ( t0 ) + x0 cos ( t0 ) (17.148)
s
where we have introduced the notation

c = cos (t t0 ) , s = sin (t t0 ) (17.149)

1372
We now determine the action integral associated with the path ((17.148),(17.149))
Zt  
1 1
S[xcl (t)] = d mx2cl ( ) m 2 x2cl ( ) (17.150)
2 2
t0

We assume that t0 = 0. From (17.148) the velocity along the classical path is
x x0 c
xcl ( ) = sin x0 cos (17.151)
s
and for the kinetic energy we get
2
m 2 m 2 (x x0 c)
xcl ( ) = cos2
2 2 s2
x x0 c m 2 x20
m 2 x0 cos sin + sin2 (17.152)
s 2
Similarly, we get the potential energy
2
m 2 2 m 2 (x x0 c)
xcl ( ) = sin2
2 2 s2
x x0 c m 2 x20
m 2 x0 cos sin + cos2 (17.153)
s 2
Using
1 1 1 1
cos2 = + cos 2 , sin2 = cos 2 (17.154)
2 2 2 2
1
cos sin = sin 2 (17.155)
2
the Lagrangian, considered as a function of , reads
1 1
g( ) = mx2cl ( ) m 2 x2cl ( )
2 2 !
2 2
m (x x0 c)
= x20 cos 2 m 2 sin 2 (17.156)
2 s2

Evaluation of the action integral

Zt
S[xcl (t)] = d g( ) (17.157)
0

requires the integrals

Zt Zt
sin 2 sc 1 cos 2 s2
d cos 2 = = , d sin 2 = = (17.158)
2 2
0 0

1373
where we have used the definitions (17.149). We finally get
m  2
x0 + x2 cos (t t0 ) 2x0 x


S[xcl (t)] = (17.159)
2 sin (t t0 )

The expression for the propagator of the harmonic oscillator is then given by

K(x, t; x0 , t0 ) =
 1/2  
m im  2 2


exp x0 + x cos (t t0 ) 2x0 x
2i~ sin (t t0 ) 2~ sin (t t0 )
(17.160)

17.7 Problems
17.7.1 Path integral for a charged particle moving on a
plane in the presence of a perpendicular magnetic
field
Consider a particle of mass m and charge e moving on a plane in the presence of
an external uniform magnetic field perpendicular to the plane and with strength
B. Let ~r = (x1 , x2 ) and p~ = (p1 , p2 ) represent the components of the coordinate
~r and of the momentum p~ of the particle. The Lagrangian for the particle is
 2
1 d~r e d~r ~
L= m + A(~r)
2 dt c dt

1. Find the relation between the momentum p~ and the coordinate ~r and
explain how the momentum is related to the velocity ~v = d~r/dt in this
case.

2. Show that the classical Hamiltonian of for this problem is

1  2 e ~ 2
H(q, p) = p~ A(~r)
2m c
~ r) is the vector potential for a uniform magnetic field, normal to
where A(~
the plane, and of magnitude B. In what follows, we will always write the
~ r) = 0, where it is given by
vector potential in the gauge A(~

B B
A1 (~r) = x2 , A2 (~r) = x1
2 2

3. Use canonical quantization to find the quantum mechanical Hamiltonian

and the commutation relations for the observables.

4. Derive the form of the path integral, as a sum over the histories of the
position ~r(t) of the particle, for the transition amplitude of the process in

1374
 which the particle returns to its initial location ~r0 at time tf having left
that point at ti , i.e.,
h~r0 , tf | ~r0 , ti i
where ~r0 is an arbitrary point of the plain and |tf ti | . What is the
form of the action? What initial and final conditions should be satisfied
by the histories ~r(t)?

17.7.2 Path integral for the three-dimensional harmonic

oscillator
Consdider a harmonic oscillator of mass m and frequency in three dimensions.
We will denote the position vector of the oscillator by ~r = (x, y, z). The classical
Hamiltonian is
p~ 2 1
H(~r, p~) = + m 2~r 2
2m 2
Derive an expression for the path integral for the matrix element

h~rf = 0, tf | ~ri = 0, ti i

for this three-dimensional oscillator, where tf + and ti . Make sure

you explain how this limit is taken. HINT: you will find it convenient to write
the path integral in terms of the histories of the three components x(t), y(t)
and z(t).

17.7.3 Transitions in the forced one-dimensional oscillator

Consider a one-dimensional oscillator of mass m and frequency , labeled by
the coordinate q(t) on an infinite line. The oscillator is subject to an external
force J(t) of the form

J(t) = W 2
t + 2
1. What are the units of W ? Use W and m to construct a quantity with
units of energy.

2. Using path integral methods calculate the amplitude

hq = 0, tf | q = 0, ti i

for ti and tf +.

3. How does the expression you found depend on W , , m and ? Give

a physical interpretation to this dependence by looking at the extreme
regimes of large and small (relative to what?).

4. What dependence on W would you have expected to find in the Born

approximation? And in higher orders in perturbation theory?

1375
17.7.4 Greens Function for a Free Particle
The Greens function for the single-particle Schrodinger equation is defined as
the solution of the equation
h i
i~t H G(~r, t; ~r0 , t0 ) = i~(t t0 )(~r vecr0 )

Find explicitly the expression for the greens function, G0 (t t0 , ~r ~r0 ), of a

free particle in one and two spatial dimensions in real space-time representation.
HINT: Use the Fourier transform to solve the above equation. Shift the pole in
the Greens function, G(, p~), (  + i0) and use the inverse Fourier transform
to obtain the representation of interest.

17.7.5 Propagator for a Free Particle

The single-particle propagator that appears in the derivation of the Feynman
path integral is defined as the solution of the equation:
h i
i~t H K(x, t; xi , ti ) = i~(t ti )(x xi )

which is to be complemented with the initial condition:

K(x, t + 0; xi , ti ) = (x xi )

We have derived the following path integral expression for the propagator:
Z  
i
K(x, t; xi , ti ) = N [Dx(t)]exp S[x(t)]
~

where N is a normalization constant and S is the classical action understood

as a functional of x(t).

Using the above definition of the path integral, calculate explicitly the propaga-
tor for a free particle in one spatial dimension. Compare your result with that
of Problem 17.7.4.

1376
Chapter 18

Solid State Physics

18.1 Crystal Structure and Symmetry

Crystalline solids are regular periodic arrays of atoms and are characterized by
having long-range spatial order. The periodicity of crystals and the resultant
high symmetry makes it easier to study these systems.

We will use the Born-Oppenheimer approximation scheme we used earlier with

molecules. In this case, we assume that the motion of the electrons in the solid
can be dealt with separately from that of the nuclei.

The nuclear part of the problem involves several types of motion. For a solid
with N nuclei and hence 3N nuclear degrees of freedom we have

1. Translation of the solid as a whole (3 degrees of freedom). This is motion

of the center of mass of the entire crystal.

2. Rotation of the solid as a whole (3 degrees of freedom). This is rigid-body

rotation of the crystal (no nuclear motion) about the center of mass.

3. Vibration of the nuclei in the solid (3N 6 degrees of freedom). This

motion is called lattice vibrations.

To study the electron motions in the crystal, we ignore the lattice vibrations
and the motion of the crystal as a whole. We assume that the nuclei are fixed at
their equilibrium positions (the crystal ). The electronic Schrodinger equation is
then
~ ) = Ee e (~ri ; R
H 0 e (~ri ; R ~ ) (18.1)

1377
where

H 0 = Te + Vee + Ven (18.2)

Te = electron kinetic energy
Vee = electron electroninteraction , Ven = electron - nucleus interaction
~ ) = electron wavefunctions
e (~ri ; R
Ee = electron energies
~ = nuclear coordinates
~ri = electroncoordinates , R

Although we can formally write these equations, we have a problem. The solid
contains a very large number of electrons (and nuclei), i.e., for a solid 1 cm3
in volume we have on the order of 1023 electrons, which leads to an enormous
number of coupled equations. We cannot solve these equations.

We will use a one-electron approach to this multielectron system. We assume

the so-called orbital approximation where we write the full electronic wave func-
tion in terms of single-particle wave functions. This means that we approximate
the electron-electron interaction terms in the Hamiltonian with a smooth av-
erage potential, which reflects the periodicity of the crystal. It turns out that
the details of the approximation are not crucial as long as the periodicity is
accurately incorporated. This approach leads eventually to the band theory of
solids. Real crystals are extraordinarily complicated environments which make
the study of crystals very difficult. They are finite so that there are surface
effects and they contain defects and impurities which destroy the symmetry of
the crystal structure. While these complications lead to interesting effects, they
are not important for the study of the energy level structure. We will assume
ideal crystals, i.e., we confine our attention to regions deep inside the crystal
so that we can neglect surface effects. These are called bulk regions and their
properties are independent of any surface details. Essentially, we are assume
that the crystals are infinite in extent. We also assume that the crystals are pure
and without defects so that the crystal symmetry is perfect. Finally, we only
include electrostatic terms in the Hamiltonian. We will restrict our attention to
those features of the crystal structure and associated symmetry that are needed
to study the electronic energy level structure.

18.1.1 Symmetry of the Crystal System

A symmetry operation is a coordinate transformation that leaves the system
unchanged or invariant. In crystals, we say that the system and the associated
Hamiltonian are invariant under some symmetry operations. Symmetry opera-
tions provide a scheme for classifying crystal structures and , in addition, will
give us a powerful method for determining the energy level structures.

1378
Translational Symmetry
A crystal(ideal) has translational symmetry. This means that we can pick it
up and move it through a specified vector and we cannot distinguish the final
crystal from the original. For example, we consider a one-dimensional crystal
as shown in Figure 18.1 below.

Figure 18.1: 1-dimensional crystal

In the one-dimensional crystal, which extends the entire length [, +] of

the xaxis, each atom is separated from its nearest neighbor by a distance d.
The choice of origin is arbitrary. If we translate the crystal in either direction
by an amount nd where n = 1, 2, 3, .... the resulting crystal is indistinguishable
from the original. The crystal is then said to have translational symmetry and
the vectors
T~ = nex , n = 1, 2, ...... (18.3)
are called lattice translation vectors. The word lattice as used here refers to the
set of mathematical points which characterize the translational symmetry of the
crystal.

We say that two points on the are equivalent if the surrounding crystal looks
the same independent of which of the two points we stand on. Clearly, any two
such points must be connected by a lattice translation vector. The points do
not have to be lattice points.

We generalize to 3 (or 2) dimensions by defining the general translation

vector as
T~ = n1~a + n2~b + n3~c (18.4)
where

n1 , n2 , n3 are any integers

~a , ~b , ~c are any 3 noncoplanar lattice translation vectors in the crystal

All of these vectors are defined with respect to some origin(arbitrary). This is
illustrated for a 2dimensional crystal in Figure 18.2 below.

1379
 Figure 18.2: 2-dimensional crystal

In Figure 18.3 below, translational invariance means that the interactions felt
by an electron at position ~r + T~ are identical to those felt at position ~r.

Figure 18.3: Translation Invariance

The statement of translational invariance of the crystal in terms of the Hamil-

tonian operator means that

H(~r + T~ ) = H(~r) (18.5)

The choice of the three vectors ~a, ~b and ~c is not unique. In Figure 18.4 below
we illustrate several choices (all equally good).

Figure 18.4: Alternate Basis Vectors

1380
    
The vector pairs ~a, ~b and ~a0 , ~b0 are primitive lattice translation vectors(not
unique). The appellation primitive is attached because using the
 pair as a basis
we can reach all lattice points. On the other hand the pair ~a00 , ~b00 are not
primitive since a point like P cannot be reached.

The lattice, which is the set of all mathematical points that are equivalent
under translation by T~ = n1~a + n2~b + n3~c where ~a, ~b and ~c are primitive lattice
translation vectors, defines the structure of the crystal. The crystal consists of
lattice + basis, where a basis is one or more atoms that are attached to all the
lattice points.

As we have seen, the choice of basis is arbitrary. Sometimes one chooses a

region of the crystal as the basis, where the region chosen can be stacked to
generate the entire crystal. Such a region is called a unit cell. In Figure 18.4
above, the three sets of basis vectors each define a unit cell (a parallelogram in
2 dimensions).

A more detailed definition goes as follows: A volume of the crystal that when
translated through all translation vectors, just fills all of space without either
overlapping itself or leaving any gaps is called a primitive cell or primitive  unit

cell. It must contain exactly one lattice point. In Figure 18.4 above, ~a, ~b
   
and ~a0 , ~b0 generate primitive unit cells, but ~a00 , ~b00 does not (it contains two
atoms).

In addition, every primitive cell has the same area (volume). In Figure18.5
below we have drawn several unit cells for a rectangular-centered lattice. The
cells labelled (a), (b) , and (d) are primitive while (c) is not.

Figure 18.5: Primitive Cells

Point Symmetry

There are another group of symmetry operations for a crystal which leave one
point fixed in space. They are called point symmetry operations. The three most
important point symmetry operations are rotation(R), reflection(F) and inver-
sion(V). Not all crystal structures are characterized by all of these operations.
In fact, they can be used to classify lattices and crystals.

1381
Digression to Group Theory
A group is a set of elements A, B, C, .... such that a group multiplication is
defined which associates a third element with any ordered pair of two elements.
The multiplication must satisfy these requirements:

1. The product of any two elements in the set is in the set also, i.e., the set
is closed under group multiplication.

2. The associative law holds, i.e., A(BC) = (AB)C.

3. There exists an identity element E such that EA = AE = A for any

element A.

4. There exists an inverse A1 for every element in the group, i.e., AA1 =
A1 A = E.

The order of the group is equal to the number of elements.

An example is the set of four elements (group of order 4) with the multiplication
table below

E A B C
E E A B C
A A E C B
B B C E A
C C B A E

Table 18.1: Group with 4 elements

If group multiplication is commutative, i.e., AB = BA for any pair of elements,

then the group is Abelian. The above table represents a 4 element Abelian
group.

Another example is the non-Abelian group of order 6 (6 elements) given by the

table below

E A B C D F
E E A B C D F
A A E D F B C
B B F E D C A
C C D F E A B
D D C A B F E
F F B C A E D

Table 18.2: Group with 6 elements

1382
If the group multiplication operation were ordinary matrix multiplication, then
a representation of this group (an explicit set of elements) is given by
   
1 0 1 0
E= , A=
0 1 0 1
   
1 1 3 1 1 3
B= , C=
2 3 1 2 3 1
   
1 1 3 1 1 3
D= , F =
2 3 1 2 3 1

Notice that each row/column only contains each element once. This is an ex-
ample of the rearrangement theorem, which says

In the sequence EAk , A1 Ak , A2 Ak , A3 Ak , ....., Ak Ak , ....... each group element Aj

appears exactly once (as one of the products). The elements are merely rear-
ranged by multiplying each by Aj .

Returning to point symmetries, we first consider the 1dimensional crystal,

which is particularly simple. Its point symmetry operations form an Abelian
group of order 2 given by

E F
E E F
F F E

Table 18.3: Abelian group of order 2

In this case, the two elements are

E = identity does nothing

F = ref lection takes x x

In 2 dimensions, the possible point symmetry operations are

1. The identity operation.

2. Reflection through a mirror line in the plane of the crystal.

3. An n-fold rotation (through angle 2/n) about an axis perpendicular to

the plane of the crystal where n = 2, 3, 4, or6.

Notice that n = 5isnotallowed. The symmetry elements corresponding to these

operations are the mirror line and the nfold axes (inversion is equivalent to a
twofold rotation).

In 3 dimensions, the possible point symmetry operations are

1383
 1. The identity operation.

2. Reflection through one or more mirror planes.

3. An nfold rotation about an axis where n = 2, 3, 4, or6.

4. Inversion through a center of symmetry (~r ~r).

The last operation, inversion, is actually equivalent to a rotation by followed

by reflection through a plane perpendicular to the axis of rotation. Because, of
its importance, however, we keep it as a separate symmetry operation.

Since each point symmetry operation is a coordinate transformation, when we

say that the crystal has a particular symmetry element (e.g., a center of sym-
metry), we mean that the corresponding symmetry operation (e.g., inversion)
leaves the Hamiltonian of the crystal unchanged.

Classification of Two-Dimensional Lattices and Crystals

Lattices having different sets of point symmetries are labelled as distinct Bra-
vais lattices. We can understand what distinguishes Bravais lattices using
2dimensional lattices as an example. In two dimensions there are 5 Bravais
lattices:

the oblique lattice, the primitive rectangular lattice,

the centered rectangular lattice, the square lattice,
and the hexagonal lattice.

The oblique lattice is the Bravais lattice of lowest symmetry. It is shown in

Figure 18.6 below:

Figure 18.6: Oblique Lattice

The only symmetry operations for this lattice is twofold rotations about axes
perpendicular to the plane of the crystal and passing through each lattice point
and about axes halfway between each pair of lattice points as shown above. The
primitive lattice translation vectors ~a and ~b have no simple relationship to each
other.

1384
Each of the other four Bravais lattices are special cases of the oblique lattice.
Each has more symmetry operations than the oblique lattice and, in each, the
primitive lattice translation vectors ~a and ~b have some definite relationships.

The primitive rectangular lattice shown in Figure 18.7 below has the same
twofold rotation axes as the oblique lattice and in addition has several mir-
ror line as shown. Reflection through mirror lines is a symmetry operation of

Figure 18.7: Primitive Rectangular Lattice

the lattice. The primitive lattice translation vectors ~a and ~b are orthogonal to
each other and their lengths are unrelated.

Another type of rectangular lattice is the centered rectangular lattice shown in

Figure 18.8 below.

Figure 18.8: Centered Rectangular Lattice

This lattice has the twofold rotation axes, mirror lines and glide lines as shown.

1385
A glide line is a symmetry element corresponding to a combined operation: to
get from one point in the crystal to an equivalent point via a glide line, we first
reflect across the glide line, and then translate along the glide line by 1/2 the
repeat distance of the crystal. The angle between the lattice translation vector
is as shown and their lengths are unrelated.

The square
  lattice is a special case of the primitive rectangular lattice where
|~a| = ~b. It is shown in Figure 18.9 below. In addition to the twofold rotation
 

Figure 18.9: Square Lattice

axes, mirror lines and glide lines of the primitive rectangular lattice, the square
lattice has the fourfold rotation axes as shown in the diagram. The primitive
lattice translation vectors are orthogonal and equal in length.

The hexagonal
  lattice is derived from the centered rectangular lattice by setting
|~a| = ~b as shown in in Figure 18.10 below. This lattice has twofold , threefold
 
and sixfold axes of rotation as shown. It has a large number of mirror line and
glide lines as shown. The primitive lattice translation vectors are of equal length
and make an angle of 60 with respect to each other.

1386
 Figure 18.10: Hexagonal Lattice

Symmetry of the Crystal

Up to this point we have dealt exclusively with the classification of lattices ac-
cording to their point symmetry. As we stated earlier, however, a crystal is
composed of a lattice plus a basis at each lattice point. Because of the basis, the
point symmetry of the crystal may not be the same as that of the lattice, i.e.,
the symmetry of the crystal may be lower than that of the lattice. The lattice
may possess some symmetry elements not possessed by the crystal. This is a
very important point. Let us illustrate it with an example.

We consider a square lattice and, in particular, we look at the mirror lines

through a lattice point as shown in Figure 18.11 below.

Figure 18.11: Mirror Lines in Square Lattice

Reflection through any of these mirror lines yields a lattice that is indistinguish-
able from the original lattice.

1387
Now associating with each lattice point of this square lattice one of the two
different bases as shown in Figure 18.12 below

Figure 18.12: Possible Bases

we get the crystals shown in Figure 18.13 below.

Figure 18.13: Possible Crystals

It is clear that the crystal on left possesses all of the same mirror lines as the
original square lattice while the crystal on the right does not. The symmetry of
the crystal on the right is lower than that of its lattice.

Three-Dimensional Crystals
Our discussions will mostly involve one- and two-dimensional crystals. Three-
dimensional crystals are significantly more difficult to study and we will only
say some general things about them at the end.

In three dimensions there are 14 Bravais lattices. Three examples are shown in
Figure 18.14 below.

Figure 18.14: 3-Dimensional Crystals

1388
From each Bravais lattice we can construct crystals by associating a basis with
each lattice point.

Miller Indices
In order to discuss three dimensional crystals we need to be able to specify
particular directions (axes) and particular planes in the crystal.

A convention has been developed for dealing with this problem. Axes and planes
in a crystal are described by Miller indices.

For example, to specify a particular axis in a crystal, we first write a lattice

translation vector
T~ = n1~a + n2~b + n3~c (18.6)
that points along this axis. We then determine any integer factor N common
to n1 , n2 and n3 .

The Miller indices of the axis are then given by the set of integers
hn n n i
1 2 3
, , = [h, k, l] (18.7)
N N N
where negative indices are denoted by a bar over the integer.

An example is shown below for the body diagonal of a cubic lattice where

T~ = ~a + ~b + ~c N = 1 [1, 1, 1] (18.8)

as shown in Figure 18.15 below.

Figure 18.15: [111] axis in cubic lattice

To specify a particular plane in a crystal, we first determine the intercepts of

the plane on three noncoplanar lattice translation vectors ~a, ~b and ~c and call
these intercepts a, b and c.

We then form the reciprocals of these intercepts. Finally, we multiply these

1389
reciprocals by a common factor that reduces them to the three smallest integers.
These are the Miller indices of the plane, written (h, k, l). Some examples are
shown in Figure 18.16 below.

Figure 18.16: [111] axis in cubic lattice

A slightly more complicated example is one where the intercepts are (3, 4, 12)

1390
so that
 
1 1 1
(a, b, c) = (3, 4, 12) , , 12 = (4, 3, 1) = (h, k, l) (18.9)
3 4 12

The Miller indices for an axis represent all equivalent axes (all parallel). The
Miller indices for a plane represent all equivalent planes (all parallel).

This concludes our short introduction to crystal structure and symmetry. The
symmetry of the crystal is important beyond what it tells us about the clas-
sification of the crystal. It reflects the fundamental properties of the crystal
Hamiltonian and in particular, the crystal potential energy. We will use these
properties to deduce facts about the energy eigenfunctions for the electrons in
the crystal.

18.2 Bloch Theorem, the Reciprocal Lattice and

Brillouin Zones
We first study the general properties of electron wavefunctions in crystals that
are a consequence of the periodicity of the crystal.

The electronic Schrodinger equation for a single electron in the crystal is given
by
HE (~r) = EE (~r) (18.10)
In the orbital approximation, the potential energy of the electron consists of
terms describing its interaction with the static array of nuclei and average po-
tential energy terms that take into account interactions with the other electrons
in the crystal. The interaction potential with the nuclei clearly has the periodic-
ity of the crystal and we assume that the average potential energy is constructed
so that it also has the periodicity of the crystal. This means that the total po-
tential energy is invariant under translation, i.e.,

V (~r + T~ ) = V (~r) (18.11)

for any lattice translation vector T~ . Since the kinetic energy operator is also
translation invariant we then have

H(~r + T~ ) = H(~r) (18.12)

i.e., the electronic Hamiltonian is invariant under translations.

18.2.1 Translation Operators in Configuration Space

We now introduce the translation operator Top which we define by its action on
any function f (~r) as
Top f (~r) = f (~r + T~ ) (18.13)

1391
Each lattice translation vector T~ = n1~a + n2~b + n3~c has a corresponding trans-
lation operator Top = Top (n1 , n2 , n3 ).

We define the product of two translation operators by the relation

0
Top 0
Top f (~r) = Top f (~r + T~ ) = f (~r + T~ + T~ 0 ) = Top
00
f (~r) (18.14)

i.e., the product of two translation operators is another translation operator

00
given by successive application of the two operators. Top corresponds to the
translation T~ + T~ . This means that
0

0
Top (n1 , n2 , n3 )Top (n01 , n02 , n03 ) = Top
00
(n1 + n01 , n2 + n02 , n3 + n03 ) (18.15)
T~ 00 = T~ + T~ 0 = (n1 + n0 )~a + (n2 + n0 )~b + (n3 + n0 )~c
1 2 3 (18.16)

This result will have important consequences later.

The invariance of H under translation by any lattice translation vector T~ is

expressed by the commutator relation
h i
H, Top = 0 (18.17)

for any Top . In addition, from the definition we can see that

0 0
T op Top = Top T op (18.18)

i.e., the order of application is unimportant. This gives another commutator

relation h i
0
Top , Top =0 (18.19)
 n o
These commutators say that the set of operators H, Top not only form a
closed commutator algebra but also that the Hamiltonian H and all lattice trans-
lation operators Top are a set of mutually commuting operators. This means
that we can find a set of simultaneous eigenfunctions for all these operators. If
we assume that E is a particular non-degenerate eigenfunction we can write
the eigenvalue equation

Top E = E = (n1 , n2 , n3 )E (18.20)

where is a constant(the eigenvalue). This property of the energy eigenfunc-

tions is a direct consequence of the translational invariance of the crystal.

What can we say about the constant ?

1392
18.2.2 Derivation of Blochs Theorem
All the energy eigenfunctions are normalized so that
Z
2
|E (~r)| d3 r = 1 (18.21)

This says that the eigenfunction

E (~r + T~ ) = Top E (~r) = E (~r) (18.22)
must satisfy
Z  2 Z
2 2 2
E (~r + T~ ) d3 r = || |E (~r)| d3 r = || = 1 (18.23)
 

Thus, we can always write = ei where = (n1 , n2 , n3 ) is a constant. There

is one (or ) corresponding to each lattice translation vector T .

Applying successive translation operators we find that

0 0
Top Top E (~r) = Top E (~r) = 0 E (~r)
00
= Top E (~r) = 00 E (~r) (18.24)
00 0
00 = 0 ei = ei ei
(n1 , n2 , n3 ) + 0 (n01 , n02 , n03 ) = 00 (n1 + n01 , n2 + n02 , n3 + n03 ) (18.25)
i.e., not only is 00 the sum of 0 and but also the arguments of 00 are the
sum of the arguments of 0 and .

This says that we must have (n1 , n2 , n3 ) = k1 n1 + k2 n2 + k3 n3 , i.e., linear in

all of its arguments (k1 , k2 and k3 are constants). This, in turn, says that
is proportional to T~ , i.e., = ~k T~ , where ~k is a constant vector that is not
depend on n1 , n2 or n3 . We can then write for the eigenvalue of the translation
operator
~ ~
= ei = eikT (18.26)
and the eigenvalue equation takes the form
~ ~
Top E (~r) = E (~r + T~ ) = E (~r) = eikT E (~r) (18.27)
This is Blochs theorem of the first form.

It tells us that we can choose solutions of the Schrodinger equation such that
for every Hamiltonian eigenfunction E (~r) there exists a vector ~k called a wave
vector, such that (18.27) holds for any translation operator Top , that is, that
~ ~
E (~r + T~ ) differs from E (~r) only by a multiplicative phase factor eikT . Any
function that satisfies Blochs theorem is called a Bloch function or Bloch wave.

Blochs theorem is a direct consequence of the translational symmetry of the

crystal.

1393
The Wave Vector
The proportionality constant ~k has units of inverse length. For each Hamilto-
nian eigenfunction E (~r), there is a wave vector ~k. The relationship of ~k to the
wave function is given by Blochs theorem. This means that ~k can be viewed as
a quantum number for the function E (~r) and we can write E,~k (~r).

What are some of the properties of the wave vector ~k?

We first consider the one-dimensional version of Blochs theorem. Suppose that

we have a one-dimensional crystal with equally spaced atoms (separation dis-
tance = d). We ask for the relation ship between E (~r) and k for a wave function
of the form
E (x) = eix (18.28)
where is a real constant. In this case, Blochs theorem becomes

E (x + nd) = ei(x+nd) = eind eix = eind E (x) = eiknd E (x) (18.29)

It is clear that Blochs theorem is satisfied for an infinite set of kvalues given
by
2
k =+ m , m = 0, 1, 2, ...... (18.30)
d
where we have used the fact that ei2m = 1. This suggests that we cannot
uniquely define the wave vector ~k, i.e., there are an infinite number of wave
vectors that are equivalent to each other in the way that they are related to
the wave function E (x). This result generalizes to three dimensions. For each
single-particle Hamiltonian eigenfunction E (~r) we can find a set of different
wave vectors ~k that impose a similar restriction on E (~r), i.e., for which
~ ~
E (~r + T~ ) = eikT E (~r) (18.31)

In general, we choose the smallest of these wave vectors to label the eigenfunction
E,~k (~r).

Alternative Form of Blochs Theorem

We define a new function by
~
uE,~k (~r) = eik~r E,~k (~r) (18.32)

Now from the definition of the translation operator we have

~ ~
Top uE,~k (~r) = uE,~k (~r + T~ ) = eik(~r+T ) E,~k (~r + T~ ) (18.33)

Using Blochs theorem we get

~ ~ ~ ~ ~
Top uE,~k (~r) = eik(~r+T ) eikT E,~k (~r) = eik~r E,~k (~r) = uE,~k (~r) (18.34)

1394
This says that the new function uE,~k (~r) is a periodic function invariant under
translation by T~ , i.e., it has the periodicity of the crystal. Using this result we
rewrite the Hamiltonian eigenfunction as
~
E,~k (~r) = eik~r uE,~k (~r) (18.35)

which is the second form of Blochs theorem. This form shows that the electronic
eigenfunction of the crystal can be chosen to have the form of a periodic function
~
uE,~k (~r) modulated by a plane-wave envelope eik~r .

Generally, this envelope varies more slowly with ~r then does uE,~k (~r). The real
part of E,~k (~r) is shown in Figure 18.17 below, where it has been assumed that
uE,~k (~r) is a real function.

Figure 18.17: Second form of Bloch function

We have determined the qualitative form of the Hamiltonian eigenfunction for

any crystal using only translational symmetry of the crystal. Blochs theorem
in the first form links the wave function and the wave vector and in the second
form describes an important property of the wave function.

The Reciprocal Lattice

The strangest feature of Blochs theorem is the proportionality constant ~k. The
name (wave vector) and the symbol (~k) used for this vector suggest(in analogy
with the wave vector for free electrons) that ~k will be related to the energy.
This is, in fact, true as we shall show during the rest of this chapter. At this
point, however, we will just label the energy eigenfunctions by both energy and
wave vector.

Up to this point we have not restricted the magnitude of ~k in any way. Thus,

1395
there exists a continuum of values of ~k . We call this continuum ~kspace or
reciprocal space.

Each vector in ~kspace has three components, in general, and we write

~k = kx x + ky y + kz z (18.36)

where x, y and z are mutually orthogonal unit vectors in regular or coordinate or

direct space. The crystal exists in direct space and its lattice is called the direct
lattice. A lattice constructed in reciprocal space is called a reciprocal lattice.

Reciprocal Lattice Vectors

What do we mean precisely by the term reciprocal lattice?

Each crystal lattice in direct space has an associated reciprocal lattice in ~kspace.
Blochs theorem says that for a particular energy eigenfunction of the electronic
Hamiltonian, there exists a vector ~k (actually an infinity of such vectors) such
that
~ ~ ~ ~ ~
Top E,~k (~r) = Top eik~r uE,~k (~r) = eik(~r+T ) uE,~k (~r) = eikT E,~k (~r) (18.37)

~ that satisfies the condition

We now consider a vector, say G,
~ ~
eiGT = 1 (18.38)

~ is a special wave vector such that u ~ (~r) is invariant under any transla-
i.e., G E,k
tion Top . The condition implies
~ ~ ~ T~ ) + i sin(G
~ T~ ) = 1 G
~ T~ = 2n
eiGT = cos(G , n = integer (18.39)

or that
~ T~ = n1 G
G ~ ~b + n3 G
~ ~a + n2 G ~ ~c = 2n (18.40)
The integers n1 , n2 and n3 which define a particular T~ , can take on any values.
For each set {n1 , n2 ,n3 } there exists some integer n for which the above equality
holds. Thus, n depends on n1 , n2 and n3 .

For example, suppose that n2 = 0, n3 = 0 and n1 some arbitrary nonzero

integer. Then we have
~ ~a = 2n G
n1 G ~ ~a = 2na (18.41)

since the relation must be valid for any value of n1 . n1 is another integer given
by n = na n1 . Similarly, we must have
~ ~b = 2nb and G
G ~ ~c = 2nc (18.42)

1396
 ~
These three relations allow us to write a general form for G
~ = na A
G ~ + nb B
~ + nc C
~ (18.43)

Although this looks like the expression for T~ , remember, however, that G
~ is a
~
vector in kspace, i.e., that it is a vector such that
~ ~
eiGT = 1 (18.44)

Putting everything together we get the relations

~ ~a = 2 , B
A ~ ~a = 0 , C~ ~a = 0
~ ~b = 0 , B
A ~ ~b = 2 , C~ ~b = 0
~ ~c = 0 , B
A ~ ~c = 0 , C~ ~c = 2

~ is orthogonal to both ~b and ~c or that

The first column of relations say that A
we can write
~ = A~b ~c
A (18.45)
and
~ ~a = 2 = A~a (~b ~c)
A (18.46)
so that
~b ~c
~ = 2
A (18.47)
~a (~b ~c)
Similarly,
~ = 2 ~c ~a ~ = 2 ~a ~b
B , C (18.48)
~a (~b ~c) ~a (~b ~c)
The three vectors A,~ B ~ and C ~ are the primitive reciprocal lattice translation
~
vectors. Used together with G = na A ~ + nb B
~ + nc C ~ (the reciprocal lattice trans-
lation vector) they generate a lattice of points in ~kspace, called the reciprocal
lattice. By construction, we see that each reciprocal lattice is defined with re-
spect to a particular direct lattice(given by ~a, ~b and ~c) in coordinate space.

~ ~
Since eiGT = 1, all vectors ~k 0 = ~k + G
~ are equivalent in ~kspace, i.e.,
~0 ~ ~ ~ ~ ~ ~ ~ ~ ~ ~
eik T = ei(k+G)T = eikT eiGT = eikT (18.49)

There are an infinite number of such equivalent wave vectors where

~0 ~
E,~k0 (~r + T~ ) = eik T E,~k0 (~r) (18.50)

The Reciprocal Lattice for a Square Direct Lattice

We consider a crystal with a square direct lattice as shown inn Figure 18.18
below.

1397
 Figure 18.18: Square direct lattice

 
where |~a| = ~b. We have
 

~a = dx and ~b = dy (18.51)
The vector defining the reciprocal lattice in two dimensions is
~ = na A
G ~ + nb B
~ (18.52)

In order to calculate expressions for A ~ in terms of ~a and ~b, the primitive

~ and B
lattice translation vectors of the square direct lattice, we temporarily introduce
a third vector of the same length ~c = dz. This vector, which is orthogonal to
the plane of the direct lattice, is only a mathematical tool useful for the stated
purpose. Its properties will not enter the final answers.

We then have

~a (~b ~c) = d3 , ~b ~c = d2 x , ~c ~a = d2 y (18.53)

so that
~ = 2 x , B
A ~ = 2 y (18.54)
d d
This result shows that the reciprocal lattice of a square direct lattice is also a
square lattice with the spacing of adjacent lattice points in the x and y direction
given by 2/d.

This is, in fact, a general property of two-dimensional Bravais lattices . They are
self-reciprocal, i.e., the reciprocal lattice of a particular two-dimensional Bravais
lattice is another lattice of the same Bravais type.

The reciprocal lattice is shown in Figure 18.19 below.

1398
 Figure 18.19: Reciprocal Lattice for Square direct lattice

Some Other Properties

The reciprocal lattice of a particular direct lattice is unique, i.e., it is indepen-
dent of the choice of primitive lattice translation vectors. We note that, if we
start with a particular direct lattice in coordinate space, construct from it the
reciprocal lattice in ~kspace, and then construct the reciprocal of the reciprocal
lattice, we obtain the original direct lattice in coordinate space.

Brillouin Zones Made Simple

The reciprocal lattice is a periodic array of mathematical points in ~kspace,
where each point is defined by a lattice translation vector G. ~ Because of its
relationship to the electronic wave function E,~k (~r) via Blochs theorem, each
vector ~k even if it is not a G
~ vector, is important.

As we noted earlier, a particular ~k vector is not unique. The infinity of vectors

~k 0 = ~k + G
~ are all equivalent for each vector ~k.

We choose to label the wave functions by the smallest or most significant vector
in each of the sets of equivalent vectors.

We now ask the following question: Is there a particular region of the reciprocal
lattice that contains all of the significant ~k vectors for a particular crystal?

The First Brillouin Zone

n o
Since ~k is the smallest vector in the set ~k 0 = ~k + G ~ , we know that
    n o
~ 0  ~ 
k  k  for any ~k 0 in ~k 0 = ~k + G
~ (18.55)

1399
Consider the two different wave vectors ~k1 and ~k2 for a particular two-dimensional
crystal shown in Figure 18.20 and Figure 18.21 below. We also show the vectors
~k 0 = ~k1 G
~ and ~k20 = ~k2 G
~ (18.56)
1

~ is a particular reciprocal lattice vector for the crystal.

where G

   
Figure 18.20: ~k10  > ~k1 
   

       
In the Figure 18.20 above, ~k10  > ~k1  and in Figure 18.21below ~k10  > ~k1 .
       

   
Figure 18.21: ~k10  > ~k1 
   

Thus, ~k1 may be the smallest vector in its set, but ~k2 is certainly not the small-
est vector in its set.

One difference between ~k1 and ~k2 is that the tip of ~k1 lies to the left of the per-
~ whereas the tip of ~k2 lies to the right of the bisector.
pendicular bisector of G,

It turns out that any wave vector whose tip lies on the side of the bisector away

1400
from the origin, is equivalent to at least one smaller wave vector, which can be
generated by subtracting the lattice vector that was bisected. All the significant
wave vectors (i.e., those not equivalent to some smaller wave vector), lie on the
side of the perpendicular bisector closest to the origin.

~ , we bisect
Now, suppose that instead of bisecting one reciprocal lattice vector G
all such vectors from the origin to nearby points. Then these bisectors enclose
a particular region of reciprocal space that includes the origin. Figure 18.22
below gives this construction for the square lattice(in reciprocal space)

Figure 18.22: Construction of first zone

The enclosed region (shaded) is called the first Brillouin zone. The solid arrows
are lattice vectors and the dotted lines are their bisectors.

Each vector in the first Brillouin zone is the smallest vector in the set of equiv-
alent vectors since it lies on the side toward the origin of every perpendicular
~ Any vector whose tip lies outside the first zone is equivalent to
bisector of G.
some smaller vector whose tip is inside the first zone. The first Brillouin zone
contains all ~kvalues closer to the origin than any other point of the reciprocal
lattice.

Higher Brillouin Zones

The ~kvalues inside the first Brillouin zone have special significance. We will
generally restrict our attention to these values and ignore the infinity of other
vectors outside this zone. Each vector outside the first Brillouin zone is, of
course, equivalent to a smaller vector inside the first zone.

1401
We will, however, have to refer to these other vectors in certain situations and
it is useful to introduce a terminology for describing them. Suppose we wish,
for example, to discuss the second-smallest ~k vector for a particular electronic
state, i.e., the ~k vector that is smaller than all but one of the vectors in the set
of equivalent ~k vectors for this state. Where is this vector located in reciprocal
space? In Figure 18.23 below we have added several more bisectors to the Figure
18.22 that we used to describe the first zone. These extra bisectors also enclose
various regions of ~k space.

Figure 18.23: Construction of higher zones

Consider the shaded region in Figure 18.23. We reach this region from the origin
by crossing one boundary of the first zone. Each ~k vector in this region is inside
all Brillouin zone boundaries (the bisectors) except one - the one we crossed.
Therefore, for each vector in this region there is one and only one smaller ~k
vector, namely, ~k G,
~ where G
~ is the reciprocal lattice translation vector whose
bisector we crossed. This region is part of the second Brillouin zone.

The second Brillouin zone is defined as that region of ~k space containing each
wave vector that is smaller than all but one of the vectors equivalent to it.
Looking at Figure 18.23, there are four parts of the second zone (labelled by II
in Figure 18.23). Crossing one more boundary - one of the boundaries of region

1402
II - we enter a region of ~k space where each ~k vector is smaller than all but two
equivalent ~k vectors. This region is the third Brillouin zone (regions labelled
by III in Figure 18.23). In a similar manner the fourth zone can be identified
(labelled by IV in Figure 18.23).

Properties of Brillouin Zones

Central to the construction of the first Brillouin zone was the choice of a point
as the origin. The selection of the origin point, however, is arbitrary. We could
have chosen any point of the reciprocal lattice as the origin and constructed a
zone identical to the first Brillouin zone.

Alternatively, we could have generated these other regions by translating the

first Brillouin zone for some arbitrary origin by various lattice translation vectors
~ eventually filling all of reciprocal space as shown in Figure 18.24 below. Thus,
G,

Figure 18.24: First Brillouin zone as primitive cell

the first Brillouin zone is the primitive unit cell of ~k space.

Another property of Brillouin zones that are higher than the first zone is that,
although they consist of several disjoint segments, using an appropriate choice
~ we can translate each segment of
of reciprocal lattice translation vectors G,
any higher zone into the first zone as shown in Figure 18.25 below. When
all segments have been translated, they will fully cover the first zone with no
overlap, i.e., the area (volume in 3 dimensions) of each zone is identical.

1403
 Figure 18.25: Translating higher zone to first zone

Brillouin Zones in Three Dimensions

We can use the same methods to construct the Brillouin zones for three-dimensional
lattices. The main difference is that the perpendicular bisectors of the recipro-
~ are planes instead of lines. As an example we consider a
cal lattice vectors G
simple cubic lattice.

Brillouin Zones of a Simple Cubic Lattice

The three-dimensional analog of the square lattice is the simple cubic lattice
which is described by three mutually orthogonal primitive lattice translation
vectors
~a = dx , ~b = dy , ~c = dz (18.57)
The reciprocal lattice of a simple cubic lattice is also simple cubic. The sepa-
ration distance between reciprocal lattice points is 2/d. The primitive lattice
translation vectors defining the reciprocal lattice are

~ = 2 x
A , ~ = 2 y
B , ~ = 2 z
C (18.58)
d d d
To construct the first Brillouin zone, we choose a point of the reciprocal lattice
as the origin and draw reciprocal lattice translation vectors to all adjacent points
of the lattice.

The perpendicular bisectors (planes) of these vectors defines a volume which is

1404
the first Brillouin zone. The bisectors, in this case (see figure below) are six
planes perpendicular to the

x(100) , y(010) , z(001) (18.59)

axes and intersecting these axes at /d. The first zone, as shown in Figure
18.26 below, forms a cube.

Figure 18.26: First Brillouin zone of Simple Cubic Lattice

The second Brillouin zone of the simple cubic lattice is a little more difficult
to visualize because it involves more intersecting planes. It is defined by these
boundaries
1. the 6 planes that enclose the first Brillouin zone
2. the 12 planes
that are orthogonal to the (110) axes and intersect these
axes at 2(/d) as shown in the Figure 18.27 below.

1405
 Figure 18.27: Second Brillouin zone of Simple Cubic Lattice

Figure 18.28 below show three typical segments of the second Brillouin zone.

Figure 18.28: Parts of the Second Brillouin zone

The second zone is a dodecahedron with a cube(first zone) removed from its
center.

1406
Electronic States in a Brillouin Zone
Keep in mind that the significance of each vector in ~k space resides in its effect
on the electronic state of the crystal E,~k (~r). In order to determine an analytic
form for this function we must solve the Schrodinger equation for the crystal.

Whatever the form of the wave function, however, we know that it satisfies
Blochs theorem
~ ~
E,~k (~r + T~ ) = eikT E,~k (~r) (18.60)

Each different ~k vector in the first Brillouin zone corresponds to a different elec-
tron state. We now ask the question: How many distinct electronic states are
represented by wave vectors in the first Brillouin zone?

If the crystal were actually infinite, then there are an infinite number of different
wave vectors in the first zone and hence the number of electronic states is also
infinite. However, no real crystal is truly infinite. We now see how to take this
fact into account.

We denote by N the number of primitive unit cells in the crystal. We will now
prove that there are exactly N electronic state in the first Brillouin zone, i.e., N
distinct ~k vectors in that zone. Since each electronic state can be occupied by
at most two electrons, one with spin up and one with spin down , the electronic
states in the first zone can be occupied by at most 2N electrons.

We do the proof in one dimension (it can easily be extended to 2 and 3dimensions).
We consider a 1dimensional crystal of length L with lattice spacing d (so that
L = N d since there are N primitive cells). We associate a basis with each lat-
tice point. We are only interested in bulk properties so we neglect edge effects
and assume that L, the length of the crystal, is chosen so that we have periodic
boundary conditions
E,~k (0) = E,~k (L) (18.61)
i.e., the crystal repeats itself forever!

This gives, using Blochs theorem

~ ~
~ = eikL ~ (L)
~
E,~k (0) = E,~k (L) E,k (18.62)

or that
~ ~
eikL = 1 ~k L
~ = kL = kN d = 2n , n = integer (18.63)

Thus,
2n 2  n  N
k= = , n = 0, 1, 2, ......., (18.64)
L d N 2
The reciprocal lattice of this direct lattice is a lattice in ~k space of equally spaced
points with separation of adjacent point equal to 2/d.

1407
The first Brillouin zone extends from /d to +/d, such that

k = /d n = N/2 (18.65)
k = /d n = N/2 (18.66)

Thus, for the values of ~k in the first zone, the integer n can assume all integral
values in the range [N/2, +N/2] and there are N integers in this range.

This result says that there are N electronic states in each Brillouin zone.

All of these ideas form the basis of the detailed discussion of the electronic states
and energies in a crystal, which we now begin.

18.3 Free-Electron and Weak-Binding

Approximations in 1-Dimension
The complexity of the potential energy function for a real crystal makes direct
solution of the associated exact Schrodinger equation impossible. We must make
approximations in order to learn something about the energies of the electronic
states.

We will use two approximations in these notes.

First, we will use the so-called free-electron approximation in which the crystal
potential energy function is replaced by a constant. This is a very simple and
a very crude approximation. It reflects only the fact that the electron is con-
strained to move within the crystal.

After learning as much as we can from the free-electron approximation, we

reintroduce the periodic nature of the crystal potential energy in the so-called
weak-binding approximation. In using the second approximation we will only
look for qualitative solution. We also start with a simple one dimensional sys-
tem for simplicity.

We will not attempt to find explicit forms for the electronic wave functions,
but instead concentrate on determining the energy eigenvalues. In particular,
we shall only attempt a determination of graphs for the energy versus the wave
vector and the density of states as a function of the energy. These features
will enable us to get a clear picture of what is happening in crystals due to
translational symmetry.

18.3.1 The Free-Electron Approximation

The potential energy of an electron in a crystal is a periodic function of vecr
and has singularities at the nuclear sites. The dominant effect of the potential

1408
on the electron is to keep it confined inside the crystal. If this is the case,
then we can neglect the periodicity and the singularities in a the free-electron
approximation and determine what we might call a zero-order picture of the
energy as a function of the wave vector. This picture will then be modified in a
weak-binding approximation by reimposing periodicity and the singularities so
that the basis structure of the energy versus wave vector curve remains basically
unchanged except where the added constraints are dominant. That is the reason
for the name weak-binding.

We illustrate this approximation in Figure 18.29 below.

Figure 18.29: Free Electron Approximation

The dashed line in the figure represents V (x) in the free-electron approximation.
It is just a finite square well in one dimension. It is equivalent to replacing the
periodic potential with its average value.

Electronic Wave Functions

In this finite square well approximation, the Schrodinger equation inside the
crystal is that of a free particle
~2 2
E (~r) = EE (~r) (18.67)
2me
The solutions are the plane-wave eigenfunctions of the free-particle Hamiltonian
~0
E (~r) = Aeik ~r (18.68)
with energy eigenvalues
~2 k 02
E= (18.69)
2me
where ~k 0 is the wave vector related to the linear momentum of the particle by
p~ = ~~k 0 .

The solution is in the form of a Bloch function

~0
E,~k0 (~r) = eik ~r uE,~k0 (~r) uE,~k0 (~r) = A (18.70)

1409
At this point, no limit has be imposed on ~k 0 . It is called an extended-zone-
scheme wave vector. In this case, ~k 0 is a redundant label because the E and ~k 0
are simply related by the (18.69).

As we said earlier, however, we usually choose the wave vector ~k in Blochs

theorem to be the smallest of the set of equivalent wave vectors, which is called
a reduced-zone-scheme wave vector. This wave vector ~k is related to ~k 0 by
~k 0 ~k = G,
~ where G
~ is a reciprocal lattice translation vector.

In the reduced zone scheme we can also write the wave function as a Bloch wave
using
~ ~ ~
E,~k (~r) = AeiG~r eik~r uE,~k (~r) = AeiG~r (18.71)

In the reduced zone scheme, both the energy and ~k labels are required since these
two quantities are no longer simply related as in the extended zone scheme.

E(k) for a One-Dimensional Crystal

We will use an E versus k graph to show the relationship between the energies
of the electronic states of a crystal and the allowed values of the wave vector
~k. For example, in the 1dimensional crystal, if we let k take on all values,
i.e., < k < + and plot E = ~2 k 2 /2me , we obtain the parabola shown in
Figure 18.30 below.

Figure 18.30: E versus k - Extended Zone Scheme

On the horizontal axis - the kaxis - we find values of k from every Brillouin
zone of the lattice and corresponding to each k value there is an E value.

The plot of E versus k above is in the extended zone scheme , i.e., it contains
values of k from all Brillouin zones. The labeled k values correspond to the
Brillouin zone boundaries.

1410
Our earlier discussions imply that as far as Blochs theorem is concerned, the
only significant values of k are those in the first Brillouin zone.

As we saw earlier, for a 1dimensional crystal with N primitive unit cells, the
allowed values of k in the first zone are given by

2  m  N
k= , m = 0, 1, 2, ......., (18.72)
d N 2
Since N is a very large number, in general, the values of k are very closely
spaced.

We can restrict our attention to the first Brillouin zone by plotting E versus k
in the reduced zone scheme. To do so we translate all ~k vectors in the higher
zones on the plot in the extended zone scheme (Figure 18.30 above) into the
first zone given by

k+ (18.73)
d d
That means we add the appropriate reciprocal lattice translation vector to ~k 0
to generate a wave vector in this range,

2 0
k = k0 n , n0 = integer (18.74)
d
In Figure 18.31 below we show explicitly how to generate the reduced zone plot.

Figure 18.31: E versus k - Reduced Zone Scheme

1411
The arrows labeled a correspond to n0 = 1 and the arrows labeled b correspond
to n0 = 2.

The final plot in the reduced zone scheme is shown in Figure 18.32 below.

Figure 18.32: E versus k - Final Plot in Reduced Zone Scheme

We note that in the plot for the reduced zone scheme an infinite number of
energies correspond to each value of k in the first zone. There is an electronic
wave function corresponding to each energy.

For each k, we can specify a particular wave function by giving the energy,
i.e., E,k , or we can introduce a new quantum number to distinguish different
electronic states with the same value of k. This new quantum number, n,
is shown in Figure 18.32 above. We say that each of the ranges of energy
corresponding to k in the first Brillouin zone is a band and n labels the bands
and is called the band index. We write n,k (x). Figure 18.32 above shows the
first four bands and part of the fifth band of the 1dimensional crystal.

The wave function for an electron in this crystal with energy E ~2 2 /2me d2
is
ZL
ikx 2
1,k (x) = Ae , A L= 1,k (x)1,k (x)dx = 1 A = 1/ L (18.75)
0
0
For an energy in the second band, we have 2,k (x) = Aeik x where
2 2
k0 = k + , k<0 , k0 = k , k>0 (18.76)
d d

1412
and so on for higher bands.

Summarizing the results we have found in this free-electron approximation:

1. We can represent the energy eigenvalues of the electron in a 1dimensional
crystal in two ways, namely, the extended-zone (wave vectors take on all
values) and the reduced-zone (wave vectors in first Brillouin zone) schemes.
2. We defined bands, each representing N values of k and corresponds to
states that can be occupied by at most 2N electrons.
3. The bands are connected at the Brillouin zone boundaries.

The Density of States in the Free-Electron Approximation

We define w(k)dk = number of states with wave vectors between k and k + dk.
As we have seen, there are two states for each wave vector and the wave vectors
are uniformly spaced along the kaxis since
2  m  N
k= , m = 0, 1, 2, ......., (18.77)
d N 2
This says that w(k) is a constant (neglecting the discreteness of k). Since there
are 2N states in the first Brillouin zone, where k = 2/d, we have
2 Nd L
w(k)k = w(k) = 2N w(k) = = , L = length of crystal
d
(18.78)
A more interesting physical quantity is D(E), the density of states as a function
of the energy. We define
D(E)dE = number of states with energy between E and E + dE (18.79)
For each interval dE we can find a corresponding interval in k space dk. Then
D(E) and w(k) are related by
D(E)dE = 2w(k)dk , k>0 (18.80)
where the factor of 2 arises because each value of E corresponds to two values
of k - one positive and the other negative. We then have
2w(k)
D(E) = dE
(18.81)
dk

In the free-electron approximation, E = ~2 k 2 /2me so that

dE ~2
= k (18.82)
dk me
and the density of states in the free-electron approximation is

L 2me 1/2
D(E) = E (18.83)
~

1413
 Figure 18.33: Density of states D(E)

Figure 18.33 above shows that the density of states is largest at small energies
and falls off as E increases. Note that we are not restricting k to the first
Brillouin zone in this discussion, although we are taking it to be positive.

The Fermi Energy

Each band in a crystal can be occupied by as many as 2N electrons. Elec-
trons have half-integral spin and are fermions. In the orbital approximation the
occupation of electronic states by these electrons is governed by Fermi-Dirac
statistics. We thus define f (E) = probability that an electronic state with en-
ergy E will be occupied by an electron(the distribution function). We then
have
1
f (E) = (E)/k T (18.84)
e B +1
where kB = 1.380622 1016 erg/ K = Boltzmanns constant, T is the temper-
ature, and is a constant that is independent of T . The distribution function
is shown in Figure 18.34 below.

Figure 18.34: Fermi-Dirac Distribution Function

The dashed line corresponds to T = 0 and the solid line corresponds to kB T <<
.

The product D(E)f (E) is equal to the density of occupied electronic states. As
can be seen from Figure 18.34 above, at T = 0 we find that there is an energy
EF such that all states with E < EF are occupied but no states with E > EF

1414
are occupied. EF is called the Fermi energy. At T = 0 the Fermi energy is the
maximum energy of the electrons in the crystal and is equal to the constant
in the distribution function. We can define a wave vector kF corresponding to
the Fermi energy (the Fermi wave vector).

To obtain an expression for kF we introduce a new quantity , defined to be the

number of electrons per primitive unit cell in the crystal. The total number of
electrons in the crystal is given by N . At T = 0, the total number of occupied
states is given by
ZkF
2 w(k)dk = N (18.85)
0

Using
Nd
w(k) = (18.86)

we get

kF = (18.87)
2d
The corresponding Fermi energy in the free-electron approximation is then

~2 2 ~2
EF = kF = 2 (18.88)
2me 32me d2
In the figures below we show the location of the Fermi energy on plots of E
versus k in the extended and reduced zone schemes and on a plot of D(E)
versus E. We also show the Fermi energies (and Fermi wave vectors) for various
values of .

Figure 18.35: Fermi Energy in Extended Zone Scheme - Various values

1415
The arrows label the Fermi energies and Fermi wave vectors in the figures above
and below.

Notice that as the number of electrons per primitive unit cell increases, the
Fermi energy, and hence the number of occupied states, increases. The Fermi
energy enables us to describe the occupation of electronic states in the crystal
(at T = 0), and in the 1-dimensional free-electron approximation it can be
calculated knowing only the repeat distance d and the number of electrons per
primitive unit cell .

Figure 18.36: Fermi Energy in Reduced Zone Scheme - Various values

Figure 18.37: Occupied states on D(E) plot - Various values

The shaded region shows the occupied states for a crystal with = 2.

1416
18.4 Introduction to the Weak-Binding Approx-
imation
We must now introduce the periodicity of the crystal potential energy. We write
H = T + V (x) (18.89)
where V (x) has the periodicity of the crystal. We assume that the electron in-
teractions due to the potential energy is weak and they still move easily within
the crystal, although they are now influenced by the periodic potential. This is
the so-called weak-binding approximation.

We still cannot solve the exact Schrodinger equation. How can we introduce the
weak periodic potential and also use all the results of the free-electron approxi-
mation.

One possible approach is to use perturbation theory, where the free-electron ap-
proximation is used as the zeroth-order solution and V (x) as the perturbation
Hamiltonian. For example, using
(0) 1
E,k (x) = eikx (18.90)
L
as the zeroth-order wave functions, the first-order perturbed wave function can
be written as
(0)
X hk 0 | V |ki (0)
E,k (x) = E,k (x) + (0) (0) 0
E,k0 (x)
k0 6=k
E (k) E (k )
 
1 ikx X Vk 0 k 1 ik0 x
= e + e (18.91)
L (~2 k 2 /2me ) (~2 k 02 /2me ) L
k0 6=k

where
+L/2
Z
0 1 0
Vk0 k = hk | V |ki = ei(kk )x V (x)dx (18.92)
L
L/2
0
This expression represents a mixing of the functions eik x with the zeroth-order
function eikx to form the first-order approximate wave function.

If perturbation theory is valid in this case, then the above result is a good
approximation to the electronic wave function.

Let us assume that this approximation is valid and investigate the summation
term. The summation is over all allowed wave vectors so it represents a large
number of terms. We ask the question - are there any terms where Vk0 k = 0? If
so, then these functions do not mix with
(0) 1
E,k (x) = eikx
L

1417
We can answer this question by applying the translation operator Top on the
integrand of Vk0 k . This leaves Vk0 k unchanged since it simply corresponds to
redefining the zero of the dummy variable of integration, x, and we may consider
the integral to extend from to + if nd  L. For a 1dimensional crystal,
any lattice translation vector can be written as nd so we have

+L/2
Z +L/2
Z
1 i(kk0 )x 1 0
Vk0 k = Top e V (x)dx = ei(kk )(x+nd) V (x + nd)dx (18.93)
L L
L/2 L/2

Since V (x) is invariant under translation, we have

+L/2
Z
1 0 0
Vk 0 k = ei(kk )(x+nd) V (x)dx = ei(kk )nd Vk0 k (18.94)
L
L/2

This result says that the matrix element Vk0 k = 0 unless

0
ei(kk )nd = 1 (18.95)

This is the definition of a reciprocal lattice translation vector. Therefore, we

(0)
have shown that the only states that mix with E,k (x) in a first-order pertur-
0
bation calculation are states with wave vectors k differing from k by reciprocal
0
lattice vectors. All other functions eik x can be discarded from the summation,
which can then be written
 
1 ikx X VG 1 i(kG)x
E,k (x) = e + e
L (~2 k 2 /2me ) (~2 (k G)2 /2me ) L
G6=0
(18.96)
where
+L/2
Z
1
VG = eiGx V (x)dx (18.97)
L
L/2

Since both V (x) and eiGx are periodic, VG can be written as the integral over
one primitive unit cell

+d/2
Z +d/2
Z
1 iGx 1
VG = N e V (x)dx = eiGx V (x)dx (18.98)
L d
d/2 d/2

This important result - that the only free-electron wave functions that mix with
(0)
E,k (x) are those wave vectors k 0 satisfying k 0 k = G for some reciprocal
lattice vector G - can be shown to hold in any order of perturbation theory. It
is true, in fact, even if perturbation theory breaks down!

1418
The energy to second order can be written as
2
~2 k 2 X |VG |
Ek + (18.99)
2me (~2 k 2 /2me ) (~2 (k G)2 /2me )
G6=0

where k is the wave vector in the extended zone scheme and hence ~2 k 2 /2me is
the energy in zeroth order.

Validity of Perturbation Theory

This analysis depends on the question: Is perturbation theory valid for this
problem?

The first-order expression for the wave function is a good approximation to the
true electronic wave function provided that none of the states in the summation
(0)
over G mixes strongly with E,k (x), i.e., provided that
 
 VG 
 (~2 k 2 /2me ) (~2 (k G)2 /2me )  << 1 (18.100)
 

for all G 6= 0. If there are states that mix strongly, i.e., for which the above in-
equality is violated, then these states cannot be handled by perturbation theory.
We can rewrite the inequality as

~2
|VG | << |G(2k G)| (18.101)
2me

Since this inequality is violated for any state with k G/2 there exist states for
which we cannot use perturbation theory (if VG 6= 0). So, as long as VG is small
enough (V (x) is weak enough), there exist some states, i.e., those for which k is
not nearly equal to any G/2, for which the perturbation expansion can be used.
For these states the above expressions for the first-order wave function and the
second-order energy are valid and show that the wave functions and energies
differ from their zeroth-order counterparts, although not very much.

In the weak-binding approximation, we assume that the energies of these states

is essentially the same as in the free-electron approximation.

On the other hand, for states with k = G/2, perturbation theory cannot be
used since the function ei(kG)x mixes strongly with eikx . We must treat these
states some other way.

It is precisely these cases for which the energy will differ most from the free-
electron energy in the dependence of E on k when we move from the free-electron
approximation to the weak-binding approximation.

1419
Strongly Mixed States
Let us look back at the free-electron E versus k plots we developed earlier
and locate the states that are strongly mixed. We have found that states
with wave vector near k = G/2 will strongly mix with states with wave vector
k 0 = k G G/2, i.e., the strongly mixed states are those with wave vectors
near opposite Brillouin zone boundaries. The plots below show (in both the ex-
tended zone scheme and the reduced zone scheme) the states that are strongly
mixed.

The strongly mixed states are connected by an arrow (see Figure 18.38 and
18.39) and numbered for convenience(triangles indicate the state energies). Be-
cause of the restriction k 0 k = G, there are at most two electronic states that
are close to one another in energy and strongly mixed.

Figure 18.38: Strongly mixed states in Extended Zone Scheme

Figure 18.39: Strongly mixed states in Reduced Zone Scheme

1420
The strongly mixed states are

12 , 34 , 56 (18.102)

Notice that states 1 and 6 are not strongly mixed even though their wave vectors
are near zone boundaries and related by a reciprocal lattice translation vector.
These states are widely separated in energy, so the denominator in the pertur-
bation expansion term corresponding to these two states is large, and the states
are weakly mixed.

Calculation of the Energy Shifts

We now do a simple calculation to show how the energies of the strongly

mixed states are affected by the periodic potential energy. We start with the
Schrodinger equation

(H E)E (x) = 0 , H = T + V (x) (18.103)

We are interested in an energy corresponding to two strongly mixed states with

wave vectors
k G/2 and k 0 = k G (18.104)

In this case, we can expand E (x)i n the complete set of free-particle states,
summing only over equivalent wave vectors, and keep only the terms with large
expansion coefficients for the energy of interest

1 1 0
E (x) a eikx + b eik x (18.105)
L L

Note that the two free-particle states (see the reduced zone energy plot above)
kept in this expansion are nearly degenerate.

We solve this problem by substituting the two state expansion for the wave
function into the Schrodinger equation

(H E)E (x)dx = 0 = (H E) (a |ki + b |k 0 i) (18.106)

0
hk| (H E) (a |ki + b |k i) = 0 = a(Hkk E) + bHkk0 (18.107)
hk 0 | (H E) (a |ki + b |k 0 i) = 0 = aHk0 k + b(Hk0 k0 E) (18.108)

These two homogeneous equation have a nontrivial solution only if

 
 Hkk E Hkk0 
 =0 (18.109)
 Hk0 k Hk0 k0 E 

1421
where
Hkk = hk| H |ki
L/2
~2 d2 ~2 k 2
Z  
1 ikx ikx
= e + V (x) e dx = (18.110)
L 2me dx2 2me
L/2

Hk0 k0 = hk 0 | H |k 0 i
L/2
~2 d2 ~2 k 02
Z  
1 0 0
= eik x 2
+ V (x) eik x dx = (18.111)
L 2me dx 2me
L/2
L/2
~2 d2
Z  
1 0 ikx 0
H kk0 = hk| H |k i = e + V (x) eik x dx
L 2me dx2
L/2
L/2
Z
1 0
= ei(k k)x V (x)dx = VG = VG (18.112)
L
L/2
L/2
~2 d2
Z  
0 1
Hk0 k = hk | H |ki = eikx + V (x) eikx dx
L 2me dx2
L/2
L/2
Z
1 0
= ei(kk )x V (x)dx = VG (18.113)
L
L/2

Solving for the energy eigenvalues we get

(Hkk E)(Hk0 k0 E) Hkk0 Hk0 k = 0 (18.114)
2
E (Hkk + Hk0 k0 )E Hkk0 Hk0 k = 0 (18.115)
or s 2
1 ~2 k 2 ~2 k 02 ~2 k 2 ~2 k 02
 
1 2
E= + + 4 |VG | (18.116)
2 2me 2me 2 2me 2me
We are especially interested in the behavior of these energies at or near a Bril-
louin zone boundary. At a zone boundary, we have
|k| = |k 0 | = G/2 (18.117)
and thus
~2 k 2
|VG | E= (18.118)
2me
This result shows that the energies of the two strongly mixed states at a zone
boundary are split about their average energy
~2 k 2
(18.119)
2me

1422
by an energy separation of 2 |VG | which is called a band gap. As we shall see
shortly, it separates two nearly free-electron bands.

What happens near a zone boundary?

There we have k G/2 and k 0 G/2. If we introduce a new variable

defined by
1
= Gk (18.120)
2
we can write
 2 s 2
~2 G ~2 2 1 ~2 G 2
E= + + 4 |VG | (18.121)
2me 2 2me 2 2me

This equation shows that as we move away from a zone boundary, and in-
creases, the splitting between the two states with wave vectors k and k 0 increases.
In addition, the average energy of the two states increases.

As we get further away from the zone boundary, we eventually leave the region
of strongly mixed states.

Once far enough from the boundary so that perturbation theory is valid and
can be used to calculate the energy shift, we know that the energies of the
electronic states in the weak-binding approximation are nearly the same as the
corresponding energies in the free-electron approximation.

If we are only interested in the qualitative (or semiquantitative) behavior of E

versus k , we can simply use the free-electron energies far from the boundaries.

Figures 18.40 and 18.41 below show the effect of the periodic potential energy on
the first two bands of the free-electron E versus k plot of the one-dimensional
crystal in the reduced and extended zone schemes. The dashed lines are the
free-electron result.

Figure 18.40: Band Gaps in Reduced Zone Scheme

1423
 Figure 18.41: Band Gaps in Extended Zone Scheme

We can see from the figures that in the weak-binding approximation the energy
splitting calculated using the results above, leads to a gap between the first and
second bands. The magnitude of the gap, 2 |VG | is dependent on the size of the
matrix element VG . This means that the magnitude of the gap between bands
2 and 3 will differ from that of the gap between 1 and 2 and so on.

Bands and band gaps for the first three bands are shown in both the extended
and reduced zone schemes in Figures 18.42 and 18.43 below. The dashed lines
are the free-electron result.

Figure 18.42: Band Gaps in Reduced Zone Scheme

1424
 Figure 18.43: Band Gaps in Extended Zone Scheme

Qualitative Explanation of Band Gaps

Consider the approximate wave function
1 1 0
E (x) a eikx + b eik x (18.122)
L L
At the first Brillouin zone boundary, we have

|k 0 | = |k| = (18.123)
d
so that the two free-electron wave functions that mix are degenerate and are
given by
(0) 1
E,/d (x) = eix/d (18.124)
L
where the zero superscript reminds us that these are free-electron approxima-
tions to the wave function. The two weak-binding
wave functions E (x) at
the zone boundary, where |a| = |b| = 1/ 2 from the solution of the eigenvalue
equation can be written as normalized standing-wave eigenfunctions

1  (0)  r2  
(0)
E,c (x) = E,+/d (x) + E,/d (x) = cos x (18.125)
2 L d
1  (0)  r2  
(0)
E,s (x) = E,+/d (x) E,/d (x) = sin s x (18.126)
2 L d
2 2
The probability densities |E,c (x)| and |E,s (x)| for wave vectors at the first
Brillouin zone boundary in the case of one atom per primitive unit cell (with
the origin at one of the atoms) is shown in Figure 18.44 below.

1425
 Figure 18.44: Probability Densities

We see that E,c (x) has a maximum of probability density near each singularity
of the strongly attractive potential energy of the crystal. On the other hand,
E,s (x) has a minimum of probability density near each singularity. Conse-
quently, we expect to see a splitting between the energies for these two wave
functions. This splitting is just the band gap 2 |VG |.

Density of States in the Weak-Binding Approximation

How do we modify D(E) in the weak-binding approximation? Our original

derivation of w(K) is not dependent on the approximation V (x) = constant.
Thus, that derivation is also valid in the weak-binding approximation and we
still have
Nd L
w(k) = = (18.127)

Similarly, the derivation of the Fermi wave vector is only based on

ZkF
2 w(k)dk = N (18.128)
0

and does not use the assumption of free-electron behavior (it does assume a
monotonic increase of E with k in the extended zone scheme). So, in the weak-
binding approximation, we still have

kF = (18.129)
2d

where is the number of atoms per primitive unit cell. The expression for
the Fermi energy cannot be carried over into the weak-binding approximation
because we used the fact the the energy E was quadratic in the wave vector k
in the derivation.

1426
Metallic and Nonmetallic Behavior

Now we consider a 1-dimensional crystal with one electron per primitive unit
cell ( = 1). The Fermi wave vector is kF = /2d. For this wave vector the
Fermi energy is near the middle of the first band (see the earlier figure). For
= 1, the first band is half-filled. Since the Fermi energy is the maximum
energy for the electrons at absolute zero, no states with E > EF are occupied.
This means that there exist empty states infinitesimally close to the energy of
the highest occupied state and an infinitesimal increment of energy can excite an
electron to one of these empty states (turn on a small electric field). Thus, for
infinitesimal electric fields we get infinitesimal currents (extra kinetic energy).

A crystal that exhibits such behavior is called a metal.

In contrast, consider a crystal, with two electrons per primitive unit cell. In
this case, kF = /d, which corresponds to the edge of the band - the first band
is full, i.e., all the electronic states in this band are occupied at absolute zero.
Since the next nearest empty state is separated from the first band by an energy
gap of magnitude 2 |VG |, at least this much energy is required in order to excite
an electron into one of the empty states.

A crystal that exhibits such behavior is called an insulator.

18.5 The Kronig-Penney Model

18.5.1 Exact Analysis
The Kronig-Penney model of a crystal consists of an infinite series of square
wells as shown in Figure 18.45 below. Each well is of width d s and depth V0 .

Figure 18.45: Kronig-Penney Potential

Adjacent wells are separated by a distance s, the width of each barrier. The
repeat distance for this crystal - the length of the primitive unit cell - is d. The

1427
only restrictions we place on the well parameters at this time are

s > 0, s < d, V0 > 0 (18.130)

We have not imposed either free-electron or weak-binding behavior. This will

be an exact solution.

The Eigenvalue Equation

The Kronig-Penney potential energy is a periodic function, which means that
we can use Blochs theorem to connect the wave functions from different regions
of the crystal. In one dimension, the second form of Blochs theorem says

E,k (x) = eikx uE,k (x) (18.131)

where uE,k (x) is a periodic function invariant under translation by any lattice
translation vector T = nd, i.e.,

uE,k (x + nd) = uE,k (x) (18.132)

the relationship between E and k is not known yet.

From the graph(Figure 18.45) of the potential function, we see that there are
two regions of potential energy:
1. ranges of x for which V = 0 nd x nd + (d s)
2. ranges of x for which V = V0 nd + (d s) x (n + 1)d
where n = 0, 1, 2, ...... The form of the wave function is different in these
two regions.

In the region 0 x (d s) the potential energy is zero and the electronic

Schrodinger equation is
d2 E,k (x) 2me
+ 2 EE,k (x) = 0 (18.133)
dx2 ~
with solutions
1p
E,k (x) = Aeix + Beix , = 2me E (18.134)
~
The form of uE,k (x) is then

uE,k (x) = Aei(k)x + Bei(+k)x (18.135)

Using the same sort of arguments for the region (d s) x d where V = V0

we have the electronic Schrodinger equation
d2 E,k (x) 2me
2 (V0 E)E,k (x) = 0 (18.136)
dx2 ~

1428
with solutions
1p
E,k (x) = Cex + Dex , = 2me (V0 E) (18.137)
~

The form of uE,k (x) is then

uE,k (x) = Ce(ik)x + De(+ik)x (18.138)

There are four unknowns in these solutions A, B, C and D, but there are also
four connecting equations - two for the periodicity conditions and two for the
continuity of u and du/dx.

The periodicity conditions give

u(0) = u(d) (18.139)

ikd d d
A+B =e (Ce + De ) (18.140)
 
du  du 
= (18.141)
dx 
0 dx  d
i(A B) = eikd (Ced Ded ) (18.142)

The continuity conditions give

lim u(d s + ) = lim u(d s ) (18.143)

0 0
i(ds)
Ae i(ds)
+ Be = Ce(ds) + De(ds) (18.144)
 
du  du 
lim = lim (18.145)
0 dx  0 dx 
ds+ ds

i(Aei(ds) Bei(ds) ) = (Ce(ds) De(ds) ) (18.146)

These four equations have nontrivial solutions only if the determinant of the
matrix of the coefficients of A, B, C and D vanishes

eikd ed eikd ed
 
 1 1 
eikd ed eikd ed
 
 i i 
=0
 i(ds) (18.147)
 e
 ei(ds) e(ds) e(ds) 

 iei(ds) iei(ds) e(ds) e(ds) 

Some algebra gives

2 2
cos kd = cos (d s) cosh s + sin (d s) sinh s (18.148)
2

This is a transcendental equation relating the electronic energy E to the quan-

tum number k in the Kronig-Penney model.

1429
Limiting Cases
1. In the limit V0 0, the particle will be confined to one of the square wells
in the model crystal. In this case the energy eigenvalue equation becomes

cos kd 2 2
= cos (d s) + sin (d s) tanh s
cosh s 2

0 = cos (d s) + sin (d s)
2
sin (d s) = 0 (d s) = n
n2 2 2me E
2 = =
(d s)2 ~2
2 2 2
n ~
E= (18.149)
2me (d s)2

which is the familiar result for a particle in an infinite box.

2. Now we keep V0 finite, fix the width of each well, and let the well separation
s . In this case, wells will be completely isolated from one another
and we expect the eigenvalue equation to reduce to the equation for a
particle in a single finite square well. We find

2 2
0 = cos (d s) + sin (d s)
2
2 2
cot (d s) = (18.150)
2
which is the standard transcendental equation of the finite square well.

3. Let us remove the periodicity of V (x). There are three ways to do this

V0 0 or s 0 or s d (18.151)

In each case, we get

~2 k 2
E= (18.152)
2me
which is the expected result for a free particle.

Confident that we have arrived at a sensible eigenvalue equation we now

look at solutions of the exact equation.

Solutions of the Eigenvalue Equation

A form of the eigenvalue equation that is particularly easy to work with is
obtained by considering cases where

V0 , s 0 such that V0 s remains constant (18.153)

1430
In this case we have

cos kd = cos d + sin d sinh s (18.154)
2
1
2me V0 1p
cos kd = cos d + ~ sin d sinh 2me V0 s (18.155)
2 ~
1
2me V0 1p
cos kd = cos d + ~ sin d 2me V0 s (18.156)
2 ~
me V0 sd sin d
cos kd = cos d + (18.157)
~2 d
If we choose
me V0 sd
=1 (18.158)
~2
we have
sin d
cos kd = cos d + (18.159)
d
This result clearly shows the band structure, that is, since the value of the left
hand side of this equation is between 1 and 1, solutions only exist for energies
E such that the magnitude of the right hand side is 1.

Energies violating this condition do not correspond to possible states of the

particle.

In Figure 18.46 below we plot the right hand side versus d for d 3.

Figure 18.46: Kronig-Penney Model - Transcendental Plot

1431
We find that the lowest allowed energy of the particle in this Kronig-Penny
crystal is
2 ~2
E = 0.174 (18.160)
2me d2
Looking at the regions lying between +1 and 1 we see the structure of energy
bands and band gaps clearly revealed. In particular, the particle can have any
energy in the following ranges

0.174 E 1 band 1
1.370 E 4 band 2
4.490 E 9 band 3

States of the system exist for any energy in one of these bands. No solution
exist if

1 E 1.37 gap 1
4 E 4.49 gap 2

These results are very different from those obtained by solution of the Schrodinger
equation for an atom or molecule. Instead of discrete bound levels, we find con-
tinuous bands of allowed energies and intervening gaps of forbidden energies.

A plot of E versus k can be made from the eigenvalue equation. An example is

shown in Figure 18.47 below.

Figure 18.47: Energy Bands and Gaps in the Kronig-Penney Model

1432
The dashed line shows the free-electron approximation, i.e., treating the Kronig-
Penney model as a giant square well.

Weak-Binding Limit
In this limit we let the well separation s get small. In particular, we assume that
s is sufficiently small that the barriers can be treated as a small perturbation
on a constant potential energy. In this way, we can approach the weak-binding
limit through the exact energies. If s is small we assume that

s << 1 and s << 1 (18.161)

In this case, we can write the electronic energy as

~2 k 2
E = E0 + = + (18.162)
2me
We want to expand the exact eigenvalue equation in the small parameter . We
write
1p 1p
= 0 + 1 = 2me E0 + 1 = 2me E (18.163)
~ ~
1p 1p
= 0 + 1 = 2me (V0 E0 ) + 1 = 2me (V0 E) (18.164)
~ ~
We then have
1p 1p 1p
1 = 2me (E0 + ) 2me E0 = 2me E0 ((1 + /E0 ) 1)
~ ~ ~
1p me m2 2
= 2me E0 ((1 + /2E0 + ....) 1) = 2 e4 3 (18.165)
~ ~ k 2~ k
and
1p 1p
1 = 2me (V0 E) 2me (V0 E0 )
~ ~
0 0 2
= (18.166)
2 V0 E0 8 (V0 E0 )2

Using trigonometric identities for sin(a b) , cos(a b) and Taylor series expan-
sions of cos, sin, cosh and sinh function, the eigenvalue equation becomes

2 + 2
cos kd = cos d + s sin d (18.167)
2
We can also write
2 + 2 2 + 02 2 + 2
= 0 0 2 0 1 (18.168)
2 20 20
This only retains terms to order , but when multiplied by s sin d the second
order terms are obtained.

1433
Substitution gives

2 + 02
 
1
12 d2 + 0 s1 d cos kd
2 20
02 + 02 02 + 02
 
+ 1 d + s 1 s sin kd = 0 (18.169)
20 202

Using this expression we can solve explicitly for the energy for wave vectors at
the Brillouin zone boundaries and determine the band gaps.

In the extended zone scheme the boundaries occur at

kd = n , n = integer (18.170)

For these values of k we have

sin kd = 0 , cos kd = (1)n (18.171)

Therefore, at zone boundaries the second term is zero and the first term gives
two conditions for 1

02 + 02
1 d = 0 or 1 d = s (18.172)
0

The first choice gives = 0 and the second choice gives = 2V0 s/d and thus
we get a band gap of energy E = 2V0 s/d at the zone boundary in the weak-
binding approximation.

18.6 Free-Electron and Weak-Binding

Approximations in 2-Dimensions
We will now study two-dimensional crystals, examining the behavior of energy
as a function of wave vector and of the density of states as a function of energy.

The study of electron band theory in one dimension was comparatively simple
for several reasons. First, there is only one Bravais lattice in one dimension;
thus the only difference between various one-dimensional problems arises from
such factors as the number of atoms per primitive unit cell, the strength of
the potential, and the separation of the lattice points. This situation contrasts
strikingly with that in two and three dimensions, where there are 5 and 14
Bravais lattices, respectively, each of which poses its own special problem.

Thus, there is a considerable increase in the complexity of any calculation in

band theory as the number of dimensions increases.

1434
18.6.1 The Free-Electron Approximation
Let us consider a two-dimensional crystal in the free-electron approximation,
that is, we shall neglect the periodic part of the crystal potential energy V (x, y)
and assume that the electrons are free to move about within the confines of the
crystal. In this approximation the equation relating the energy of an electronic
state to the wave vector is
~2 k 2
E= (18.173)
2me
where the two-component wave vector can be expressed as
~k = kx x + ky y (18.174)

In order to see the basic structure of such a crystal, we must examine plots of E
versus k. Since k is a two-component vector, E(k) is a paraboloid of revolution,
as shown in Figure 18.48 below in the extended zone scheme.

Figure 18.48: E(k) paraboloid

Eventually we will want to use the reduced zone scheme and identify E versus
k graphs so as to incorporate the periodic nature of V (~r). It is clear that each,
of these steps is rather difficult if we use the three-dimensional graph in Figure
18.48.

Moreover, even if we could draw the resulting diagrams, their complexity might
obscure some of the information they contain.

For this reason, we will now introduce two other ways of representing the energy
of the electronic states of a crystal.

E versus Path Distance

The first requires that we pick a path in k space (i.e., a continuous connected
set of line segments) and plot the energy as a function of the distance along this

1435
path. We will represent the distance along the path by the integral
Z
dk (18.175)
c

For example, suppose that we choose the path shown in Figure 18.49 below.

Figure 18.49: Path in k space

Using the quadratic relation between E and k for the energy in the free-electron
approximation, we can sketch E as a function of distance along this path, ob-
taining the curve shown in Figure 18.50 below (in the extended zone scheme).

Figure 18.50: E versus path in k space

We shall return to this technique frequently.

Constant Energy Contours

An alternate representation of the energy can be obtained by choosing several
different values of energy and plotting curves in k space that correspond to
these constant values of E. Such curves are called constant energy contours.
In the free-electron approximation, where E(k) is quadratic in k, each curve of
constant energy is a circle of radius
1p
k= 2me E (18.176)
~

1436
As E increases, the radius of the circle increases. Thus we obtain a series of
concentric circles; several constant energy contours are shown in Figure 18.51
below, where the extended zone scheme is used.

Figure 18.51: Constant energy contours in k space

Notice that E1 < E2 < E3 < E4 < E5 .

Now, in going to the reduced zone scheme, we must specify a particular type
of Bravais lattice. Recall from our earlier discussion that there are five Bravais
lattices in two dimensions: the oblique lattice, the primitive rectangular lattice,
the centered rectangular lattice, the square lattice, and the hexagonal lattice.
To illustrate the preparation of graphs of E versus path distance and constant
energy contours, we choose the square lattice. Let d denote the spacing of adja-
cent lattice points in the x or y direction in the square lattice. Then the spacing
of adjacent points in the x or y direction in the reciprocal lattice is 2/d.

The Square Lattice

In order to plot E versus path length for the square lattice, we must first select
a path. The choice of path is somewhat arbitrary, but common sense can guide
us. Since we want to generate a plot in the reduced zone scheme, where only
values of k in the first Brillouin zone are considered, we should choose a path
contained in that zone or on one of its boundaries. We would like the path to
be as simple as possible, and, ideally, each segment of the path should provide
some new information about the behavior of E(k).

The reciprocal of the square lattice is square and possesses fourfold symmetry.

1437
A path that passes through all points of high symmetry in the first zone but
that does not contain segments equivalent to one another under the symmetry
operations of the square lattice is shown in Figure 18.52 below.

Figure 18.52: Path in k space through high symmetry points

We will now compute the value of E at points on this path and on equivalent
paths in the surrounding regions of the reciprocal lattice. In Figure 18.53 below,
we have translated the first Brillouin zone and the chosen path throughout a
portion of the reciprocal lattice.

Figure 18.53: Equivalent paths in k space by translation

The resulting squares are numbered (1) through (9) for purposes of identifica-

1438
tion. We have also indicated the value of E in the free-electron approximation
at several points in k space - specifically, at the corners of each path (units of
energy are ~2 2 /2me d2 ).

A typical point, say the 13 along the right edge in the (9) box has

~k = 3 x + 2 y (18.177)
d d
so that its energy, in the free-electron approximation, is
~2 k 2 ~2 2
E= = 13 (18.178)
2me 2me d2
The origin is at the center of the (0) square.

The reason for considering values of k outside the first zone is that we wish
to take into account energies greater than ~2 2 /2me d2 , the maximum value of
energy that corresponds to points all of which lie within the first zone. Notice
that each path in a region other than the first zone is equivalent to the path in
the first zone. The wave vectors k 0 of these equivalent paths will be related to
wave vectors in the first zone by some reciprocal lattice translation vector
~k 0 = ~k + G
~ (18.179)

Several different methods exist to generate data like that shown in the above
figure - and hence the resulting free-electron energy bands. The following steps
will enable us to obtain the necessary information in a fairly organized way:

(1) Draw all paths with |k 0 | less than some maximum value corresponding to
an upper limit on the energy for which the resulting bands will be accurate (we
choose 3/d for the upper limit on k 0 in this example).
R
We will guarantee an accurate sketch of E versus dk only for energies below
c
~2 kmax
2
/2me , where kmax is the maximum value of |k 0 |. Above this energy, we
may not have included all the necessary path segments.

(2) Calculate the energy E in reduced units ~2 2 /2me d2 .

!2  2  2
E ~k kx ky
= = + (18.180)
~2 2 /2me d2 /d /d /d

for each important point on the path. (In this case, the three corners of the
triangular paths are the important points,
R for it is there that the path changes
direction and the relationship of E to dk is altered.
c

We now use the data in Figure 18.53 to plot the energy in the reduced zone

1439
scheme for each path. These energies are then connected by segments of a
parabola. This process yields the desired plot of E versus path distance.

For this example, the plot is shown in Figures 18.54-18.60 below. Notice that
in the free-electron approximation portions of some of the paths are degenerate
in energy. Most of this degeneracy will be lifted in the weak-binding approxi-
mation( degenerate curves in this figure have more than one number associated
with them). We have plotted a sequence of graphs showing each path being
added.

Figure 18.54: Step #1

Figure 18.55: Step #2

1440
Figure 18.56: Step #3

Figure 18.57: Step #4

Figure 18.58: Step #5

1441
 Figure 18.59: Step #6

Figure 18.60: Step #7

Above, we have shown the energy versus path distance for the nine paths in the
free-electron approximation. The energy is plotted in reduced units. The little
numbers beside each path segment correspond to the path indices (1) through
(9) of the earlier figure. The arrow indicates the upper energy limit correspond-
ing to |k 0 | = 3/d.

1442
Notice carefully that the figure does not show the energy bands explicitly. How-
ever, we can extract them from this sketch.

The curve of E versus path distance that is everywhere lowest in energy corre-
sponds to the first band, the curve that is second lowest in energy corresponds
to the second band, and so on.

Thus we need only examine the figure in each region of (d/)(kx , ky ) (0, 0)
to (1, 0), (1, 0) to (1, 1), and (1, 1) to (0, 0) and select the first, second....
lowest curves to obtain plots of E versus path distance for the first, second, ...
bands. We have extracted the first nine bands in Figure 18.61 below.

Figure 18.61: First 9 Bands

The section of this graph between (0,0) and (1,0) shows the first nine bands in
the reduced zone scheme for the free-electron approximation.

We see from the figure that paths 1, 2, and 3 each correspond to a single energy
band, a fortuitous consequence of the triangular path chosen.

However, path 4 does not correspond everywhere to the fourth lowest energy.
Part of path 5 falls below path 4, and so the fourth energy band is made up of
contributions from paths 4 and 5.

In general, all bands higher than the first (lowest) have a complicated structure,
with contributions from parts of several paths. (This situation occurs because

1443
a general path will cross at least one Brillouin zone boundary and so cannot
correspond to a single energy band).

Constant Energy Contours

Consider now the second representation of the E versus k relationship for the
square lattice, namely, constant energy contours.

The objective of this method is to display the variation of energy with k for
each band in the reduced zone scheme.

We will begin in the extended zone scheme, with several Brillouin zones, and
draw constant energy contours. These contours can be translated into the first
zone by appropriate reciprocal lattice translation vectors. Such translation of
segments of higher zones into the first zone was discussed earlier.

The first four Brillouin zones of the square lattice were shown earlier in Figure
18.23.

In Figure 18.62 below, we have extended this sketch to show the first seven
zones and have also drawn contours of constant energy that lie entirely within
these zones in the free-electron approximation for the square lattice.

Figure 18.62: First 7 zones with energy contours - square lattice

In Figure 18.63 below we show the portions of these contours that lie in the
first zone, together with the free-electron energies (in reduced units) at several
points of the first zone. These are constant energy contours for the first band
of energy.

1444
 Figure 18.63: Energy contours in first band

To obtain contours for the second band, we translate the segments of the second
Brillouin zone from the figure above into the first zone, carrying along the
appropriate arcs.

The sequence of steps and the resulting reduced zone scheme constant energy
contours are shown in Figures 18.64-18.66 below.

Figure 18.64: Energy contours in extended scheme

1445
 Figure 18.65: Translating contours to second band regions

Figure 18.66: Energy contours in second band

Similar constructions yield the higher bands.

Notice the behavior of the energy for each band. For example, the second band
has a minimum in energy at the center of the edge of the Brillouin zone, As we
approach a corner of the zone, the energy increases monotonically to 2; as we
go toward the center of the zone, the energy increases to its maximum value of
4.

The Fermi Surface

Earlier we defined an energy EF called the Fermi energy, with the property
that, at absolute zero, states with energies E > EF are not occupied but states
with energies E < EF are occupied. Corresponding to the Fermi energy is the
Fermi wave vector kF . We saw that both quantities depend on , the number
of electrons per primitive unit cell.
 
The wave vectors of the occupied states each satisfy ~k  < kF (in the ex-
 

1446
tended zone scheme). Therefore we can view the points kF , as defining a
one-dimensional surface enclosing values of k that correspond to occupied states.
Similarly, we can define a Fermi surface in two dimensions corresponding to the
Fermi energy EF . In the free-electron approximation this energy corresponds
to one of the constant energy circles introduced earlier (in the extended zone
scheme). The radius of this circle is kF , the Fermi wave vector. Again, each state
with an energy less than EF is occupied at absolute zero. The corresponding
constant energy contour will lie within the Fermi surface. Let us derive ex-
pressions for kF and EF in the free-electron approximation and then locate the
Fermi surfaces for the square lattice. The derivations will be similar to their
one-dimensional counterparts we carried out earlier. We denote by q the area
in k space of the first Brillouin zone. The number of states in this zone is 2N ,
where N is the number of primitive unit cells in the crystal. Thus w(~k) is equal
to 2N/q, a result completely equivalent to our earlier result. The total number
of electrons in the crystal is [from earlier]
ZkF
2N
w(~k)d~k = kF2 (18.181)
q
0

But there are N electrons in the crystal, so we have

2N
N = kF2 (18.182)
q
which implies that the Fermi wave vector is
r
q
kF = (18.183)
2
The Fermi energy in the free-electron approximation is
~2 kF2 ~2 q
EF = = (18.184)
2me 2me 2
These results apply to any two-dimensional crystal. To illustrate, we will apply
them to the square lattice.

The area of the first Brillouin zone of the square lattice is

 2
2
q= (18.185)
d
where d is the lattice separation in coordinate space. Therefore, the Fermi wave
vector is r
2
kF = (18.186)
d
and the Fermi energy is
~2  2 2
EF = (18.187)
2me d

1447
 kF (/d) EF (~2 2 /2me d2
1 0.798 0.637
2 1.128 1.273
3 1.384 1.910
4 1.596 2.546
5 1.784 3.183
6 1.954 3.820

Table 18.4: Two-dimensional square lattices values

Table 18.4 above shows values of kF and EF for = 1 = 6 for the two-
dimensional square lattice.

The corresponding Fermi surfaces are shown in the extended zone scheme in
Figure 18.67 below.

Figure 18.67: Fermi surfaces

The circle with the smallest radius corresponds to = 1; the one with the
largest radius to = 6. In Figure 18.68 below these Fermi surfaces are trans-
lated into the first Brillouin zone. The shaded parts correspond to occupied
states at absolute zero (wave vectors with energies less than EF .

1448
 Figure 18.68: Fermi surfaces translated to first zone

Let us look at these sketches. For q = 1, the entire Fermi surface lies within
the first Brillouin zone. Thus all occupied states have energies within the first
band. Notice that there are states with energies in the first band that are not
occupied. This fact is also reflected by the E versus path distance graph for
the first four bands. This graph is reproduced in Figure 18.69 below, where we
show the location of the Fermi energies.

1449
 Figure 18.69: First 4 bands with location of Fermi surfaces

For crystals with 2 or 3 electrons per primitive unit cell, we see from (b) in
Figure 18.68 and Figure 18.69 above that there are occupied states in the first
and second bands; neither of these bands is completely filled. In contrast, the
first band is filled for crystals with 4.

The Weak-Binding Approximation

We will now improve on the free-electron approximation by introducing the
periodicity of the crystal potential energy. We approach this task via the weak-
binding approximation discussed earlier, which assumes that the electrons are
not too strongly bound to the individual nuclei of the crystal. In particular, we
will ignore states that can be treated by time-independent perturbation theory
and concentrate on the strongly mixed states.

As earlier, we can determine which states are strongly mixed by examining the
matrix elements of the crystal potential energy. As earlier, this matrix element
is defined as Z
D   E 1 ~0 ~
V~k0~k = ~k 0  V ~k = d~reik ~r V (~r)eik~r (18.188)
 
A
where D  E 1 ~ D  E 1 ~0
~r  ~k = eik~r , ~r  ~k 0 = eik ~r (18.189)
 
A A
are normalized free-electron wave functions of energy
~2 k 2 ~2 k 02
, (18.190)
2me 2me

1450
respectively, and where ~r = xx+y y is a vector in the two-dimensional coordinate
space of the direct lattice. Recall from earlier that these matrix elements also
appear in the calculation of the energy eigenvalues due to mixing of the strongly
mixed states.

An analysis identical to the perturbation discussions earlier reveals that the

(0)
only states that mix at all, strongly or weakly, with ~ (~r) are those with
E,k
wave vectors ~k 0 that are related to ~k by a reciprocal lattice translation vector
~k ~k 0 = G.
~ If this relation is not satisfied, then V~ 0~ = 0.
k k

We can write the nonzero matrix elements as

Z
1 ~ ~k ~k 0 = G
~
VG
~ = eiG~r V (~r)d~r if (18.191)
A

or in terms of the integral over one primitive unit cell (PUC) of area Q as
Z
1 ~
VG~ = eiG~r V (~r)d~r (18.192)
Q
P UC

The strongly mixed states satisfying ~k ~k 0 = G~ are those with wave vectors
near Brillouin zone boundaries. All other states can be treated by perturbation
theory. Let us be precise about what this statement means. If we introduce the
periodic potential energy as a perturbation, we obtain an expression for the first-
(0)
order perturbed wave function in terms of the ~ (~r) and a sum over all other
E,k
free-electron wave functions with wave vectors ~k 0 = ~k G ~ for every reciprocal
~ ~
lattice translation vector G(as we found earlier). If k is not near a Brillouin
(0)
zone boundary, all the other free-electron states mix weakly with ~ (~r) and
E,k
will not greatly affect the energy or wave function. If ~k is near a zone boundary,
(0)
one or more other states will mix strongly with (~r).
These states satisfy the
E,~
k    
selection rule ~k k = G
~0 ~ and have energies very close to each other ~k  ~k 0 
so that the energy denominator in the perturbation theory summation is small.
Thus, at or near a single Brillouin zone boundary, two states are strongly mixed.
At or near the intersection of two or more zone boundaries, more than two
states can be strongly mixed. For example, if ~k is near the intersection of two
boundaries, then we can associate two other equivalent wave vectors with ~k,
(0)
each of which corresponds to a state nearly degenerate with ~ (~r). (Notice
E,k
that this situation did not arise in one dimension). For example, in Figure 18.70
below, we show the location of several strongly mixed states for the first four
Brillouin zones of the square lattice.

1451
 Figure 18.70: Strongly mixing states

In one case, ~k is near a single zone boundary, and there is only one state that
(0)
mixes strongly with ~ (~r). In the other case shown, ~k is near the intersection
E,k
of three zone boundaries, and we must contend with four strongly mixed states.
In Figure 18.70, the first four Brillouin zones of the square lattice in the reduced
zone scheme are shown. The arrows connect the ~k values for strongly mixed
states. Two cases are considered, one near the edge of one zone boundary (two
states) and one near the intersection of three boundaries (four states).

Calculation of Band Gaps

The calculation of the band gap for a two-dimensional crystal at any single
Brillouin zone boundary is analogous to the one-dimensional calculation we did
earlier. However, a somewhat more elaborate calculation is required when two
or more Brillouin zone boundaries intersect.

Let us focus on one of these cases in the square lattice and calculate the energy
gaps that arise from this strong mixing. In particular, suppose that we choose
~k at the corner of the first Brillouin zone, ~k = (/d) (1, 1). At this point,
three Brillouin zone boundaries intersect, so four free-electron states are strongly
mixed. These states correspond to the ~k vectors

(/d) (1, 1) , (/d) (1, 1) , (/d) (1, 1) , (/d) (1, 1) (18.193)

and their negatives. In general, we can write these vectors as

~ = (2/d) (m, n)
G (18.194)

and can denote VG

~ by Vmn .

weak
The weak-binding wave function E can be written as a linear combination
of the strongly mixed free-electron wave functions
4
(0)
X
weak
E (~r) = ai E,i (~r) (18.195)
i=1

1452
 (0)
where the E,i (~r) are the four free-electron wave functions corresponding to
the wave vectors of above. (Remember that these four free-electron states are
degenerate). Substituting this expansion into the Schrodinger equation
HE = EE (18.196)
we obtain
4 4
(0) (0)
X X
H ai E,i (~r) = E ai E,i (~r) (18.197)
i=1 i=1
The corresponding eigenvalue equation is obtained by multiplying this equation
(0)
by E,j (~r) and integrating over d~r = dxdy (using orthonormality of the free-
electron wave functions)
4
X
Hji ai = Eaj , j = 1, 2, 3, 4 (18.198)
i=1

where Hji is the matrix element

Z
(0) (0)
Hji = hj| H |ii = d~rE,j (~r)HE,i (~r) (18.199)

This result yields the determinant equation

det (Hji Eji ) = 0 (18.200)
For convenience, we will measure the energy from V00 , the constant part of
the crystal potential energy (taken as the zero of energy in the free-electron
approximation). The diagonal elements of the Hamiltonian matrix (Hji , i = j)
are due solely to the free-electron term corresponding to the various ~k values.
Thus the determinant equation becomes
 ~2 k 2 
 2me E V01 V10 V11 

~2 k 2
V01 2me E V11 V10 

=0 (18.201)

2 2
 ~ k
V10 V11 2me E V01 

~2 k 2

 V 11 V 10 V 01 E  2me

where k = 2(/d). From an (18.191) or (18.192), Vmn is given by
Z   
1 2 2
Vmn = 2 exp i mx + ny V (x, y)dxdy (18.202)
d d d
P UC

Expanding the exponentials, we obtain

Z
1 2 2
Vmn = 2 cos mx cos nyV (x, y)dxdy (18.203)
d d d
P UC
sin 2 2 2 2
Z  
1 d mx sin d ny + i cos d mx sin d ny
+ 2 V (x, y)dxdy
d +i sin 2 2
d mx cos d ny
P UC

1453
Clearly, we must know more about the form of V (x, y) in order to further
evaluate this matrix element. Since the crystal potential energy has fourfold
rotational symmetry, it satisfies

V (x, y) = V (y, x) = V (x, y) = V (y, x) (18.204)

If we consider only a square crystal with mirror symmetry such that

V (x, y) = V (x, y) = V (x, y) (18.205)

the second integral in the expression for Vmn is zero, and we conclude that

Vmn = Vnm = Vmn = Vnm (18.206)

Using this result, we can define two real quantities V1 and V2 by

V1 = V10 = V01 = V10 = V01 , V2 = V11 = V11 = V11 = V11

(18.207)
Let us also define a quantity equal to the actual energy as measured from the
free-electron value,
~2 k 2
=E (18.208)
2me
Written in terms of V1 , V2 and , the determinant equation takes on the far
simpler form

4 2(2V12 + V22 )2 8V12 V2 + V22 (4V12 + V22 ) = 0 (18.209)

Factoring this equation, we have

2
( + V2 ) ( V2 + 2V1 ) ( V2 2V1 ) = 0 (18.210)

This equation is a quartic equation, so it has four roots. The eigenvalues are
( 2 2
~ k
V2 twice
E = ~2m e
2 2
k
(18.211)
2me + V2 V1

These are the energies of the four weak-binding states whose wave functions are
given by the linear combinations above. They arise from the mixing of four free-
electron states at the corner ~k = (/d) (1, 1) of the first Brillouin zone. Notice
that a twofold degeneracy remains in the weak-binding approximation.

E versus Path Distance

We can usually assume that |V2 | < |V1 |, that is, that the matrix elements are
smaller for longer G~ vectors. Then we obtain four energies at the corner of the
first Brillouin zone, one above and one below a pair of degenerate levels.

1454
This result induces gaps in the E versus path distance plot for the square lattice
(remember that our earlier plot had no gaps). The modified plot of weak-binding
energy versus path distance for paths of the first four bands of the square lattice
(energy is shown in reduced units) is shown in Figure 18.71 below. The bands
are labeled.

Figure 18.71: Band Gaps create Distinct Bands

Compare this plot to the earlier plot in Figure 18.69.

Here splittings appear between every band for all values of ~k except at the zone
corner. But values of E do not exist for which no electronic states are allowed,
that is, there are no actual band gaps. In fact, the bands overlap in this example.
These features are characteristic of two- (and three-) dimensional systems.

Constant Energy Contours

We have seen that only near a Brillouin zone boundary is the weak-binding
energy appreciably different from the free-electron energy. Consequently, we
expect that, in the weak-binding approximation, constant energy contours for
a two-dimensional lattice would be altered significantly from the free-electron
contours we described earlier only near a zone boundary. For any given wave
vector ~k ear a single zone boundary, the energy in the weak-binding approxi-
mation is lower than the free-electron energy if ~k is inside the zone boundary.
Energies for ~k outside the Brillouin zone boundary are raised from their free-
electron values due to the mixing of the two free-electron states. In addition, if
we are close to only one zone boundary, the constant energy contours will strike
the boundary at a right angle. These effects bend the contours in toward the
boundaries, as shown in Figure 18.72 below.

1455
 Figure 18.72: Constant energy contours bending toward zone boundaries

Above, we have shown a comparison of a weak-binding constant energy contour

(solid curve) and a free-electron constant energy contour (dashed curve) near a
Brillouin zone boundary.

Returning to the square lattice, we show in Figure 18.73 below how to modify
the free-electron constant energy contours. Contours are shown in the weak-
binding approximation for the first four bands in the reduced zone scheme.

Figure 18.73: Modified constant energy contours

Fermi Surfaces
The Fermi surfaces are simply particular constant energy contours correspond-
ing to the Fermi energies. Fermi surfaces for several values of shown in Figure
18.74 below. These curves are obtained by modifying the surfaces we drew
earlier, using the bending at the boundary method shown above.

1456
 Figure 18.74: Fermi surfaces for = 1

For a crystal with a square lattice and = 1, the Fermi surface in the weak-
binding approximation is a circle whose area is half that of the first zone. Since
this circle is far from the zone boundaries, it is unaltered from its free-electron
behavior. This feature is not true of other values of , as the figure shows.

18.7 Born-Oppenheimer description of two atoms

in a combined oscillator and lattice trap
In this section, we analyze the quantum states of two atoms in a combined har-
monic oscillator and periodic lattice trap in one spatial dimension. In the case
of tight-binding and only nearest neighbor tunneling, the equations of motion
are conveniently represented in the momentum representation. We then show
that in the case of strong attraction between the particles, the different time
scales of relative and center-of- mass motion validate a separation of the prob-
lem similar to the Born-Oppenheimer approximation applied in the description
of electronic and nuclear motion in molecules.

18.7.1 Introduction
We will now study cold atoms(lattice trap) in periodic potentials formed by
standing wave laser beams.

1457
18.7.2 Lattice Hamiltonian
A. One-body Hamiltonian and Wannier states

We consider a particle moving in a sinusoidal potential, so that the Hamiltonian

can be written
Hlat = P 2 + V0 sin2 ( X) (18.212)
where the scaled position and momentum operators have the dimensionless com-
mutator
[X, P ] = i (18.213)
Since the potential is periodic with unit period, Blochs theorem ensures that we
can choose energy eigenstates
 E  E
Hlat q(n) = Eq(n) q(n) (18.214)
 

with quasi-momenta q [, ] and band indices n = 0, 1, ..... Another basis

- the Wannier states -can be obtained as the Fourier transform over a single
Brillouin zone of the eigenstates
Z +

 (n)
E 1  E
= dqeikq q(n) (18.215)

wk
2
For sufficiently deep lattices, the Wannier states with different n are localized
around different lattice potential minima and are identical up to translation. We
note that, within each energy band, the overlap between the quasi-momentum
eigenstates and the Wannier states
D  E
 (n) m,n
q(n)  wk = eikq (18.216)
2
is similar to the usual overlap between position eigenstates and momentum
eigenstates.

The Hamiltonian is block-diagonal in the basis of Wannier states, and the cou-
pling of Wannier states at different locations is given by
D   E
(m)   (n) (n)
wk  Hlat wk = m,n J|jk| (18.217)

where Z +
(n) 1
Jk = dqeikq Eq(n) (18.218)
2
(n)
This shows that Jk is the Fourier transform of the energy bands as a function
of q and the dispersion relations can be written as

(n) (n) (n)
X X
Eq(n) = Jk eikq = J0 2 Jk cos (kq) (18.219)
k=0 k=1

1458
For deep potentials the energy bands are relatively flat, and the higher order
cosine terms are suppressed. This justifies the tight binding approximation in
which one retains only the nearest lattice site coupling, and in the following we
will suppress the band index (n), and focus on the lowest band described by the
tight binding Hamiltonian

X
HT B = J1 {|wk1 i hwk | + |wk+1 i hwk |}
k=

= 2J1 cos (P ) (18.220)

B. Harmonic confinement

Adding a harmonic confinement to the lattice potential is adequately described

by adding the term k X 2 with the spring constant k to the Hamiltonian. For
deep lattice potentials, the Wannier state Ketwj is well localized at X = j, so
we make the approximation to replace X by the discrete quasi-position operator
of the lowest band
X
W = j |wj i hwj | (18.221)
j=

Introducing a rescaling of the Hamiltonian by 4J1 and defining = k/4J1 we

end up with the Hamiltonian

HT B k W 2 cos (P )
H = + = W 2 (18.222)
4J1 4J1 2
Similar to the usual relationship between continuous position and momentum
operators, the discrete position operator W acts as a differentiation in the con-
tinuous quasi-momentum representation

hq | W |i = i hq | i (18.223)
q
which is easily derived by inserting a resolution of the identity in Wannier states
and using the overlap formula (18.216). Therefore, we arrive at the quasi-
momentum expression of the single particle Hamiltonian

2
 
cos (q)
hq | H |i = 2 hq | i (18.224)
q 2

At this point we make the curious observation that, after having restricted the
Hilbert space to the lowest energy band and having added a quasi-harmonic
confinement, the Hamiltonian in momentum space (18.224) has the same form
as the original optical lattice Hamiltonian (18.214) in position space. In both
cases, the Schrodinger equation takes the form of the Mathieu equation, but
instead of looking for eigenstates of (18.214) with any quasi-momentum, we will
only look for periodic eigenstates for (18.224), i.e. with zero quasi-position.

1459
C. Interacting particles

In an ultra-cold gas of bosons, the interaction between the particles is adequately

described by the two-particle contact interaction operator Uint with the matrix
elements
hX1 ; X2 | Uint |X3 ; X4 i = g(X1 X3 )(X2 X4 )(X3 X4 ) (18.225)
for some interaction strength g. In the tight binding approximation, the Wannier
states are localized at different lattice sites, and one may neglect matrix elements
of the interaction potential with Wannier product states located on different
sites. We thus end up with the following effective interaction operator acting
on two-particle states
ef f
X
Uint =G |wj ; wj i hwj ; wj | (18.226)
j

where the strength parameter is given by

Z
4
G = g dX |w0 (X)| (18.227)

Using the relation (18.216), we can calculate the matrix elements of the effective
interaction operator in quasi-momentum space
ef f G
hq1 ; q2 | Uint (q3 + q4 q1 q2 )
|q3 ; q4 i = (18.228)
2
which shows that the interaction conserves the total quasi-momentum and is
independent of its value.

A system of two identical particles in an optical lattice with harmonic confine-

ment, which interact by the contact interaction is described by the Hamiltonian
cos (P1 ) cos (P2 )
H = (W12 + W22 ) + U (18.229)
2 2
ef f
with U = Uint /4J1 .

18.7.3 Relative- and Center-of-Mass Quasi-Momenta

In the quasi-momentum representation
cos (P1 ) + cos (P2 )
|q1 ; q2 i
2
cos (q1 ) + cos (q2 )
= |q1 ; q2 i
 2   
q1 + q2 q1 q2
= cos cos |q1 ; q2 i
2 2
! !
Q+ Q
cos cos |q1 ; q2 i (18.230)
2 2

1460
where we have defined new operators by their action on quasi-momentum eigen-
states,
q iQ /2 |q1 ; q2 i ei(q1 +q2 )/2 |q1 ; q2 i (18.231)
The introduction of these operators suggest to reparameterize the quasi-momentum
basis states |q1 ; q2 i in terms of their sum and difference:

q = q2 q1 (18.232)

The quasi-momentum eigenstates states are defined for pairs of q1 and q2 in the
set
S12 = [; ] [; ] (18.233)
corresponding to a diamond shaped area in the coordinate plane of q as shown
in figure 18.75 below.

Figure 18.75: Quasi-momentum of the two particles vs. relative and center-of-
mass quasi-momentum. Left: The first Brillouin zone S12 in the (q1 , q2 )-plane is
emphasized and repeated in each direction. The color coding indicates the values
of a function that is periodic in both variables with period 2 and illustrates the
required periodicity. The set S which contains exactly one representative of
each point from S12 is shown by the gray rectangle. Right: The same function
is shown but in the (q+ , q? )-coordinate system. The set S is emphasized and
repeated, but with a different tiling than for S12 in the left panel.

1461
If we choose the values of (q+ , q? ) in the set

S = [; ] [2; 2] (18.234)

then each point from S12 is represented exactly once as is evident from figure
18.75. This means that we can reparametrize the quasi-momentum eigenstates
as
1  
|q+ , q i = (q+ q )/2 ; (q+ +q )/2
2

|q1 ; q2 i = 2 |q1 + q2 , q2 q1 i (18.235)

where the front factor is chosen to preserve orthonormality, such that we have
the resolution of identity
Z + Z +2
1 = dq+ dq |q+ , q i hq+ , q | (18.236)
2

The corresponding discrete relative and center-of-mass position coordinates

W2 + W1
W (18.237)
2
act in the following way

hq+ , q | W |i = i hq+ , q | i (18.238)
q

and the interaction operator U has the following matrix elements in terms of
the relative and center-of-mass quasi-momentum states
Z +2
0 0 

 
hq+ , q | U |i = dq q+ , q (18.239)
2

with = G/(16J1 ).

The two-atom Hamiltonian can now be written,

! !
2 2 Q+ Q
H = 2(W+ + W ) cos cos + U (18.240)
2 2

The Schrodinger equation with the Hamiltonian (18.240) can be solved accu-
rately for a wide range of parameters as in 18.241 below. The resulting eigenen-
ergies and the wave functions (18.246 below) will be used as reference for our
analysis by the Born-Oppenheimer separation of the motional degrees of free-
dom.

1462
Solving the equations numerically

1. Solving the non-approximated equation

To solve the two-atom Schrodinger equation in the tight binding approximation

we expand the state as X
|i = Fj,k |wj ; wk i (18.241)
j,k

The stationary Schrodinger equation with the Hamiltonian (18.240) yields the
equation for the expansion coefficients

EFj,k = 4j,k Fj,k + (j 2 + k 2 )Fj,k

Fj1,k + Fj+1,k + Fj,k1 + Fj,k+1
(18.242)
4
The original Hamiltonian is invariant under parity inversion of both particles
so we can find a complete set of solutions of even and odd wave-functions. In
terms of the expansion (18.241) this means that we can find solutions where

Fj,k = pFj,k (18.243)

where p can assume the values 1. In addition, since we are dealing with two
identical bosons, only symmetrized wave functions are physically meaningful,
with implies that we have the symmetry

Fj,k = Fk,j (18.244)

For numerical purposes we enforce these requirements by hand in the following

way. Instead of looking at all pairs (j, k) Z2 , we restrict our attention to those
in the subset
T = {(j, k) Z2 ||k| j jmax } Z2 (18.245)
for some manually chosen jmax . Using the symmetries we reformulate the re-
currence equation such that it only involves coefficients from T . The equation
can be expressed as a matrix eigenvalue equation which is amenable to standard
numerical diagonalization routines. When all coefficients have been found - and
properly normalized - the wave function in relative and center-of-mass quasi-
momenta is given by
X
hq+ , q | i = Fj,k hq+ , q | wj ; wk i
j,k
1 X
= Fj,k ei(j+k)q+ /2 ei(kj)q /2 (18.246)
2(2) j,k

1463
18.7.4 Born-Oppenheimer Separation
A. Derivation

We write the Hamiltonian in (18.240) as

H = H + 2W+2 (18.247)

where ! !
Q+ Q
H = 2W2 cos cos + U (18.248)
2 2

We note that eiQ+ /2 commutes with H and we define their joint eigenstates
by |q+ , ni:
H |q+ , ni = n (q+ ) |q+ , ni (18.249)
eiQ+ /2 |q+ , ni = eiq+ /2 |q+ , ni (18.250)
with the following orthogonality relations

 0 0 0
q+ , n  q+ , n = n,n0 (q+ q+ ) (18.251)

The states |q+ , ni can be expanded

Z +2
0 0
|q+ , ni = dq A(q
n
+)
(q ) |q+ , q i (18.252)
2

and the orthogonality relation (18.251) implies

Z +2 h i
(q )
dq A(qn
+)
(q ) An0+ (q ) = n,n0 (18.253)
2

Any eigenstate of the full Hamiltonian (18.247) can be expanded as

Z + X
0 0  0
C (n) (q+
 
|i = dq+ ) q+ , n (18.254)
n

0
The expansion coefficients C (n) (q+ ) are found by applying the Hamiltonian
(18.247) to the expanded wave function (18.254) and using (18.248)
Z + X
0 0 0
 0 
H |i = dq+ C (n) (q+ ){n (q+ ) + 2W+2 } q+ ,n (18.255)
n

In the (q+ , q? )-representation for the state vector, the eigenvalue equation takes
the form of coupled differential equations
X
0
E An(q+ ) (q )C (n) (q+ )
n
X 2

0 0
= n (q+ ) 2 2 A(q
n
+)
(q )C (n) (q+ ) (18.256)
n
q+

1464
The goal of the following analysis is to find an approximation for the eigen-
states, which is easier to apply numerically and which offers insights into
 0their
internal structure and dynamics. To this end, we assume that the states q+ ,n
(q )
described by q wave functions An + (q ) depend only weakly on the argument
(q )
q+ . Eliminating thus the partial derivatives of An + (q ) with respect to q+ in
the evaluation of the right hand side of (18.256), and using the orthogonality
(q )
of the An + (q ) functions, we arrive at the following approximate equation for
the expansion coefficients

2 C (n) (q+ )
n (q+ )C (n) (q+ ) 2 2 = EC (n) (q+ ) (18.257)
q+

This has the form of a Schrodinger equation for a single particle in the poten-
(n)
tial n (q+ ). For each energy potential we can find discrete eigenenergies Em
(n)
and associated eigenfunctions
E Cm that solve (18.257) and yield approximate
 (n)
eigenstates m for the full Hamiltonian (18.247)
D  E
 (n) (n)
q+ , q  m = Cm (q+ )A(q
n
+)
(q ) (18.258)

Note the formal similarity of this reduction of the problem with the use of the
Born-Oppenheimer approximation in chemistry. In the latter, the wave function
is expanded as a product of wave functions in nuclear and electronic coordinates,
and due to the large difference in mass and hence in energy and time scales,
the electronic wave functions are supposed to follow changes in the slow nuclear
coordinates adiabatically.

In our case, the two particles have identical masses, and in the absence of mutual
interaction, the relative and center-of-mass motion occur on similar time scales,
and the Born-Oppenheimer approximation should not be valid. But, as we
increase the attractive interaction between the atoms, bounds states are formed,
and the relative position develops a new, faster time scale given by the binding
energy. Our separation is carried out and motivated in the quasi-momentum
picture, where a further observation may be in order: a strongly bound state in
the relative position coordinate corresponds to a very extended wave function
in the relative momentum, while the center-of-mass momentum may be well
defined. This supports the assumption that the dominant contribution to the
(n)
second derivative in (18.256) stems from the q+ wave function Cm (q+ ), and
(q+ )
hence that the derivative of An with respect to q+ may be neglected.

B. Application

Before we apply the Born-Oppenheimer approximation, let us consider how we

expand states onto the center-of-mass and relative quasi-momentum eigenstates.
Every state can be expanded in both the two-particle quasi-momentum basis,

1465
and in the basis of relative and center-of-mass quasi-momenta.
R + R +
dq1 dq2 (q1 , q2 ) |q1 ; q2 i

|i = (18.259)
R +
R +2

dq+ 2 dq (q+ , q ) |q+ , q i

While |q1 , q2 i and |q+ , q i are defined for (q1 , q2 ) S12 and (q+ , q ) S ,
respectively, we can look for functions defined on the entire R2 and restrict the
solution afterwards. The function is periodic in both variables with period
2, and this enforces to obey the symmetry
(q+ + 2, q 2) = (q= , q ) (18.260)
c.f. the tiling of R2 with replicas of S in the right panel of figure 18.75. Thus, a
necessary - but not sufficient - condition is that is periodic in both q+ and q
with periodicity 4. We are considering bosons and the state must be symmetric
under the exchange of the two particles, (q+ , q ) 7 (q+ , q ), which implies
the further constraint
(q+ , q ) = (q+ , q ) (18.261)
Using these arguments on (18.258) we conclude that we are looking for solutions
(q )
such that An + (q ) is even and periodic in q with period 4, and such that
(n) (q )
the product of Cm (q+ ) and An + (q ) is periodic in q+ with the same period.
Furthermore, the product must satisfy the relation (18.260).

1. The first Born-Oppenheimer equation

To apply the Born-Oppenheimer approximation, we must first find the eigen-

states of H and their eigenenergies, and using the formal expansion of the
states (18.252), the eigenvalue equation (18.249) leads to the equation
2
  q 
(q )
n (q+ )An = 2 2 F (q+ ) cos A(q
n
+)
(q )
q 2
Z +2
0
+ dq A(q
n
+)
(q ) (18.262)
2
q+

where F (q+ ) = cos 2 . For each value of q+ this equation has the
form of a
Schrodinger equation with argument q , and with a periodic cos q2 potential
with amplitude F (q+ ) and a non-local potential with strength . Solutions
which are periodic in q with period 4 are readily found by Fourier expansion
(q )
of An + (q ) as in 18.246.

When solving (18.262) we are looking for solutions to a Schrodinger like equation
with a cosine potential with period 4. Solutions which are periodic with the
same period - referred to as zero quasi-momentum states for periodic problems
in position space - can be chosen to be real-valued. The front factor F (q+ ) of
the cosine potential is itself a cosine function of q+ leading to two observations:

1466
 1. F (q+ ) is an even function of q+ so equation 18.262 is unaltered under the
transformation q+ 7 q+ . Thus the solutions must be identical up to a
complex factor, and since they are real-valued we can choose the solutions
as
A(q
n
+)
(q ) = A(q
n
+)
(q ) (18.263)
(q )
We could not have chosen a minus sign, since this would have made An +
vanish for q+ = 0.

2. F (q+ ) changes to values of opposite sign when q+ is increased by an

amount of 2 and the cosine potential cos (q /2) in (18.262) is effectively
translated by half a period. For this translated potential the eigenvalues
are the same, while the eigenfunctions are translated and scaled

n (q+ ) = n (q+ + 2) (18.264)

An(q+ ) (q ) = n An(q+ +2) (q 2) (18.265)

(q )
The factor n may take the values 1 sin An + (q ) is real-valued for all
values of q+ and q .

Applying the relations (18.263) and (18.265) for q+ = we get the relation

An(+) (q ) = n A(+)
n (q 2) (18.266)
(+)
so we can determine n from the translational symmetries of AN .

2. The second Born-Oppenheimer equation

Eq. (18.262) yields the potential n (q+ ) which is periodic with period 2, and
(n)
we are looking for functions Cm (q+ ) that are periodic in q+ with period 4.
Therefore, Blochs theorem tells us that we can choose a complete set of solutions
as
(n)
Cm (q+ ) = ein q+ /2 Dm
(n)
(q+ ) (18.267)
(n)
where Dm is periodic with period 2, and n = 0, 1. For n = 0 the solution
(n)
Cm (q+ ) is thus periodic with period 2, whereas for n = 1, it is antiperiodic.
(n) (q )
We require that the product of Cm (q+ ) and An + (q ) satisfies the symmetry
(18.260), and if we combine this with (18.265), we get the relation
(n)
Cm (q+ )A(q
n
+) (n)
(q ) = n Cm (q+ + 2)A(q
n
+)
(q ) (18.268)
(n)
from which we conclude that Cm (q+ ) must fulfill the symmetry
(n) (n)
Cm (q+ + 2) = n Cm (q+ )A(q
n
+)
(18.269)

Comparing to (18.267) we see that this implies that for n = 1 we must choose
n = 1 and for n = +1,we can use = 0. We can solve (18.257) by a Fourier

1467
 (n)
expansion of Dm and n (as shown below in 18.270).

2. Solving the first Born-Oppenheimer equation

(q )
The solutions of (18.262) are functions An + (q ) which are periodic in q with
period 4. Therefore, for each value of q+ we make the expansion
1 X (q+ ) ijq /2
A(q
n
+)
(q ) = j,n e (18.270)
4 j

and obtain the tridiagonal recurrence relation,

F (q+ ) h (q+ ) (q+ )
i
j1,n + j+1,n
2  
(q )
= j 2 + 4j,0 n (q+ ) j,n+ (18.271)
2
To accommodate the bosonic nature of the particles, we only look for even
(q )
solutions to (18.262), so we only need to consider terms j,n+ with j 0, and
for j = 0 we use
(q ) (q )
F (q+ )1,n+ = [4 n (q+ )]0,n+ (18.272)
By expressing the recurrence relation as a matrix eigen- value equation, this can
be truncated and solved with good accuracy.

2. Solving the Second Born-Oppenheimer equation

We solve the second Born-Oppenheimer equation using the results from the first
Born-Oppenheimer equation. First, coefficients in the expansion
X
n ijq+
n (q+ ) = K e (18.273)
k

are determined by a discrete Fourier transformation. We then use the expansion

(n) 1 X m,n i(l+ 2n )q+

Cm (q+ ) = e (18.274)
2 l l

in (18.257) together with the expansion (18.273) which yields the following equa-
tion for the -coefficients
 2
X
n m,n n
k lk + 2 l + lm,n = Em
(n) m,n
l (18.275)
2
k

Since the potential energy curves n (q++ ) are even functions, the solutions can
be chosen to be either even or odd, and the coefficients then fulfill lm,n = l
m,n
.
It suffices to only consider coefficients with m 0 and solve the recurrence
equations.

1468
18.7.5 Exact and Born-Oppenheimer Approximate Solu-
tions
A. Wave functions

When solving the first Born-Oppenheimer equation (18.256) we find eigenval-

(q )
ues n (q+ ) and eigenfunctions An + (q ) for each value of q+ [, ]. In the
left-most panel in figure 18.76 below is shown the lowest few potential curves
n (q+ ).

Figure 18.76: Energies and eigenfunctions found by solving the two Born-
Oppenheimer equations for = 0.5 and = 0.5. Left panel: The six low-
est potential curves n (q+ ) found from the first Born-Oppenheimer equation.
Upper panels: Magnification of four of the potential curves in the left panel.
(q )
Lower panels: Eigenfunctions An + (q ) for the first Born-Oppenheimer equa-
tion shown for all values of q+ for the corresponding n-values. In the upper
panels are shown (horizontal dashed blue/red lines) the two lowest energies
(n)
Ej for j = 1, 2 found from solving the second Born-Oppenheimer equation
in the potential n (q+ ) and the corresponding wave functions (solid blue/red
lines).

The lowest potential curve is well separated from the higher ones which lie
closer. Each of the potential curves has an energy variation which is typically
small compared to the energy distance between the bands. In the upper panels,
a magnified view of the curves are shown. In the lower panels eigenfunctions
(q )
An + (q ) to the first Born-Oppenheimer equation are shown for four different
values of n.

1469
The second Born-Oppenheimer equation uses the energies n (q+ ) as potential
functions in a Schrodinger like equation, and each of the upper panels in Figure
18.76 show the energy levels of the two lowest eigenstates (m = 1, 2) in these
(n)
potentials along with their eigenfunctions Cm (q+ ). As we saw in the previous
(n)
section, the function Cm (q+ ) should be chosen periodic or anti-periodic de-
(q )
pending on the value of n . By studying the behavior of An + (q ) at q+ =
(q )
one can see if n is +1 or 1 depending on whether the wave function An + (q )
(n)
changes sign when translated by or not. For n = 0, 2 the solutions Cm (q+ ) to
the second Born-Oppenheimer equation must be periodic with period 2, while
(n)
for n = 1, 5 the solutions Cm (q+ ) must be chosen anti periodic.

Total Born-Oppenheimer solutions to the two-atom Hamiltonian are shown in

Fig. 18.77.

Figure 18.77: Upper panels: Quasi-momentum wave functions found by the

Born-Oppenheimer approximation for = 0.5 and = 0.5. Columns 1-2 show
the two lowest eigenstates in the lowest potential curve 0 (q+ ) corresponding to
the n = 0 column in figure 18.76. Columns 3-4 correspond to the n = 1 column
in figure 18.76 and columns 5-6 correspond to the n = 2 column in figure 18.76.
Lower panels: The corresponding exact solutions.

Panels (A1-2) corresponds to the approximate solutions from the n = 0 case of

figure 18.76, panels (A3-4) corresponds to the n = 1 case, and panels (A5-6)
corresponds to the n = 2 case. In the lower panels of figure 18.77 the corre-
sponding exact two-atom eigenstates are shown. There is a good agreement
between the exact and approximate solutions, especially for the low excitations
of the lowest bands.

1470
B. Energies

In figure 18.78 both the exact and the approximative energies are shown for fixed
as functions of the scaled interaction strength . Except in the region where
is numerically small, there is reasonable agreement between the exact and
the approximated energy levels. For negative there is a clear grouping of the
energy levels in two groups: Those that are nearly constant as a function of and
those that depend linearly on . Comparing to the approximate energies found
by the Born-Oppenheimer approximation we see that the linear dependence
comes from the fact that the position of the lowest potential curve varies linearly
with .

Figure 18.78: Exact and approximate energies as a function of for = 1.

Exact energies are plotted by the black dashed lines, and or n = 0, 1, 2, 3 the
(n)
energies Ej found from the Born-Oppenheimer equation is plotted in four
different colors.

1471
C. Approximate solution of the first Born-Oppenheimer equation

To understand the behavior of the energy spectra, we start by analyzing the

system in the limit where at least one of the two coefficients , is (numerically)
much larger than unity, so that we can find analytical approximations. This
limit enables us to treat the term
q 

F (q+ ) cos (18.276)
2
in the first Born-Oppenheimer equation (18.256) as a perturbation. When we
neglect this term, we can choose a complete set of eigenfunctions as plane waves
with wave number k/2 for k = 1, 1, 2, .... and with energies

k = k 2 + 4k,0 (18.277)
2
Only the even linear combinations are physically relevant, and note that the k
does in general not coincide with the excitation number n as used in the first
Born-Oppenheimer equation, where the energy curves n were sorted by energy.

The term 4 contributes only for k = 0 since all other plane waves integrate
to zero in the second line of Eq. (18.256). Even when the omission of (18.276)
(q )
is not valid, the integral still becomes substantial if An + (q ) has no nodes,
(q )
whereas it is suppressed when there are sign changes in An + (q ).

In figure 18.78 we notice some discontinuities in the approximate energies, which

can be explained in the following way. Due to the linear dependence of the
energy for the k = 0 plane wave, its energy becomes degenerate with the higher
levels, when varies. More precisely, 0 will cross k at the -value
k 2
k = (18.278)
8
Without the symmetry requirement (18.260) we could find two families of solu-
tions to the second Born-Oppenheimer equation for each potential curve n (q+ ).
(q )
Depending on the symmetries of the solution An + (q ) discussed in B. Energies
section earlier, we can only choose one of these families, and at each side of the
energy crossing (18.278), we must discard one or the other and thus obtain a
discontinuous energy dependence.

Even though figure 18.76 is obtained with moderate values of and , we see
(q )
the similarity between the numerically determined An + (q ) and plane waves,
while the energy levels n (q+ ) are clearly not constant. This is due to the term
(18.276), the effect of which we will approximate using non-degenerate pertur-
bation theory. Due to the orthogonality between the cosine functions, there are
no first-order corrections. The second order corrections, on the other hand, give
contributions of the form
k (q+ ) = ak F (q+ )2
(18.279)

1472
where the amplitude ak can be calculated
1
8 k = 0

1
ak = 216 k = 1 (18.280)
1

otherwise

(4k2 1)

as shown below.

Calculation of perturbation terms

The second order perturbation terms for the potential curves k (q+ ) are found
by calculating the matrix elements of the terms (18.276) between pairs of un-
perturbed eigenfunctions which are plane waves:

F (q+ ) +2
Z  
q )
Il,k dq ei(kl)q /2 cos (18.281)
4 2 2

Using the orthogonality of the cosine functions we see that only coefficients with
neighboring values of l and k are coupled

F (q+ )
Il,k = [kl+1 + kl1 ] (18.282)
2
The resulting perturbative corrections then takes the form

X |Ilk |2

k (q+ ) = = ak F (q+ )2 (18.283)
k m
l6=k

where the amplitude of the oscillation is

X [kl+1 + kl1 ]2
ak = (18.284)
2(k 2 l2 ) + 16(k,0 l,0 )
l6=k

Here we can distinguish between the three cases k = 0, k = 1 and |k| 2,

where we get the results summarized in Eq. (18.280).

This gives the perturbative approximation to the potential curves

2
k (q+ ) +
k (q+ ) = k + 4k,0 + ak F (q+ )2
2
ak
= k 2 + 4k,0 + (1 + cos (q+ )) (18.285)
2 2
Due to the term 4 in the expression for 0 (q+ ), the energies of the eigenfunc-
tions in this potential change linearly with . For k 1 the position of n (q+ )
depends less strongly on and the eigenstates in these potentials have almost
constant energy.

1473
D. Approximate solution of the second Born-Oppenheimer equation

(n)
To analyze in more detail how the eigenenergies Em are distributed we must
take a closer look at the second Born-Oppenheimer equation which has the form
of a Schrodinger equation for a particle of mass ~2 /4k in the potential n (q+ ).
When the above perturbative treatment is valid, this potential is a cosine with
amplitude |ak |/2, so in order to estimate the eigenstates and energies, we must
compare and |ak |. In the limit where we can neglect the q+ -dependence
of
k (q+ ), the solutions can be well approximated by plane waves eimq+ / 2 with
box potential -energies
(k) 2
Em = k + 4k,0 + 2m2 (18.286)
2
which depend quadratically on m. In the opposite limit where k (q+ ) s a deep
potential in (18.257), we can approximate the cosine potential by a quadratic ex-
pansion around its minimum. The resulting equation is a Schrodinger equation
for a particle with mass ~2 /4k in a harmonic oscillator of frequency
1p
k = 2|ak | (18.287)
~
For the lower part of the energy spectrum the solutions are then well approxi-
mated by the usual harmonic oscillator eigenstate wave functions and the ener-
gies are equidistantly spaced with spacing ~k :
  
(k) ak  1 p
Em = k 2 + 4k,0 + + m+ 2|ak | (18.288)
2 2 2

Figure 18.79 illustrates the transition between the particle in a box and the
harmonic oscillator regimes by showing the exact and approximate energies
(0)
Em as functions of for fixed negative . Since the harmonic oscillator ap-
proximation is valid when the potential in (18.257) is deep, it requires that
 |a0 | = | 8|1 , so to capture the whole transition, the -axis is loga-
rithmic. The energies are plotted after subtracting the ground state energy E0
and scaling by the energy difference E1 E0 between the first excited state and
the ground state. For  1 the harmonic oscillator spectrum is then revealed
as levels with unit spacing. For 1, on the other hand, the curves become
constant at 1, 4, 9, ... showing the quadratic dependence on m. We note that
there is a perfect agreement between the exact and approximate energies shown
in the figure. In the transition from the harmonic oscillator regime to the par-
ticle in a box regime the energy levels group in pairs, which has the following
explanation: For a deep potential curve n there is a significant energy differ-
ence between the first excited even and odd states, but when the potential curve
is nearly constant, then even and odd solutions with a given wave number has
almost the same energy. This is exemplified in the eigenfunctions in the upper
panels of figure 18.76.

1474
Figure 18.79: Exact and approximate energies as a function of for = 10.
The black dashed curves show the exact energies En , while the solid red curves
(0)
show the approximate energies Em found from the lowest potential-curve in
the second Born-Oppenheimer equation. The green curve shows the position
of the maximum of the lowest potential curve 0 (q+ ) within the perturbative
approximation. The exact ground state energy E0 , which varies with , has
been subtracted from all energies, and afterward, the energies is scaled by the
energy difference between the two lowest exact energy levels E1 E0 .

No matter how deep the potential curve k (q+ ) is, the harmonic approximation
is not perfect, and above some energy the spectrum is ill-described by a harmonic
oscillator spectrum. A simple estimate suggests that the description is good for
eigenstates whose energies lie below the maximum of the potential curve, which
is approximated by the unperturbed energies plus a term depending on the sign
of ak
ak + |ak |
tk (, ) = k 2 + 4k,0 + (18.289)
2 2
In figure 18.79 this (solid green) curve is shown for k = 0 and agrees sys-

1475
tematically with the border where the harmonic oscillator energy spectrum is
significantly altered.

18.7.6 Conclusions
In the present section we have considered two identical bosons on an infinite,
discrete lattice with an additional harmonic confinement. In the tight bind-
ing approximation, the single particle physics in terms of quasi- momenta is
described by the same equation as a single particle in a continuous cosine po-
tential - namely the Mathieu equation. Adding a contact interaction yields a
Hamiltonian which does not separate in relative and center-of-mass coordinates,
even though the two-body interaction problem separates in both a homogeneous
discrete lattice Hamiltonian and in a continuous harmonic oscillator.

By formulating the problem in quasi-momentum representation we can make

an approximation which is mathematically equivalent to the Born-Oppenheimer
approximation performed in position space in molecular physics: We thus find
approximate solutions by first solving an equation for the relative quasi-momentum
wave function that depends parametrically on the center-of-mass quasi-momentum.
This yields potential curves for a Schrodinger equation for the center-of-mass
coordinate, which is readily solved. Contrary to the Born-Oppenheimer approx-
imation used to separate slow nuclear and fast electronic motion in molecules,
in our system we have a tunable adiabaticity parameter, namely the strength
of the inter-particle interaction.

In the solution of both the first and second Born-Oppenheimer equations we can
identify the excitation degrees of freedom in the system. This provides physi-
cally motivated quantum numbers valid also for the exact eigenstates together
with rules for which quantum numbers are allowed by symmetry considerations.

Finally, from the good agreement between the exact and approximate solutions
we conclude that the Born-Oppenheimer is well justified when the energy scales
for the relative and the center-of-mass motion of the two-particle quantum state
are well-separated. We imagine that a similar separation may be useful for ap-
proximate first principle calculations on many other cold atom systems, e.g.,
with more particles and possibly with mixtures of different species.

18.8 Problems
18.8.1 Piecewise Constant Potential Energy
One Atom per Primitive Cell
Consider a one-dimensional crystal whose potential energy is a piecewise con-
stant function of x. Assume that there is one atom per primitive unit cell - that
is, we are using the Kronig-Penney model as shown below.

1476
 Figure 18.80: Piecewise Constant Potential - 1 Atom per Primitive Cell

(a) Let s = d/2 (same spacing as in the text) and explain why no energy
gap occurs at the second Brillouin zone boundary in the weak-binding
limit, using physical argument based on sketches of the electron probability
density.

(b) For s = d/3, what are the magnitudes of the lowest six band gaps in the
weak binding limit?

18.8.2 Piecewise Constant Potential Energy

Two Atoms per Primitive Cell
Consider a one-dimensional crystal with two atoms per primitive unit cell as
shown below.

Figure 18.81: Piecewise Constant Potential - 2 Atoms per Primitive Cell

(a) Using the weak-binding approximation, determine the band gap for an
arbitrary Brillouin zone boundary.

(b) Use the results of part (a) to obtain an expression for the band gaps for
w  d and zone boundaries corresponding to small values of G. Are any
of these band gaps zero? Use physical arguments to explain why or why
not.

1477
 (c) Use the results of part(a) to determine the magnitude of the lowest eight
band gaps for w = d/4. Are any of these band gaps zero? Use physical
arguments to explain why or why not.
(d) In the weak-binding approximation, the energies for wave vectors k that
are far from the Brillouin zone boundaries are given by the free-electron
energies E = ~2 k 2 /2me . In relation to the zero of V (x) above, from
what value of the energy are the free-electron energies measured? Does
anything unusual happen when the energies exceed zero - the beginning
of the continuum for the isolated atoms? Determine how many band
gaps occur below E = 0. Answer these questions using the weak-binding
approximation.

18.8.3 Free-Electron Energy Bands for a Crystal with a

Primitive Rectangular Bravais Lattice
Consider a two-dimensional crystal with a primitive rectangular Bravais lattice.
Take the ratio of sides of the rectangular primitive cell to be 2:1, where the
larger side is along the yaxis.
(a) Working in the reduced zone scheme, sketch the free-electron energy for
the four lowest bands as a function of the distance in ~k space (starting at
~k = 0) along the path in the first Brillouin zone shown in the figure below.

Figure 18.82: Paths in the First Brillouin Zone

(b) Sketch free-electron constant energy contours in the reduced zone scheme
for the lowest four bands.
(c) Sketch the free-electron density of states, D(n) (E), for each of the four
lowest bands individually. Sketch D(E) for the total of the four lowest
bands.
(d) Sketch the free-electron Fermi surfaces in the reduced zone scheme for
= 1 to 6. Indicate the positions of the various Fermi energies on the
density-of-states graphs of part (c). Use quantitatively correct values for
kF and EF in this part.

1478
18.8.4 Weak-Binding Energy Bands for a Crystal with a
Hexagonal Bravais Lattice
Consider a two-dimensional crystal with an hexagonal rectangular Bravais lat-
tice oriented so that two nearest lattice points can lie along the yaxis but not
along the xaxis.

(a) Using the reduced zone scheme, sketch the energy versus distance in ~k
space (starting at ~k = 0) along the path in the first Brillouin zone shown
in the figure below. Do this for the six lowest bands in both the free-
electron and weak-binding approximations (assuming (incorrectly) that
all degeneracies are absent in the latter case).

Figure 18.83: Paths in the First Brillouin Zone

(b) Sketch constant energy contours in the reduced zone scheme for the lowest
six bands. Do so in both the free-electron and weak-binding approxima-
tions. Indicate the location in ~k space of all distinct maxima, minima and
saddlepoints (only one of a set that are equivalent by symmetry need be
shown).

(c) Sketch the Fermi surface in the reduced zone scheme for = 1 to 7.
Do so in both the free-electron and weak-binding approximations. Use
quantitatively correct values for kF in the free-electron sketches.

(d) Sketch the density of states, D(n) (E), for each of the five lowest bands
individually and D(E) for the total of the five lowest bands.. Do so in
both the free-electron and weak-binding approximations. Assume all de-
generacies are absent in the latter case and make reasonable assumptions
about the sense of the energy shifts from the free-electron values at the
singular points.

(e) For which integral value of would insulating properties be most likely to
first occur as the strength of the periodic potential energy is increased?
Why?

1479
18.8.5 A Weak-Binding Calculation #1
Consider a two-dimensional crystal with a primitive rectangular Bravais lattice
and two identical atoms per primitive unit cell. Take the structure to be as
shown below with a : b : c :: 4 : 2 : 1. Take the potential energy to be the sum of
the potential energies for the individual atoms located at the atom sites given
in the figure. Use the weak-binding approximation.

Figure 18.84: 2-Dimensional Crystal - Rectangular Bravais Lattice with 2 Atoms

per Primitive Unit Cell

(a) Find expressions for the matrix elements VG that describe the band gaps
in the weak-binding limit. Under what circumstances, if any, is VG = 0?

(b) Use the results of part (a) to draw qualitatively correct constant energy
contours in the reduced zone scheme for the lowest three bands.

(c) Sketch qualitatively correct individual band densities of states for the low-
est three bands.

18.8.6 Weak-Binding Calculations with Delta-Function Po-

tential Energies
Consider a two-dimensional crystal for which the potential energy consists of
delta functions, one for each atom. Use the weak-binding approximation .

(a) For the case of one atom per primitive cell, obtain a general expression for
the energy difference between adjacent bands at a Brillouin zone bound-
ary where they would be degenerate in the free-electron approximation
(ignoring the intersections of two or more boundaries). How does this
result depend on the Bravais lattice (assuming the area of a primitive cell
is the same for each different case)?

(b) For a crystal with a square lattice and one atom per primitive unit cell,
what are the energies of the lowest four bands at ~k = (/d)(1, 1)? Explain
your result in a physical and qualitative way.

1480
 (c) For a crystal with a centered rectangular Bravais lattice and two different
delta-function atoms per primitive unit cell as shown in the figure below,
evaluate the energy splittings between the bands for all zone boundaries in
the extended zone scheme for the five lowest bands (ignore all intersections
of two or more boundaries).

Figure 18.85: 2-Dimensional Crystal - Centered Rectangular Bravais Lattice

with 2 Atoms per Primitive Unit Cell

18.8.7 Is the spectrum of the harmonic crystal example

really thin?
Check that the thin spectrum of a harmonic crystal is indeed thin. That is,

(a) Show that only the lowest N total momentum states are not exponen-
tially suppressed in the symmetrybroken wavefunction (18.300). (This
result implies that only the lowest N total momentum states contribute
to the symmetry broken wavefunction, and these states all be-come de-
generate in the thermodynamic limit).
(b) Calculate the
partition function of the thin spectrum states and show that
it scales as N , so that the contribution of these states to the free energy
vanishes in the thermodynamic limit.

18.8.8 Are the limits really noncommutative in the har-

monic crystal example?
Show the noncommutativity of the limits in Eq. (18.304) explicitly, by going
through the following steps:
(a) Formulate the Hamiltonian of Eq. p (18.299) in terms of the boson (rais-

ing
p and lowering) operators b = C/(2~)(xtot (i/C)ptot ) and b =
C/(2~)(xtot + (i/C)ptot ) (where C is some constant).
(b) Choose C such that the Hamiltonian becomes diagonal and find its ground
state.

1481
 (c) Evaluate the limits of Eq. (18.304) by expressing x2tot in terms of boson
operators and taking the expectation value with respect to the ground
SB
state of Hcoll .

18.8.9 The Bogoliubov transformation in the harmonic

crystal example.
Work out the Bogoliubov transformation of Eqs. (18.294) and (18.295) explic-
itly.
(a) Write the Hamiltonian of Eq. (18.294) in terms of the trans- formed
bosons k = cosh (uk )bk + sinh (uk )bk .

(b) Which value should be chosen for uk in order for the Bogoliubov trans-
formation to yield the diagonal Hamiltonian of Eq. (18.295)? [Answer:
tanh (2uk ) = Bk /Ak ]

1482
Chapter 19

Second Quantization

19.1 Identical Particles Revisited

When dealing with a system that contains only a few identical particles, we
were easily able to explicitly construct appropriately symmetrized state vectors.
However, when the number of identical particles in the system gets very large
as with electrons in a metal, superfluids or superconductors and so on, these
methods become too cumbersome to use effectively.

To see how to proceed in these cases, we will step back, revisit the subject
of identical particles and look for an alternative way of thinking about such
systems.

19.1.1 Indistinguishability
As we saw earlier, in quantum mechanics, the state of a system with identical
particles can be described by a set of quantum numbers corresponding to the
eigenvalues of a commuting set of single-particle operators representing single-
particle observables.

When we specify a state vector we designate how many particles have certain
sets of quantum numbers, i.e.,

n1 particles have quantum number set K1

n2 particles have quantum number set K2
and so on

More complicated states are then superpositions of these states.

The important fact is that, for identical or indistinguishable particles, it is

impossible to state which particle has K1 , which particle has K2 , etc. As we

1483
saw in atoms, this indistinguishability has measurable effects on energy levels
arising from particle exchange symmetries.

19.1.2 State Vectors

When we define a state vector for a system of n identical particles we are work-
ing under the following assumption

any complete set of quantum numbers (observables) that can be

used to describe the behavior of a single particle can also be
used for n non-interacting particles of the same type

This assumption is postulated to be true even if the n particles are interacting.

The assumption implies that composite systems somehow retain the information
related to single-particle properties.

Mathematically it says that

for each set of quantum numbers Ki there exists an

occupation number operator Ni such that the
eigenvectors of Ni imply states in which a definite
number, ni , of particles has the quantum numbers Ki
and that the eigenvalues of the Ni are the occupation
numbers ni .

We make the fundamental postulate that the set of all Ni forms a complete set
of commuting Hermitian operators for any system of identical particles.

We now construct the state vector space appropriate for the many particle
system by generalizing one-particle quantum mechanics and building in indis-
tinguishability from the start.

Our postulates imply that the state vector space for the many particle system
(called Fock space) has the basis vectors

|n1 , n2 , n3 , ........i (19.1)

where the notation implies that

n1 particles have quantum number set K1

n2 particles have quantum number set K2
and so on

As we said, we assume this set is a complete orthonormal basis.

1484
In this vector space, we define

|0i = |0, 0, 0, 0, ............i = zero particle or vacuum state (19.2)

and one-particle states are of the form

|0, 0, 0, ......., 0, ni = 1, 0, 0, ........i (19.3)

These one-particle states span the one-particle subspace of the much larger state
space of the many particle system.

Most of the quantum mechanics we have developed so far applies to these one-
particle states.

We now ask whether the postulate implies enough formalism to construct a

quantum mechanics of a many-particle interacting system?

To answer this question yes, we must show that a consistent framework exists
that makes predictions in agreement with experiment.

The fundamental assumption that the states of an interacting system can be

written in terms of states for noninteracting single particles comes from our
experience with perturbation theory. It will fail to work if the noninteracting
and interacting states are two inequivalent representations of the Hilbert space
for the many particle system. Standard perturbation theory implicitly assumes
that the zero-order or unperturbed states and the exact states are equivalent
representations of the Hilbert space.

19.2 Occupation Number Space

19.2.1 Creation and Annihilation Operators
Earlier we studied the a , a+ operators in conjunction with the harmonic oscil-
lator problem. In that system we found that a lowered the energy of the system
by 1 quantum = ~, while a+ raised the energy of the system by ~.

In terms of the eigenvalues and eigenvectors of the number operator N = a+ a

where N |ni = n |ni we found that

a |ni = n |n 1i and a+ |ni = n + 1 |n + 1i (19.4)

In some sense, we can think of the a , a+ as annihilation/creation operators for

a quantum of the energy associated with a single particle system.

Another appearance of operators of this type with similar properties will occur
when we study the interaction of radiation with matter in Chapter 20. We will
be able to introduce photon annihilation/creation operators which remove/add a

1485
single photon with particular quantum numbers (~k, ~) corresponding to photon
momentum and polarization. The photon operators will have same mathemat-
ical structure(commutators) as the a , a+ operators of the harmonic oscillator
system.

In the photon case, as we will see, the states of the system will be the photon
number states given by
 E
N~k1~1 , N~k2~2 , N~k3~3 , .......... (19.5)


We will base our generalization on these examples. The generalization will al-
low us eventually to define the most general Fock space for any many particle
system.

Suppose that we have a potential well V (~r) with single particle energy eigen-
states given by

0 (~r) = h~r | 0i , 1 (~r) = h~r | 1i , 2 (~r) = h~r | 2i , ......... (19.6)

We start the discussion by considering a system of n bosons.

We assume that all n particles are in the lowest level (ground state) 0 (~r) of
the well. We label this state by the symbol |ni where n = 0, 1, 2, 3, ......, i.e., |0i
is the state with no particles in the lowest level.

We now introduce the operators a0 , a+

0 such that

a0 |ni = n |n 1i and a+ 0 |ni = n + 1 |n + 1i (19.7)

By definition, these operators relate states of the nboson system with all n
particles in 0 (~r) to those states of an n 1 particle system with all particles
in 0 (~r). In this sense, we say that

a0 = a particle annihilation operator

which removes a particle in the state 0 (~r) from the system
a+
0 = a particle creation operator
which adds a particle in the state 0 (~r) to the system

These operators, by construction, have the same algebra as the harmonic oscil-
lator operators, i.e.,

a0 , a+ [a0 , a0 ] = 0 = a+ +
   
0 =1 , 0 , a0 (19.8)

and

n
a+
|ni = 0 |0i , a0 |0i = 0 (no particles to be annihilated) (19.9)
n!

1486
This says that

the state with n particles in the lowest level, |ni, is generated

by adding n particles, in 0 (~r), to the vacuum state |0i

Acting to the left (instead of to the right) these operators reverse their roles,
i.e.,

hn| a0 = n + 1 hn + 1| adds particles in the state 0 (~r)

hn| a+
0 = n hn 1| removes particles in the state 0 (~r)
The operator
N0 = a+
0 a0 (19.10)
measures the number of particles in a state since
N0 |ni = a+
0 a0 |ni = n |ni (19.11)
This is a very appealing picture, but really all we have done is rewrite the
harmonic oscillator story using a lot of new words. We do not have any new
physics yet!

Before introducing the new physical ideas, we carry out this same discussion for
fermions.

In this case, the only allowed states are

|0i = no particles in 0 (~r)
|1i = 1 particle in 0 (~r)
since we cannot, according to the PEP, have two particles with the same quan-
tum numbers (in the same state).

We can still introduce annihilation/creation operators but they must have a

very different algebra. We must have
a0 |0i = 0 there are no particles to be annihilated
a0 |1i = |0i we can remove a particle when there is a particle
a+
0 |0i = |1i we can add a particle when there are no particles
+
a0 |1i = 0 we cannot have two particles in the same state
The last relation is necessary to satisfy the PEP.

We can figure out the operator algebra by using a special representation of these
states and operators. We let the two states (only allowed states) be a basis and
select the 2-dimensional representation
   
1 0
|0i = and |1i = (19.12)
0 1

1487
In this representation we have

h0| a0 |0i = 0 , h1| a0 |0i = 0 , h0| a0 |1i = 1 , h1| a0 |1i = 0

 
0 1
a0 =
0 0
h0| a+ + + +
0 |0i = 0 , h1| a0 |0i = 1 , h0| a0 |1i = 0 , h1| a0 |1i = 0
 
0 0
a+
0 = 1 0

Another way of representing these operators is via outer products,i.e.,

a0 = |0i h1| , a+
0 = |1i h0| (19.13)

The operator algebra is then given by

a0 a+
0 = |0i h0| = projection operator on the |0i state
a+
0 a0 = |1i h1| = projection operator on the |1i state
a0 a+
0 + a+
0 a0 = |0i h0| + |1i h1| = I

This last relation was derived, in general, earlier and is just the sum over all
projection operators.

Therefore we get
a0 , a+ = a0 a+ +

0 0 + a0 a0 = I (19.14)
Thus, the algebra involves anticommutators instead of commutators. That is
the only change we need to make!!

We also have
{a0 , a0 } = 0 = a+ +

0 , a0 (19.15)
These anticommutators imply that
2
(a0 ) = 0 we cannot remove two fermions from the
same state (maximum of one allowed)

2
a+
0 =0 we cannot put fermions into the
same state (maximum of one allowed)

Summary:

For one single particle level in a potential well, we can define annihilation/creation
operators such that:

bosons have

a0 , a+ , [a0 , a0 ] = 0 = a+ +
   
0 =I 0 , a0 (19.16)

1488
 and

fermions have

a0 , a+ = I , {a0 , a0 } = 0 = a+ +
 
0 0 , a0 (19.17)

We now expand our view and consider the case where particles can occupy two
levels of the potential well, say 0 (~r) and 1 (~r).

For bosons, we write the state of a many particle system as

|n0 , n1 i (19.18)

which implies

n0 particles in 0 (~r) and n1 particles in 1 (~r) (19.19)

We now define a pair of boson annihilation/creation operators by the relations

a0 |n0 , n1 i = n0 |n0 1, n1 i

a+
0 |n0 , n1 i = n0 + 1 |n0 + 1, n1 i

a1 |n0 , n1 i = n1 |n0 , n1 1i

a+
1 |n0 , n1 i = n1 + 1 |n0 , n1 + 1i

which imply that

a0 annihilates a particle in the state 0 (~r)

a+
0 creates a particle in the state 0 (~r)
a1 annihilates a particle in the state 1 (~r)
a+
1 creates a particle in the state 1 (~r)

In the same manner as in the one level case, we must have the commutator
algebra
a0 , a+ [a0 , a0 ] = 0 = a+ +
   
0 =1 , 0 , a0 (19.20)
and
a1 , a+ [a1 , a1 ] = 0 = a+ +
   
1 =1 , 1 , a1 (19.21)
For bosons, the order in which we create or annihilate particles in a state does
not matter, i.e.,
a0 a1 |n0 , n1 i = a1 a0 |n0 , n1 i (19.22)
which says
[a0 , a1 ] = 0 (19.23)
In a similar way all the other mixed commutators are also zero

a0 , a+ +
     + +
1 = 0 = a1 , a0 = a1 , a0 (19.24)

1489
All allowed states can be constructed from the vacuum state |0, 0i by using

n1 +
n0
a+ a
|n0 , n1 i = 1 0 |0, 0i (19.25)
n1 ! n0 !
Finally,

N0 = a+
0 a0 the number of particles in state 0 (~r)
N1 = a+
1 a1 the number of particles in state 1 (~r)

and
N = N0 + N1 the total particle number operator (19.26)
Therefore,

N0 |n0 , n1 i = n0 |n0 , n1 i
N1 |n0 , n1 i = n1 |n0 , n1 i
N |n0 , n1 i = (n0 + n1 ) |n0 , n1 i

Thus, for bosons we are able to just glue two single level many particle systems
together to create a two-level many particle system. We are really constructing
direct product states.

For fermions,however, there are some extra complications that we have to deal
with.

We start off by following a similar procedure. For the two-level system we have
only four possible fermion states, namely,

|0, 0i , |0, 1i , |1, 0i , |1, 1i (19.27)

We are completely free to define one set of creation/annihilation operators.

We define the a1 , a+
1 by

a1 |0, 0i = 0 , a1 |1, 0i = 0
a1 |0, 1i = |0, 0i , a1 |1, 1i = |1, 0i
a+
1 |0, 0i = |0, 1i , a+
1 |1, 0i = |1, 1i
a+
1 |0, 1i = 0 , a+
1 |1, 1i = 0

For a0 , a+
0 we can freely define the operation on state with no particles in level
1 (~r)

a0 |0, 0i = 0 , a0 |1, 0i = |0, 0i

a+
0 |0, 0i = |1, 0i , a+
0 |1, 0i = 0

1490
We must take care, however, in the fermion case when a particle exists in level
1 (~r), i.e., for the operations

a0 |0, 1i , a0 |1, 1i
a+
0 |0, 1i , a+
0 |1, 1i

The reason we must worry about these cases is connected with our earlier dis-
cussion of a totally antisymmetric state vector for fermions,i.e., if we interchange
any two identical fermions we must get a minus sign.

In this formalism, how do we interchange two fermions in the state |1, 1i? Using
only the defined relations we have.....
Step 1: |1, 1i |1, 0i = a1 |1, 1i remove a particle from state 1 (~r)
using a1
Step 2: |1, 0i |0, 1i = a+ 1 a0 |1, 0i transfer the particle from state
0 (~r) to 1 (~r) by applying a+
1 a0 (a0 followed by a1 )

Step 3: put the leftover particle back into 0 (~r) using a+

0 which gives the
state relationship
a+ + +
0 a1 a0 a1 |1, 1i = a0 |0, 1i (19.28)

However, all we have done is switch the particles in the original state |1, 1i,
which means that a minus sign must appear or

a+
0 |0, 1i = |1, 1i (19.29)

In this way, the state is completely antisymmetric under particle exchange.

In a similar manner, the other relations that complete the definition of the
annihilation and creation operators are

a0 |1, 1i = |0, 1i , a0 |0, 1i = 0 = a+

0 |1, 1i (19.30)

+
Since a0 = a+
0 we have

a+
0 |0, 1i = |1, 1i , a0 a+
0 |0, 1i = a0 |1, 1i = |0, 1i (19.31)

so that a0 completely undoes the operation of a+

0.

These definitions, which are now consistent with complete antisymmetry corre-
spond to the anticommutation relations

a0 , a+
  + +
0 =1 {a0 , a0 } = 0 a0 , a0 = 0
+
  + +
a1 , a1 = 1 {a1 , a1 } = 0 a1 , a1 = 0
+
  +
{a0 , a1 } = 0 a0 , a1 = 0 a0 , a1 = 0
 + +
a0 , a1 = 0

1491
We get anticommutators instead of commutators because of the complete anti-
symmetry under particle interchange(see argument below).

The rule for constructing the allowed states is

n1 +
n0
|n0 , n1 i = a+
1 a0 |0, 0i (19.32)

Note the order: a+

0 acts first and we have no factorial factors since 0! = 1! = 1.

The connection between the minus sign and the anticommutators is now clear,
i.e.,

a0 |1, 1i = a0 a+ + + +
1 a0 |0, 0i = a1 a0 a0 |0, 0i
= a+ + +
1 [1 a0 a0 ] |0, 0i = a1 |0, 0i
= |0, 1i

So we could have assumed the anticommutators and derived the state operations
instead of going the other way.

The generalization to the many particle system where particles can occupy all
of the levels is now straightforward.

In generalizing, we will not only let the particles occupy all of the levels, but
also have all spin orientations.

Once again the states will be labeled

|n0 , n1 , n2 , ........i (19.33)

where now

ni = the number of particles in i (r) with a

given spin orientation (if the particles have spin)

We define creation and annihilation operators a+

i and ai for each different single
particle state.

For boson operators we have commutation relations

ai , a+ , [ai , aj ] = 0 = a+ +
   
j = ij i , aj (19.34)

and
n2 +
n1 +
n0
a+
2 a a
|n0 , n1 , n2 , ....i = .......... 1 0 |0i (19.35)
n2 ! n1 ! n0 !
where
|0i = |0, 0, 0, 0, .........i = the vacuum state (19.36)

1492
To within numerical factors, as we will see, these are the same relations we will
find in Chapter 20 for photons. Therefore, photons must be bosons !!

For fermion operators we have anticommutation relations

ai , a+
  + +
j = ij , {ai , aj } = 0 , ai , aj = 0 (19.37)

and

n2
n1
n0
|n0 , n1 , n2 , ....i = .......... a+
2 a+
1 a+
0 |0i , ni = 0, 1 only (19.38)

In both cases X X
N = Ni = a+
i ai (19.39)
i i
and h i
Ni , Nj = 0 (19.40)

19.2.2 An Example
We now consider the complete set of plane wave states in a box using periodic
boundary conditions. We have
p~2op
H = (19.41)
2m
Since h i
H, p~op = 0 (19.42)
we have a common eigenbasis that we will label by |~
pi. We then have
p~2
p~op |~
pi = p~ |~
pi , H |~
pi = |~
pi = E |~
pi (19.43)
2m
so that
p~2
E= (19.44)
2m
The corresponding wave functions are
~
eik~r
p~ (~r) = h~r | p~i = (19.45)
V

where p~ = ~~k and V = volume of the box. The factor 1/ V normalizes the
wavefunction in the box.

Periodic boundary conditions are imposed by the relations

p~ (0, y, z) = p~ (Lx , y, z)
p~ (x, 0, z) = p~ (x, Ly , z)
p~ (x, y, 0) = p~ (x, y, Lz )

1493
which imply that

2nx
eikx Lx = 1 kx = , nx = 0, 1, 2, ....
Lx
2ny
eiky Ly = 1 ky = , ny = 0, 1, 2, ....
Ly
2nz
eikz Lz = 1 kz = , nz = 0, 1, 2, ....
Lz

We now define

a+
~s = the operator that creates (adds) a particle of
p
momentum p~ and spin orientation s in(to) the box
ap~s = the operator that annihilates (removes) a particle of
momentum p~ and spin orientation s in(from) the box

The probability amplitude for finding the particle added by a+

~s to the box at
p
position ~r 0 is
~ 0
eik~r
= h~r 0 | p~i (19.46)
V
The operator
X ei~k~r
s+ (~r) = a+ (19.47)
p
~
V p~s

adds a particle to the system in a superposition of momentum states (a+

~s ) each
p
with amplitude
~
eik~r
(19.48)
V
This implies that the probability amplitude for finding the particle added to the
box by s+ (~r) at the position ~r 0 is

X ei~k~r X ei~k~r
h~r 0 | a+ p
~s |0i = h~r 0 | p~i
p
~
V p
~
V
X ei~k~r ei~k~r 0
= = (~r ~r 0 ) (19.49)
p
~
V V

This says that the operator s+ (~r) adds all the amplitude at the position ~r or
we say

s+ (~r) adds(creates) a particle at position ~r with spin orientation s

1494
In a similar manner, the operator

+ X ei~k~r
s (~r) = s+ (~r) ap~s (19.50)
p
~
V

removes (annihilates) a particle at point ~r.

The s+ (~r) and s (~r) are called field operators.

In this new formalism, position and momentum are once again just numbers,
but the wave functions are now operators. Hence the name second quantization.

For bosons we have

[s (~r), s0 (~r0 )] = 0 = s+ (~r), s+0 (~r0 )
 
(19.51)
and for fermions we have
{s (~r), s0 (~r0 )} = 0 = s+ (~r), s+0 (~r0 )

(19.52)
These relations imply that

for bosons adding(removing) a particle at ~r

is an operation that commutes with
adding(removing) a particle at ~r 0

and

for fermions adding(removing) a particle at ~r

is an operation that anticommutes with
adding(removing) a particle at ~r 0 (change of sign)

Finally, for bosons we get

0
+ 0
 X ei~k~r ei~k~r h i
ap~s , ap+

s (~r), s0 (~r ) = 0
~ s 0
0
V
p
~p~

X ei~k~r ei~k~r 0
= p~p~ 0 ss0 = (~r ~r 0 )ss0 (19.53)
V
p
~p~0

and similarly for fermions we get

0
X ei~k~r ei~k~r n o
s (~r), s+0 (~r 0 ) = ap~s , a+

p 0
~ s 0
0
V
p
~p~

X ei~k~r ei~k~r 0
= p~p~ 0 ss0 = (~r ~r 0 )ss0 (19.54)
V
p
~p~0

1495
These relations imply that creating particles commutes(bosons) or anticom-
mutes(fermions) with annihilating particles unless the two, operations occur at
the same point in space.

In this case, if there are no particles at ~r, then

s+ (~r)s (~r) 0 (19.55)

i.e., we cannot annihilate a particle at a point if none exists there.

On the other hand

s (~r)s+ (~r) does not 0 (19.56)
since s+ (~r) adds a particle that s (~r) then removes!

Suppressing spin indices for simplicity, the state vector

1
|~r1 , ~r2 , ......., ~rn i = + (~rn )....... + (~r2 ) + (~r1 ) |0i (19.57)
n!

represents the state with one particle at ~r1 , one particle at ~r2 , and so on.

We will use these states as a basis for the many particle, many level system.
The states have the properties:

1. for bosons
|~r2 , ~r1 , ......., ~rn i = |~r1 , ~r2 , ......., ~rn i (19.58)
due to the commutation relations which imply

+ (~r1 ) + (~r2 ) = + (~r2 ) + (~r1 ) (19.59)

2. for fermions
|~r2 , ~r1 , ......., ~rn i = |~r1 , ~r2 , ......., ~rn i (19.60)
due to the anticommutation relations which imply

+ (~r1 ) + (~r2 ) = + (~r2 ) + (~r1 ) (19.61)

Since

n+1 +
+ (~r) |~r1 , ~r2 , ......., ~rn i = (~r) + (~rn )....... + (~r2 ) + (~r1 ) |0i
n + 1!

= n + 1 |~r1 , ~r2 , ......., ~rn , ~ri (19.62)

the commutation/anticommutation properties of the + imply that the new

state is automatically correctly symmetric or antisymmetric. This is one of the

1496
great advantages of the annihilation/creation operator formalism. We can show
this important property this way:
1
(~r) |~r1 , ~r2 , ......., ~rn i = (~r) + (~rn )....... + (~r2 ) + (~r1 ) |0i (19.63)
n!
1 
(~r ~rn ) + (~rn )(~r) + (~rn1 )....... + (~r2 ) + (~r1 ) |0i

=
n!
where  
bosons
(19.64)
f ermions
We now continue commuting (~r) with the + s to the right until we have

............(~r) |0i = 0 (19.65)

since (~r) |0i = 0. If we actually carry out this process we obtain

1
(~r) |~r1 , ~r2 , ......., ~rn i = [(~r ~rn ) |~r1 , ~r2 , ......., ~rn1 i
n!
(~r ~rn1 ) |~r1 , ~r2 , ......., ~rn2 , ~rn i
n1
+ ........... + (1) |~r2 , ......., ~rn i] (19.66)

This says that

removing a particle at ~r is only

possible if ~r = ~rn or ~r = ~rn1 or ............. or ~r = ~r1

and if one of these conditions is true, then

what remains is the correctly symmetrized

combination of (n 1) particle states
+
Because + = () , i.e., they are Hermitian conjugate operators, we have

(~r) |~r1 , ~r2 , ......., ~rn i removes a particle
when acting to the right
+ (~r) |~r1 , ~r2 , ......., ~rn i adds a particle
but

h~r1 , ~r2 , ......., ~rn | (~r) adds a particle
when acting to the left (19.67)
h~r1 , ~r2 , ......., ~rn | + (~r) removes a particle
i.e.,
 +
1
h~r1 , ~r2 , ......., ~rn | = + (~rn )....... + (~r2 ) + (~r1 ) |0i
n!
1
= h0| (~r1 )(~r2 ).......(~rn ) (19.68)
n!

1497
Note the reversal in the order of the operators.

According to these rules the basis states are normalized as follows:

h~r10 , ~r20 , ......., ~rn0 0 | ~r1 , ~r2 , ......., ~rn i

nn0 X P
= (1) P [(~r1 ~r10 )(~r2 ~r20 )......(~rn ~rn0 )] (19.69)
n!
P

where
X
= sum over all permutations if the coordinates (19.70)
P

and
+1 bosons

P
() = +1 fermions - even permutation (19.71)

1 fermions - odd permutation

What is the state |i where the particles have a wave function (~r1 , ....., ~rn )?

Since the correctly symmetrized wave function must be

h~r1 , ~r2 , ......., ~rn | i (19.72)

we must have
Z
|i = d3~r1 0 d3~r2 0 ....d3~rn 0 (~r1 0 , ....., ~rn 0 ) |~r1 0 , ~r2 0 , ......., ~rn 0 i (19.73)

which implies that

1 X P
h~r1 , ~r2 , ......., ~rn | i = (1) P (~r1 , ....., ~rn ) (19.74)
n!
P

and thus the true wave function is properly symmetrized.

This result is true even if (~r1 , ....., ~rn ) is not already properly symmetrized.
When it already properly symmetrized, then all n! terms are identical and

h~r1 , ~r2 , ......., ~rn | i = (~r1 , ....., ~rn ) (19.75)

We must have
h | i = 1 (19.76)
if (~r1 , ....., ~rn ) is symmetrized and
Z
1 = d3~r1 d3~r2 ....d3~rn (~r1 , ....., ~rn )(~r1 , ....., ~rn ) (19.77)

1498
i.e.,
Z
h | i = d3~r1 d3~r2 ....d3~rn (~r1 , ....., ~rn )
Z
h~r1 , ~r2 , ......., ~rn | d3~r1 0 d3~r2 0 ....d3~rn 0 (~r1 0 , ....., ~rn 0 ) |~r1 0 , ~r20 , ......., ~rn 0 i
Z
= d3~r1 d3~r2 ....d3~rn (~r1 , ....., ~rn )(~r1 0 , ....., ~rn 0 )
1 X P
(1) P [(~r1 ~r1 0 )(~r2 ~r2 0 )......(~rn ~rn 0 )]
n!
P
Z
= d3~r1 d3~r2 ....d3~rn (~r1 , ....., ~rn )(~r1 , ....., ~rn ) = 1

Now h~r1 , ~r2 , ......., ~rn | i is the amplitude for observing particles at ~r1 , ~r2 , ......., ~rn .
It implies that
Z
|i = d3~r1 d3~r2 ....d3~rn |~r1 , ~r2 , ......., ~rn i h~r1 , ~r2 , ......., ~rn | i (19.78)

which says that

Z
d3~r1 d3~r2 ....d3~rn |~r1 , ~r2 , ......., ~rn i h~r1 , ~r2 , ......., ~rn | = In (19.79)

as it should for a complete set within the nparticle subspace, i.e., it is In only
when operating on properly symmetrized nparticle states.

We then have for |i and nparticle states

In0 |i = nn0 |i (19.80)
which implies that

X
X
I = In = |0i h0| + In (19.81)
n=0 n=1
is the identity operator when acting on properly symmetrized states for any
number of particles.

19.3 Second Quantized Operators

How do we write operators in this formalism?

As a first example, let us consider the operator

(~r) = + (~r)(~r) (19.82)
To see what this represents physically we calculate the matrix element
h0 | (~r) |i (19.83)

1499
where |i and |0 i are nparticle states. We obtain
r) |i
h0 | (~r) |i = h0 | + (~r)(~r) |i = h0 | + (~r)I(~

!
X
= h0 | + (~r) |0i h0| + In0 (~r) |i (19.84)
n0 =1

Now the state

(~r) |i an (n 1) particlestate (19.85)
and therefore,
|0i h0| (~r) |i = 0 (19.86)
and

X
In0 (~r) |i = 0 unless n0 = n 1 (19.87)
n0 =1

We thus obtain

h0 | (~r) |i = h0 | + (~r)In1 (~r) |i

Z
= d3~r1 ........d3~rn1 h0 | + (~r) |~r1 , ........, ~rn1 i h~r1 , ........, ~rn1 | (~r) |i
Z
= d3~r1 ........d3~rn1 h0 | ~r1 , ........, ~rn1 , ~ri h~r1 , ........, ~rn1 , ~r | i
(19.88)

Now, since the h~r1 , ........, ~rn | i are completely symmetrized (or antisymmetrized)
this can be written as
Z n
X
h | (~r) |i = d3~r1 ........d3~rn h0 | ~r1 , ........, ~rn i
0
(~r ~ri ) h~r1 , ........, ~rn | i
i=1
n
X
= h0 | (~r ~ri ) |i (19.89)
i=1

Since these two objects have all their matrix elements are identical, they must
be equal. Therefore,
X n
(~r) = (~r ~ri ) (19.90)
i=1
or
(~r) = a representation of the density operator in this formalism
The way to think about this operator is as follows:

+ (~r)(~r) tests the density of particles at ~r

by attempting to remove a particle located
at ~r and then to putting it back

1500
If the particles have spin, the density operator for particles at ~r with spin s is

s+ (~r)s (~r) (19.91)

and
X
(~r) = s+ (~r)s (~r) = the total density operator (19.92)
s
Z XZ
3
N = d ~r(~r) = d3~rs+ (~r)s (~r) = total number operator (19.93)
s

This agrees with our earlier result as can be seen below:

XZ XZ X ei~k~r X ei~k0 ~r
N = d3~rs+ (~r)s (~r) = d3~r a+ ap~0 s
s s p
~
V p~s p
~
V
Z k0 ~
i(~ k)~
r
3 e
XX XX X
= a+ a
~s p
p
0
~s d ~
r = a+
~s ap
p ~0 s p
~p~0 = a+
~s ap
p ~s
s 0
V s 0
p
~p~ p
~p~ p
~s

Now to figure out other operators

Any operator that is given by the relation f (~ pop ) is easily written down in
this formalism, i.e., such an operator is given by the number of times p~ occurs
(Np~ = ap+
~s ap
~s ) times the value of the operator (f (~
p)) summed over (~ p s) or
X
f (~
pop ) = p)a+
f (~ ~s ap
p ~s (19.94)
p
~s

Therefore, the kinetic energy operator T is given by

X p~2
T = a+ ap~s (19.95)
2m p~s
p
~s

We can rewrite this expression in a form involving the field operators that will
then lead to a prescription for writing any operator in this formalism.

If we invert the equations for s+ (~r) and s (~r) we get

~ ~
eik~r eik~r
Z Z
a+
~s =
p d3~r s+ (~r) , ap~s = d3~r s (~r) (19.96)
V V
Physically, the first of this pair of equations implies that

to create a particle with momentum p~ we create particles

~
at different points ~r with relative amplitudes eik~r / V

and so on for the second equation of the pair, which is exactly what we have

1501
been assuming all along!

Using these equations we get

Z
1 1 X 
~
 
~ 0

T = d3~rd3~r,0 p~eik~r p~eik~r, s+ (~r)s (~r,0 )
2m V
p
~s

~2 1 X
Z    
~ ~ 0
= d3~rd3~r0 ~r eik~r ~r,0 eik~r, s+ (~r)s (~r,0 )
2m V
p
~s

We now integrate by parts, assuming that s+ (~r) 0 and s (~r) 0 as r

so that surface terms 0. We get

~2 1 X
Z
~ ~ 0
T = d3~rd3~r,0 eik~r eik~r, (~r s+ (~r)) (~r,0 s (~r,0 )) (19.97)
2m V
p
~s

Using
1 X i~k~r i~k~r 0
e e = (~r ~r 0 ) (19.98)
V
p
~s

and doing the ~r 0 integration we get

~2
Z
T = d3~r + (~r) (~r) (19.99)
2m

This is very similar to the expectation value of the kinetic energy operator for
a single particle, i.e.,

~2
D E Z
T = d3~r (~r) (~r) (19.100)
2m

Similarly, the density operator resembles the probability density (~r)(~r) for
finding a single particle with wave function at the point ~r.

This leads to the general concept of second quantization where

one-particle wavefunctions appear to have become

operators that create/annihilate particles

single-particle expectation values appear to have

become operators for physical observables

position and momentum appear to have become

ordinary numbers

We can now exploit this similarity to write down other operators, for example,

1502
the current density operator is given by
1  +
(~r)((~r)) ( + (~r))(~r)

j(~r) = (19.101)
2im
the operator for the density of spin at point ~r is given by

~ r) = 1
X
S(~ + (~r)~ss0 s0 (~r) (19.102)
2 0 s
ss

where
~ = (x , y , z ) (19.103)

19.3.1 Example - Gas of N Non-Interacting Spin 1/2 Fermions

We assume that the system is in its ground state.

As we discussed earlier, the ground state |0 i corresponds to all momentum

states (energy levels) being filled starting from the lowest level. Since we have
a finite number of particles, there will be a maximum momentum value, pF =
the Fermi momentum.

In the new formalism, we describe the ground state via a set of occupation
numbers
(
+ 1 |~
p| p F
np~ = np~ = h0 | ap~ ap~ |0 i = (19.104)
0 |~
p| p F

Now X X
N= np~s = 2 1 (19.105)
p
~s |~
p|pF

Converting the sum to an integral we have

ZpF
d3 p~ p3F
N = 2V = V (19.106)
(2~)3 3 2 ~3
0

where we have used the properties

2~
spacing between px values is
Lx
2~
spacing between py values is
Ly
2~
spacing between pz values is
Lz
which imply Z
X Lx
dpx (19.107)
px
2~

1503
and, thus, Z
X V
d3 p~ (19.108)
(2~)3
p
~

We then get

1/3
pF = ~ 3 2 n (19.109)
where
N
n=
= average particle density (19.110)
V
The expectation value of the density operator is
X
h(~r)i = h0 |s+ (~r)s (~r) |0 i
s
X ei~k~r ei~k 0 ~r
= h0 |ap+ r)ap~ 0 s (~r) |0 i
~s (~ (19.111)
0
V
s~
pp~

Now
h0 | a+ r)ap~ 0 s (~r) |0 i = p~p~ 0 np~s
~s (~
p (19.112)
since if we remove a particle of momentum p~ from the ground state, we can only
get the ground state back if we add a particle of the same momentum p~. Where
we have used
np~s = a+
~s (~
p r)ap~s (~r) (19.113)
Therefore,
1 X
h(~r)i = np~s = n = constant (19.114)
V
p
~s

Thus, the density of the non-interacting fermion gas is uniform.

A useful physical quantity is defined by

Gs (~r ~r 0 ) = h0 | s+ (~r)s (~r 0 ) |0 i (19.115)
which is

the amplitude for removing(annihilating) a particle

with spin s at ~r 0 from the ground state and then
returning to the ground state by replacing(creating)
a particle with spin s at ~r

We have
X ei~k~r ei~k 0 ~r
Gs (~r ~r 0 ) = h0 |ap+ r)ap~ 0 s (~r) |0 i
~s (~
V
p
~p~0
~ ~0
1 X eik~r eik ~r 1 X i~k(~r~r 0 )
= p~p~ 0 np~s = e np~s (19.116)
V 0
V V
p
~p~ p
~

1504
Changing to an integral we obtain
ZpF Z Z ZpF Z1
d3 p~ i~k(~r~r 0 )
 
0 1 
r 0 
ik r ~

Gs (~r ~r ) = e = p2 dp de
(2~)3 4 2 ~3
0 0 1
ZpF      
1 
r 0 
ik r ~
 
r 0 
ik r ~

= p2 dp e e
4 2 ~3
0
3n sin x x cos x
= (19.117)
2 x3
where
pF 1 p3F
x=|~r ~r 0 | , n = (19.118)
~ 3 2 ~3
As a function of x this looks as shown in Figure 19.1 below (for n = 1000)

Figure 19.1: G(x) - Single Particle Correlation Function

The single-particle correlation function oscillates with a characteristic period

1/kF under an envelope which falls to zero. We have for ~r = ~r 0 , Gs (0) = n/2 =
density of particles with spin orientation s. For small |~r ~r 0 | we have
3n sin x x cos x
Gs (~r ~r 0 ) =
2  x3   
x3 x5 2
x4

3n x 6 + 120 x 1 x2 + 24

2 x3

3n 13 x3 301 5
x2
  
x n
= 1
2 x3 2 10
" #
2
1 pF |~r ~r 0 |
 
= 1 (19.119)
10 ~

1505
Gs (~r ~r0 ) is called the one-particle density matrix.

19.4 Pair Correlation Function

As we have seen earlier, in a system of fermions, there is a tendency, due to the
PEP, for particles with the same spin to avoid each other. This says that the
amplitude for being close together must be small.

How do we calculate the relative probability of finding a particle at ~r 0 if we

know there is a particle at ~r?

One way is as follows:

1. Remove (mathematically) a particle with spin s at ~r from the system. We
are then left in the N 1 particle state
|0 (~r, s)i = s (~r) |0 i (19.120)

2. Calculate the density distribution of the particles with spin s0 in the new
state. The density is
h0 (~r, s)| s+0 (~r0 )s0 (~r0 ) |0 (~r, s)i
 n 2
= h0 | s+ (~r)s+0 (~r0 )s0 (~r0 )s (~r) |0 i = gss0 (~r ~r0 ) (19.121)
2
where gss0 (~r ~r 0 ) the pair correlation function.

An equivalent way of asking the same question is the following:

1. Remove(annihilate) a particle from ~r using s (~r)
2. Remove(annihilate) a particle from ~r 0 using s0 (~r) 0 The relative amplitude
for ending up in some N 2 particle state |00i i is
h00i | s0 (~r 0 )s (~r) |0 i (19.122)

3. Sum over a complete set of N 2 particle states

This gives the total probability of removing two particles and ending up in any
N 2 particle state. We get
2
X
|h00i | s0 (~r 0 )s (~r) |0 i|
i
X
= h0 | s+ (~r)s+0 (~r 0 ) |00i i h00i | s0 (~r 0 )s (~r) |0 i
i
 n 2
= h0 | s+ (~r)s+0 (~r 0 )s0 (~r 0 )s (~r) |0 i = gss0 (~r ~r 0 )
2
= total probability of removing two particles (19.123)

1506
We can evaluate gss0 (~r ~r 0 ) by shifting to the creation/annihilation operator
formalism. We get
 n 2 X ei(~p~p 0 )~r/~ ei(~q~q 0 )~r0 /~
0
g ss0 (~r ~r ) = 2
h0 |a+ +
~s aq
p ~s0 aq ~ 0 s |0 i
~ 0 s0 ap
2 0 0
V
p
~p~ q
~q~
(19.124)
Now
h0 | ap+ r)aq+
~s (~ r 0 )aq~ 0 s0 (~r 0 )ap~ 0 s (~r) |0 i = 0
~s0 (~ (19.125)
unless we put back particles with the same spin and same momentum that we
remove, i.e.,
if s 6= s0 , then p~ 0 = p~ and ~q 0 = ~q (19.126)
This implies that

h0 | a+
p
+
~s aq
~s0 aq
~ 0 s0 ap
+
~ 0 s |0 i = h0 | ap
~s ap
+
~s aq ~s0 |0 i = np
~s0 aq ~s nq
~s0 (19.127)

We then get
 n 2 1 X  n 2
gss0 (~r ~r 0 ) = 2
np~s nq~s0 = ns ns0 = (19.128)
2 V 2
p
~q~

or
gss0 (~r ~r 0 ) = 1 for s 6= s0 (19.129)
This implies that the relative probability for finding particles at ~r and ~r 0 for
different spins is independent of |~r ~r 0 |. This is the same result as one obtains
in a classical non-interacting gas.

The PEP does not influence particles of different(opposite in this case) spins.

On the other hand, if the spins are the same, s = s0 , then we have two possibil-
ities, namely,

p~0 = p~ and ~q0 = ~q or p~0 = ~q and ~q0 = p~

Note that if p~ 0 = ~q 0 , then h0 | a+ +

~s aq
p ~s aq ~ 0 s |0 i = 0 since
~ 0 s ap

2
aq~ 0 s ap~ 0 s (ap~ 0 s ) = 0 (for fermions) (19.130)

Therefore, we have

h0 | a+
p
+
~s aq
~s aq ~ 0 s |0 i = p
~ 0 s ap ~p~ 0 q
~q
+ +
~ 0 h0 | ap
~s aq
~s aq ~s |0 i
~s ap

+ p~q~ 0 q~p~ 0 h0 | a+
p
+
~s aq
~s ap ~s |0 i
~s aq (19.131)

We must use the superposition of both (indistinguishable)possibilities.

1507
This becomes

h0 | a+
p
+
~s aq
~s aq ~ 0 s |0 i = (p
~ 0 s ap ~p~ 0 q
~q~ 0 p
~q~ 0 q
~p
+
~ 0 ) h0 | ap
~s ap
+
~s aq ~s |0 i
~s aq
= (p~p~ 0 q~q~ 0 p~q~ 0 q~p~ 0 )np~s nq~s (19.132)

where we have used the anticommutation relations for fermion operators

n o
~q 6= p~ ap~s , aq+
~s = 0 = {ap ~s }
~s , aq

~q = p~ h......i = 0

We finally obtain
 n 2 1 Xh i
q 0 )(~ r 0 )/~
gss0 (~r ~r 0 ) = 1 e i(~
p~ r ~
np~s nq~s
2 V2
p
~q
 n 2
2
= [Gs (~r ~r 0 )] (19.133)
2
where Gs (~r ~r 0 ) is the single particle density function. This then becomes
9 2
gss0 (~r ~r 0 ) = 1 (sin x x cos x) (19.134)
x6
where
pF
|~r ~r 0 |
x= (19.135)
~
As a function of x this looks as shown in Figure 19.2 below.
This result implies a substantial reduction in the probability for finding two
fermions of the same spin at distances less than ~/pF .

The PEP causes large correlations in the motion of the particles with the same
spin. It seems like fermions of the same spin repel each other at short distances.
This effective repulsion is due to the exchange symmetry (PEP) of the wave
function and not from any real additional potentials.

1508
 Figure 19.2: g(x) - Two Particle Correlation Function - Fermions

At large |~r ~r 0 |, gss gss0 = 1 as we might have guessed since at large

separations the PEP should have no effect and spin effects should drop out.

What happens if we consider a system of non-interacting bosons instead?

Suppose the system is in the state

|i = |np~0 , np~1 , .........i (19.136)

The density in this state is

1 X
h| + (~r)(~r) |i = np~ = n (19.137)
V
p
~

The calculation of the pair correlation function is the same as for fermions up
to this point.
X ei(~p~p 0 )~r/~ ei(~q~q 0 )~r0 /~
0
2
n g(~r ~r ) = h0 |a+ +
~ aq
p ~ aq ~ 0 |0 i (19.138)
~ 0 ap
0 0
V2
p
~p~ q
~q~

In this case,
h0 | ap+ +
~ aq
~ aq ~0 |0 i =
~0 ap 6 0 (19.139)

1509
only if
p~ = p~ 0 , ~q = ~q 0 or p~ = ~q 0 , ~q = p~ 0
These two cases are not distinct if p~ = ~q.

Therefore, in the same manner as before, we have

h0 | a+ +
~ aq
p ~ aq ~ 0 |0 i
~ 0 ap
h i
= (1 p~q~ ) p~p~ 0 q~q~ 0 h0 | ap+ +
~ aq
~ aq ~ |0 i + p
~ ap ~q~ 0 q
~p
+ +
~ 0 h0 | ap
~ aq
~ ap ~ |0 i
~ aq

+ p~q~ p~p~ 0 q~q~ 0 h0 | a+ +

~ ap
p ~ ap ~ |0 i
~ ap

= (1 p~q~ ) (p~p~ 0 q~q~ 0 + p~q~ 0 q~p~ 0 ) np~ nq~ + p~q~ p~p~ 0 q~q~ 0 np~ (np~ 1) (19.140)
h i
where we have used ap~ , a+ ~ = 1 in the last term. Therefore, we obtain
p

n2 g(~r ~r 0 ) = h| + (~r) + (~r 0 )(~r 0 )(~r) |i

 2
 
2
 1 X ~
ik(~r ~ 0 
r ) 1 X
=n +  np~ e 2 np~ (np~ + 1) (19.141)
 V p~ V

 p
~

This differs from the fermion result in

1. the sign of the second term (+ instead of ), which is due to different
exchange symmetry properties.
2. the existence of the third term, which is due to the fact that we can have
many bosons in the same state.
If all the particles are in only one state, p~0 , then we have

n2 g(~r ~r 0 ) = h| + (~r) + (~r 0 )(~r 0 )(~r) |i

1 N (N 1)
= n2 + n2 N (N + 1) = (19.142)
V2 V2
The pair distribution function is position independent. The above result implies
that the relative amplitude for removing the first particle is N/V , while the
amplitude for removing a second particle is (N 1)/V since there are only
N 1 remaining after removing the first particle.

Now suppose that

p~
(~ p0 )2
np~ = smoothly varying distribution = Ce 2 (19.143)

This could be a beam of particles of momentum centered at p~0 with a Gaussian

spread about p~0 .

If we let V large, with N/V = n fixed, then the third term is of order 1/V

1510
smaller than the first two terms and we can neglect it.

Converting the remaining terms to integrals we have

n2 g(~r ~r0 ) = h| + (~r) + (~r0 )(~r0 )(~r) |i

2
d3 p~
Z  
(rr0 )2
2 i~
k(~ r0 ) 
r ~ 2
 
=n +  np~ e  =n 1+e (19.144)
(2~)3

A plot of this result is shown in Figure 19.3 below.

Figure 19.3: g(r) - Two Particle Correlation Function - Bosons

0 2
The e(~r~r ) / term is due to exchange symmetry. In this case, exchange
symmetry increases the probability that two bosons will be found at small sep-
arations.

In fact,

the probability of finding two bosons on top of each other is equal

to two times the probability of finding two bosons with a large sep-
aration

19.5 Hanbury-Brown and Twiss Experiment

Do bosons really tend to clump together?

At this point we will discuss the Hanbury-Brown and Twiss(HBT) experiment

that shows that they do tend to clump. The HBT experiment measures the
probability of observing 2 photons simultaneously at different points in a beam
of incoherent light. Incoherent light, as we saw earlier, can be described in terms
of occupation numbers of photon states. The apparatus is shown in Figure 19.4.

1511
 Figure 19.4: Hanbury-Brown and Twiss Boson Clumping Experiment

The half-silvered mirror splits the beam into two identical beams. The
am-
plitude for a photon to be transmitted/reflected by the mirror is 1/ 2 (the
probability is 1/2).

HBT measured the following

I1 (t) = light intensity at detector 1 at time t

I2 (t + ) = light intensity at detector 2 at a later time t +

and then they averaged the quantity I1 (t)I2 (t + ) over t, keeping fixed.

This is the same as measuring the relative probability of observing two photons
separated by a distance c in the beams (c = speed of light).

The experimental result is just g(~r ~r 0 ) for |~r ~r 0 | = c . This confirmed the
theory or so one thought at that time.

We ask the following question: Is this experiment a verification of the quantum

mechanical theory for identical bosons? The answer might, in fact, be no!

Can we understand the result completely using classical electromagnetic theory

and wave superposition? The experiment might imply that the boson properties
of the photon follow from the superposition principle obeyed by classical fields
and we might not need the quantum mechanical concept of a photon in this
experiment.

Let us see how. Consider the setup shown in Figure 19.5 below:

1512
 Figure 19.5: Classical or Quantum?

We have two sources of photons, A and B. A emits coherent light with amplitude
and wave number k. B emits coherent light with amplitude and wave number
k 0 . We assume that the relative phase of the two coherent beams is random and
that they have the same polarization.

The amplitude for light (A 1) eikr1 and the amplitude for light (B
0 0
1) eik r1 . Therefore, the superposition principle implies that

a1 = total amplitude at detector 1

0 0
= eikr1 + eik r1 (19.145)

and that

I1 = intensity at detector 1
0 0 2
   0 0

2 2
= eikr1 + eik r1  = || + || + 2Real ei(k r1 kr1 ) (19.146)


Therefore,

I1 = I1 averaged over the random phase

2 2
= || + || (19.147)

Similarly,

a2 = total amplitude at detector 2

0 0
= eikr2 + eik r2 (19.148)

and that

I1 = intensity at detector 2
0 0 2
   0 0

2 2
= eikr2 + eik r2  = || + || + 2Real ei(k r2 kr2 ) (19.149)


1513
Therefore,

I1 = I2 averaged over the random phase

2 2
= || + || (19.150)

The product of the averaged intensities I1 I2 is clearly independent of the sepa-

ration distance between the two detectors.

The product of the intensities, however, behaves very differently, however, i.e.,
2
I1 I2 = |a1 a2 |
 0 0 0
 0 0 0 0
2
= 2 eik(r1 +r2 ) + 2 eik (r1 +r2 ) + eikr1 eik r2 + eik r1 eikr2  (19.151)
 

Multiplying out and averaging over

 the random
  phases of and eliminates
the terms proportional to 2  and  2 , etc and we get
 0 0 0 0
2
4 4 2 2
I1 I2 = || + || + || || eikr1 eik r2 + eik r1 eikr2 


2 2
= I1 I2 + 2 || || cos [k 0 (r10 r20 ) k(r1 r2 )] (19.152)

For well-collimated beams r10 r20 r1 r2 , which gives

2 2
I1 I2 = I1 I2 + 2 || || cos [(k 0 k)(r1 r2 )] (19.153)

Therefore, the correlated intensities have a term that depends on the detector
separation.

This term is a maximum when the detectors are at the same point.

If we average over all the different k and k 0 present in the beam (using a Gaus-
sian distribution) we get the same form as the quantum result.

This seems to imply that the photon bunching effect seen in the HBT experi-
ment is a consequence of the superposition principle applied to light from noisy
sources.

The quantum mechanical interpretation of the classical result is

1st term = amplitude for both photons from A

2nd term = amplitude for both photons from B
3rd term = amplitude for one photon from A and the other from B

There are two ways to do this

1. A 1 , B 2
2. A 2 , B 1

1514
These two ways are indistinguishable and the interference between them gives
the cosine term.

So, we corroborate superposition and bunching, but we do not seem to need the
quantum concept of a photon to do it. When one studies this problem in more
detail one can prove that a photon with quantum properties must exist.

19.6 The Hamiltonian

Finally, we write the Hamiltonian in second quantized form.

Suppose the particles interact via a two-particle potential V (~r ~r 0 ). The inter-
action energy operator then becomes
Z
1X
= d3~rd3~r 0 V (~r ~r 0 )s+ (~r)s+0 (~r 0 )s0 (~r 0 )s (~r) (19.154)
2 0
ss

The order of the operators in this expression is very important. This form for
can be confirmed by comparing its matrix elements to the matrix elements in
the standard formalism.

We interpret in this way:

1. it tries to remove particles from ~r and ~r 0
2. if successful it counts V (~r ~r 0 ) and then replaces the particles
3. it replaces the last particle removed first
4. it sums over all pairs (the factor 1/2 avoids double counting of pairs) of
points ~r and ~r 0
The second quantized Hamiltonian for particles of mass m with such a pairwise
interaction potential is then
XZ ~2
H= d3~r s+ (~r) s (~r)
s
2m
Z
1X
+ d3~rd3~r 0 V (~r ~r 0 )s+ (~r)s+0 (~r0 )s0 (~r0 )s (~r) (19.155)
2 0
ss

We now calculate the ground state energy of our gas of spin = 1/2 fermions.

We will treat the interaction potential as a perturbation. To lowest(zeroth)

order we have
X p2 X p2 ZpF 3
(0) d p~ p2
E =T = np~ = 2 = 2V
2m 2m (2~)3 2m
p
~s p
~ 0
3 p2F 3
= N = EF (19.156)
5 2m~2 5

1515
The first order energy correction E (1) is the expectation value of in the un-
perturbed ground state. We get
Z
1 X
E (1) = d3~rd3~r 0 V (~r ~r0 ) h0 |s+ (~r)s+0 (~r0 )s0 (~r 0 )s (~r) |0 i
2
ss0
Z X  n 2
1
= d3~rd3~r 0 V (~r ~r 0 ) gss0 (~r ~r 0 )
2 0
2
ss
Z " #
1 3 3 0 0 2
X
2 0
= d ~rd ~r V (~r ~r ) n Gs (~r ~r ) (19.157)
2 s

If we let Z
v0 = d3~rV (~r) (19.158)

then we have Z
1 N nv0
d3~rd3~r 0 V (~r ~r 0 )n2 = (19.159)
2 2
This is the average interaction of a uniform density of particles with itself (no
correlations). It is called the direct or Hartree energy.

The second term represents the exchange energy

Z
1 X
Eex = d3~rd3~r 0 V (~r ~r 0 ) G2s (~r ~r 0 ) (19.160)
2 s

This term takes account of the tendency of particles of the same spin to stay
apart. The effects of the short-range part of V (~r ~r 0 ) are overcorrected in the
direct energy and fixed up in exchange energy.

We have
pF r 2
sin pF~ r pF r


~ cos ~
Z
Eex 9n 3
= d r pF r 6
V (r) (19.161)
N 4


~

and to first order

3 p2F nv0 Eex
E0 = + + (19.162)
5 2m 2 N

19.6.1 An Example
Consider a gas of electrons of average density n interacting via a Coulomb
potential
e2
V (~r ~r 0 ) = (19.163)
|~r ~r 0 |
The conduction electrons in a metal form such a gas.

We note that in a real physical system of this type, we never have an isolated

1516
electron gas. There always exists enough positive charges to make the overall
system electrically neutral.

To first approximation in a metal or a plasma we can replace the positive ions

by a uniform background of positive charge of density +ne.

The electrostatic self-energy of this background

e2 n2
Z
1
d3 rd3 r 0 (19.164)
2 |~r ~r 0 |

plus the average electrostatic interaction between the positive background and
the electrons
e2 n2
Z
d3 rd3 r 0 (19.165)
|~r ~r 0 |
exactly cancels the Hartree energy as it must because the electrostatic energy of
a neutral system can only be proportional to the volume for a large system(not
a higher power of the volume!).

Therefore, the net interaction energy of the electron gas (to first order) is
Z
Eex 9ne2 dx 2
= 2 (sin x x cos x) (19.166)
N pF x5
0

For a typical electron gas, this energy is written in terms of a parameter

d average interparticle spacing
rs = = (19.167)
a0 Bohr radius
Now, we also have
~2
 
4 3
a0 = , n d =1 (19.168)
me2 3
Therefore,
1/3
me2

9
rs = (19.169)
4 pF
and
e2
 
2.21 0.916
E= (19.170)
rs2 rs a0
where the first term is the kinetic energy and the second term is the exchange
energy.

Finally, if we write in the creation/annihilation operator formalism we get

1 XXX
v= Vp~ 0 ~p p~+~q,~p 0 +~q 0 ap+ +
~ 0 s aq
~ 0 s0 aq
~s0 ap
~s (19.171)
2V 0 0 0
p
~p~ q
~q~ ss

1517
where Z
~
V~k = d3 r eik~r V (~r) (19.172)

This is a sum over scattering processes. Feynman developed a clever diagram-

matic method of thinking about such terms.
1. The delta-function represents conservation of energy and momentum in
the scattering process.
2. The process itself is represented by the diagram in Figure 19.6 below.

Figure 19.6: Feynman Diagram

We shall return to this idea later in Chapter 21 and develop a method of

thinking about the universe as expressed by these diagrams.

19.7 Bogoliubov Transformation, Quasiparticles

and Superfluidity
Let us consider a weakly interacting Bose gas. From our earlier discussions we
have, in general,
1 XX
v= Vp~ 0 ~p p~+~q,~p 0 +~q 0 a+
p
+
~ 0 aq
~ 0 aq
~ ap
~
2V 0 0 p
~p~ q
~q~
1 XX
= Vq~ a~+ a~+ ap~ aq~ (19.173)
2V k+~
q k~q
~
k~p q
~

so that
X 1 XX
H = ~~k a~+ a~k + Vq~ a~+ a+ q ap
~~ ~ a~ (19.174)
k 2V q p
k+~ k
~
k ~
k~p q
~

where we have used the delta function (which corresponds to momentum con-
servation). This corresponds to the process
~k + p~ (~k + ~q) + (~
p ~q) (19.175)

or we have momentum transfer of ~q between the two interacting bosons.

1518
At low temperatures, a Bose-Einstein condensation takes place in the ~k = 0
mode, i.e., the ~k = 0 mode is macroscopically occupied or
N0 = h0 | a+ a |0 i N , N N0 = # excited particles << N0 (19.176)
This means that we can neglect the interaction of excited particles with one
another and restrict our attention to the interaction of the excited particles
with the condensed particles.

This gives
X 1
H = ~~k a~+ a~k + V0 a+ +
0 a0 a0 a0
k 2V
~
k
1 X 1 X
+ (V0 + V~k )a+ +
0 a0 a~ a~k + V~k (a~+ a+~ a0 a0 + a+ +
0 a0 a~
k a~
k ) + ....
V k 2V k k
~
k6=0 ~
k6=0
(19.177)

The effect of a+
0 and a0 on the state with N0 particles in the condensate is
p
a0 |......, N0 , ....i = N0 |......, N0 1, ....i (19.178)
+
p
a0 |......, N0 , ....i = N0 + 1 |......, N0 + 1, ....i (19.179)
Since N0 is a verylarge number ( 1023 ), both of these relations correspond to
multiplication by N0 . It is physically clear that the removal or addition of one
particle from the condensate will make no difference to the physical properties
of the system.

In comparison to N0 , the effect of the commutator a0 , a+

 
0 = 1 is negligible.
+
This saysthat, in this case, the operators a0 and a0 can be approximated by a
number N0 .

We then have
X ~2 k 2 1 2
H = a+ a~ + N V0
2m ~k k 2V 0
~
k6=0
N0 X N0 X
+ (V0 + V~k )a~+ a~k + V~k (a~+ a+~ + a~k a~k ) (19.180)
V k 2V k k
~
k6=0 ~
k6=0

We can write X
N = N0 + a~+ a~k (19.181)
k
~
k6=0

which says that (carrying out the necessary algebra)

1 2 1 2 N X V0 X +
N0 V0 = N V0 + V~k a~+ a~k + a~ a~k a~+0 a~k0 (19.182)
2V 2V V k 2V k k
~
k6=0 ~
k,~
k0 6=0

1519
The Hamiltonian then becomes
X ~2 k 2 N X
H = a~+ a~k + V~k a~+ a~k
2m k V k
~
k6=0 ~
k6=0
2
N N X
+ V0 + V~k (a~+ a+~ + a~k a~k ) (19.183)
2V 2V k k
~
k6=0

up to terms with 4 creation/annihilation operators, which are of order n02 where

N N0
n0 = = density of particles not in condensate (19.184)
V
We make the approximation of neglecting these anharmonic terms, which is
good for n0  n.

The remaining Hamiltonian is a quadratic form which we need to diagonalize

to solve the problem.

We use the so-called Bogoliubov transformation. We assume that

a~k = u~k ~k + v~k +~ , a~+ = u~k ~+ + v~k ~k (19.185)
k k k

with real coefficients. We then require that the operators satisfy Bose com-
mutation relations
 h i h i
~k , ~k0 = ~+ , ~+0 = 0 , ~k , ~+0 = ~k~k0

(19.186)
k k k

This requires that u~2k v~k2 = 1. The inverse transformations are

~k = u~k a~k v~k a+~ , ~+ = u~k a~+ v~k a~k (19.187)

k k k

We also have that

a~+ a~k = u~2k ~+ ~k + v~k2 ~k +~ + u~k v~k (~+ +~ + ~k ~k )
k k k k k
+ +
a~ a ~ = u~k ~ ~ + v~k ~k ~k + u~k v~k (~+ ~k + ~k +~ )
2 + + 2
k k k k k k
a~k a~k = u~2k ~k ~k + v~k2 ~+ +~ + u~k v~k (+~ ~k + ~k ~+ )
k k k k

The Hamiltonian becomes

1 2
H = N V0
2V
X  ~2 k 2 h i
+ + nV~k u~2k ~+ ~k + v~k2 ~k ~+ + u~k v~k (~+ +~ + ~k ~k )
2m k k k k
~
k6=0
N X h 2 i
+ V~k (u~k + v~k2 )(~+ +~ + ~k ~k ) + 2u~k v~k (~+ ~k + ~k ~+ )
2V k k k k
~
k6=0
(19.188)

1520
In order for the non-diagonal terms to vanish, we must have
 2 2 
~ k N
+ nV~k u~k v~k + V~k (u~2k + v~k2 ) = 0 (19.189)
2m V
This equation together with u~2k v~k2 = 1 is sufficient to determine u~2k and v~k2 . If
we define
" 2 #1/2
~2 k 2
2
~~k = + nV~k nV~k
2m
" 2 #1/2
~2 k 2 nk 2 V~k
= + (19.190)
2m m

then we get
   
~2 k 2 ~2 k 2
~~k + 2m + nV~k ~~k + 2m + nV~k
u~2k = , v~k2 = (19.191)
2~~k 2~~k
and
2
nV~k nV~k
u~k v~k = , v~k2 = ~2 k 2

(19.192)
2~~k 2~~k ~~k + 2m + nV~k
Finally, the Hamiltonian becomes
1 X ~2 k 2
  X
1 2
H = N V0 + nV~k ~~k + ~~k ~+ ~k (19.193)
2V 2 2m k
~
k6=0 ~
k6=0

where
~2 k 2
 
1 2 1X
N V0 + nV~k ~~k = ground state energy E0 (19.194)
2V 2 2m
~
k6=0
X
~~k ~+ ~k = sum of oscillators or excitations (19.195)
k
~
k6=0

The excitations (oscillators) or quanta that are created by the ~+ are called
k
quasiparticles.

All the excited states correspond to different numbers of noninteracting bosons

where each boson has the excitation energy
" 2 #1/2
~2 k 2
2
E~k = ~~k = + nV~k nV~k
2m
" 2 #1/2
~2 k 2 nk 2 V~k
= + (19.196)
2m m

1521
Quasiparticles appear in all kinds of physical systems at all energy scales.

The ground state of the system |0i is fixed by the condition that no quasiparticles
are excited,
~k |0i = 0 for all ~k (19.197)
The number of particles outside the condensate(the ground state) is given by
X X X
N 0 = h0| a~+ a~k |0i = h0| ~k ~+ |0i = v~k2 (19.198)
k k
~
k6=0 ~
k6=0 ~
k6=0
 
For small ~k  we have
 

E~k = ~~k cs ~k (19.199)

where r
nV0
cs = = speed of sound in fluid (19.200)
m
This says that the long wavelength excitations have a linear dispersion relation
(E versus k).
 
For large ~k  we have
 

~2 k 2
E~k = ~~k = + nV~k (19.201)
2m
This corresponds to the dispersion relation for free particles whose energy is
shifted by a mean potential nV~k .

We can now understand qualitatively how superfluidity comes about.

Consider a small particle(call it a cluster) which could be several dirt atoms or a

piece of a surrounding wall moving through a quantum liquid with the dispersion
relations just described for the quasiparticle excitations, i.e., the quasiparticle
energy rises linearly with k at low k and quadratically with k at large k.

The cluster can lose energy (experience friction) only by causing excitations in
the fluid. For T > 0, there are already excitations present in the fluid at which
the cluster may scatter and thus lose energy, but at T = 0, this is not the
case. Let the initial momentum of the cluster be ~~q and the momentum of the
excitations be ~~k. In a scattering event of the cluster with the fluid, energy and
momentum are conserved.
~2 q 2 ~2 q 02
~~q = ~~q 0 + ~~k , = + E(k) (19.202)
2m 2m
where ~~q 0 is the momentum of the cluster after the scattering. The elementary
excitations of the fluid (quasiparticles) are given by
" 2 #1/2
~2 k 2 nk 2 V~k
E~k = ~~k = + (19.203)
2m m

1522
Consider the energy conservation equation. We have

~2 q 2 ~2 (~q ~k)2
= + E(k) (19.204)
2m 2m
or
~2 ~2 k 2
0= ~q ~k + + E(k) (19.205)
2m 2m
This says that (let be the angle between ~q and ~k)

1 E(k) 1 ~k
cos = + + E(k) (19.206)
v ~k v 2m
where
~q
v= = initial velocity of the cluster (19.207)
m
Now for the quasiparticle excitations

E(k)
> cs (19.208)
~k
therefore, for the excitation(emission) of a quasiparticle the cluster velocity must
be larger than cs , i.e., v > cs . This follows from the above relation as k 0 or
the angle becomes imaginary!

A cluster moving with v < vcritical = cs (in this model) cannot lose energy to
the fluid. Thus, there is no friction and one has superfluidity. For liquid helium
vcritical << cs and the physics is even more dramatic.

19.8 Spontaneous Symmetry Breaking in Quan-

tum Mechanics
In quantum mechanics symmetry has a much more powerful role than in classi-
cal mechanics. Translational invariance in a classical system causes momentum
to be conserved; in quantum mechanics it immediately implies that all eigen-
states of the Hamiltonian are spread out with equal amplitude over all of space.
It could be argued that because a chair is built of many microscopic particles
that all obey the rules of quantum mechanics, the chair as a whole should also
respect the symmetry of its Hamiltonian and be spread out over all of space.
Clearly this situation is not physically realized. The way out of the seeming
paradox is the spontaneous symmetry breaking of the collective system. The
description of spontaneous symmetry breaking in macroscopic systems that are
constructed from microscopic, quantum mechanical constituents is one of the
highlights of condensed matter theory. It is used to explain the classical features
of macroscopic systems ranging from crystals and antiferromagnets to supercon-
ductors.

1523
The general idea behind spontaneous symmetry breaking is easily formulated:
as a collection of quantum particles becomes larger, the symmetry of the system
as a whole becomes more unstable against small perturbations. In the limit of
an infinite system an infinitesimal perturbation is enough to cause the system
to break the underlying symmetry of the Hamiltonian. The fact that the sym-
metry breaking can happen spontaneously is signaled by a set of noncommuting
limits: In the complete absence of perturbations even a macroscopic system
should conform to the symmetry of the Hamiltonian. However, in the presence
of an infinitesimal perturbation a macroscopic system will be able to break the
symmetry and end up in a classical state. This intuitive picture of spontaneous
symmetry breaking is not always easy to demonstrate in an equally clear math-
ematical description of the process.

In this section we present a simple mathematical procedure that can be applied

to the spontaneous breaking of any continuous symmetry and that naturally em-
phasizes the roles of the key players (the symmetry-breaking field, the thermo-
dynamic limit, and the global excitations of a thin spectrum) in this symmetry
breaking. The procedure is described by considering the example of a quantum
harmonic crystal that spontaneously breaks translational symmetry. All of the
methods, that is, bosonization, using the Bogoliubov transformation to identify
the thin spectrum of states involved in spontaneous symmetry breaking, intro-
ducing a symmetry-breaking field in the collective dynamics, and considering a
noncommuting order of limits, can be applied to other systems as well.

19.8.1 The Harmonic Crystal

As a basic example of spontaneous symmetry breaking, we consider how trans-
lational symmetry is broken in a crystalline lattice. Consider a harmonic crystal
with the Hamiltonian
X p2j X 2
H= + (xj xj+1 ) (19.209)
j
2m 2 j

where j labels the N atoms in the lattice, which have mass m, momentum pj ,
and position xj . We consider here only a one- dimensional chain of atoms, but
all of the following can be straightforwardly generalized to higher dimensions.
The parameter gives the strength of the harmonic potential between neigh-
boring atoms. The results on spontaneous symmetry breaking that follow are
equally valid for anharmonic potentials.

In the standard treatment of the harmonic oscillator a Fourier transformation

of the Hamiltonian is used to identify its eigenstates. We follow a slightly
longer route by introducing boson (phonon) operators from the outset and di-
agonalizing them using the Bogoliubov transformation. This approach has the
advantages that it naturally brings to the fore the thin spectrum of the crystal
and enables us to keep track of the center of mass motion of the crystal as a

1524
whole. The momentum and position operators can be expressed in terms of
boson operators as
r r
~ +
1 ~ +

pj = iC b bj , xj = b + bj (19.210)
2 j C 2 j
2
so
that the commutation relation [xj , pk ] = i~jk is satisfied. We choose C =
2m so that the Hamiltonian reduces to
r
~ 2 X 
2 b+ + +


+

H= j bj + bj bj bj + bj bj+1 + bj+1 (19.211)
4 m
k

After a Fourier transformation we have

r
~2 X
 
+ Bk + +

H= Ak bk bk + bk bk + bk bk + 1 (19.212)
2m 2
k

where Ak = 2 cos (ka), Bk = cos (ka), and a is the lattice constant. This
Hamiltonian is still not diagonal, because the terms b+ +
k bk and bk bk create
and annihilate two bosons at the same time. We get rid of these terms by
introducing Bogoliubov transformed operators k = cosh (uk )bk + sinh (uk )b+
k
and choosing uk such that the resulting Hamiltonian is diagonal. After this
Bogoliubov transformation, the Hamiltonian in terms of transformed bosons is
given by
r "   #
X ka + 1 2
H=~ 2 sin | | k k + + cos (ka)
m 2 2 4
k
r  
X ka 1
= 2~ sin | | nk + (19.213)
m 2 2
k
P R
because k cos k = (N/2)
dk cos k = 0.

19.8.2 The Thin Spectrum

The form Eq. (5) of the Hamiltonian in terms of phonon operators coincides
with the standard textbook result. The use of the Bogoliubov transformation
to obtain Eq. (5) has the advantage that it draws attention to a rather subtle
point. When k 0 the excitation energy k 0, and the two parameters
in the Bogoliubov transformation diverge [sinh (uk ) and cosh (uk ) .
This at k = 0 the canonical transformation is no longer well defined, and we
should treat the k = 0 part of the Hamiltonian Eq. (1) separately from the rest.
The excitations with k = 0 are the ones that describe the collective dynamics of
the crystal as a whole, and therefore they are also the states that are involved
in the collective symmetry breaking. The k = 0 part of the Hamiltonian, writ-
ten in terms of the original operators, is given by
p2tot
Hcoll = + constant (19.214)
2N m

1525

where ptot = N pk=0 is the total momentum of the system. It can easily be
checked that this part of the Hamiltonian, which describes the external dy-
namics of the crystal as a whole, commutes with the rest of the Hamiltonian,
which describes the internal dynamics of the phonon modes inside the crystal.
We therefore focus on the collective part of the Hamiltonian and disregard the
phonon spectrum given by Eq. (5).

Notice that by considering only the collective part of the Hamiltonian, we have
effectively reduced the problem to a single-particle problem. The single particle
in Eq. (6) with mass N m and momentum ptot describes the dynamics of the
crystal as a whole. Its momentum and position are the center of mass momen-
tum and position of the crystal. In contrast, Eq. (5) describes the internal
degrees of freedom of the crystal and includes all many-body effects that arise
from the coupling of the N individual atoms.

The relevant eigenstates of the collective Hamiltonian Eq. (6) are very low in
energy: their excitation energies scale as 1/n, where N is the number of atoms
in the crystal. In the thermodynamic limit all of these states thus become
nearly degenerate. Because of this property a combination of these states that
break the symmetry of the Hamiltonian can be spontaneously formed in the
thermodynamic limit. At the same time, these collective eigenstates are so few
in number and of such low energy that their contribution to the free energy
completely disappears in the thermodynamic limit. This vanishing contribution
can be seen by looking at their contribution to the partition function.

Digression on the Partition Function

In statistical mechanics, the partition function Z is an important quantity that
encodes the statistical properties of a system in thermodynamic equilibrium. It
is a function of temperature and other parameters, such as the volume enclosing
a gas. Most of the aggregate thermodynamic variables of the system, such as
the total energy, free energy, entropy, and pressure, can be expressed in terms
of the partition function or its derivatives.

There are actually several different types of partition functions, each correspond-
ing to different types of statistical ensemble (or, equivalently, different types of
free energy.) The canonical partition function applies to a canonical ensem-
ble, in which the system is allowed to exchange heat with the environment
at fixed temperature, volume, and number of particles. The grand canonical
partition function applies to a grand canonical ensemble, in which the system
can exchange both heat and particles with the environment, at fixed temper-
ature, volume, and chemical potential. Other types of partition functions can
be defined for different circumstances; see partition function (mathematics) for
generalizations.

Canonical Partition Function - Definition

1526
As a beginning assumption, assume that a thermodynamically large system is in
constant thermal contact with the environment, with a temperature T , and both
the volume of the system and the number of constituent particles fixed. This
kind of system is called a canonical ensemble. Let us label with s (s = 1, 2, 3, ...)
the exact states (microstates) that the system can occupy, and denote the total
energy of the system when it is in microstate s as Es. Generally, these mi-
crostates can be regarded as discrete quantum states of the system.

The canonical partition function is

X
Z= eEs
s

where the inverse temperature is conveniently defined as

1
=
kB T

where kB denotes Boltzmanns constant.

In quantum mechanics, the partition function can be more formally written as

a trace over the state space (which is independent of the choice of basis):

Z = T r eH

where H is the quantum Hamiltonian operator. The exponential of an operator

can be defined using the exponential power series.

Meaning and Significance

It may not be obvious why the partition function, as we have defined it above, is
an important quantity. Firstly, let us consider what goes into it. The partition
function is a function of the temperature T and the microstate energies E1 , E2 ,
E3 , etc. The microstate energies are determined by other thermodynamic vari-
ables, such as the number of particles and the volume, as well as microscopic
quantities like the mass of the constituent particles. This dependence on mi-
croscopic variables is the central point of statistical mechanics. With a model
of the microscopic constituents of a system, one can calculate the microstate
energies, and thus the partition function, which will then allow us to calculate
all the other thermodynamic properties of the system.

The partition function can be related to thermodynamic properties because it

has a very important statistical meaning. The probability Ps that the system
occupies microstate s is
1
Ps = eEs
Z

1527
The partition function thus plays the role of a normalizing constant (note that
it does not depend on s), ensuring that the probabilities sum up to one:
X 1 X Es 1
Ps = e = Z=1
s
Z s Z

This is the reason for calling Z the partition function: it encodes how the
probabilities are partitioned among the different microstates, based on their
individual energies.

Calculating the Thermodynamics Total Energy

In order to demonstrate the usefulness of the partition function, let us calculate

the thermodynamic value of the total energy. This is simply the expected value,
or ensemble average for the energy, which is the sum of the microstate energies
weighted by their probabilities:
X 1 X 1 Z ln Z
hEi = Es Ps = Es eEs = =
s
Z s Z

or, equivalently,
hEi = kB T 2 ln ZT
Incidentally, one should note that if the microstate energies depend on a pa-
rameter in the manner

Es = Es(0) + As for all s

then the expected value of A is

X 1
hAi = A s Ps = ln Z(, )
s

This provides us with a method for calculating the expected values of many mi-
croscopic quantities. We add the quantity artificially to the microstate energies
(or, in the language of quantum mechanics, to the Hamiltonian), calculate the
new partition function and expected value, and then set to zero in the final
expression.

Relation to Thermodynamic Variables

We now state the relationships between the partition function and the various
thermodynamic parameters of the system. These results can be derived using
the method of the previous section and the various thermodynamic relations.
As we have already seen, the thermodynamic energy is
ln Z
hEi =

1528
The variance in the energy (or energy fluctuation) is

2 ln Z
h(E)2 i = h(E hEi)2 i =
2
The heat capacity is
hEi
Cv = = f rac1kB T 2 h(E)2 i
T
The entropy is
X F
S = kB Ps ln Ps = kB (ln Z + hEi) = (kB T ln Z) =
s
T T

where F is the Helmholtz free energy defined as F = U T S, where U = hEi

is the total energy and S is the entropy, so that

F = hEi T S = kB T ln Z

Returning to our discussion. We have

X
Zthin = eHcoll N
Fthin = kB T ln (Zthin ) ln (N ) (19.215)

The free energy of the total system is an extensive quantity, so that Fthin /Ftot
ln (N )/N disappears in the limit N , which is the so-called thermodynamic
limit. The states of this part of the spectrum are thus invisible in thermodynam-
ically measurable quantities such as for instance the specific heat of macroscopic
crystals, and it is consequently called the thin spectrum of the quantum crystal.

To see how the states in the thin spectrum can conspire to break the translational
symmetry, we need to add a small symmetry-breaking field to the Hamiltonian:

SB p2tot B
Hcoll = + x2tot (19.216)
2N m 2
Here the symmetry-breaking field B is introduced as a mathematical tool and
need not actually exist. We will send the value of B to zero at the end of the
calculation. The Hamiltonian Eq. (8) is the standard form of the Hamiltonian
for a quantum harmonic oscillator, and its eigenstates are well known. The
ground state wavefunction can be written as
 1/4
mN 2
0 (xtot ) = e(mN/2~)xtot (19.217)
~
p
where = B/mN . This ground state is a wavepacket ofthe total momentum
states that make up the thin spectrum. Apart from the ground state configura-
tion there are also collective eigenstates that are described by the excitations of

1529
the harmonic oscillator Eq. (8). These excitations describe the collective motion
of the crystal as a whole. As N becomes larger, the ground state wavepacket
becomes more and more localized at the position xtot = 0, until it is completely
localized as N . That this localization can occur spontaneously without
the existence of a physical symmetry-breaking field B can be seen by considering
the noncommuting limits

lim lim |0 (xtot )|2 = constant (19.218)

N B0

lim lim |0 (xtot )|2 = (xtot ) (19.219)

B0 N

If we do not include any symmetry-breaking field, then the crystal is always com-
pletely delocalized and respects the symmetry of the Hamiltonian. If we do allow
for a symmetry-breaking field, then it turns out that in the limit of having in-
finitely many constituent particles, an infinitesimally small symmetry-breaking
field is enough to completely localize the crystal at a single position. This math-
ematical instability implies that the symmetry breaking happens spontaneously
in the thermodynamic limit(spontaneous symmetry breaking).

Notice that once the crystal has been localized at a specific position and the
unphysical symmetry-breaking field has been sent to zero, the delocalization of
the crystal due to the spreading of its wavefunction will take a time proportional
to N and can thus never be observed.

To see more rigorously whether or not the crystal as a whole is localized, we

should look at the spatial fluctuations hx2tot i. The absolute size of these fluc-
tuations by itself is meaningless. Their size becomes meaningful only if it is
compared to the size of the crystal. Because the size of the crystal is propor-
tional to the number of particles in the system, the appropriate order parameter
in this case is of the crystal hx2tot i/N . This order parameter has a noncommuting
order of limits as
lim lim hx2tot i/N = (19.220)
N B0

lim lim hx2tot i/N = 0 (19.221)

B0 N

which again signals the spontaneous localization of the crystal as a whole.

19.8.3 Subtleties
In the derivation of the spontaneous symmetry breaking of a harmonic crystal
we have been somewhat sloppy in the definition of the symmetry-breaking field.
After all, the collective model of Eq.(6) was only the k = 0 part of the full blown
Hamiltonian in Eq. (1), but we did not consider the symmetry-breaking field
to be only the k = 0 part of some other field acting on all atoms individually.
It would therefore be better to start with a microscopic model, which already

1530
includes a symmetry-breaking field such as
" #
X p2j
SB 2
H = + (xj xj+1 ) + B(1 cos (xj )) (19.222)
j
2m 2

SB p2tot B 2
Hcoll + x (19.223)
2N m 2N tot
In Eq. (15) we again consider only the k = 0 part of the Hamiltonian and
have expanded the cosine to quadratic order. The fact that the symmetry-
breaking field now scales as 1/N is a direct consequence of our definition of
the microscopic symmetry-breaking field. The factor 1/N cannot be avoided
if we insist that the microscopic Hamiltonian be extensive. This factor might
seem to imply an end to the localization of the total wavefunction 0 (xtot ),
but spontaneous symmetry breaking is still possible as long as we consider the
correct order parameter. Even though the wavefunction itself does not reduce to
a Dirac delta function anymore, the spatial fluctuations of the crystal compared
to its size still become negligible in the thermodynamic limit if an infinitesimal
symmetry-breaking field is included:

lim lim hx2tot i/N = (19.224)

N B0

lim lim hx2tot i/N = 0 (19.225)

B0 N

Once again the disappearance of fluctuations in the thermodynamic limit sig-

nals the spontaneous localization of the crystal as a whole.

This digression into extensivity and the correct choice for the symmetry-breaking
field seems unnecessary for understanding the essential ingredients of sponta-
neous symmetry breaking, and therefore we have ignored these subtleties in our
main treatment of quantum spontaneous symmetry breaking. In the application
of this procedure to other systems, such as antiferromagnets and superconduc-
tors, these issues dont arise because we are forced to consider extensive models
from the outset. In these cases, however, the mathematics of diagonalizing the
collective Hamiltonian is a bit more involved.

Summarizing, this section provides a simple way of explaining the appearance

of spontaneous symmetry breaking in quantum systems. The procedure starts
with the bosonization of the microscopic Hamiltonian. The quadratic part of
the bosonized Hamiltonian can in principle be diagonalized using a Bogoliubov
transformation, but in doing so we find that there are some modes for which
the transformation is ill-defined. These singular modes are the ones that de-
scribe the dynamics of the system as a whole (in contrast to the dynamics of
constituent particles within the system). These collective excitations should be
treated separately from all other modes, and together they define the collective
part of the Hamiltonian of the system. The eigenstates of this collective Hamil-
tonian that scale as 1/N form the thin spectrum. It is a combination of these

1531
states that make up the symmetry broken wavefunction. As a mathematical
tool necessary to be able to see the symmetry breaking explicitly, we introduced
the symmetry-breaking field B. If we look at the new ground state wavefunction
or at a suitably defined order parameter for the system, we see that in the ther-
modynamic limit an infinitesimally small field B is enough to completely break
the symmetry of the underlying Hamiltonian. It is thus argued that symmetry
breaking can happen spontaneously in the limit N .

This method as presented can be adapted to describe rotors, antiferromagnets,

and even superconductors, and should in principle be applicable to all quantum
systems that spontaneously break some continuous symmetry.

19.9 Problems
19.9.1 Bogoliubov Transformations
Consider a Hamiltonian for Bosonic operators b+ k , bk of the form
h i
H = E(k)b+ + +
k bk + A(k) bk bk + bk bk

Define a Bogoliubov transformation to new Bosonic operators k+ , k as follows:

+
bk = cosh 2k k + sinh 2k k , b+ +
k = cosh 2k k + sinh 2k k

(a) Assume that E(k), A(k), K are all even functions of k and find the form
of sinh 2k as a function of A(k), E(k) so that
H = (k)k+ k + F (k)
and find (k) and F (k).

(b) Show that if [b+ + + +

k , bk ] = 1, [bk , bk ] = [bk , bk ] = 0, then [k , k ] = 1,
+ +
[k , k ] = [k , k ] = 0.

(c) Show that if the operators b+

k , bk are Fermionic instead of Bosonic, the
Bogoliubov transformation must have cosh k cos k and sinh k
sin k so that the new operators k+ , k now obey anticommutation rules.

19.9.2 Weakly Interacting Bose gas in the Bogoliubov Ap-

proximation
In this case we have the Hamiltonian
X 1 X X + +
H = (k)a+
k ak + Vq ap+q akq ak ap
2V q
k p,k

a+ N0 ei +
P
Consider the operator K = H N where N = k k ak . Definea0 =
b0 , ak6=0 = bk6=0 .

1532

(a) Separate the terms of order N02 , N0 N0 , N0 in the interaction term, show

that these are quadratic in b, b+ , and show that terms oforder N0 and
1 are cubic and quartic in b, b+ . Neglect the terms of O( N0 ) and O(1),
which is the Bogoliubov approximation, and write down K only keeping
terms up to quadratic order in b, b+ .

(b) Show that in this Bogoliubov approximation that K = KQ +Kcl where KQ

is quadratic and linear in b, b+ and Kcl is purely classical and independent
of b, b+ . Establish a relation = (N0 ) by minimization of Kcl , i.e.,

Kcl 
 =0
N0 

This is the Gross-Pitaevskii equation. Show that imposing this condition

leads to the cancellation of the terms linear in b, b+ in KQ .

(c) Diagonalize the resulting quadratic form for KQ by a Bogoliubov trans-

formation:
b+ i +
k e = ck cosh k + ck sinh k

Find cosh k , sinh k by requesting the cancellation of terms c+ +

k ck , ck ck .
Show that in this Bogoliubov transformation,
X
KQ = c+
k ck ~(k) + K0
k

Find (k) and K0 , consider Vq = V0 constant and evaluate the integral

for K0 .

19.9.3 Problem 19.9.2 Continued

(a) Invert the Bogoliubov transformation in part (c) of Problem 19.9.2 and
show that
ck = bk cosh k b+
k sinh k

where bk = bk ei . Use

1
eA BeA = B + [A, B] + [A, [A, B]] + .....
2!
to show that
ck = U ()bk U 1 ()
where U () is the unitary operator

k (b+ +
k bk bk bk )
P
U () = e k>0

1533
 (b) Show that the  ground state
 of K in the Bogoliubov approximation   is
|GSi = U () 0 where 0 is the vacuum of the operators bk : bk 0 = 0
for all k. Argue that |GSi is a linear superposition of states with a pair
of momenta ~k, ~k respectively. This is a squeezed quantum state (see
Chapter 14 example). These states are ubiquitous in quantum optics and
quantum controlled nanoscale systems.

19.9.4 Mean-Field Theory, Coherent States and the Gross-

Pitaevkii Equation
Consider the pair potential V (~x ~y ) = V0 3 (~x ~y ) and introduce the coher-
ent states of the Bosonic operator |(~x)i such that (~x) |(~x)i = (~x) |(~x)i.
Include a one-body trap potential in the Hamiltonian H:
 2 2 
~
Z
H = d3 x + (~x) + U (~x) (~x)
2m
Z Z
1
+ d3 x d3 y + (~x) + (~y )V (~x ~y )(~y )(~x)
2

(a) Minimize the energy E() = h| K |i where |i is the coherent state

above with h | i = 1 and show that

E
=0
(~x)

leads to the Gross-Pitaevskii equation

 2 2 
~
+ U (~x) (~x) + V0 |(~x)|2 (~x) = 0
2m

with the constraint that |(~x)|2 d3 x = N .

R

(b) Define new operators (~x) (~x) + (~x) where (~x) is the solution
to the Gross-Pitaevskii equation and write K up to quadratic order in
(~x), + (~x). Show that terms linear in (~x), + (~x) are cancelled by (~x)
being a solution to the Gross-Pitaevskii equation.

(c) Introduce the Bogoliubov transformation:

(~x) = u(~x)(~x) + v(~x) + (~x) , + (~x) = u (~x) + (~x) + v (~x)(~x)

h i
Show that (~x), + (~y ) = 3 (~x ~y ) if |u(~x)|2 + |v(~x)|2 = 1.

(d) Write K up to quadratic order in , + found in part (b) in terms of , + .

What is the equation that U ,V must obey so that the terms of the form
2 , + 2 are cancelled? These are the Bogoliubov-DeGennes equations!

1534
19.9.5 Weakly Interacting Bose Gas
Consider a homogeneous, weakly interacting Bose gas with Hamiltonian
 2 2
~
Z Z Z
H = d3 x + (~x) (~x)+ d3 x d3 y + (~x) + (~y )V (~x ~y )(~y )(~x)
2m
(a) Consider V (~x ~y ) = |V0 | 3 (~x ~y ) and assume a condensate with N0
particles. Obtain the operator for K = H N in the Bogoliubov ap-
proximation. By a canonical Bogoliubov transformation bring it to the
form X
K = ~(k)c+k ck + K0
k

Show that (k) becomes imaginary for some values of 0 < k < kmax .
(b) What is kmax ? What is the physical reason for this imaginary value and
what do they mean?

19.9.6 Bose Coulomb Gas

Consider the same problem as Problem 19.9.1, but now with
e2
V (~x ~y ) = (19.226)
|~x ~y |
which is the Coulomb potential. Namely, consider a weakly interacting Bose ga
of charged particles (and assume a homogeneous neutralizing background like
in the so-called Jellium model).
(a) Obtain the energy eigenvalues ~(k) in the Bogoliubov approximation.
Show that now limk0 (k) = (0) 6= 0. What is (0)? Compare to the
result for plasma oscillations in an electron gas.
(b) Give the behavior of the Bogoliubov coefficients for small k. This is related
to long-range behavior of the forces. Goldstones theorem states that
any theory with an exact symmetry other than that of the vacuum must
contain a massless particle. Does this case violate Goldstones theorem?

19.9.7 Pairing Theory of Superconductivity

The B(ardeen)C(ooper)S(chrieffer) Hamiltonian in the mean-field approxima-
tion is
X   Xh i
+
K = H N = (k) a+
k ak + a+
k ak + a+
k ak + ak a k
k k

where a+
k, are creation operators of an electron of spin up or down and mo-
mentum ~k and
g X0
= hGS| ak ak |GSi
V 0
k

1535
 P 0
|GSi is the ground state of K and k0 is a sum over states with

~2 k 2
0 (k 0 ) ~m and (k 0 ) = (k) =
2m
(a) Diagonalize K by a Bogoliubov transformation, i.e., introduce new oper-
ators
Ak = uk ak vk a+ +
k , Bk = vk ak + uk ak
and their respective Hermitian conjugates with uk , vk and even in k. Show
that the transformation is canonical, namely, that A, B obey the usual
commutation relations if u2k + vk2 = 1. It is convenient to write
 1/2  1/2
1 1
uk = (1 + k ) , vk = (1 k )
2 2

Invert the Bogoliubov transformation and write K in terms of A, A+ , B, B + .

Choose k so that K becomes
X h i
K = E(k) hatA+ k A k + B +
k B k + K0
k

that is, choose k to make terms like AB vanish. Find E(k).

(b) Obtain a self-consistent equation for by evaluating hGS| ak ak |GSi
and solve this equation by replacing
Z ~m
1 X0
N () d
V 0 0
k

where N () is the density of states so that

d3 k
N () d =
(2)3
and assume that N () N (0).
(c) Obtain the distribution function hGS| a+
k ak |GSi

(d) Show that E(0) = = gap and evaluate the resulting integral in part
(b) to give the gap as a function of gN (0) for gN (0)  1.

19.9.8 Second Quantization Stuff

1. Quantum Chain of Oscillators

Consider a chain of atoms with masses m connected by springs of rigidity

:
 2 
X pn
Hph = + (un un+1 )2
n=
2m 2

1536
 where un are the displacements of atoms from their equilibrium positions,
and pn are the corresponding conjugate momenta.

Consider the problem in quantum mechanics, i.e., treat un and pn as op-

erators satisfying the canonical commutation relation [pn , un0 ] = i~n,n0

Diagonalize the quantum Hamiltonian above. In order to do this, first do

the Fourier transform: un uk , pn pk and then introduce the creation
and annihilation operators of phonons a+ k and ak by the following formula:
s r
~ + ~m(k)
uk = (ak + ak ) , pk = i (ak a+
k)
2m(k) 2

Write the Hamiltonian in terms of a+

k and ak and determine the phonon
spectrum (k). Calculate the ground state energy of the system.
2. Interaction between Phonons

Suppose the springs have small anharmonicity 0 , so the Hamiltonian of

the system has the additional term

X
0
Hph = 0 (un un+1 )3
n=

Rewrite the Hamiltonian in terms of the phonon operators a+

k and ak
introduced in part (1). What can you say about momentum conservation
of the phonons in the new Hamiltonian?
3. Electron-Phonon Interaction

Suppose electrons are also present on the same chain of atoms. Suppose
that the electrons can make transitions between neighboring lattice sites
with the probability amplitude tn so that

X
+
Hel = tn n+1 n + h.c.
n=

where n+ and n are the fermion operators creating and annihilating

electrons on the site n.

In the case tn = t = constant, diagonalize the electron Hamiltonian by

Fourier transform n k , and determine the spectrum (k) of electronic
excitations.

In general, the amplitude of electron tunneling tn depends on the relative

diosplacement of the nearest neighboring atoms un un+1 . Let us expand

1537
 tn as a function of (un un+1 ) to the first order: tn = t + (un un+1 )t0 .
When substituted into the electron Hamiltonian, the second term gives
the following Hamiltonian:

X
Helph = t0 +
(un un+1 )n+1 n + h.c.
n=

Rewrite this last Hamiltonian in terms of phonon and electron operators

ak and k and their conjugates. Comment on conservation of momentum.
This Hamiltonian describes the electron-phonon interaction. Phonons are
excitations of the lattice or lattice vibrations.

19.9.9 Second Quantized Operators

Write down the second-quantized form of the following first-quantized operators
describing N particles in both a position space basis ((~r)) and a momentum
space basis (ak ):
P
(a) particle density at ~r: (~r) = ` (~r ~r` )
P
(b) total number of particles: `1=N

(c) charge current density at ~r: ~je (~r) = 2m

e
P
p` (~r ~r` ) + (~r ~r` )~
` [~ p` ]
P
(d) magnetic moment density at ~r: m(~ ~ r) = (g/2) ` ~` (~r ~r` )

19.9.10 Working out the details in Section 19.8

(1) Check that the thin spectrum of a harmonic crystal is indeed thin, that
is,

(a) Show that only the lowest N total momentum states are not expo-
nentially suppressed in the symmetry broken
wavefunction (19.217).
(This result implies that only the lowest N total momentum states
contribute to the symmetry broken wavefunction, and these states
all become degenerate in the thermodynamic limit).
(b) Calculate the partition
function of the thin spectrum states and show
that it scales as N , so that the contribution of these states to the
free energy vanishes in the thermodynamic limit.
(2) Show the noncommutativity of the limits in Eq.(19.221) explicitly, by
going through the following steps:
(a) Formulate the Hamiltonian of Eq.(19.216)
p in terms of the boson
+
raising
p and lowering operators b = C/(2~)(x tot (i/C)ptot ) and
b = C/(2~)(xtot + (i/C)ptot ) where C is some constant.
(b) Choose C such that the Hamiltonian becomes diagonal and find its
ground state.

1538
 (c) Evaluate the limits of Eq. (19.221) by expressing x2tot in terms of
boson operators and taking the expectation value with respect to the
SB
ground state of Hcoll .
(3) Work out the Bogoliubov transformation of Eqs.(19.212) and (19.213) ex-
plicitly.

(a) Write the Hamiltonian of Eq. (19.212) in terms of the transformed

bosons k = cosh (uk )bk + sinh (uk )b+
k.
(b) Which value should be chosen for uk in order for the Bogoliubov
transformation to yield the diagonal Hamiltonian of Eq. (19.213)
[Answer: tanh (2uk ) = Bk /Ak ].

1539
1540
Chapter 20
Relativistic Wave Equations

Electromagnetic Radiation in Matter

20.1 Spin 0 particles: Klein-Gordon Equation

A classical nonrelativistic free particle has an energy-momentum relation E =
p2 /2m. Under a Galilean transformation to a new coordinate system traveling
with ~v with respect to the first system, we have
~r 0 = ~r + ~v t , t0 = t (20.1)
which gives
d~r 0 d~r
~u 0 = = + ~v = ~u + ~v
dt0 dt
m0 = m
p~ 0 = p~ + m~v
We then have
d~p0 d~p
F~ 0 = 0 = = F~ (20.2)
dt dt
which implies the invariance of the form E = p2 /2m, i.e.,
p~ 02
Z Z
0 ~ 0 0 1
E = F d~r = 0 p~ 0 d~
p0 =
m 2m0
Z Z 2
1 p~
E = F~ d~r = p~ d~p=
m 2m
and
p~ 02 p + m~v )2
(~ 1
E0 = 0
= = E + p~ ~v + mv 2 (20.3)
2m 2m 2
Thus, the final Galilean transformation relations are
1
E 0 = E + p~ ~v + mv 2 , p~ 0 = p~ + m~v (20.4)
2

1541
which, as shown, leaves the quadratic form E = p2 /2m invariant, i.e., if E =
p~2 /2m, then E 0 = p~ 02 /2m0 and we derive the transformation rules for E and p~
from that condition.

The non-relativistic Schrodinger equation for the free particle then follows from
the standard identifications
~
E i~ , p~
t i
p2
H = E , H=
2m
which gives
~2 2
i~ = (20.5)
t 2m
It is clear from the form of the Schrodinger equation for a free particle that the
equation cannot be invariant under Lorentz transformation (Lorentz covariant),
i.e., the time derivative is first order and the space derivatives are second order.

20.1.1 How to find correct form of relativistic wave equa-

tion?
Before proceeding let us recall some results from special relativity. Components
of spacetime four-vectors will be labeled by Greek indices and the components
of spatial three-vectors will be labeled by Latin indices and we will use Einstein
summation convention.

Starting from x (s) = (ct, ~x) = (x0 , ~x), the contravariant 4-vector representa-
tion of the worldline as a function of the proper time s, we first obtain the
4-velocity, i.e.,
dx (s) dx (s)
 0 
dx d~x
x (s) = = 1 0 = , = (1, ~v /c) (20.6)
ds dx
dx0 dx0

where
1 d~x d~x
=q , ~v = =c 0 (20.7)
1 v2 dt dx
c2

The 4-momentum vector is then given by

p = mcx (s) = m (c, ~v ) = (E/c, p~) (20.8)

where we have used the fact that

E
p0 = = mc , m = rest mass (20.9)
c
This says that the energy E and momentum p~ transform as the components of a
contravariant 4-vector and we know that the square of any 4-vector is invariant.

1542
The metric tensor defined by

1 0 0 0
0 1 0 0
g = (20.10)
0 0 1 0
0 0 0 1

allows the construction of the covariant components(using Einstein summation

convention)  
E
p = g p = , ~
p (20.11)
c
We can then calculate the invariant square or the invariant scalar product of
the 4-momentum with itself as
E2
p p = g p p = p~2 = m2 c2 (20.12)
c2
We therefore have the relativistic energy momentum relation

1/2
E = p~2 c2 + m2 c4 (20.13)

If we use this expression to construct a new wave equation by operator substi-

tution we would have

1/2
i~ = ~2 c2 2 + m2 c4 (20.14)
t
Although the energy formula is now relativistically correct, the time and space
derivatives still do not appear symmetrically. In fact, a Taylor expansion of
the square root gives infinitely high-order derivatives leading to a very difficult
mathematical equation to deal with.

This fact, in itself, is not a valid reason for rejecting the equation.

There are, however, strong physical reasons for rejecting this equation. The
equation says that the momentum space amplitude
Z
p~ (t) = d3 r ei~p~r/~ (~r, t) (20.15)

obeys the equation

p~ (t)
1/2
i~ = p2 c2 + m2 c4 p~ (t) (20.16)
t
If we Fourier transform both sides back to position space we get
Z
(~r, t)
i~ = d3 r0 K(~r ~r 0 )(~r 0 , t) (20.17)
t

1543
where
d3 p i~p(~r~r 0 )/h 2 2
Z

1/2
K(~r ~r 0 ) = 3
e p c + m2 c4 (20.18)
(2~)
This equation for (~r, t) is nonlocal, which means that the value of the integral
at ~r depends on the value of at the other points vecr 0 . The function K(~r ~r 0 )
is large as long as ~r 0 is within a distance
~
= Compton wavelength (20.19)
mc
from ~r. As a consequence of the nonlocality, the rate of change in time of at
the spacetime point (~r, t) depends on the values of at points (~r 0 , t) outside
the light cone centered on (~r, t).

If we construct a wave packet localized well within a Compton wavelength of

the origin, then the packet will be nonzero an arbitrarily short time later at
points as distant as the Compton wavelength.

Thus, this equation leads to violations of relativistic causality when used to

describe particles localized to within more than a Compton wavelength, which
is unacceptable.

Instead, we will start from an equation involving E 2 , i.e., we have

 2

E 2 = i~ = p~2 c2 + m2 c4 = ~2 c2 2 + m2 c4



t
1 2
  mc 2 
2 = 2 2 (20.20)
~ c t
which looks like a classical wave equation with an extra term of the form
 mc 2
(20.21)
~
It is called the Klein-Gordon equation. In 4-vector notation it looks like
  mc 2 

+ = 0 , = (20.22)
~ x
The equation can be generalized in a relativistically invariant way to include the
coupling of charged particles to the electromagnetic field by the substitutions
(corresponding to minimal coupling we discussed earlier in Chapter 8)
~ ~ e~
i~ i~ e , A (20.23)
t t i i c
to obtain
 2  2 !
1 ~ e~
i~ e(~r, t) (~r, t) = A(~r, t) + m2 c2 (~r, t) (20.24)
c2 t i c

1544
The Klein-Gordon equation has several unusual features.

First, it is second-order in time (space and time derivatives are now the same
order). This means we need to specify twice as much initial information (the
function and its derivative) at one time to specify the relativistic solution as
compared to the nonrelativistic solution which only required specification of the
function at one time. .

This will mean that the equation has an extra degree of freedom. We will see
shortly that this extra degree of freedom corresponds to specifying the charge
of the particle and that the Klein-Gordon equation actually describes both a
particle and its antiparticle together. .

Second, since the equation is second order in time, the functions

= ei(~p~rEt)/~ (20.25)

satisfy the free particle equation with either sign of E, i.e.,

1/2
E = p~2 c2 + m2 c4 (20.26)

The Klein-Gordon equation has negative energy solutions for a free particle!
For these solutions when we increase the magnitude of the momentum p~, then
the energy of the particle decreases! As we will see later, these negative energy
solutions are real and will correspond to antiparticles, while the positive energy
solutions will be particles.

In nonrelativistic Schrodinger theory we were able to interpret as a positive

probability density that was conserved in time (no sinks or sources of probability
in nonrelativistic Schrodinger theory). Let us see what happens in the case of
the Klein-Gordon equation.

For the Klein-Gordon equation

Z
d3 r (20.27)

changes in time and thus, we cannot interpret (~r, t)(~r, t) as being the prob-
ability of finding a particle at ~r at time t.

We can, however, construct a different conserved density as follows. We write

  mc 2 

+ =0 (20.28)
~

and   mc 2 
+ = 0 (20.29)
~

1545
which give(subtracting)
= 0 (20.30)

( ) = 0 (20.31)
Expanding these expressions, we have the continuity equation

+ ~j = 0 (20.32)
t
where

 
i~
(~r, t) = (20.33)
2mc2 t t
~
~j(~r, t) = ( ) (20.34)
im
We have inserted a multiplicative constant so that the current density vector
~j(~r, t) is identical to the nonrelativistic case. Because this density (~r, t) satis-
fies a continuity equation, its integral over all space does not change in time.
Clearly, however, it is not necessarily positive. In particular, < 0 for a nega-
tive energy free particle eigenstate.

This means that we cannot interpret this new (~r) as being the particle (prob-
ability) density at ~r and we cannot interpret ~j(~r) as a particle current.

The interpretation that will eventually emerge is that for charged particles e(~r)
represents the charge density at ~r, which can have either sign and e~j(~r) repre-
sents the electric current at ~r.

20.1.2 Negative Energy States and Antiparticles

How do we interpret the Klein-Gordon equation and its solutions?

Consider a free particle at rest, i.e., p~ = 0. The wave function for the positive
energy solution is
2
(~r, t) = eimc t/~ (20.35)
where the energy of a particle at rest is E = mc2 . The density for this state is
(~r, t) = +1.

Now make a Lorentz transformation to a new frame moving with velocity ~v

with respect to the particle at rest. The particle now appears to have a velocity
~v in this new frame. It, therefore, has
momentum = p~ = m~v and energy = E = mc2 (20.36)
where
1
=p (20.37)
1 v 2 /c2

1546
This result follows because

p x 0 = Lorentz scalar
= mc2 t in the rest frame
= p~ ~r 0 Et 0 in the moving frame

The new wave function

0
Et0 )/~ 0 0
(~r 0 , t0 ) = ei(~p~r = ei(~p~r Ep
~ t )/~
(20.38)

is the result we expect for a particle of momentum p~ and energy Ep~ . If we

calculate the density for this wave function we get
Ep~
(~r 0 , t0 ) = (20.39)
mc2
and the current is
2
~j(~r 0 , t0 ) = p~ = p~c (~r 0 , t0 ) = ~v (~r 0 , t0 ) (as expected) (20.40)
m Ep~

where
p~c2
~v = (20.41)
Ep~
We see that (~r, t) transforms like Ep~ or as the time component of a 4-vector,
which makes physical sense. Since a unit volume in the rest frame appears
smaller by a factor p
1/ = 1 v 2 /c2 (20.42)
when observed from the moving frame, a unit density in the rest frame will
appear as a density
1 Ep~
= (20.43)
mc2
in a frame in which the particle is moving.

What about the negative energy solutions? For a particle at rest we have, in
this case,
2
(~r, t) = eimc t/~ (20.44)
where the energy of this particle at rest is E = mc2 .

The density for this state is (~r, t) = 1.

It turns out that one way to interpret a state with a negative particle density
is to say that it is a state with a positive density of antiparticles.

We will make the interpretation that a particle at rest with energy E = mc2
is actually an antiparticle with positive energy E = mc2 . As we shall see,

1547
this interpretation of negative energy states will lead to a consistent theoretical
picture that is confirmed experimentally.

In a Lorentz frame traveling with velocity ~v with respect to the antiparticle,

the wave function is
0
Et0 )/~ 2
(~r 0 , t0 ) = ei(~p~r = eimc t/~
(20.45)

where
momentum = p~ = m~v and energy = E = mc2 (20.46)
In this new frame the particle has velocity ~v , momentum p~ and energy Ep~ . The
wave function, however, describes a particle of energy Ep~ and momentum ~ p.

The density in the moving frame is

Ep~
(~r 0 , t0 ) = (20.47)
mc2
and the current is
2
~j(~r 0 , t0 ) = p~ (~r 0 , t0 ) = p~c (~r 0 , t0 ) (20.48)
m Ep~

Thus, an antiparticle moving with velocity ~v has associated with it a current

moving in the opposite direction, i.e., a flow of antiparticles in one direction is
equivalent to a flow of particles in the opposite direction.

For a charged particle e(~r, t) is the charge density. It is positive for a free
particle with e > 0 and negative for a free antiparticle, which has opposite
charge to the particle.

The quantity e~j(~r, t) is the electric current of the state . For a particle the
electric current is in the direction of the particle velocity. For the antiparticle
with e < 0, the electric current is opposite to the velocity.

This says that the interpretation of the negative energy solutions as antiparticles
is consistent with the interpretation of the density as a charge density and ~j
as an electric current.

Is this interpretation consistent with the way charged particles interact with the
electromagnetic field?

The Klein-Gordon equation with an electromagnetic field present is given by

 2  2 !
1 ~ e~
i~ e(~r, t) (~r, t) = A(~r, t) + m2 c2 (~r, t) (20.49)
c2 t i c

1548
Taking the complex conjugate we have
 2  2 !
1 ~ e ~ r, t) + m2 c2 (~r, t)
i~ + e(~r, t) (~r, t) = + A(~
c2 t i c
(20.50)
These equations say that if (~r, t) is a solution to the Klein-Gordon equation
with a certain sign of the charge, then (~r, t) is a solution of the Klein-Gordon
equation with the opposite sign of the charge and the same mass.

Thus, the relativistic theory of a spin zero particle predicts the existence of its
antiparticle with the opposite charge and same mass, i.e., the theory contains
solutions for both particles and antiparticles.

Relativistic invariance requires

the existence of antiparticles

The complex conjugate of a negative energy solution is a positive energy solu-

tion with the opposite sign of the charge. The operation of taking the complex
conjugate of the wave function will be called charge conjugation. Charge conju-
gation changes particles into antiparticles and vice versa. If we label quantities
calculated with the complex conjugate wave function by a subscript c , we find
(~r, t) = c (~r, t) , ~j(~r, t) = ~jc (~r, t) (20.51)
as expected.

The solutions are normalized by the requirement that the total associated charge
equals 1 unit, i.e.,
Z Z
d3 r(~r, t) = +1 = d3 rc (~r, t) (20.52)

This normalization is conserved in time and invariant under a Lorentz transfor-

mation.

20.2 Physics of the Klein-Gordon Equation

We first transform the Klein-Gordon equation into two equations, each first
order in time.

We define
 
ie
0 (~r, t) = + (~r, t) (~r, t) (1st - order equation #1) (20.53)
t ~
We then have
   2
ie ie
+ (~r, t) 0 (~r, t) = + (~r, t) (~r, t) (20.54)
t ~ t ~

1549
Now using the Klein-Gordon equation we have
2 2 !
m2 c4
 
ie ie ~
+ (~r, t) (~r, t) = c2 + A(~r, t) 2 (~r, t) (20.55)
t ~ ~c ~

so we get
 
ie
+ (~r, t) 0 (~r, t)
t ~
2 !
2 4

ie ~ r, t) m c
= c2 + A(~ (~r, t) (1st - order equation #2)
~c ~2
(20.56)

These two new first-order equations involve the two functions 0 (~r, t) and
(~r, t).

Now define the linear combinations

   
1 i~ 0 1 i~ 0
= + , = (20.57)
2 mc2 2 mc2

Substitution then gives the more symmetric equations

   2
1 ~ e~
i~ e = A ( + ) + mc2 (20.58)
t 2m i c
   2
1 ~ e~
i~ e = A ( + ) + mc2 (20.59)
t 2m i c

Now define a two-component wave function

 
(~r, t)
(~r, t) = (20.60)
(~r, t)

and three 2 2 matrices

     
0 1 0 i 1 0
1 = , 2 = , 3 = (20.61)
1 0 i 0 0 1

The two symmetric equations can then be combined into the single equation
" 2 #
~ e~ 2
i~ = A (3 + i2 ) + mc 3 + e (20.62)
t i c

This equation is completely equivalent to the original Klein-Gordon equation

where
mc2
= + , 0 = ( ) (20.63)
i~

1550
The internal degree of freedom represented by these two components is the
charge of the particle(one component represents the particle and the other the
antiparticle).

Using the two component equation, we can write the density as

2 2
(~r, t) = || || = + 3 (20.64)

where + = ( , ). A very simple expression. The current density, however,

becomes less transparent

~j(~r, t) = ~ + 3 (3 + i2 ) (+ )3 (3 + i2 )
 
2im
eA~
+ 3 (3 + i2 ) (20.65)
mc
The normalization condition becomes
Z
d3 r + 3 = 1 (20.66)

The scalar product between two such wave functions and |P si0 is defined by
Z
h | 0 i = d3 r + (~r, t)3 0 (~r, t) (20.67)

Finally, the wave equation is of the form

i~ = H (20.68)
t
where the Hamiltonian is
 2
~ e~
H = A (3 + i2 ) + mc2 3 + e (20.69)
i c

Since (3 + i2 )+ = 3 i2 , we find that H + 6= H, which seems to indicate that

H is not Hermitian. It is Hermitian, however, when we use the proper scalar
product definition of hermiticity, i.e.,
  h  i
h| H |0 i = h0 | H |i (20.70)

This relation requires that

3 H + 3 = H (20.71)
All required properties of Hermitian operators, i.e., real eigenvalues and so on,
follow from the scalar product definition so that is all that is actually required.

Under the charge conjugation operation


c = 1 (20.72)

1551
which is the form of the charge conjugation operation in two-component lan-
guage.

We can now see the physical meaning of charge conjugation. Using

 
~ ~
p~ = = = ~ p (20.73)
i i
and 3 + i2 = real matrix, we find that
h e ~ i2
H (e) = ~
p A (3 + i2 ) + mc2 3 + e (20.74)
c
We then have
h e ~ i2
1 H (e)1 = p~ + A (3 + i2 ) mc2 3 + e = H(e) (20.75)
c
This means that, if solves the equation

i~ = H(e) (20.76)
t
we have

i~ = H (e) = 1 H(e)1 (20.77)
t
Multiplying by 1 we get
c
i~ = H(e)c (20.78)
t
which is the two-component statement of the fact that c solves the Klein-
Gordon equation with the opposite sign of the charge.

What can we say about the two-component solutions for free particles and
antiparticles?

The wave function of a free particle (positive energy solution) of momentum p~

(normalized to 1) is given by
s
(+) mc2 i(~p~rEp~ t)/~
p~ (~r, t) = e (20.79)
Ep~

where p
Ep~ = p2 c2 + m2 c4 (20.80)
Using  
0 ie
(~r, t) = + (~r, t) (~r, t) (20.81)
t ~
and    
1 i~ 0 1 i~ 0
= + , = (20.82)
2 mc2 2 mc2

1552
we find (in two-component language)
(+) (+)
p~ (~r, t) = p~ ei(~p~rEp~ t)/~ (20.83)

(+)
where the two-component vector p~ is given by

mc2 + Ep~
 
(+) 1
p~ = p (20.84)
2 Ep~ mc2 mc2 Ep~

In a similar manner, we can write for the negative energy solutions (free an-
tiparticles) s
() mc2 i(~p~rEp~ t)/~
p~ (~r, t) = e (20.85)
Ep~
() ()
p~ (~r, t) = p~ ei(~p~rEp~ t)/~ (20.86)

mc2 Ep~
 
() 1 (+)
p~ = p = 1 p~ (20.87)
2 Ep~ mc2 mc2 + Ep~
We note that in the nonrelativistic limit
1/2
p2 p2
p   
Ep~ = p2 c2 + m2 c4 = mc2 1 + 2 2 mc2 1 + (20.88)
m c 2m2 c2
2mc2

mc2 Ep~ (20.89)
p2 /2m

so that
v 2 /4c2
   
(+) 1 ()
p~ = , p~ = (20.90)
v 2 /4c2 1
This shows that in the nonrelativistic limit

is v 2 /c2 times for a particle (20.91)

If we drop , then satisfies the nonrelativistic Schrodinger equation with the

constant mc2 included in the energy.

Similarly, dropping in the antiparticle solution shows that satisfies the

nonrelativistic Schrodinger equation for the opposite charge with the constant
mc2 included in the energy.

The particle and antiparticle solutions are orthogonal in the sense that
(+) () () (+)
p~ 3 p~ = 0 = p~ 3 p~ (20.92)

which should be the case since they represent different energy eigenstates of the
same Hamiltonian.

1553
The free particle solutions form a complete set since any wave function can be
expanded as a linear combination of the free particle and antiparticle solutions.

We first write as a Fourier transform

d3 p i~p~r/~ p~
Z  
(~r, t) = e (20.93)
(2~)3 p~

(+) ()
Since the two vectors p~ and p~ are linearly independent, we can write
 
p~ (+) ()
p~ (t) = = up~ (t)p~ + v~p (t)~
p (20.94)
p~

Good reasons for the choice of ~

p and will appear shortly.

Substituting we get

d3 p i~p~r/~ h
Z i
(+) ()
(~r, t) = e u p
~ (t) p
~ + v ~p (t) ~p
(2~)3
Z 3
d p h
(+) i~p~
r /~ () i~ p~
r /~
i
= u p
~ (t) p
~ e + v p
~ (t) p
~ e (20.95)
(2~)3

where a change of variables was made in the second term. From the form of this
result, up~ (t) is the amplitude for a particle in the state to have momentum p~
and positive charge and vp~ (t) is the amplitude for a particle in the state to
have momentum p~ and negative charge.

()
Using the orthonormality of p~ we get
Z
(+)+
up~ (t) = d3 rp~ ei~p~r/~ 3 (~r, t) (20.96)
Z
()+
vp~ (t) = d3 rp~ ei~p~r/~ 3 (~r, t) (20.97)

The normalization integral for then becomes

d3 p 
Z 
2 2
3
|up~ | |vp~ | = 1 (20.98)
(2~)

This says that there is no restriction on the magnitude of either up~ or vp~ . Only
the integral of the difference(above) is fixed.

Physically, we can then say that one can have a state with an arbitrarily large
amplitude for finding a particle with a certain momentum, which is the first
indication that we are dealing with bosons or that spin zero particles must be
bosons.

1554
We can write some expectation values in this formalism, i.e.,

p2
H0 = (3 + i2 )2 + mc2 3 = Kineticenergy (20.99)
2m
d3 p
Z Z  
+ 3 2 2
(~r)3 H0 (~r)d r = E p
~ |u p
~ | + |v p
~ | (20.100)
(2~)3

and
~
p~ = = momentum (20.101)
i
d3 p 
Z   Z 
+ ~ 3 2 2
(~r)3 (~r)d r = p
~ |u p
~ | + |v p
~ | (20.102)
i (2~)3

20.3 Free Particles as Wave Packets

A wave packet formed from the positive energy solutions is given by

d3 p
Z
(+)
(+) (~r, t) = up~ ei(~p~rEp~ t)/~ p~ (20.103)
(2~)3

Let us assume that up~ is peaked about p~ = p~ 0 . Then, using arguments similar to
our earlier discussions on stationary phase, the center of the wave packet moves
with a group velocity
p~ 0 c2
~vg = (p~ Ep~ )p~=~p 0 = (20.104)
Ep~ 0
and similarly for a free wave packet made of the negative energy solutions for
antiparticles.

Can we construct a free particle wave packet perfectly localized at the origin?
It would have the form  
a
(~r) = (~r) (20.105)
b
We then have
Z
(+)+ i~
p~
up~ = d3 rp~ e r /~
3 (~r)
Z    
3 (+)+ i~
p~
r /~ a (+)+ a
= d rp~ e 3 (~r) = p~ 3
b b
+
mc2 + Ep~
     
1 1 0 a
= p
2 Ep~ mc 2 mc2 Ep~ 0 1 b
Ep~ (a + b) + mc2 (a b)
= p (20.106)
2 Ep~ mc2

1555
and similarly

Ep~ (a + b) mc2 (a b)
 
()+ a
vp~ = p~ 3 = (20.107)
b
p
2 Ep~ mc2

Looking at these results we can see that independent of the choice of a and
b, the wave packet will always have both particle and antiparticle components.
This means that it is impossible to construct a perfectly localized wave packet
from positive energy solutions alone.

Suppose that we take a general wave packet made up of positive energy solutions
and try to squeeze it(make it more localized) with real-world devices such as
collimators. To see what might happen we multiply the wave packet by the
position operator ~r. We then have

d3 p
Z
(+)
~r(+) (~r, t) = up~ (t)p~ ~rei~p~r/~
(2~)3
d3 p
Z
(+) ~
= up~ (t)p~ p~ ei~p~r/~ (20.108)
(2~)3 i

Integrating by parts we have

d3 p
Z
(+)
~r(+) (~r, t) = (i~p~ up~ (t))p~ ei~p~r/~
(2~)3
d3 p
Z
(+)
+ up~ (t)(i~p~ p~ )ei~p~r/~ (20.109)
(2~)3

Using
() p~c2 ()
p~ p~ = (20.110)
2Ep~2 p~
we get
~r(+) (~r, t) = ~r+ (+) (~r, t) + ~r (+) (~r, t) (20.111)
where
d3 p
Z
(+)
~r+ (+)
(~r, t) = (i~p~ up~ (t))p~ ei~p~r/~ (20.112)
(2~)3
d3 p pc2 () i~p~r/~
Z
(+) i~~
~r (~r, t) = u p
~ (t) e (20.113)
(2~)3 2Ep~2 p~

This says that multiplying a wave packet of positive energy states by the position
operator mixes in negative energy solutions, i.e.,

~r+ generates positive energy solutions while ~r generates negative

energy solutions or changes free particles in free antiparticles and
vices versa

1556
The same result occurs for any function of the position operator.

Suppose that ~r(+) (~r) = ~r0 (+) (~r), i.e., it is an eigenstate of ~r with eigenvalue
~r0 . This says that
i~p~ up~ = ~r0 up~ up~ = ei~p~r0 /~ (20.114)
and the state
d3 p i~p(~r~r0 )/~ (+)
Z
~r+0 (~r) = e p~ (20.115)
(2~)3
is an eigenstate of ~r+ .

This eigenstate is not normalizable. It is large over a region of space within

~/mc (a Compton wavelength) of ~r0 or, in other words, the theory with positive
energy solutions cannot describe particles localized to a region smaller than a
Compton wavelength.

The presence of the ~r part in the position operator says that putting a wave
packet made from positive energy solutions (a particle) through a potential
(~r) (which multiplies by functions of r) causes the creation of antiparticles
and because charge must be conserved, creates new particles also.

Thus, the relativistic spin-zero theory of the Klein-Gordon equation has built
into it the mechanism of particle-antiparticle production by external potentials.

An example of this phenomenon is Kleins paradox. Suppose we have a beam of

positively charged particles with momentum p hitting an electrostatic potential
barrier of height e from the left as shown in Figure 20.1 below.

Figure 20.1: Electrostatic Potential Barrier - Klein Paradox

The solution follows the same lines as the nonrelativistic problem. For x < 0
we have
(x) = aeipx/~ + beipx/~ , Energy = Ep (20.116)
This corresponds to incident and reflected waves. For x > 0, the Klein-Gordon
equation is

2 (x)
(Ep V )2 (x) = ~2 c2 + mc2 (x) , V = e (20.117)
x2

1557
The solution takes the form (x) = deikx where substitution gives

(Ep V )2 = ~2 c2 k 2 + m2 c4 (20.118)

We have the boundary conditions at x = 0 (since potential only has a finite

discontinuity)

(x) , continuous (20.119)
x
Note that 0 is given by
 
0 ie
(~r, t) = + (~r, t) (~r, t) (20.120)
t ~

is not continuous at x = 0.

We obtain
p ~k 2p
b= a , d= a (20.121)
p + ~k p + ~k
We consider three cases:

1. If Ep > V + mc2 , then the particle can pass the over the barrier and the
results are identical to the nonrelativistic case, i.e., part of the wave is
reflected and part is transmitted.

2. If we have a stronger potential such that Ep + mc2 > V > Ep mc2 , then
k must be imaginary so that the wave function goes to zero as x .
We then have p
m2 c4 (Ep V )2
k = i = (20.122)
~c
and the wave is totally reflected at the barrier. The charge density on the
right (x > 0) is given by

Ep V 2
(x) = |d| e2x (20.123)
2mc2
For Ep > V , there exists a positive, exponentially decaying charge density
to the right of the barrier. For Ep < V , however, the density is negative
(remember it is a beam of positive particles). We reflect positively charged
particles from the barrier and find negative particles inside the barrier.

3. We make the potential even stronger so that V > Ep + mc2 . Nonrela-

tivistically it would be even more impossible for the particles to pass over
the barrier. In the relativistic case, however, k is real again, This says
that once again there is a particle current to the right of the barrier. The
group velocity of the waves for x > 0 is

Ep
vg = (20.124)
(~k)

1558
 Using (Ep V )2 = ~2 c2 k 2 + m2 c4 we get

Ep ~c2 k
(Ep V ) = ~c2 k vg = (20.125)
k Ep V

Since Ep V < 0, we must have k < 0 (negative!) in order to have a

wave(packet) traveling from the barrier towards positive x.

This says that the reflection coefficient b/a is greater than one, i.e., more
wave is reflected than is incident! In addition, the charge density on the
right is
Ep V 2
(x) = |d| < 0 (20.126)
2mc2
and the current on the right is negative.

One possible explanation is to say that the incident particle induces the
creation of particle-antiparticle pairs at the barrier. The created antiparti-
cles, having the opposite charge, find x > 0 a region of attractive potential
and thus travel towards the right, which explains the negative current on
the right. The created particles travel to the left and together with the
incident particles (wave) which are(is) totally reflected, they add up to an
outgoing current on the left that is larger than the incident current.

The total outgoing current on the left and right equals the incident current
since total charge must be conserved.

This pair creation solution does not violate conservation of energy. The
energy of a created particle on the left is Ep . The energy of a created
antiparticle on the right is ~2 c2 k 2 + m2 c4 V since the electrostatic po-
tential energy has the opposite sign for a particle of opposite charge.

Adding the two energies we get Ep + ~2 c2 k 2 + m2 c4 V = 0, i.e., it takes
zero energy to create a particle-antiparticle pair. This happens because
the potential V is so large that the energy of the antiparticle on the right
is not only less than mc2 but is negative.

20.4 Bound State Problems

We now study the bound states of spin zero relativistic particles in a static
potential (~r). For a positively charged particle with energy E the bound state
wave function is
(~r, t) = eiEt/~ (~r) (20.127)
and the charge density of the bound state particle is
e [E e(~r)] 2
e(~r) = |(~r)| (20.128)
mc2

1559
This says that in regions where E > e(~r), which includes classically accessible
regions, the charge density is positive. But, in regions where E < e(~r), the
charge density is negative. The way to think about this is to say the particle in
the potential is a linear combination of free particle and free antiparticle states.

Because we are considering an electrostatic potential, the positively charged

parts will be found mainly in regions of smaller e(~r) while the negatively
charged parts will be found in regions of larger e(~r). A relativistic particle
seems to have an internal structure that can be polarized by and electric field.

Alternatively, we might say that the potential produces particle-antiparticle

pairs in the vacuum, The positively charged particles are attracted to regions of
smaller e(~r) while the negatively charged particles are attracted to regions of
larger e(~r). We say that the electric potential has polarized the vacuum. This
polarization modifies the effective potential felt by the bound particle.

This interaction cannot be taken into account in the present one-particle rela-
tivistic theory(requires quantum field theory).

We now turn to the problem of a spin zero particle bound in a Coulomb poten-
tial. An example is a bound to a nucleus. We have
Ze2
e(~r) = (20.129)
r
which leads to the Klein-Gordon equation
" 2 #
Ze2 2 2 2 2
E+ + ~ c mc (~r) = 0 (20.130)
r

Since this is a central potential, we can assume that the eigenstates have definite
values of total orbital angular momentum. We then have
 2
1 2 `(` + 1) (Z)2 2Ze2 E
   
E 2 2 2
m c + ~ r + (r) = 0
c2 r r2 r2 r c
(20.131)
or
1 2 `(` + 1) (Z)2
 2
E m2 c4
 
2ZE
r+ (r) = 0 (20.132)
r r2 r2 ~cr ~2 c2
where
e2
= = fine structure constant (20.133)
~c
Now we define
= Z , ` 0 (` 0 + 1) = `(` + 1) 2
2E 4(m2 c4 E 2 )
= , = 2 , = r
~c ~2 c2

1560
and we get
d2 ` 0 (` 0 + 1)
 
2
+ 1 + () = 0 (20.134)
d(/2)2 /2 (/2)2
which is identical to the radial equation for the nonrelativistic Coulomb problem
for the function u = (). The difference is that ` 0 is not necessarily an
integer (remember that it is an integer in the nonrelativistic problem), which
causes the orbits of the relativistic Coulomb (Kepler) problem to no longer be
closed, i.e., the orbits precess. This also means that the extra degeneracy of
the nonrelativistic problem which causes the energy to be independent of ` is
broken in the relativistic problem. We now solve this equation in the standard
way. For
0
0 () `
() e/2
Therefore, we guess a solution of the form
 `0 +1
u = () = e/2 w(/2) (20.135)
2
The solution method is identical to the nonrelativistic hydrogen atom. We get
a power series which must terminate (so that the solution is normalizable) when
= N + ` 0 + 1 , N = 0, 1, 2, 3, ......
1/2
2 mc2

E = mc2 1 + 2 E= v (20.136)
u
u1 + "
2
#2
r
t 1 1
N+ 2 + (`+ 2 )2 2

If we define the principal quantum number n = N + ` + 1 = integer, then we

have
mc2
En` = v (20.137)
u
u1 + " 2
r #2
t 1 1 2 2
n(`+ 2 )+ (`+ 2 )

The principal quantum number has the possible values n = 1, 2, 3, ...... For
a given n the possible values of the total orbital angular momentum are ` =
0, 1, 2, 3, ...., n 1.

The degeneracy that was present in the nonrelativistic theory with respect to
orbital angular momentum ` is clearly removed.

If we expand the energy in a power series in the fine structure constant (or
) we get
2
 
2 1 1 3
En` = mc Ry 2 Ry 3 + O(Ry 4 ) (20.138)
n n ` + 21 4n

1561
The first term is the rest energy. The second term is the nonrelativistic Rydberg
formula. The third term is the relativistic correction due to using the relativistic
form of the kinetic energy, which as we saw earlier in Chapter 10 took the form

p4
Hrel = (20.139)
8m3 c2
It is this correction that removes the degeneracy in `, i.e.,

4 2 n 1
En,`=0 En,`=n1 = Ry (20.140)
n3 2n 1
As we shall see later when we derive the Dirac equation, there are more correc-
tions to this formula due to the fact that the electron has spin = 1/2.

20.4.1 Nonrelativistic Limit

The Klein-Gordon equation in the presence of an electromagnetic field is
 2  2 !
1 ~ e~ 2 2
i~ e(~r, t) (~r, t) = A(~r, t) + m c (~r, t)
c2 t i c
(20.141)
and using  
(~r, t)
(~r, t) = (20.142)
(~r, t)
we have as earlier
   2  2
1 ~ e~ 2 1 ~ e~
i~ e = A + mc + A (20.143)
t 2m i c 2m i c
   2
1 ~ e~
i~ e = A ( + ) mc2 (20.144)
t 2m i c

Remember that in the nonrelativistic limit the dominant term in the energy will
be mc2 so that we expect the zeroth order equation for to be

i~ = mc2 (20.145)
t
which then implies in the next approximation that
 2
1 ~ e~
= 2 2 A (20.146)
4m c i c

and that satisfies the equation

   2  4
1 ~ e~ 2 1 ~ e~
i~ e = A + mc A
t 2m i c 8m3 c2 i c
(20.147)

1562
The operator on the right-hand side is just the kinetic energy operator
s  2
2 4 2
~ e~
m c +c A (20.148)
i c

expanded to second order in 1/mc2 . This agrees with our earlier result that the
first relativistic correction for a spinless particle is entirely due to the relativistic
modification of the kinetic energy.

For a weak magnetic field B ~ this becomes (to order (v/c)3 ) after much algebra

~2 2 ~2 2 ~2 2
   
2 e ~ ~
i~ = 1+ + (mc + e) BL 1+
t 2m 2m2 c2 2mc 2m2 c2
(20.149)
~
where L is the orbital angular momentum of the particle. The term
~2 2 p2
   
1+ 1 (20.150)
2m2 c2 2m2 c2
represents the relativistic correction to the magnetic moment.

20.5 Relativistic Spin 1/2 Particles - The Dirac

Equation
20.5.1 Lorentz Transformation of Spin
The contravariant and covariant components of the position 4-vector in space-
time are:
x : x0 = ct , x1 = x , x2 = y , x3 = z
x : x0 = ct , x1 = x , x2 = y , x3 = z
The flat spacetime metric tensor is defined by

1 0 0 0

0 1 0 0
g = (g ) = (g ) = (20.151)
0 0 1 0
0 0 0 1
The metric tensor relates covariant and contravariant components by
x = g x , x = g x (20.152)
We also have

1 0 0 0
0 1 0 0
g = g g ,
( ) =
0
(20.153)
0 1 0
0 0 0 1

1563
Under the action of a Lorentz transformation along the z-axis with velocity
v = c, a 4-vector (any type) since it is a first-rank tensor, transforms as

V 0 = V (20.154)

where

0 0

0 1 0 0 1 v
( ) =
0
, =p , = (20.155)
0 1 0 1 2 c
0 0

This corresponds to the standard transformation relations for the position and
momentum 4-vectors

ct0 = (ct z) , x0 = x , y 0 = y , z 0 = (z ct)

E0
   
E 0 0 0 E
= pz , px = px , py = py , pz = pz
c c c

Spin is an angular momentum corresponding to internal degrees of freedom of

the system. This means, as we showed earlier, that spin must have the same
transformation properties as any other angular momentum.

Nonrelativistically, we think of angular momentum as a vector and, in fact,

under a simple spatial rotation it does transform as a vector (as we saw earlier).
Consider, however, the behavior of an orbital angular momentum

~ = ~r p~ Li = ijk xj pk
L (20.156)

under the action of a Lorentz transformation along the zdirection. We find

that
L0z = x0 p0y y 0 p0x = xpy ypx = Lz (20.157)
since the components of vectors orthogonal to the zaxis are unchanged. This
is clearly not the transformation property of a vector.

~ is the product of two vectors and therefore should have the transfor-
In fact, L
mation properties of a second-rank tensor, i.e., as

Q0 = Q (20.158)

Relativistic electrodynamics can be written in terms of a second-rank field tensor

0 1 2 3
1 0 B3 B2
( F ) =
2 B3
(20.159)
0 B1
3 B2 B1 0

1564
as Maxwells equations
F 4
= J
x c

F F F
+ + =0
x x x
where the current density 4-vector is

J = (c, Jx , Jy , Jz ) (20.160)

and the Lorentz force law is

dp q
= p F (20.161)
d m
The transformation rule then says that the fields transform according to the
relations

F 0 01 = 01 = F = 0 1 F = 0 11 F 1 = 0 F 1
= 00 F 01 + 03 F 31 = 1 B2 = (1 ((~v /c) B) ~ 1)

or
~ 1 ) where ~v = vez
01 = (1 + ((~v /c) B) (20.162)
Similarly, we find
~ 2) ,
02 = (2 + ((~v /c) B) 03 = 3
B10 = (B1 ((~v /c) ~)1 ) , B20 = (B2 ((~v /c) ~)2 ) , B30 = B3

We can summarize these results for an arbitrary (direction) Lorentz transfor-

mation applied to a second-rank tensor by

B||0 = B|| , 0|| = || || = component parallel to ~v

 
~
B 0
= B ~ (~v /c) ~
 
~0 = ~ + (~v /c) B ~ = component perpendicular to ~v

Thus, a pure magnetic field in one frame is a mixture of magnetic and electric
fields in the new frame.

Now, under a spatial inversion transformation, we have

~ ~ , ~ B
B ~
~r ~r , p~ ~ ~ L
pL ~

Therefore, an angular momentum(including spin) has the same transformation

properties as the magnetic field.

1565
Since spin,
1
S = ~ (20.163)
2
must transform as an angular momentum, which transforms like a magnetic
field and the magnetic field is part of second-rank tensor with the electric field,
we must conclude that there exists another set of dynamical variables generated
by the internal degrees of freedom of the particle that will be analogous to
the electric field. Do not think of the operator ~ as the standard 2 2 Pauli
matrices; we shall see later that ~ will need to be represented by 4 4 matrices
relativistically.

We define these new variables as i~

/2 where the i/2 factor is chosen for later
convenience. We then have that S~ and i~
/2 or ~ and i~ ~ and ~,
transform as B
i.e.,

0|| = || , ~ 0 = (~ (~v /c) i~

)
i 0|| = i|| , 0 = (i~
i~ + (~v /c) ~ )

and they form a second-rank tensor analogous to F , i.e.,

0 i1 i2 i3
i1 0 3 2
( ) =

i2 3
(20.164)
0 1
i3 2 1 0

We must now investigate the dynamical properties of the new variables ~ and
also ask this question - where have these objects been hiding in all of previous
discussions?

Since spin is an angular momentum, we know its algebraic properties (commu-

tators). In addition, spin generates rotations of the internal degrees of freedom.
Spin commutes with spatial degrees of freedom like ~r and p~ and, thus, so does

~.

Since
~ behaves like a vector under spatial rotations (it is like the electric field
vector), it must have the standard commutation relations with S ~

[i , Sj ] = iijk k [i , j ] = 2iijk k (20.165)

Since ~ is angular momentum, it satisfies the relations

i j = iijk k + ij (20.166)

which must be true in all Lorentz frames, i.e., since 12 = 1, we must have
102 = 1.

Let us now determine all the properties of

~.

1566
For a Lorentz transformation along the zdirection we have
0x = (x + ivy /c)
0y = (y ivx /c)
Squaring 0x we get
x02 = 1 = 2 x2 + iv(x y + y x )/c v/c)2 y2

= 2 1 v/c)2 y2 + 2 (iv(x y + y x )/c)

Since this must be true for all v, the coefficient of v/c must vanish. Thus,
x y + y x = 0 (20.167)
We then have
1 = 2 1 v/c)2 y2



(20.168)
Since
1 = 2 1 v/c)2 y2



(20.169)
we must also have
y2 = 1 (20.170)
These results generalize to the following:
i 6= j i j = j i (20.171)
i=j [i , i ] = 0 (20.172)
Multiplying y0 by x0 we get

x0 y0 = iz0 = iz = 2 (x + ivy /c) (y ivx /c)

iz = 2 x y + (v/c)2 y x + i(v/c)(y y x x

Multiplying x0 by y0 we get

y0 x0 = iz0 = iz = 2 (y ivx /c) (x + ivy /c)

iz = 2 y x + (v/c)2 x y + i(v/c)(y y x x

Adding, we have
(y x + x y ) + (v/c)2 (x y + y x ) = 0
(x y + y x ) = 0 (i j + j i ) = 0 i 6= j
Continuing, we find these other relations
y x = iz = x y i j j i = 2iijk k (20.173)
or summarizing we have
{i , j } = 2ij , [i , j ] = 2iijk k (20.174)
[i , j ] = 2iijk k , {i , j } = 0 , i 6= j (20.175)

1567
So obeys exactly the same algebraic relations as . How do we know that
is not equal to ? If we apply a parity transformation, we find that ~ ~
since angular momentum is unchanged by spatial inversion, i.e., the space-space
components of a second-rank tensor do not change sign under parity. On the
other hand, the time-space components such as the electric field or i~
do change
~ ~
sign, i.e., . So they cannot be the same operator!

Let be the operator that corresponds to the parity transformation in spin

space. Now two successive inversions brings us back to the starting configura-
tion. Remember, however, that the spin representation of rotations is doubled
valued, i.e., a rotation by 2 produces a minus sign. We have a choice of letting
the square of the parity operation include a 2 rotation(about any axis) or not.
This means that we can have 2 = +1 or 1.

In the first case, the eigenvalues of are 1 and in the second case i. We
choose 2 = +1 1 = . The properties under parity become

1~ = ~ ~ = ~ (20.176)
1
~ = ~
~
= ~
(20.177)

We can now construct an explicit matrix representation for the operators ~,

and ~ similar to the two-dimensional Pauli matrix representation in the nonrel-
ativistic case.

Consider the determinant of the matrix for 1 i . Using ~

= ~
we have

det( 1 i ) = det( 1 i ) = det(i ) = (1)N det(i ) (20.178)

where N is the dimension of the matrix representation. However, using the

cyclic property of determinants, i.e.,

det ABC = det BCA = det CAB (20.179)

we get
det( 1 i ) = det( 1 i ) = det(i ) (20.180)
Putting these results together we get

(1)N det(i ) = det(i ) (1)N = 1 N = 2, 4, 6, ..... (20.181)

We have used the fact that det(i ) 6= 0, since i2 = 1.

Now, N = 2 is not possible as we show below.

All 2 2 matrices can be constructed from the set {I, ~ } and [, ~ ] = 0. This
means that would have to commute with all 2 2 matrices. Since ~ would
then have to commute with , we would then violate the relation ~ = ~ .

1568
This means N must be at least as large as 4. This says that a relativistic spin
1/2 particle would have 4 internal states (the nonrelativistic case has 2). This is
similar to the Klein-Gordon case and it will turn out here also that this doubling
signals the appearance of antiparticles.

An explicit representation(not unique) using the 2 2 matrices

       
1 0 0 1 0 i 1 0
I= , 1 = , 2 = , 3 =
0 1 1 0 i 0 0 1

where the the last three matrices are the standard Pauli matrices. It is given
by
     
~ 0 0 ~ I 0
~ = , ~= , = (20.182)
0 ~ ~ 0 0 I

Note that the trace of each of these matrices is zero, which is a general property
of matrices that obey anticommutation relations.

We can make the following physical interpretations of the components of the

tensor .

It follows from earlier discussions that the space-space components, i.e., the
spin operators j , generate (in the spin degrees of freedom) a rotation of the
coordinate system.

This implies that the operators ~ 0 ,

~ 0 , 0 in a spatially rotated frame are given
by the operator relations

~ 0 = R~ R
1
, ~ 0 = R
1
~ R , 0 = R R
1
(20.183)

where
R = ei~n/2 (20.184)
and

n = unit vector in direction of axis of rotation

= angle of rotation

We then assume that time-space components generate a rotation of the space

axes with the time axis, which is a Lorentz transformation and that the operators
in the new frame are given by

~ 0 = Lv ~ L1 ~ 0 = Lv
v , ~ L1 0 1
v , = Lv Lv (20.185)

where
Lv = ei(i~)~/2 = e~ ~/2 L1
v =e
~
~
/2
(20.186)

1569
and

~ = vector in direction of velocity of primed frame
with respect to unprimed frame and of magnitude
v
tanh() =
c
Proof : First,

||0 = L|| L1 = e|| /2 || e|| /2 = e|| /2 e|| /2 || = || (20.187)

where we have used

 
[i , j ] = 2iijk k || , || = 0 (20.188)

This agrees with our earlier result.

Second,
0
= L L1 = e|| /2 e|| /2 = e|| /2 e|| /2 = e|| = e~ ~
(20.189)
where we have used

{i , j } = 0 , i 6= j || , = 0 (20.190)

)2 = 1 we get
~ = , = 1 and (~
Now using

e~ ~ = cosh +
~ sinh (20.191)

This derivation is the analog of

ei~n = cos + i~ n sin (20.192)

Therefore,
0
= e~ ~ = cosh [1 +
~ tanh ] = [1 +
~ tanh ] (20.193)

where we have used

1 1
cosh2 = = = 2 (20.194)
1 tanh2 1 (v/c)2

Using tanh = ~v /c we then get

0
~ tanh ] = [1 +
= [1 + ~ ~v /c] (20.195)

Finally, assuming ~v = vez and using i j = iijk k , i 6= j we get

~v
~v /c) = i
(~ ~ (20.196)
c

1570
so that  
0 ~v
= + i
~ (20.197)
c
which agrees with our earlier result. Thus, ~ transforms correctly. A similar
calculation shows that
~ transforms correctly also and, thus, our interpretation
is correct.

What about the operator ? In a new Lorentz frame we get

0 = Lv L1
v =e
~ ~

(20.198)

since anticommutes with

~ . We then get (as above)

0 = [ (~v /c) ~
] (20.199)

where we have used ~ = ~ . From the form of this transformation relation,

it looks like transforms as the time-component of a 4-vector of which ~ is
the space part.

Some algebra shows that

0
0
= L L1 = (20.200)

since both and anticommute with L and therefore commutes with L.

In addition, we can show that

0 ||0 = || (v/c)
 
(20.201)

Therefore, (,
~ ) does transform like a 4-vector.

This 4-vector is called = (,

~ ). In our earlier notation the space part is
 
I ~
~ = ~
= (20.202)
~ I

~ is anti-Hermitian and 0 is Hermitian.

Some properties

( 0 )2 = 1 , ( i )2 = 1 , i = 1, 2, 3 (20.203)

{ , } = 0 , 6= (20.204)

which are summarized by the relation

{ , } = 2g (20.205)

We also have
i
= [ , ] (20.206)
2

1571
In fact, any 4 4 matrix can be written as a unique linear combination of the
. The set of 16 matrices

I , , , 5 , 5 , where 5 = 0 1 2 3 (20.207)

are linearly independent and complete. All are traceless except for the identity
matrix.

The new operator 5 commutes with the . This implies that it commutes
with i = 0 i and is invariant under a Lorentz transformation. It is not a
scalar, however, since under parity

5 = 5 (20.208)

This means that it is a pseudoscalar.

Similarly, 5 is a pseudovector (or axial vector ), which is a 4-vector whose

space part does not change sign under parity and whose time component does.

We see that the set I , , , 5 , 5 transforms as a scalar or zeroth-

rank tensor, a vector or first-rank tensor, a second-rank tensor, a pseudovector
or axial vector or a first-rank pseudotensor and a pseudoscalar or a zeroth-rank
pseudotensor.

This is a clear indication that they are linearly independent.

20.6 The Dirac Equation

The scalar product of two 4-vectors is a Lorentz invariant. We have identified
two 4-vectors, namely,

= (, ~
) , p = (E/c, p~) (20.209)

Their scalar product is

E
~
p~
(20.210)
c
Since it is an invariant, it has the same value in all frames. What is that value?
We can find out by looking at its square
 2  2
E E 2 p~E
p~ = 2
~ + (~ p~) (~+ ~ )
c c c
 2 X  2 X
E 2 p~E E 2
= (1) + (i pi ) (0) = (i ) p2i
c i
c c i
 2 X  2
E 2 E
= (1) p2i = p2 (20.211)
c i
c

1572
This says that s 2
E E
~
p~ = p2 = mc (20.212)
c c
The sign depends on the sign we choose for . If we had interpreted 2 = 1 to
mean = 1 instead of +1, which is equivalent to choosing the parity operator
as , no physics would have changed. This means we are free to choose the
sign. We choose
E
~ p~ = +mc (20.213)
c
or
E c~ p~ = mc2 (20.214)
This operator equation involves 4 4 matrices which implies that any physical
state vectors must be 4-component spinors.

We make the standard operator correspondence

E i~ , p~ i~ (20.215)
t
and obtain a wave equation
~
i~ + mc2
= c~ (20.216)
t i
or multiplying by we get
 
~
i~ + mc2
= c~ (20.217)
t i

which is the Dirac equation for a relativistic spin 1/2 particle.

The form of the result says that the Hamiltonian of a relativistic spin 1/2 particle
is
H = c~ p~ + mc2 (20.218)
In the presence of an electromagnetic field we use minimal coupling to get
     
~ e~
i~ e (~r, t) = c~ A + mc2 (~r, t) (20.219)
t i c

We note that the vector potential A ~ (corresponding to spatial degrees of free-

dom) is directly coupled to
~ corresponding to internal degrees of freedom).

20.6.1 Nonrelativistic Limit

First, we separate time using

(t) = eiEt/~ (20.220)

1573
to get    
~ e~
(E e) = c~
A + mc2 (20.221)
i c
We then write  
A
= (20.222)
B
where A and B are still two-component functions and use the explicit Dirac
matrices to obtain
       I 0   
A 0 ~ ~ 2 A
(E e) = c~
p eA + mc
B ~ 0 0 I B
(20.223)
This is equivalent to two coupled equations
 
~ c~
p eA~ B + mc2 A = (E e) A (20.224)
 
~ c~
p eA~ A mc2 B = (E e) B (20.225)

Letting E = E 0 + mc2 the second equation of the pair becomes

1  
~ A
B = 0 2
~ c~
p eA (20.226)
E e + 2mc
Inserting this result into the first equation of the pair we get
1 
~
 1  
~ A = (E 0 e) A
~
c~
p eA 0 e
~
c~
p e A (20.227)
2mc2 1 + E2mc 2

These last two equations are exact and very useful substitutes for the Dirac
equation.

We now make some approximations relevant to the nonrelativistic case. We

assume that

E 0 << mc2 , e << mc2

eigenvalues of p~ are of order mv << mc

This says that the components satisfy

v
B A (20.228)
c
Or that the 4-component wavefunction has two large components A and two
small components B .

If we ignore terms of order (v/c)2 the equation for A becomes

1   2
~
c~
p e ~
A A + eA = E 0 A (20.229)
2mc2

1574
Now, earlier we derived the identity
 
(~ ~a)(~ ~b) = ~a ~b + i~ ~a ~b (20.230)

We then have
  2  2    
~ c~ ~
p eA ~ + i~ c~
p eA
= c~ ~ c~
p eA ~
p eA (20.231)

Now   2  2
~ c~ ~
p eA ~
p eA
= c~ (20.232)
and
       
~ c~
p eA
c~ ~ = ec p~ A
p eA ~+A
~ p~ = +ie~c A
~+A
~
(20.233)
Now
~+A
~ )i A = ijk j Ak Ak j A


( A
= ijk (j Ak )A + Ak (j A ) Ak (j A )

~ A = B
= ijk (j Ak )A = ( A) ~ A (20.234)
Putting everything together we get
1  e ~ 2 e ~ ~ + eA = E 0 A
p~ A A ~ B (20.235)
2m c mc 2
This the Pauli equation. The term involving the magnetic field has the form of
a magnetic dipole interaction energy
e ~ ~
SB (20.236)
mc
with a gyromagnetic ratio
e
2 g=2 (20.237)
mc
The full time-dependent form of the nonrelativistic limit is given by
 2
1 ~ e~ e ~ ~ + (e + mc2 )A = i~ A
A A ~ B (20.238)
2m i c mc 2 t

20.6.2 Currents and Continuity Equations

Going back to the full equation
     
~ e~ 2
i~ e (~r, t) = c~ A + mc (~r, t) (20.239)
t i c
we take the Hermitian conjugate to get
   
~ e~
i~ e + (~r, t) = c A + (~r, t)~
+mc2 + (~r, t) (20.240)
t i c

1575
Note that the Hermitian conjugate operation reverses matrix order. Now multi-
ply the first equation by + (~r, t) on the left and the second equation by (~r, t)
on the right and subtracting we get the continuity-type equation
( + )
+ ( + c~
) = 0 (20.241)
t
This says that the quantity + is a positive conserved quantity that can be
interpreted as a probability density and then
~j = + c~
(20.242)
is the corresponding probability current. The operator c~ corresponds to the
velocity operator, which is the derivative of the Hamiltonian with respect to p~.

What happens to the Dirac equation under a Lorentz transformation?

In one frame we have

~
i~ + mc2
= c~ (20.243)
t i
and in a new frame we have
(~r 0 , t0 ) ~
i~ 0 = 0 c~
0 0 (~r 0 , t0 ) + mc2 (~r 0 , t0 ) (20.244)
t0 i
where (~r 0 , t0 ) is the wave function in the new frame.

We already determined, however, that

~ ~
i~ 0 0 c~
0 0 = i~ c~
(20.245)
t0 i t i
since the scalar product of two 4-vectors is an invariant. This implies that
(~r 0 , t0 ) = (~r, t) = Lorentz scalar (20.246)
i.e., they both satisfy the same equation when ~r 0 , t0 and ~r, t are the same space-
time point.

It turns out, however, that a more convenient equation to use in the new frame
~ matrices, i.e., 0 and
is one that still involves the old and ~ 0 are represented
by the same matrices as and ~ . We can find this other equation as follows.
We have
(~r 0 , t0 ) ~
i~ 0 = 0 c~ 0 0 (~r 0 , t0 ) + mc2 (~r 0 , t0 )
t0 i
0 0
(~r , t ) ~ 0
i~Lv L1
v 0
= Lv c~ L1v (~ r0 , t0 ) + mc2 (~r 0 , t0 )
t i
(~r 0 , t0 ) ~ 0
i~L1
v = c~ L1
v (~ r 0 , t0 ) + mc2 L1 v (~r 0 , t0 )
t0 i

1576
If we define
0 (~r 0 , t0 ) = L1 r 0 , t0 ) = L1
v (~ v (~
r, t) (20.247)
we have the equation
0 (~r 0 , t0 ) ~
i~ 0
0 0 (~r 0 , t0 ) + mc2 0 (~r 0 , t0 )
= c~ (20.248)
t i
This form of the equation has the same matrices and
~ in all frames with the
wave function in the new frame related to the wave function in the old frame
by the Lorentz transformation.

Alternatively, we can write the Dirac equation in covariant form. The Dirac
equation is  
~
i~ = c~ + mc2 (20.249)
t i
which we can rewrite as

i~0 i~i i + mc = 0 (20.250)

using the definition of the Dirac gamma matrices we have

 mc 
i + =0 (20.251)
~
which is clearly covariant.

20.6.3 Free Particle Solutions

We start off by constructing solutions for a particle at rest. In this case, we
have
(~r, t) = eiEt/~ u (20.252)
where u is a spinor independent of space and time. Substituting into the Dirac
equation we have
Eu = mc2 u (20.253)
The eigenvalues of are 1. If u is an eigenstate of with eigenvalue +1, then
E = +mc2 and if u is an eigenstate of with eigenvalue 1, then E = mc2 .
So we find negative energy solutions again and we will associate them with
particles and antiparticles as before. Their properties in the spin 1/2 case will
be different, however.

We choose to write four linearly independent solutions to the free particle Dirac
equation as:

1 0 0 0
(+) 0 (+) 1 () 0 () 0
u0 = 0 , u0 = 0 , u0 = 1 , u0 = 0

0 0 0 1

1577
where the upper index () denotes the eigenvalue of , the 0 denotes that the
particle is at rest p~ = 0, and the arrow denotes the value of the spin associated
physically with these states.

(+) () (+)
The spinors u0 and u0 are eigenstates of z with eigenvalue +1 and u0
()
and u0 are eigenstates of z with eigenvalue 1.

()
We are saying here that while u0 is the spinor of a negative energy particle
with spin up, we will associate it with a positive energy antiparticle with spin
down.
2
The states with = +1 vary in time as eimc t/~ and those with = 1
2
vary in time as e+imc t/~ . The positive and negative states have opposite par-
ity(intrinsic).

We can now construct states for a particle with momentum p~ by starting with
the particle at rest and applying a Lorentz transformation to take us to a frame
moving with velocity

p~c2 p
~v = where Ep = + p2 c2 + m2 c4 (20.254)
Ep

We showed earlier that

2
t/~ ()
0 (~r 0 , t0 ) = L1 r, t) = e~~/2 (~r, t) = e~~/2 eimc
v (~ u0, (20.255)

Now Ep0 t0 p~ 0 ~r 0 is a Lorentz scalar (scalar product of two 4-vectors). In the

rest frame it is equal to mc2 t. Therefore we can write
0
2
r 0 Ep0 t0 )/~
eimc t/~
= ei(~p ~
(20.256)

Dropping the superfluous primes we then have

()
(~r, t) = ei(~p~rEp t)/~ e~~/2 u0, (20.257)

as the wave function for nonzero momentum. The new spinors are given by
() ()
h i ()
up~, = e~~/2 u0, = cosh
~ v sinh u (20.258)
2 2 0,
Using
p~c2
~v = (20.259)
Ep
we get
r
Ep + mc2 p~c
cosh = , v tanh = (20.260)
2 2mc2 2 Ep + mc2

1578
so that r
Ep + mc2
 
() c~p
~ ()
up~, = 1+ u (20.261)
2mc2 Ep + mc2 0,
We then have (in the standard representation)
r
Ep + mc2
 
(+) c (+)
up~, = 1 + p
~
~ u0, (20.262)
2mc2 Ep + mc2

Now

0 0 0 1 0 0 0 i
0 0 1 0 0 0 i 0
p~
~ = px + py
0 1 0 0 0 i 0 0
1 0 0 0 i 0 0 0

0 0 1 0 1
0 0 0 1 (+) 0
+ pz
1
, u0, =
0 0 0 0
0 1 0 0 0

and we get


1
0

r 0

(+) Ep + mc2 0
up~, =
2mc2 0 0 0

c
0 0 0
+ Ep +mc 2 0 + py
px + pz
0 1
1 i 0

r 1
Ep + mc2 0
= cpz (20.263)

2mc2

Ep +mc2
c(px +ipy )
Ep +mc2

and similarly
r 0
(+) Ep + mc2 1
up~, = (20.264)

c(px ipy )
2mc2

Ep +mc2
Epcp z
+mc2

cpz

r Ep +mc2
Ep + mc2 c(px +ipy )
()
up~, = Ep +mc2 (20.265)

2mc2

1
0

1579

c(px ipy )
r Ep +mc2
() Ep + mc2
Epcp z

up~, = +mc2 (20.266)

2mc2

0
1
Remember that the arrow refers to the spin associated with the state in the
rest frame, which is minus the z eigenvalue for the () spinors. We see that a
particle in a z eigenstate in its rest frame appears to be in a z eigenstate to
an observer moving with respect to the particle only if the observer is moving
along the zdirection, i.e., if px = py = 0 we have

r 1 r
Ep + mc2 Ep + mc2 (+)
 
(+) 0 cpz ()
up~, = cpz = u0, + u
2mc2 Ep +mc2
2mc2 Ep + mc2 0,
0
(20.267)
which is a sum of a particle and an antiparticle where both have spin up!

(+)
The positive energy solutions up~ ei(~p~rEp t)/~ correspond to particles with mo-
mentum p~, energy Ep and spin orientation . The negative energy solutions
()
up~ ei(~p~rEp t)/~ correspond to particles with momentum ~ p, energy Ep and
spin orientation which we will soon associate with antiparticles with mo-
mentum p~, energy Ep and spin orientation .

The nonzero momentum spinors are orthogonal but not normalized to one (as
is the case with the zero momentum spinors). Since L+ 6= L1 , in general, the
Lorentz transformations are not represented by a unitary operator and hence the
lengths of vectors or normalizations change. In particular. The normalization
is given by
()+ () Ep
up~ up~ = (20.268)
mc2
~ is Hermitian, we have L+ = L.
Since

Thus, if up~ = L1 u0 , then (up~ )+ = (u0 )+ (L1 )+ = (u0 )+ L1 and up+

~ up
~ =
u+
0 (L1 2
) u 0 .

It is possible to define a normalization that is invariant under a Lorentz trans-

formation. Since anticommutes with ~ , we can write

(L1 )+ = L1 = e~ ~/2 = e~~/2 = L (20.269)

Now, if the spinor u transforms as u0 = L1 u, then the spinor u = u+ is given

in the new frame by

u0 = u0+ 0 = u0+ = u+ (L1 )+ = u+ L1 = u+ L = uL (20.270)

1580
This means that the product u1 u2 of any two spinors is a Lorentz invariant, i.e.,

u01 u02 = (u1 L)(L1 u2 ) = u1 u2 (20.271)

In the rest frame

(b)+ (b0 )
u0 u00 = bbb0 0 , b= (20.272)
which says that the same relation is true for all momentum p~
(b)+ (b0 )
up~ up~0 = bbb0 0 , b= (20.273)

()
The spinors up~ obey the completeness relation that says that the 44 identity
matrix can be written as the sum of the outer products of the four spinors, i.e.,
X (b) (b)
bup~ up~ = 1 (20.274)
b,

()
The spinors up~ obey

() ()
p~)up~ = mc2 up~
(Ep c~ (20.275)

and
()+ ()+
up~ p~) = mc2 up~
(Ep c~ (20.276)
Multiplying the last equation on the right by we have the equation satisfied
()
by up~
() ()
p~) = mc2 up~
up~ (Ep c~ (20.277)

20.6.4 More About Currents

As we have seen, the solution of the Dirac equation (~r, t) has the following
behavior under a Lorentz transformation

(~r, t) 0 (~r0 , t0 ) = L1 (~r, t) (20.278)

The spinor (~r, t) = + (~r, t) transforms like

(~r, t) 0 (~r0 , t0 ) = [L1 (~r, t)]+ = (~r, t)L (20.279)

This says that the product (~r, t)(~r, t) transforms like a Lorentz scalar.

Now the transform as the components of a 4-vector, i.e., L L1 .

Therefore, the product
 
1
(~r, t) (~r, t) = (~r, t), ~j(~r, t) (20.280)
c

1581
transforms like a 4-vector under a Lorentz transformation. It is the particle
4-current multiplied by 1/c. In the same manner,

second -rank tensor

5 axial vector
5 pseudo - scalar

The positive density (~r, t) = + (~r, t)(~r, t) and the current ~j(~r, t) = c + (~r, t)~
(~r, t)
satisfy the continuity equation

+ ~j = 0 (20.281)
t

which implies that the quantity

Z
(~r, t)d3 r (20.282)

is a constant of the motion.

In this case, we can interpret (~r, t) as a probability density (same as in non-

relativistic case). Remember in the spin zero case this was not so since the
corresponding conserved density needed to be interpreted as a charge density
which could be positive and negative.

One important consequence in the spin zero case was that it is impossible for
a particle to make a transition from a state normalized to +1 to a state nor-
malized to 1 since the normalization remains constant in time. We associated
the negative energy states with particles and the negatively normalized states
with antiparticles. We then see that the impossibility of a transition between
positive and negative energy states just corresponds to charge conservation.

In the spin1/2 case, however, both positive and negative energy states have
positive normalization so that there is nothing in the theory (so far) that pre-
vents a particle in a positive energy state from making a transition to a negative
energy state radiating away several high energy photons in the process. A dif-
ficulty in the theory that we must return to later!

Let us say some more about the position and velocity operators in the Dirac
theory.

The position operator has strange features similar to those of the Klein-Gordon
theory. If we apply the position operator to a wave packet made up of positive

1582
energy free particle states we get
!
XZ d3 p (+)
XZ d3 p (+) ~
~r (+)
(~r) = ~r ap~ up~ ei~p~r/~ = ap~ up~ p~ ei~p~r/~

(2~)3
(2~)3 i
3
XZ d3 p
Z
d p (+)
X (+)
= 3
(i~p~ ap~ )up~ ei~p~r/~ + ap~ (i~p~ up~ )ei~p~r/~

(2~)
(2~)3

where we have integrated by parts to get the last two terms. The first term
contains only positive energy components. The second term, however, contains
(+)
the factor i~p~ up~ , which generates both positive and negative components
(explicitly do the derivatives on the column vectors we derived earlier). If we
define, as before
~r = ~r(+) + ~r() (20.283)

then, as before, the even part ~r(+) acting on the wave packet of positive energy
free particle states produces only positive energy free particle states and acting
on the wave packet of negative energy free particle states produces only negative
energy free particle states, while the odd part ~r() turns positive positive energy
states to negative energy states and vice versa.

As in the Klein-Gordon case, both positive and negative energy free particle
solutions are needed to produce a localized wave packet.

Looking at the current expression ~j(~r, t) = c + (~r, t)~

(~r, t) we see that the
operator c~
acts as a velocity operator. This interpretation also agrees with the
commutator relation
h i
i~ ~r, H = c~ (20.284)

which leads to the Heisenberg representation operator equation

d~r
= c~
(20.285)
dt

If, however, we consider the zcomponent of this velocity operator we get

(cz )2 = c2 z2 = c2 . Thus the eigenvalues of each component of the veloc-
ity operator are c, which says that a particle in an eigenstate of the velocity
operator travels at the speed of light!

This means that the velocity operator is not simply related to the momentum
operator relativistically. The eigenstates of any component of ~ are linear com-
binations of positive and negative energy free particle states and thus cannot be
realized in any physical situation! For any arbitrary state the expectation value
of c~
has a magnitude between 0 and c.

1583
20.6.5 Non-relativistic Limit
We now derive corrections to the Pauli equation. Earlier we had
 
~ c~ ~ B + mc2 A = (E e) A
p eA (20.286)
 
~ c~ ~ A mc2 B = (E e) B
p eA (20.287)

or
   
~ e~ 1
~ A B + mcA = i~ e A (20.288)
i c c t
   
~ e~ 1
~ A A mcB = i~ e B (20.289)
i c c t
The second equation of the above pair gives (an exact equation)
   
1 ~ e~ 1 2
B = A ~ A i~ mc e B (20.290)
2mc i c 2mc2 t
Now the A term is much larger than the B term on the right. Thus, we get
the first correction by iterating once, i.e.
 
1 ~ e~
B = A ~ A
2mc i c
  
1 2 ~ e~
i~ mc e A ~ A (20.291)
4m2 c3 t i c
Substituting this expression into the first equation of the pair we get the first
relativistic correction term to the Pauli equation
    
1 ~ e~ 2 ~ e~
2 3 A ~ i~ mc e A ~ A (20.292)
4m c i c t i c
which is (v/c)2 smaller than the kinetic energy term p2 /2m.

The correction term can be rewritten as

  2  
1 ~ e~ 2
A ~ i~ mc e A
4m2 c3 i c t
  
ie~ ~ e~
A ~ (~ ~ ) A
4m2 c3 i c
where
~
1 A
~ = = electric field (20.293)
c t
To lowest order in (v/c) we have
p2
 

i~ mc2 e A = A (20.294)
t 2m

1584
Using this relation with the identity
 
(~ ~a)(~ ~b) = ~a ~b + i~ ~a ~b (20.295)

the correction becomes

p4
 
e~ ie~
+ ~ (~ p~) + p~ ~ (20.296)
8m3 c2 4m2 c2 4m2 c2

The first term is the relativistic correction to the kinetic energy. The second
term is the spin-orbit coupling. The third term is new and is not even Hermi-
tian!

The reason for this non-Hermitian term is that we are only working to order
(v/c)2 . Such a non-Hermitian term in the wave equation means that the nor-
malization integral Z
+
A A d3 r

can change in time. Now the full Dirac equation obeys the normalization con-
dition Z Z
+ 3 + +
d r = [A A + B B ]d3 r = 1 (20.297)

To lowest order, however,

2
~ + + p
B = ~ A B B = A A (20.298)
2imc 4m c2
2

Thus, the integral stays constant to order (v/c)2 . It is the integral

 + 
p2 p2 p2
Z   Z   
+ 3
A 1+ A d r = 1+ A 1+ A d3 r
4m2 c2 8m2 c2 8m2 c2

to order (v/c)2 , that remains constant and equal to 1. This implies that the
correct nonrelativistic limit of the Dirac wave function (the limit whose normal-
ization remains constant in time) is

p2
  Z
(~r, t) = 1 + A (~
r , t) + d3 r = 1 (20.299)
8m2 c2

The equation for this form of the wave function will not have any non-Hermitian
terms. A large amount of algebra gives the equation for (~r, t) as

p4
 
1  e ~ 2
i~ = mc2 + p~ A (20.300)
t 2m c 8m3 c2
2
   
e~ ~ + e~ ~ (~ p~) + e + ~ (2 e)
~ B
2mc 4m2 c2 8m2 c2

1585
This is the correct nonrelativistic limit of the Dirac equation. All terms are
Hermitian.

The terms on the right-hand side are

[rest energy + kinetic energy(to order (v/c)2 )]

[Pauli magnetic moment energy + spin-orbit energy]
+ [correction to the potential energy term]

~ = 0 for simplicity we have

Spin-Orbit Term - Letting A

e~ e~
2 2
~ (~ p~) = 2 2 ~ ( p~) (20.301)
4m c 4m c
If we assume the potential is spherically symmetric, then

1 d
= ~r (20.302)
r dr
and we get

e~ e~ d e d ~ ~
2 2
~ (~ p~) = 2 2 ~ (~r p~) = 2 2 SL (20.303)
4m c 4m c r dr 2m c r dr
which is the spin-orbit energy. It correctly contains the Thomas precession
correction! We do not have to add any terms in an ad hoc manner!

Correction to the Potential - This is called the Darwin term. Now, from
Poissons equation we have

2 e(~r) = 4eQ(~r) , Q(~r) = charge density producing (~r)

For a Coulomb potential we get

~2 2 ~2
( e) = Ze2 (~r) (20.304)
8m2 c2 2m2 c2
This term tends to raise the energy of s-states since they do not vanish at the
origin.

20.6.6 The Dirac Hydrogen Atom

We start with the equations
 
~ c~ ~ B + mc2 A = (E e) A
p eA (20.305)
 
~ c~ ~ A mc2 B = (E e) B
p eA (20.306)

1586
where we have substituted ~ for ~ . The potential function is e = Ze2 /r and
~ = 0. Writing
we let A
   
u1 u3
A = , B = (20.307)
u2 u4

we get

Ze2
 
i 2 u4 u4 u3
E+ mc u1 + i + =0 (20.308)
~c r x y z
Ze2
 
i u3 u3 u4
E+ mc2 u2 + +i =0 (20.309)
~c r x y z
Ze2
 
i u2 u2 u2
E+ + mc2 u3 + i + =0 (20.310)
~c r x y z
Ze2
 
i u1 u1 u2
E+ + mc2 u4 + +i =0 (20.311)
~c r x y z

We now use another clever trick I learned from Professor Hans Bethe at Cornell
University to find a solution.

If we hconsider
i only large components, i.e., set the small components to zero,
~
then L, H , which is proportional to ~ p~, will be zero, since
~ connects the
small and large components. This means that A will be an eigenfunction of L. ~
In addition, it must contain one spin component with spin up and another with
spin down.

Of course, ~j and jz are constants of the motion. Hence, for j = ` + 1/2 we can
set
s
` + m + 21 m 12
u1 = g(r) Y` () (20.312)
2` + 1
s
` m + 21 m+ 21
u2 = g(r) Y` () (20.313)
2` + 1

where the unknown function g(r) will be the solution of some relativistic radial
equation.

To get the small components we recall the equation

1  
~ A
B = 0 ~
c~
p eA (20.314)
E e + 2mc2
and note that the operator which gives the small component from the large
component has odd parity (~ ~ = 0 and everything else is even) and
p is odd, A
commutes with ~j. Hence, B must belong to the same j value as A but must

1587
have a different `.

Corresponding to j = `+1/2 the only other possible value of the orbital angular
momentum is `0 = ` + 1. Therefore, we set (remembering the appropriate
Clebsch-Gordon coefficients)
s
` m + 23 m 21
u3 = if (r) Y`+1 () (20.315)
2` + 3
s
` + m + 23 m+ 12
u4 = if (r) Y`+1 () (20.316)
2` + 3

where the unknown function f (r) will be the solution of some relativistic radial
equation. Inserting these solution guesses into the 4 coupled equations we find
that for j = ` + 1/2 the connection between f and g is given by

Ze2
 
1 dg g
E+ + mc2 f = ` (20.317)
~c r dr r
2
 
1 Ze df f
E+ mc2 g = (` + 2) (20.318)
~c r dr r

In an analogous way for j = ` 1/2 we have

s
` m + 21 m 21
u1 = g(r) Y` () (20.319)
2` + 1
s
` + m + 21 m+ 12
u2 = g(r) Y` () (20.320)
2` + 1
s
` + m 12 m 21
u3 = if (r) Y`1 () (20.321)
2` 1
s
` m 21 m+ 12
u4 = if (r) Y`1 () (20.322)
2` 1

and

Ze2
 
1 dg g
E+ + mc2 f = + (` + 1) (20.323)
~c r dr r
2
 
1 Ze df f
E+ mc2 g = + (` 1) (20.324)
~c r dr r

We now define (
(` + 1) if j = ` + 1/2
k= (20.325)
` if j = ` + 1/2

1588
i.e., (
1, 2, .... if j = ` + 1/2
k= (20.326)
1, 2, .... if j = ` + 1/2
We can then combine the 4 equations for f and g into 2 equations as
Ze2
   
1 2 dg g
E+ + mc f + (1 + k) =0 (20.327)
~c r dr r
Ze2
   
1 df f
E+ mc2 g + + (1 k) =0 (20.328)
~c r dr r
Setting
F = rf , G = rg
mc2 + E mc2 E
1 = , 2 =
~c ~c
Ze2
= (1 2 )1/2 , = , = r
~c
we get
   
d k 1
+ G + F =0 (20.329)
d
   
d k 2
F G=0 (20.330)
d
We now solve these coupled equations using the standard series method to obtain
the positive energy bound state solutions.

We substitute
F = ()e , F = ()e (20.331)
and obtain
 
0 k 1
+ + =0 (20.332)

 
k 2
0 =0 (20.333)

We now substitute the series

X
X
= s am m , a0 6= 0 , = s bm m , b0 6= 0 (20.334)
m=0 m=0

the requirement that f and g be finite everywhere turns out to be impossible

to satisfy. Instead, we require that the integrated probability density be finite,
i.e.,
Z h i
2 2
|F ()| + |G()| d < (20.335)
0

1589
This makes sure that s 6= . Substituting the series and equating coefficients
of the same power of we get the recursion relations
1
(s + + k)b b1 a a1 = 0 (20.336)

2
(s + k)a a1 + b b1 = 0 (20.337)

For = 0 we get
(s + k)b0 a0 = 0 = (s k)a0 + b0 (20.338)
These equations have a nontrivial solution if and only if
s = (k 2 2 )1/2 (20.339)
First we look at the negative root. For small the integrand for the integrated
probability density is 2s and we must have 2s > 1 or (k 2 2 )1/2 > 1/2.
The minimum s occurs when k 2 = 1. This corresponds to Z 109. For k 2 > 1,
no value of Z will permit the negative root.

Restricting ourselves to Z < 109, we choose the positive root s = (k 2 2 )1/2 .

For k = 1, s < 1, f and g diverge at the origin. The probability density integral
converges, however.

The recursion relations lead to function of the order e2 (the probability density
integral would diverge) unless the series terminate. Suppose the series terminate
for = n0 , i.e., an0 +1 = bn0 +1 = 0. We then have from the recursion relations
that
1 an0 = bn0 , n0 = 0, 1, 2, ..... (20.340)
We now multiply the first recursion relation by and the second by 1 and
subtract them to get
b [(s + + k) 1 ] = a [1 (s + k) + ] (20.341)
Inserting = n0 and using 1 an0 = bn0 we get
2E
2(s + n0 ) = (1 2 ) = (20.342)
~c
Putting everything together we get
1/2
 2
1/2 2
2
E = mc 1 + = mc2 1 +  (20.343)

2
(s + n0 )2 p
n0 + k 2 2
1
Since |k| = j + 2 we get
1/2
2

E = mc2 1 +  , n0 = 0, 1, 2... j + 12 = 1, 2, 3, ...

q 2
n0 + (j + 12 )2 2
(20.344)

1590
where = Ze2 /~c.

Before looking at the physics in this result let us investigate an alternative ap-
proach involving a second-order Dirac Equation. The first-order Dirac equation
is
   
~ e~
i~ H (~r, t) = 0 , H = c~ A + mc2 + e (20.345)
t i c

We now define the projection operator P as

 

i~ t H + 2mc2
P = (20.346)
2mc2
and operate on the Dirac equation from the left. After some algebra we get
"  2  2 #
1 ~ e~ 2 2 e~  ~ 
i~ e A m c + ~ B i~
~ = 0
c2 t i c c
(20.347)
where we have used the relations
  2  2
~ e~ ~ e~ e~ ~
~
A = A ~ B (20.348)
i c i c c

and  
~ e~
A, i~ e = i~e~ (20.349)
i c t
The new second-order
 equation
 is just the Klein-Gordon equation with an ad-
~
ditional term ~ B i~
~ , which represents the direct coupling of the elec-
tromagnetic fields to the magnetic(and electric) moments of the particle.

Every solution of the Dirac equation is a solution of this new second-order equa-
tion, but every solution of the second-order equation in not necessarily a solution
of the Dirac equation.

If, however, is a solution of the second-order equation, then = P is a

solution of the Dirac equation. We can see this as follows. The second-order
equation can be written as
   

P i~ H (~r, t) = i~ H P (~r, t) = 0 (20.350)
t t

or the second order equation is equivalent to

   

i~ H (~r, t) = i~ H (~r, t) = 0 (20.351)
t t

1591
This says that P acts as a projection operator, which reduces solutions of the
second-order equation to solutions of the first-order Dirac equation.

Let us now use the second-order equation to find the energy levels of the Dirac
hydrogen atom (Glauber, et al PR 109,1307(1958)). For a stationary state of
energy E in the Coulomb potential the second-order equation becomes
"  2  2 #
1 Ze2 ~ 2 2 i~Ze2
E+ m c + 2 r = 0 (20.352)
c2 r i r c

~ r. We now write
where r =
2
~2 2 2 L2

~
= r + 2 (20.353)
i r2 r2 r

and get the equation

 2  
Ze2 Ze2
2
E m c 2 4
2EZe ~ 2 2 2 L2 c i~ c r
+ + 2 2 r2 = 0
c2 rc2 r r r2

(20.354)
We now use a few tricks to change this equation, which is almost in the same
form as the Klein-Gordon equation for the Coulomb potential, into exactly the
same form.

We first define the operator

!
~
L
K = 1 + ~ (20.355)
~

with these properties

h i h i h i
~ p~ = 0 ,
K, ~ ~r = 0 ,
K, K, r2 = 0 (20.356)
h i
K, J~ = 0 , J~ = L ~ + ~ ~ (20.357)
2

These imply that K commutes with the Hamiltonian

Ze2
p~ + mc2
H = c~ (20.358)
r
for the relativistic hydrogen atom.

This says that K is a constant of the motion and since it also commutes with the
total angular momentum we can label the common eigenstates or energy levels
of the hydrogen atom by the eigenvalues of K, J2 and Jz . K is a constant of the

1592
motion for any spherically symmetric, spin-independent potential and physically
it measures the degree to which the spin and the orbital angular momentum of
the particle are aligned.

Let us find the eigenvalues k of K. We note that

!2 !2
~
L ~
L ~
L
2
K = 1 + ~ = 1 + ~ + 2~
~ ~ ~
!
~ L
L ~ i ~
L
=1+ + 2 ~ (L~ L)
~ + 2~
~ 2 ~ ~
!
~ L
L ~ i ~
L
=1+ + ~
(i~ ~
L) + 2~

~2 ~2 ~
~ 2
L2

L 1 ~ ~ 1
= 1 + 2 + ~ = 2 L + ~ +
~ ~ ~ 2 4
J 2
1
= 2 + (20.359)
~ 4
where we have used
~ B)
(~ A)(~ ~ =A
~B
~ + i~ (A
~ B)
~

~ L
L ~ = i~L
~ , J~ = L
~ +S
~ , ~ = ~ ~
S
2
Therefore, we have
1
k 2 = j(j + 1) + = (j + 12 )2 (20.360)
4
n o
Now, since K, 5 = 0 we find that, if k is an eigenvalue of K, i.e.,

K |ki = k |ki (20.361)

then
K5 |ki = 5 K |ki = k5 |ki (20.362)
which says that k is also an eigenvalue of K. The eigenvalues are then

k = 1 , 2 , 3 , ....... (20.363)

since j = 1/2, 3/2, 5/2, ........ Note that zero is not an eigenvalue of K. In addi-
tion, an eigenstate of K with eigenvalue k is an eigenstate of J2 with eigenvalue
j = |k| 1/2.

We now find the energy eigenvalues. Define the operator

Ze2
= K i r (20.364)
~c

1593
with the properties
2
Ze2
h i h i 
, K = 0 , , J~ = 0 , 2
=K 2
(20.365)
~c

A little algebra then shows that

2
Ze2 Ze2
  
~2 ( + 1) = L2 i~ r (20.366)
c c

which is the operator in the last term of the second-order equation. We can
then write
" #
E 2 m2 c4 2EZe2 ~2 2 2 ~2 ( + 1)
+ + 2 2r =0 (20.367)
c2 rc2 r r r2

This is exactly the same form as the Klein-Gordon equation except that

`0 (`0 + 1) ( + 1) (20.368)

or a number has been replaced by an operator. Now if (~r) is an eigenstate of

( + 1) then the operator ( + 1) in the equation is replaced by its eigenvalue
which we can write as `0 (`0 + 1). This says that the energy eigenvalues are given
by the same formula as in the spin zero case, i.e.,

mc2
E=h , n0 = `0 + 1 + , = 0, 1, 2, ..... (20.369)
Ze2 2

i1/2
1+ ~cn0

Since , K, J2 and Jz all commute, we can construct solutions which are

eigenstates of K, J2 and Jz as well as .

does not commute with H however. This means that the solutions we have
found for the second order equation cannot directly be eigenfunctions of H.
Instead, since
H(P ) = E(P ) (20.370)
i.e., the energy eigenvalues from the second-order equation are also the eigen-
values of H, we can find eigenfunctions of H by using the projection operator
P . Since P and do not commute, the eigenfunction of H, namely P , will
generally be a linear combination of different eigenfunctions.

To find the energy eigenvalues we need to know the eigenvalues of . Consider

an eigenstate of and K with eigenvalues k. We then have
 2 2 !  2 2 !
Ze Ze
2 |ki = 2 |ki = K 2 |ki = k 2 |ki (20.371)
~c ~c

1594
or
2 !1/2
Ze2

2
= k (20.372)
~c

and the possible eigenvalues of are . When ( + 1) acts on a ( , K)

eigenstate it has the eigenvalue ( + 1) = `0 (`0 + 1). This leads to two
possible `0 values for each eigenvalue of :
(
0 , 1 for =
` = (20.373)
, 1 for =

For each eigenvalue of , the two `0 solutions add up to 1. The smaller of

the two solutions lambda and 1 are eliminated because they are not
normalizable (behavior near the origin). This leave two cases to consider:

(1) = , `0 =
(2) = , `0 = 1

The possible energy eigenvalues are given by

mc2
E=h , n0 = `0 + 1 + , = 0, 1, 2, ..... (20.374)
Ze2 2

i1/2
1+ ~cn0

or redefining some quantities

s 2
Ze2

0 1 1 2


n = n |k| + = n j 2 + j+ 2 (20.375)
~c

so that n takes on the values

n = |k| , |k| + 1 , |k| + 2 , ...........for =

n = |k| + 1 , |k| + 2 , |k| + 3 , ...........for =

the energy levels are then given by

1/2
 2
Ze2
~c
E = mc2 1 +  (20.376)

q 2
1 1 2

Ze2
2
nj 2 j+ 2 ~c

The quantum number n is just the principal quantum number of the hydrogen
atom.

The energy level structure looks like Figure 20.2 below.

1595
 Figure 20.2: Dirac hydrogen energy level structure

where the left sequence corresponds to = and the right sequence to = .

Some Features
The energy levels for the spin 1/2 particle are the same as those found for the
spin 0 particle with ` j.

The energy is real only if

1 Ze2
j+ 2 < (20.377)
~c
which corresponds to Z < 137 for j = 1/2.

The Dirac theory leads to an accidental degeneracy in `, i.e., states with the
same j but different ` have the same energy. This degeneracy is removed by the
Lamb shift, which is due to the interaction of the electron with its own field. As
we shall see later, for j = 1/2, the effect is one order of magnitude smaller then
the fine structure splitting. For j 3/2, it is two orders of magnitude smaller.

An expansion in powers of Z, where

e2
= = fine structure constant (20.378)
~c
looks like
Z 2 2 (Z)4
   
1 3
En,j = mc2 1 2
3 1 + O((Z)6 ) (20.379)
2n 2n j+ 2
4n

which agrees with the perturbation calculations we carried out earlier.

Some Details about the Energy Levels

The solutions of the Dirac equation are not eigenstates but they are K eigen-
states and K is a constant of the motion (it commutes with the Hamiltonian).
The total orbital angular momentum L ~ is not a constant of the motion and
neither is L2 . We need to come up with some way to classify the energy levels
in the relativistic hydrogen atom using the eigenvalues k.

1596
To get a handle on how to proceed we look at the nonrelativistic limit where

K K for positive energy states (20.380)

We expect the solutions of the second order equation with one sign of to
correspond to solutions of the first-order equation with the opposite sign of K.
This means that

n = |k| , |k| + 1 , |k| + 2 , ........... for = k > 0

n = |k| + 1 , |k| + 2 , |k| + 3 , ........... for = k < 0

It turns out to be convenient to still label the solutions by the ` value that they
would have in the nonrelativistic limit. To find this ` value we use

L2 ~
L L2
K 2 = 1 + 2
+ ~ = K + 2
~ ~ ~
L2
K(K ) = 2
~
In the nonrelativistic limit, 1 and we have

L2
K(K 1) = k(k 1) = `(` + 1) (20.381)
~2
so that ` becomes the total orbital angular momentum quantum number in the
nonrelativistic limit. Solving for ` in terms of k we get
(
k 1 = j 1/2 for k > 0
`= (20.382)
|k| = j + 1/2 for k < 0

Now K measures the alignment of the spin and the orbital angular momen-
tum. The above results say that for k > 0, they are essentially parallel and so
j = ` + 1/2 and for k < 0 they are essentially antiparallel so j = ` 1/2.

A detailed calculation of the wave functions shows that the upper two com-
ponents of the wave function (the large components) are eigenstates of total
orbital angular momentum with eigenvalue `, while the lower two components
(the small components) are eigenstates of total orbital angular momentum with
eigenvalue ` + 1 for k > 0 and with ` 1 for k < 0.

The complete energy level scheme for the relativistic hydrogen atom for n = 1,
2, and 3 looks like Figure 20.3 below.

1597
 Figure 20.3: Energy level structure for relativistic hydrogen

The complete degeneracy of a given n in the nonrelativistic case is lifted by

relativistic effects. The degeneracy between states like 1S1/2 , 2P3/2 , 3D5/2 ,
4F7/2 , etc is now broken. The degeneracy still remains between states like
2S1/2 and 2P1/2 , 3S1/2 and 3P1/2 , 3P3/2 and 3D3/2 , etc., levels.

All levels except 1S1/2 , 2P3/2 , 3D5/2 , etc., are 2-fold degenerate because they
are the eigenstates of of K with opposite eigenvalues, i.e., 2P3/2 k = 2,
2D3/2 k = 2.

Hyperfine Structure
The are two corrections that modify the energy level results from the Dirac
equation. The two-fold degeneracy is removed by the interaction of the electron
with vacuum fluctuations of the electromagnetic radiation field. This effect is
called the Lamb shift. In addition, there is also a hyperfine interaction which
splits every level into two, It is due to the interaction of the electron with the
magnetic moment of the proton. We consider hyperfine splitting first.

As an example we derive the hyperfine splitting of an sstate using nonrela-

tivistic first-order perturbation theory. The interaction of the electron spin with
the magnetic moment of the proton is given by
|e| ~ ~ r) = B ~ B(~
~ r)
H 0 = ~ B(~ (20.383)
2mc
~ r) is the magnetic field due to the magnetic moment of the proton.
where B(~
This magnetic moment is given by

~ p = |e| ~gp ~p = 1 gp p~p

M (20.384)
4mp c 2
where gp is the gyromagnetic ratio of the proton, mp is the proton mass and
~~p /2 is the spin of the proton. The magnetic field from this magnetic moment

1598
(assuming the proton is fixed at the origin) is given by the relations
 
~ ~ 1
A(~r) = Mp = vector potential (20.385)
r
~ r) = A(~
B(~ ~ r) (20.386)
~ r) = (~p ) gp p
B(~ (20.387)
2r
This gives
 
1
H 0 = gp B p~ (~p )
2r
1
= gp B p~ (~p ( ) (~p ))
2r
1
= gp B p ((~ ~p )( ) (~ )(~p )) (20.388)
2r
The first-order shift of the level is
Z  
D E
2 1
H 0 = gp B p d3 r |(~r)| (h~ ~p i ( ) h(~ )(~p )i)
2r
(20.389)
where the brackets h.....i denote the expectation value in the relative spin state
of the electron and proton and (~r) is the nonrelativistic wave function of the
level. If we only consider sstates, which are spherically symmetric, then
1
h(~ )(~p )i = h(~ ~p )i 2 (20.390)
3
and we get
Z  
D E
0 1 3 2 2 1
H = gp B p d r |(~r)| h(~ ~p )i
3 2r
Z
4 2
= gp B p h(~ ~p )i d3 r |(~r)| (~r)
3
4 2
= gp B p h(~ ~p )i |(0)| (20.391)
3
where we have used
1
2 = 4(~r) (20.392)
r
For the hydrogen atom sstate

2 1
|(0)| = (20.393)
(na0 )3

and we get
D E 2  e2  m 2
H 0 = gp h(~ ~p )i (20.394)
3 2a0 m p n3

1599
We then have
F~ = S
~ + I~ = total spin (20.395)
1
For S = 2 , I = 12 , we have F = 0(singlet) , 1(triplet). But

F~ = S
~ + I~ F~ 2 = S
~ 2 + I~2 + 2S
~ I~ (20.396)
2 2
~ I~ = ~ ~ ~p = ~ (F (F + 1) 3/2)
S (20.397)
4 2
We then have for a relative triplet state h(~ ~p )i = 1 and for a relative singlet
state h(~ ~p )i = 3. This says that the singlet state lies lower than the triplet.

The total splitting of the ground state is

8 e2
 
m 2
E = gp (20.398)
3 2a0 mp

between the triplet and singlet. The transition between these two levels gen-
erates radiation with a frequency of 1420 M Hz and a wavelength of 21.4 cm.
This radiation is very important in astronomy. From its intensity, Doppler
broadening, and Doppler shift, one obtains information concerning the density,
temperature, and motion of interstellar and intergalactic hydrogen clouds.

The Lamb Shift

The coupling Z
e ~ r)
Hint = d3 r~j(~r) A(~ (20.399)
c
of the electron to the quantum mechanical radiation field causes a shift in the
energy levels of the hydrogen atom. Although not an exact calculation, we can
get some idea of the fundamental difficulties in quantum electrodynamics by
doing a nonrelativistic second-order perturbation calculation.

We consider an electron in the state |ni with energy n . Because of the above
interaction (see last part of this chapter) the electron is able to spontaneously
emit a photon thereby going to some state |n0 i. This produces a second-order
shift in the energy given by
D  2
X X  n0 , ~k~ Hint |n, 0i
  
En = (20.400)
n n0 ck
n0 ~
k~

where |n, 0iE is the initial state with the electron in |ni with no photons present,
and n0 , ~k~ is the intermediate state with an electron in |n0 i and on photon of


momentum ~k and polarization ~ present. The energy of this intermediate state

is n0 + ck.

1600
From the quantum theory of electromagnetic radiation (see end of this chapter)
we have that
s
D  e 2~2 c2 0 ~ ~
n0 , ~k~ Hint |n, 0i = hn | j~k |ni (20.401)

c k V

where ~j~k is the k th Fourier component of the current ~j(~r). Therefore,

P  0 2
hn | ~j~k ~ |ni

d3 k 2~2 e2 X
Z
~

En =
(2~)3 ck 0
n n0 ck
n
 2
d hn0 | ~j~k ~ |ni
R P 
k 2 dk e2 X
Z
~

= (20.402)
4 2 ~ ck 0 n n0 ck
n

In the dipole approximation, we can use

~j~ ~j0 = p~ , p~ = electron momentum operator (20.403)

k m
The angular integration over the polarizations is given by
Z 2 Z 2
X 
~ 1 X  4 2 2
d 0 ~
hn | j~k |ni = d hn0 | p~ ~ |ni = |hn0 | p~ |ni|
 
m m3
~
~

where the factor 2/3 comes from the fact that there are only 2 independent
polarizations for each ~k value. This gives
Z X |hn0 | p~ |ni|2
2e2
En = d , = ck (20.404)
3~c3 m2 0
n n0
0 n

The first problem we encounter is that the integral diverges!! This means that
the interaction with the radiation field produces an infinite shift downward in
the energy of the electron.

This result presented theoretical physics with a great difficulty for many years.
In the late 1940s it was resolved due to the work of Feynman, Schwinger and
Tomonaga in producing new calculation rules within the context of quantum
electrodynamics and by Bethe and Weisskopf who actually carried out the cal-
culation using the new rules and got a finite number agreeing with experiment.

Let us try to understand some aspects of what happened.

If we do a similar calculation for a free electron, then one gets an infinite result

1601
again. In the dipole approximation, we can evaluate the energy shift for a free
electron in a momentum state |~
pi. We get
Z 2
2e2 X |h~q| p~ |~
pi|
Ep~ = d (20.405)
3~c3 m2 q p
0 q
~

Since this is a free particle q p = 0 and we have

Z 2 Z
2e2 |h~
p| p~ |~
pi| 2e2
Ep~ = d = p2 d (20.406)
3~c3 m2 3~c3 m2
0 0

which is infinite. What Bethe and Weisskopf noticed was that this expression
is proportional to p2 . In their development of quantum electrodynamics, Feyn-
man, Schwinger and Tomonaga had similar problems which they were able to
deal with be redefining the electron parameters that appeared in the theory
(like mass and charge). The process is called renormalization. In this process
all infinite expression are consistently incorporated into the mass or charge pa-
rameters and then these are defined to have the known experimental values.

In our case, we can interpret the infinite result as redefining the mass, i.e., as
representing a shift of the mass of the electron. In terms of the mathematics,
this means the following. If we say that m0 is the mass and p2 /2m0 is the kinetic
energy of a free electron of momentum p~ neglecting the electromagnetic interac-
tions, then the energy including the effects of the electromagnetic interactions
is given by
Z

p2 1 2e 2 2
p = 1 p
2
+ Ep~ = d (20.407)
2m0 m0 3~c3 m2 2 m 2
0

i.e., we have renormalized the electron mass. The so-called electromagnetic self-
energy of the electron can thus be interpreted as giving a shift of the mass of the
electron from its bare (no electromagnetic interactions) value m0 to its observed
(measured in the laboratory where all interactions are present) value m.

We then argue that the reason the interacting electron has an infinite energy
shift is that it includes the infinite energy change that we already have counted
once when we use the observed mass m rather than the bare mass in the cal-
culation and, thus, we are double counting. In other words, we should really
start out with the Hamiltonian for the hydrogen atom in the presence of the
radiation field given by
p2 e2
H = + Hint (20.408)
2m0 r
Then using the corrected expression for m we get
Z

p2 e2 2e2
H = + Hint + d (20.409)
2m r 3~c3 m2
0

1602
This means that if we write the observed free particle mass in the kinetic energy
(which we always do) we should not count that part of Hint that produces the
infinite mass shift, i.e., we should regard
Z
2e2
Hint + d (20.410)
3~c3 m2
0

as the effective interaction of an electron of renormalized mass m with the

radiation field. It is now finite to second-order of our calculation. Feynman, et
al, showed that this could be done to all orders of perturbation theory!!

We therefore modify our calculation by adding in the required term. We get

Z X |hn0 | p~ |ni|2 2
!
2e2 |hn| p~ |ni|
En0 = d + (20.411)
3~c3 m2 0
n n0
0 n

Using completeness, we have

2
X X
hn| p2 |ni = hn| p |n0 i hn0 | p |ni = |hn0 | p |ni| (20.412)
n0 n0

so that

2e2 n0 n
Z
2
X
En0 = 3 2
|hn 0
| p |ni| d (20.413)
3~c m 0
n n0
n 0

The integral is still divergent but only logarithmically and, in fact, not at all
in more sophisticated relativistic calculations. We can imagine that the correct
calculation would yield a similar result but with a convergent integral. We can
simulate this result by integrating to some cutoff value (and not to infinity) say
at ~ = mc2 . We then get

2e2  mc2 
 
2
X
En0 = |hn0
| p |ni| ( n 0 n )`n   (20.414)
3~c3 m2 0  n0 n 
n

where we have neglected quantities the size of n0 n in comparison to mc2 .

Bethe evaluated this result numerically and obtained En0 = +1040 megacycles
(the 2P1/2 level turns out to be shifted downward) and the observed value equals
+1057 megacycles, which is remarkable agreement!

Taking into account both the Lamb shift and the hyperfine splitting we have
the level scheme shown in Figure 20.4 below for n = 2:

1603
 Figure 20.4: n = 2 Energy level structure for relativistic hydrogen

Dirac Hole Theory

Finally, we tackle the problem of the negative energy states in the Dirac theory.

As we said earlier, there is no simple conservation law that prevents an electron

or any other spin 1/2 particle in a positive energy state from making a radiative
transition to a negative energy state. This means all atoms must be unstable!
An energy diagram is shown in Figure 20.5 below.

Figure 20.5: n = 2 Energy level structure for relativistic hydrogen

1604
The properties of the positive energy states show remarkable agreement with
experiment. Can we simply ignore the negative energy states? The answer is no
because an arbitrary wave packet, as we saw earlier will always contain negative
energy components via interactions even if we start off only with positive energy
components.

Dirac proposed a clever way out of this dilemma : since spin 1/2 particles obey
the exclusion principle, all one needs to do to insure stability is to say that the
negative energy states are completely filled. Then a particle cannot make a tran-
sition from a positive to a negative energy state for this would put two particles
into the same (negative energy) state. The vacuum state in this picture consists
of an infinite sea of particles in negative energy states. The particle and charge
density at every point is infinite. This is not a problem for the physical theory
since Dirac contended that we only measure deviations from the vacuum. In the
absence of any potential, the charge density of the negative sea is uniform and
Dirac argued that this charge density can produce no forces, since by isotropy,
the forces have no special direction to point!

Now this theory has some very useful special property. Suppose that we remove
a negative energy electron from the vacuum. What is left behind is a hole in
the negative energy sea. Measured with respect to the vacuum, the hole would
appear to have positive charge and positive energy,i.e., since it is the absence of
negative charge and negative energy. Dirac interpreted it as a positron, which
is the electron antiparticle.

Figure 20.6: A hole appears

Let me say that again..... an excited state of the vacuum arises as shown in the
figure. A negative energy electron is excited into a positive energy state, leaving
behind a hole with charge (e) = +e and the same mass as the electron, which
is the antiparticle. It looks like a positive charge since if we apply an electric
field the infinite sea of electron translates opposite to the field direction, which
is unobservable since the sea is infinite. However, the hole seems to be traveling
in the direction of the field like a positive charge!

1605
In this way, antiparticles appear in the Dirac theory as unoccupied negative
energy states, which is very different from the way they appear in the spin zero
theory.

This Dirac hole theory gives a simple description for pair production. Suppose
that a photon of energy > 2mc2 traveling through the vacuum is absorbed by
a negative energy electron and the negative energy electron gets excited to a
positive energy state. What remains, as we have said, is a hole in the negative
energy sea, i.e., a positive energy positron and a positive energy electron. This
says that pair production is simply the excitation of a particle from a negative
to a positive energy state.

Since we could exchange the roles of positrons and electrons in the entire Dirac
theory, electrons would appear as holes in a positron sea. This forces us to
conclude that negative energy seas cannot have any physical reality. The hole
theory is simply a mathematical model that allows us to do the correct book-
keeping within the framework of a single-particle Dirac theory.

With a filled negative energy sea, the Dirac theory would become a many-
particle theory in which we are unable to take into account the interactions
between these particles. The Dirac theory gives valid results only when these
interactions can be neglected. For example, in the hydrogen atom, the mod-
ification of the Coulomb potential by vacuum polarization accounts for about
2.5% of the Lamb shift.

If we second-quantized the Dirac theory, we can treat both particles and an-
tiparticles on the same basis.

The full relativistic quantum field theory of the electrons and positrons and
their interactions with photons was carried out by Feynman, et al in a theory
which is beyond the scope of these volumes.

20.7 Electromagnetic Radiation and Matter

20.7.1 Interacting with the Classical Radiation Field
We assume classical EM radiation in the transverse gauge, where

(~r, t) = 0 , ~ r, t) = 0
A(~ (20.415)

The electric and magnetic fields are given in terms of the vector potential (in
this gauge) by

~ r, t)
1 A(~
~(~r, t) = , ~ r, t) = A(~
B(~ ~ r, t) (20.416)
c t

1606
The electromagnetic energy is given by
2 (~r, t) + B 2 (~r, t)
Z
E = d3~r (20.417)
8
and the rate and direction of energy transport is given by the Poynting vector
c ~ r, t)
(~
~ r, t) = ~(~r, t) B(~ (20.418)
4
The radiation field generated by a classical current ~j(~r, t) is given by

1 2 ~
 
2 4
2 2 A(~r, t) = j (~r, t) (20.419)
c t c
where means the transverse/divergence-free part.

We first consider the monochromatic plane wave solution of this equation. It

takes the form
~ r, t) = ~ei~k~rit + ~ ei~k~r+it
A(~ (20.420)
where

= ck
 2
~ = polarization vector with ~ = 1

= amplitude = constant

~ r, t) = 0 we require ~~k = 0 which corresponds to transverse

To insure that A(~
polarizations only.

The energy per unit volume in the electromagnetic wave is

2 + B 2 2 h 2 
2 2 2i~ r 2it
k~
i
= || Re e (20.421)
8 2c2
The quantity Re(..) oscillates in time and averages to zero so that the average
energy density is
E 2 2
= || (20.422)
volume 2c2
In a similar way the time average of the Poynting vector is
2 2
|| k (20.423)
2c
Any general wave solution is a linear superposition of these monochromatic wave
solutions.
" ~
#
i~
~
X
~ eik~rit ~ e
k~
r +it
A(~r, t) = A~k~ + A~k~ (20.424)
~~
V V
k

1607
where the sum is over all allowed ~k values and over the two orthogonal ~ polar-
izations for each ~k such that ~ ~k = 0 and we have assumed that the universe is
a very large box of volume V . The total energy in this wave solution is
X 2 
A~ ~ 2

E= 2 k (20.425)
2c
~
k~

How does this classical electromagnetic field interact with a quantum mechanical
particle?

In general (no transverse gauge at this point), the classical Hamiltonian is

 2
~ r, t)
p~ ec A(~
H = + e(~r, t) + V (~r, t) (20.426)
2m
where V (~r, t) represents all the other potentials seen by the particle.

We get to quantum mechanics via the standard substitutions

~
~r ~rop , p~ p~op = (20.427)
i
Substituting, we get the Schrodinger equation for an electron in an electromag-
netic field
"  2 #
(~r, t) 1 ~ e~
i~ = A(~r, t) + e(~r, t) + V (~r, t) (~r, t) (20.428)
t 2m i c

20.7.2 Relation to Gauge Invariance

In order to have the Schrodinger equation invariant under a gauge transforma-
tion, the wave function has to change by a phase factor,i.e.,
e
0 (~r, t) = ei ~c (~r,t) (~r, t) (20.429)

where (~r, t) is the some scalar function.

This means that the solutions of the gauge-transformed Schrodinger equation

will still describe the same physical states.

The wave functions or state vectors differ by a phase factor that depends on
space and time and thus, the invariance is LOCAL rather than GLOBAL (a
phase factor independent of space and time).

It is then clear that it is NOT the canonical momentum p i~ (whose

expectation value is NOT gauge invariant), but the genuine kinetic momentum
q~
p A(r, t) (20.430)
c

1608
(whose expectation value IS gauge invariant), that represents a measurable
quantity.

In any physical system, if the momentum operator p appears, then it must

always be replaced by
e
0 (~r, t) = ei ~c (~r,t) (~r, t) (20.431)
if we turn on an electromagnetic fields. This is the only way to guarantee gauge
invariance in quantum mechanics.

Quantum mechanics + electromagnetism requires minimal coupling for gauge

invariance to be valid.

20.7.3 Interactions
We now write
H = H0 + Hint (20.432)
where
p~2
H0 = + V (~r, t) (20.433)
2m
is the Hamiltonian in the absence of electromagnetic fields and
2
e  ~ ~ r, t) p~ + e A

~ 2 (~r, t) + e(~r, t)
Hint = p~ A(~r, t) + A(~ (20.434)
2mc 2mc2
is the operator giving the interaction between matter and radiation.

One must treat the term p~ A(~~ r, t) + A(~

~ r, t) p~ with care since [xi , pj ] = i~ij .
In general, we can show that
 
~ r, t) A(~
p~ A(~ ~ r, t) p~ = i~ A(~~ r, t) (20.435)

which says that

~ r, t) = A(~
p~ A(~ ~ r, t) p~ only when A
~=0 (20.436)

or when we are operating in the transverse gauge.

For multi-electron atoms we have

 2
XN ~ ri , t)
p~i ec A(~ XN
H = +e (~ri , t) + V (20.437)
i=1
2m i=1

and
N  2

X e  ~ ri , t) p~i + e A
~ ri , t) + A(~

~ 2 (~ri , t) + e(~ri , t)
Hint = p~i A(~
i=1
2mc 2mc2
(20.438)

1609
We now define a particle number density
X
(~r) = (~r ~ri ) (20.439)
i

and a current density

 
~j(~r) = 1 p~i p~i
X
(~r ~ri ) + (~r ~ri ) (20.440)
2 i m m

where we constructed a symmetric combination of the terms so that the operator

would be Hermitian.

These quantities imply that

X XZ Z
e(~ri , t) = d ~r e(~r ~ri )(~r, t) = d3~r e(~r)(~r, t)
3
(20.441)
i i

where (~r, t) 6= operator (all operators are in (~r)) and

Z
d3~r(~r) = N = total number of particles (20.442)

Finally, we have
N n Z
X e  ~ ri , t) + A(~
~ ri , t) p~i
o e ~ r, t)
p~i A(~ = d3~r~j(~r) A(~ (20.443)
i=1
2mc c

Since
p~i e ~
~vi = A (20.444)
m mc
when an electromagnetic field is present, the true current operator is
~ r) = ~j(~r) e A(~
J(~ ~ r, t)(~r) = (paramagnetic + diamagnetic) currents
mc
(20.445)
and therefore,
e2
Z  
3 e~ ~ ~ 2
Hint = d ~r j(~r) A(~r, t) + (~r)A (~r, t) + e(~r, t)(~r) (20.446)
c 2mc2

20.7.4 Induced Absorption and Emission

We will now use the transverse gauge, which says that the (~r, t) term is zero.
We also assume that
 the radiation fields are small compared to the fields inside
 ~ ~ 2 term
the atom, i.e., A << e2 /a0 , which implies that we can neglect the A
~ term. Therefore, we have
compared to the ~j A
Z
e ~ r, t)
Hint = d3~r ~j(~r) A(~ (20.447)
c

1610
 ~ as a linear superposition of monochromatic plane waves we then have
For A
Z  !
e 3 1 X p~i p~i
Hint = d ~r (~r ~ri ) + (~r ~ri )
c 2 i m m
" #
i~ r it
k~ i~
k~
r +it
X e e
A~k~~ + A~k~~
~ ~
V V
k
~i i~ ~i i~
!
e X X A~k~~ pm e k~ri it + A~k~~ pm e k~ri +it
= ~ ~
2c V ~ ~ i +A~k~~eik~ri it pm
~i
+ A~k~~ eik~ri +it pm
~i
k
e Xh i
= A~k~~j~k ~eit + A~k~~j~k ~ eit (20.448)
c V ~~
k

where
  Z
~j~ = 1 p~i i~k~ri ~ p~i ~
X
k e + eik~ri = d3~reik~r~j(~r) (20.449)
2 i m m

As we saw in the discussion of time-dependent perturbation theory for a har-

monic perturbation, the eit term implies an absorption of radiation process
and the eit term implies and emission of radiation process. Following the same
steps as that case, we have for the absorption transition rate

2 e2  2  2
abs = (n 0 ~) 2 A~k~  hn| ~j~k ~ |0i (20.450)

0n;~
k~ ~ Vc

To find the total rate of transition we must sum over ~k and ~ (2 polarizations
for each ~k) to get

2 X e2  2  2
abs
0n = (n 0 ~) 2 A~k~  hn| ~j~k ~ |0i (20.451)

~V c
~
k~

Now we can write

Z 2 Z 2
1 X k dkd dd
= (20.452)
V (2)3 (2c)3
~
k

so that

2e2 2
Z 2
 
A~ ~ 2 hn| ~j~ ~ |0i
X
abs
0n = 2 2 d k k (20.453)
~ c 2c3
~

where
n 0
= (from the function) (20.454)
~

1611
If the incident light beam subtends a solid angle d and it is polarized with
polarization vector ~ , then the total rate of energy transport in the beam is
the time average of the Poynting vector which is given by

1 X 2  4 
Z
2 2
A~k~  = d d 4
A~k~  (20.455)
V 2c (2c)
~
k

Now
4 A~ ~ 2
 
I() = d k
(2c)4
= intensity of the incident beam per unit frequency (20.456)

In a similar way

4 2 e2  ~ ~ 2
 
ind
n0
emis
= I() j~k |0i (20.457)
~2 c 2
hn| 

Since

hn| ~j~k ~ |0i = hn| ~j~k ~ |0i (20.458)
we have
abs ind emis
0n = n0 (20.459)
(this is the origin of the Einstein A and B coefficients).

In the absorption process, the absorption of one photon of energy ~ = n 0

causes an upward transition. The electron gains energy and the electromagnetic
field loses energy. Induced emission is just the opposite.

Now a photon of frequency and energy ~ and therefore, the total energy in
the incident beam is X
E= ~N~k~ (20.460)
~
k~

where N~k~ = the number of photons in the (~k, ~) mode in the beam. But we
already have
X 2 
A~ ~ 2

E= 2 k (20.461)
2c
~
k~

which says that

2
A~ ~ 2 = 2~c N~ ~
 
k k (20.462)

and thus
X 4 2 e2  2
abs ind emis
0n = n0 = (n 0 ~) hn| ~j~k ~ |0i N~k~ (20.463)
 
V
~
k~

1612
20.7.5 Quantized Radiation Field and Spontaneous Emis-
sion
Up to this point we have been treating the electromagnetic field classically as a
wave. We have mentioned the idea of photons, but have not created any formal
quantum mechanical structure to describe them, i.e., we have been considering
what happens to the atom and ignoring what is happening to the EM field
during these processes.

To bring out the structure of the theory in terms of photons, we must now
describe these processes in terms of state vectors, such that, in the absorption
process the atom makes a transition from |0i |ni while the electromagnetic
field makes a transition from an initial state to a state with one less photon (it
has been absorbed).

All of our development so far has involved what is physically called an incoherent
beam of light.
 
We related A~k~  and N~k~ so that knowledge of the N~k~ clearly does not imply
any information about the relative phases of the A~k~ which is the meaning of
the term incoherent.

An incoherent beam, therefore, is completely specified by the photon numbers,

i.e., the N~k~ . It is in this sense that we can write the initial state(normalized)
of the electromagnetic field as
 E
N~k1~1 , N~k2~2 , ......., N~k~ , ........ (20.464)


where, as before, the N~k~ = the number of photons in the mode (~k, ~).

Any two of these states are orthogonal if they differ in the number of photons
in any mode.

The final state of the electromagnetic field after photon absorption of a photon
in the mode (~k, ~) is
 E
N~k1~1 , N~k2~2 , ......., N~k~ 1, ........ (20.465)


We assume that there exists some Hint that causes both transitions (atom and
electromagnetic field) as it couples the electromagnetic field to matter. We
define
 E
initial state = |0i N~k1~1 , N~k2~2 , ......., N~k~ , ........ (20.466)

 E
final state = |ni N~k1~1 , N~k2~2 , ......., N~k~ 1, ........ (20.467)


1613
so that
X
Einitial = 0 + ~ck 0 N~k 0~ 0 (20.468)
~
k 0~
0
X
Ef inal = n + ~ck 0 N~k 0~ 0 ~ck (20.469)
~
k 0~
0

The transition rate between the two states is given by Fermis golden rule as
2  2
(n 0 ~) hf inal| Hint |initiali (20.470)
 
~
This must be the same as our earlier result (20.463) which implies that we must
have
 2 e2  2 
 hn| ~j ~ ~ |0i
2
hf inal| Hint |initiali = A
 
~ ~ k
V c2 k
2 2 2
e 2~c 
= N~k~ hn| ~j~k ~ |0i (20.471)
 
Vc 2
This implies that as yet undetermined operator Hint must have the following
properties:
1. it must include a part ~j~k ~ that acts on the atom

2. it must have a part that decreases the number of photons in the (~k, ~)
mode by 1
3. it must be Hermitian
One way of doing this is to write
e X ~ (op) (op)+

Hint = j~k 0 ~ 0 A~ 0~ 0 + ~j~k 0 ~ 0 A~ 0~ 0 (20.472)
c V ~ 0~ 0 k k
k

(op)
where A~ ~ reduces the number of photons in the (~k, ~) mode by 1. It is a
k
photon in mode (~k, ~) annihilation operator.

The second term is required to make Hint Hermitian. Using this model we then
have

hf inal| Hint |initiali

D   E
= n; N~k1~1 , N~k2~2 , ..., N~k~ 1, .... Hint 0; N~k1~1 , N~k2~2 , ...., N~k~ , ....
 

e
= hn| ~j~k ~ |0i
c
D   E
 (op) 
N~k1~1 , N~k2~2 , ...., N~k~ 1, .... A~ ~ N~k1~1 , N~k2~2 , ...., N~k~ , ....
k
(20.473)

1614
For agreement with the earlier result we must have
D   E
 (op) 
N~k1~1 , N~k2~2 , ..., N~k~ 1, ... A~ ~ N~k1~1 , N~k2~2 , ..., N~k~ , ...
k
r
2~c 2 q
= N~k~ (20.474)

(op)
This matrix element of A~ ~ corresponds to the A~k~ term in the classical field
k
picture.

The matrix element implies that

D   E
 (op) 
N~k1~1 , N~k2~2 , ..., N~k~ 1, ... A~ ~ N~k1~1 , N~k2~2 , ..., N~k~ , ...
k
D   E
 (op)+ 
= N~k1~1 , N~k2~2 , ..., N~k~ , ... A~ ~ N~k1~1 , N~k2~2 , ..., N~k~ 1, ...
k
r
2~c 2 q
= N~k~ (20.475)

(op)+
which says that A~ ~ is an operator that increases the number of photons in
k
the (~k, ~) mode by 1. It is a photon in mode (~k, ~) creation operator. We thus
have
 E
(op) 
A~ ~ N~k1~1 , N~k2~2 , ..., N~k~ , ...
k
r
2~c2 q  E
= N~k~ N~k1~1 , N~k2~2 , ..., N~k~ 1, ... (20.476)

 E
(op)+ 
A~ ~ N~k1~1 , N~k2~2 , ..., N~k~ , ...
k
r
2~c2 q  E
= N~k~ + 1 N~k1~1 , N~k2~2 , ..., N~k~ + 1, ... (20.477)


p
This behavior is identical (aside from the 2~c2 / factor) to that of the a
and a+ operators in the harmonic oscillator problem.

This model gives a quantum mechanical picture of the electromagnetic radiation

field as an infinite number of harmonic oscillators - one per mode and the quanta
associated with these oscillators are photons.

If we define a Hermitian electromagnetic field operator as

" #
X (op) ei~k~r (op)+ e i~
k~
r
A~ (op)
(~r) = A~ ~ ~ + A~ ~ ~

(20.478)
k V k V
~~ k

we have
2
Z  2 
e ~ (op) (~r) + e (~r) A

~ (op) (~r)
Hint = d3~r ~j(~r) A (20.479)
c 2mc2

1615
 ~ (op) (~r, t) has the time dependence
In the interaction representation A

~ (op) (~r, t) = e ~i Hem t A

A ~ (op) (~r)e ~i Hem t (20.480)

where Hem = Hamiltonian for free radiation. We then have

Z  
1 3

2 ~ 2
 X
(op)+ (op) 1
Hem = d ~r ~ + B = ~ck A~ ~ A~ ~ + (20.481)
8 k k 2
~
k~

The operator algebra similarity to the a and a+ problem then allows us to write
h i 2~c2 h i
(op) (op)+ (op) (op)
A~ ~ , A~ 0~ 0 = ~k~k 0 ~~ 0 , A~ ~ , A~ 0~ 0 = 0 (20.482)
k k k k

and
 E 1  N~k~  E
(op)+
N~k1~1 , N~k2~2 , ..., N~k~ , ... = p A~ ~ , N , ..., 0, ...
 
k
N~ ~
k1 1 ~ ~
k2 2
N~k~ !
(20.483)
and H = H0 + Hem + Hint where H0 = Hamiltonian for the electrons.

We then have
i (op) i
e ~ Hem t A~ ~ e ~ Hem t ....., N~k~ , ......
 
k
i (op) 1
= e ~ Hem t A~ ~ ei(N~k~ + 2 )t ....., N~k~ , ......
 
k
i
q
Hem t i(N~k~ + 21 )t
 
=e ~ e N~k~ ....., N~k~ 1, ......
1 1
q
= ei(N~k~ 1+ 2 )t ei(N~k~ + 2 )t N~k~ ....., N~k~ 1, ......
 

(op) 
= eickt A~ ~ ....., N~k~ , ......

(20.484)
k

or
i (op) i (op)
e ~ Hem t A~ ~ e ~ Hem t = eickt A~ ~ (20.485)
k k
and similarly
i (op)+ ~i Hem t (op)+
e ~ Hem t A~ ~ e = eickt A~ ~ (20.486)
k k
Putting this all together we have
" #
X (op) ei~k~rit (op)+~ e
i~
k~
r +it
~ (op)
A (~r, t) = ~
A~ ~ + A~ ~ (20.487)
k V k V
~~ k

By construction, we have forced the quantum mechanical description

of absorption of the electromagnetic field in terms of the photon to be
identical to the description in terms of the classical electromagnetic
field for the induced absorption process.

1616
We now apply the formalism to the emission process. This corresponds to the
transition between the states
 E
initial state = |0i N~k1~1 , N~k2~2 , ......., N~k~ , ........ (20.488)

 E
final state = |ni N~k1~1 , N~k2~2 , ......., N~k~ + 1, ........ (20.489)


so that
X
Einitial = 0 + ~ck 0 N~k 0~ 0 (20.490)
~
k 0~
0
X
Ef inal = n + ~ck 0 N~k 0~ 0 + ~ck (20.491)
~
k 0~
0

The transition rate is

2 
  2
(n 0 ~ck)  0; ........N~k~ + 1, ..... Hint n; ........N~k~ , ....  (20.492)

~
where

  
0; ........N~k~ + 1, ..... Hint n; ........N~k~ , ....
e  (op)+ 
= h0| ~j~k ~ |ni ........N~k~ + 1, ..... A~ ~ ........N~k~ , ....


c V k
r
e 2~c q
= h0| ~j~k ~ |ni N~k~ + 1 (20.493)
c V
We no longer have any features that are unknown and hence ad-
justable. Forcing agreement with induced absorption makes the re-
sults for the emission process a prediction!

We get

4 2 e2  2
emis (n 0 ~ck) h0| ~j~k ~ |ni N~k~ + 1 6= abs


=
 
n0;~k~
V 0n;~k~

(20.494)
which disagrees with the classical field result but agrees with experi-
ment.

The N~k~ part corresponds to the classical result.

The +1 part is a purely quantum mechanical effect.

This term implies that there is an emission process that can take place even if
there is no external field present.

This process is called spontaneous emission. A clear victory for the quantum
approach.

1617
20.8 Problems
20.8.1 Dirac Spinors
p
The Dirac spinors are (with E = p~2 + m2 )

p
/+m p/+m 0
   
s
u(p, s) = , v(p, s) =
E+m 0 E + m s

where p/ = p , s (s = 1/2) are orthonormalized 2-spinors and similarly for

s . Prove(using u = u+ 0 , etc):

(a) u(p.s)u(p.s0 ) = v(p, s)v(p, s0 ) = 2mss0

(b) v(p, s)u(p, s0 ) = 0

(c) u(p, s) 0 u(p, s0 ) = 2Ess0

P
(d) s u(p.s)u(p, s) = p /+m
P
(e) s v(p.s)v(p, s) = p /m

(f) u(p, s) u(p0 , s0 ) = 2Ess0 = 1

2m u(p, s) [(p + p0 ) + i (p p0 ) ] u(p0 , s0 )
(T he Gordon Identity)

20.8.2 Lorentz Transformations

In a Lorentz transformation x0 = x the Dirac wave function transforms as
0 (x0 ) = S()(x), where S() is a 4 4 matrix.

(a) Show that the Dirac equation is invariant in form, i.e., i 0 m 0 (x0 ) =

0, provided
S 1 () S() =

(b) For an infinitesimal transformation = g + , where =

. The spin dependence of S() is given by I i /4. Show
that = i[ , nu ] satisfies the equation in part (a). For finite transfor-

mations we then have S() = ei /4 .

20.8.3 Dirac Equation in 1 + 1 Dimensions

Consider the Dirac equation in 1 + 1 Dimensions (i.e., one space and one time
dimension):  
0 1
i + i m (x) = 0
x0 x1

(a) Find a 2 2 matrix representation of 0 and 1 which satisfies { , } =

2g and has correct hermiticity. What is the physical reason that can
have only two components in 1 + 1 dimensions?

1618
 (b) Find the representation of 5 = 0 1 , 5 and = 21 i [ , ]. Are
they independent? Define a minimal set of matrices which form a complete
basis.

(c) Find the plane wave solutions + (x) = u(p1 )eipx and (x) = v(p1 )eipx
in 1 + 1 dimensions, normalized to uu = vv = 2m (where u = u+ 0 ).

20.8.4 Trace Identities

Prove the following trace identities for Dirac matrices using only their property
{ , } = g (i.e., do not use a specific matrix representation)

(a) T r( ) = 0

(b) T r( ) = 4g

(c) T r( ) = 0

(d) T r( ) = 4g g 4g g + 4g g

(e) T r(5 ) = 0 where 5 = i 0 1 2 3

20.8.5 Right- and Left-Handed Dirac Particles

The right (R) and left (L) -handed Dirac particles are defined by the projections

1 1
R (x) = (1 + 5 )(x) , L (x) = (1 5 )(x)
2 2
In the case of a massless particle (m=0):

(a) Show that the Dirac equation (i/ eA)/ = 0 does not couple R (x) to
L (x), i.e., they satisfy independent equations. Specifically, show that in
the chiral representation of the Dirac matrices
   
0 I 0
0 = , =
I 0 0

we have  
R ipx
= e
L

i.e., that the lower(upper) two components of R (L ) vanish.

(b) For the free Dirac equation (A = 0) show that R and L are eigen-
states of the helicity operator 21 p with positive and negative helicity,
respectively, for plane wave states with p0 > 0.

1619
20.8.6 Gyromagnetic Ratio for the Electron
(a) Reduce the Dirac equation (i/ eA
/ m) = 0 by multiplying it with
/ /
(i eA + m) = 0 to the form
h e i
(i eA)2 F m2 = 0
2
i
where = 2 [ , ] and the field strength F = A A .
(b) Show that the dependence in the magnetic field B = A in the spin-
dependent term F is of the form (ge/2m) 12 B when the kinetic
energy is normalized to 2 /2m ( = 5 0 is the spin matrix). Deter-
mine the value of the gyromagnetic ration g for the electron.

20.8.7 Dirac Schrodinger

/ m) = 0 for the Hydrogen atom (A0 =
Reduce tyhe Dirac equation (i/ eA
Ze/4r , A = 0) to the standard Schrodinger equation
2
 

i (t, boldsymbolx) = + eA0 (t, boldsymbolx)
t 2m
in the non-relativistic limit, where |p| , A0  m. HINT: You may start from
the reduced form of the Dirac equation in Problem 20.6(a). Extract the leading
time dependence by writing (x) = (t, x)eimt .

20.8.8 Positive and Negative Energy Solutions

Positive energy solutions of the Dirac equation correspond to the 4-vector cur-
rent J = 2p = 2(E, p~), E > 0. Show that the negative energy solutions
correspond to the current J = 2(E, p~) = 2(|E|, ~p) = 2p , E < 0.

20.8.9 Helicity Operator

(1) Show that the helicity operator commutes with the Hamiltonian:
h i
~ p, H = 0

(2) Show explicitly that the solutions to the Dirac equation are eigenvectors
of the helicity operator: h i
~ p =

20.8.10 Non-Relativisitic Limit

Consider  
uA
=
uB
to be a solution of the Dirac equation where uA and uB are two-component
spinors. Show that in the non-relativistic limit uB = v/c.

1620
20.8.11 Gyromagnetic Ratio
Show that in the non-relativisitc limit the motion of a spin 1/2 fermion of charge
e in the presence of an electromagnetic field A = (A0 , A)~ is described by
" #
(~ ~ 2
p eA) e
~ + eA0 = E
~ B
2m 2m

where B~ is the magnetic field, i are the Pauli matrices and E = p0 m. Identify
the g-factor of the fermion and show that the Dirac equation predicts the correct
gyromagnetic ratio for the fermion. To write down the Dirac equation in the
presence of an electromagnetic field substitute: p p eA .

20.8.12 Properties of 5
Show that:

(a) 5 is a pseudoscalar.

(b) 5 is an axial vector.

20.8.13 Lorentz and Parity Properties

Comment on the Lorentz and parity properties of the quantities:

(a) 5

(b) 5 5

(c) 5

(d) 5 5

(e)

20.8.14 A Commutator
Explicitly evaluate the commutator of the Dirac Hamiltonian with the orbital
angular momentum operator L for a free particle.

20.8.15 Solutions of the Klein-Gordon equation

Let (~r, t) be a solution of the free Klein-Gordon equation. Let us write
2
(~r, t) = (~r, t)eimc t/~

Under what conditions will (~r, t) be a solution of the non-relativistic Schrodinger

equation? Interpret your condition physically when is given by a plane-wave
solution.

1621
20.8.16 Matrix Representation of Dirac Matrices
The Dirac matrices must satisfy the anti-commutator relationships:

{i , j } = 2ij , {i , } = 0 with 2 = 1

(1) Show that the i , are Hermitian, traceless matrices with eigenvalues 1
and even dimensionality.
(2) Show that, as long as the mass term mis not zero and the matrix is
needed, there is no 2 2 set of matrices that satisfy all the above rela-
tionships. Hence the Dirtac matrices must be of dimension 4 or higher.
First show that the set of matrices {I, ~ } can be used to express any 2 2
matrix, i.e., the coefficients c0 , ci always exist such that any 2 2 matrix
can be written as:  
A B
= c0 I + ci i
C D
Having shown this, you can pick an intelligent choice for the i in terms
of the Pauli matrices, for example i = i which automatically obeys
{i , j } = 2ij , and express in terms of {I, ~ } using the relation above.
Show then that there is no 2 2 matrix that satisfies {i , } = 0.

20.8.17 Weyl Representation

(1) Show that the Weyl matrices:
   
~ 0 0 I

~= , =
0 ~ I 0

satisfy all the Dirac conditions of Problem 20.16. Hence, they form just
another representation of the Dirac matrices, the Weyl representation,
which is different than the standard Pauli-Dirac representation.
(2) Show that the Dirac matrices in the Weyl representation are
   
0 ~ 0 I
~ = , 0 =
~ 0 I 0
 
0 1 2 3 I 0
(3) Show that in the Weyl representation 5 = i =
0 I
p~ + m] = E in the particle rest frame
(4) Solve the Dirac equation [~
using the Weyl representation.
(5) Compute the result of the chirality operators
1 5
2
when they are acting on the Dirac solutions in the Weyl representation.

1622
20.8.18 Total Angular Momentum
Use the Dirac Hamiltonian in the standard Pauli-Dirac representation

~ p~ + m
H=

to compute [H, L] and [H, ] and show that they are zero. Use the results to
show that:
[H, L + /2] = 0
where the components of the angular momentum operator are given by:

Li = ijk xj pk

and the components of the spin operator are given by:

 i 
0
i =
0 i

Recall that the Pauli matrices satisfy i j = ij + iijk k .

20.8.19 Dirac Free Particle

The Dirac equation for a free particle is

|i
= cx px + cy py + cz pz + mc2 |i


i~
t
Find all solutions and discuss their meaning. Using the identity
~ B)
(~ A)(~ ~ =A
~B
~ + i~ (A
~ B)
~

will be useful.

1623
1624
Chapter 21

Interacting Fields and Feynman Diagrams

21.1 Relativistic Quantum Mechanics of Free Par-

ticles
21.1.1 Hilbert Space
In a quantum mechanical description, the state of a free, spinless particle is
completely specified by its three-momentum since its energy follows from the
energy-momentum relation. This is called the momentum-space description. A
particle with well-defined momentum (and energy) is described by a plane wave
and E = h.

In addition, we will assume superposition is valid for quantum mechanical states.

Finally, we will assume that only particles of one type exist. Let us now imple-
ment these statements mathematically. We will set ~ = 1 and c = 1 from now
on.

We assume the existence of an infinite dimensional Hilbert space and that for
every possible physical state there is a vector in the space. We choose the vec-
tors corresponding to a particle with momentum p~, |~ pi as basis vectors of length
1. The Hilbert space basis is thus:
|0i corresponding to the vacuum
|~
p1 i corresponding to a particle of momentum p~1
|~
p2 i corresponding to a particle of momentum p~2
.....................................................................
|~
p1 , ~q1 i corresponding to two particles one of momentum p~1 , the other ~q1
and so on
The basis vectors are orthonormal, i.e.,
h~
p | ~qi = p~q~ (21.1)

1625
Since the range of momenta is infinite the subspace of the single particles is
infinite dimensional.

For mathematical convenience, we will assume that the universe is a cube with
volume V and we allow only those wave functions whose value on the boundary
is the same as the value on the opposite boundary (called periodic boundary
conditions). This means that we are working with a infinite, but denumerable,
dimensional Hilbert space, where the momenta are restricted to the discrete
values (L = length of a side of the cube) given by
2n1 2n2 2n3
p1 = , p2 = , p3 = n1 , n2 , n3 = 0, 1, 2, .... (21.2)
L L L
At the end of any calculations, we take the limit V thus returning to the
true continuous case. In the limit any sum over momenta becomes an integral
by the rule Z
X V
d3 p (21.3)
(2)3
p
~

The factor
V
(21.4)
(2)3
arises because in a little box d3 p there will be that many possible states(possible
momentum values in phase space).

In addition we have
V
lim p~ p~ = (3) (~
pi p~j ) (21.5)
V (2)3 i j

What we are doing here is of course quite horrible for any relativistic theory.
We are violating Lorentz invariance.

A cubical box of volume V is not Lorentz invariant. However, we assume that

our final results will be Lorentz invariant in the limit of infinite V . This plague,
having to abandon Lorentz invariance in order to define the formalism, is com-
mon to all approaches to quantum field theory(QFT). One always needs some
kind of grid to work on. The final results, the Feynman rules, do not suffer,
however, from the breaking of Lorentz invariance.

Since quantum mechanics will form the basis of our derivation of QFT and the
Feynman rules, a review(from a different point of view) of the basic concepts is
in order.

The wave function of a non-relativistic particle is

1
(x, t) = ei(~p~xEp t) (21.6)
V

1626
where
1 p~2
Ep = non - relativistic kinetic energy = m~v 2 = (21.7)
2 2m
and the normalization is such that the total probability (integrating over the
whole universe of volume V is 1. This wave function describes one particle in
the universe and it is a solution of Schrodingers equation
1 2
= i (21.8)
2m t
The relativistic generalization of the wave function is simple:
1 1
(x) = ei(px) = ei(~p~xEt) (21.9)
V V
where p
E = relativistic kinetic energy = p~2 + m2 (21.10)
and the wave function is now a solution of the Klein-Gordon equation:
2
 
2 + 2 m2 (x) = E 2 p2 m2 (x) = 0


(21.11)
t
In making the relativistic transition, the definition of probability needs some
revision. In the non-relativistic theory the probability density to find a particle
at some point x within a small box d3 x is given by
2
|(x)| d3 x (21.12)

where the integral of this expression over the whole volume must be equal to
1. However, since the volume is not a Lorentz invariant, we cannot maintain
this definition of probability density. Probability density does not need to be
a Lorentz invariant, only the total probability must be Lorentz invariant (and
equal to 1). This is very much like electric charge. In fact, if we assume that like
electric charge density and current, the probability density (P0 ) and its associ-
ated probability current (P~ = (P1 , P2 , P3 )) satisfies a conservation (continuity)
equation, then we have (pushing on the analogy)

P = 0 j = 0 (21.13)

It then follows that the total probability (like total electric charge) is given by
Z
P = d3 x P0 (x) (21.14)

In the same manner that we can prove total charge is constant, we can then
prove that P is a constant or P/t = 0 (assumes that P is zero on the bound-
ary of a surface integral).

For consistency, we must have that the charge density is proportional to prob-
ability density.

1627
The formal definition of P is

(x) (x)
P (x) = i (x) i (x) (21.15)
x x

In this case, since, in the low energy limit, E m and P0 2m , by

normalizing as follows
1
(x) = ei(px) (21.16)
2V p0
we have normalization equal to 1 when integrating P0 over the whole volume
V.

Since particles do have a location in space, we must now consider states that
are not pure plane waves. A particle that we know is precisely at the point ~x
at time x0 (we will take x0 = 0) is described at time 0 by a function

(~x, 0) = C (3) (~x) (21.17)

where C is a normalization constant. This can be seen as a superposition of

pure momentum states:
Z
C
(~x, 0) = d3 p ei~p~x (21.18)
(2)3

which we generalize to
Z
C
(~x, t) = (x) = d3 p ei(px) (21.19)
(2)3

where
px = p0 x0 p~ ~x = Et p~ ~x , E 2 = p~2 + m2 (21.20)
so that it satisfies the Klein-Gordon equation.

If the location is not a function but more smeared out, then we have, in
general: Z
C
(x) = d3 p f (p)ei(px) , p20 = p~2 + m2 (21.21)
(2)3
For example, if f (p) = 1, we have a function in space and if f (p) = (3) (~
p ~q)
we have a state with sharply defined momentum ~q. Therefore, the vector in
Hilbert space that corresponds to a sharply defined location at some time will
be a superposition of sharp momentum states with equal weight:
X
|~x, x0 i = Cei(px) |~
pi (21.22)
p
~

What happens to a state in Hilbert space when we apply a Lorentz transforma-

tion? The Lorentz transformation L directly transforms any momentum p into

1628
p0 where Lp = p0 . It is a 4 4 matrix in a 4-dimensional space. In Hilbert space
vectors, it must transform the arguments of all of the different ket vectors,i.e.,

|~ p01 i
p1 i |~
|~ p02 i
p2 i |~
|~ p01 , ~q10 i
p1 , ~q1 i |~

and so on.

This is an transformation matrix. Luckily we never need to write it

down, but only know what it does. Let us consider a simpler example, namely,
the behavior under translations. Under a space-time translation b = (b0 , ~x)
a particle sharply located at place ~x, time x0 , will become a particle sharply
located at place ~x + ~b and time x0 + b0 . Thus

T (b) : |xi |x + bi (21.23)

and X X
ei(px) |~
pi ei(px)+i(pb) |~
pi (21.24)
p
~ p
~

pi goes over into the state eipb |~

Therefore, under a translation the state |~ pi. This
shows explicitly how the transformation of the vectors in Hilbert space differs
from what happens in ordinary space. The vector just gets a phase factor, that
is, it stays normalized and represents, physically speaking, the same state.

General Rule
To every Lorentz transformation there corresponds a transformation in Hilbert
space.

Our physics must be unique, so we insist that we have a 1-1 correspondence.

This means that labeling the Hilbert space transformation by X,

L1 X1
L2 X2
L3 = L1 L2 X1 X2 = X3

which implies that the transformations in Hilbert space are a representation of

the Lorentz group.

Matrices in Hilbert Space

In order to describe physical processes we must introduce operators (matrices)
in the Hilbert space. A typical process, for example electrons scattering off a
proton, involves different states from the point of view of Hilbert space. Thus
initially we may have an electron of momentum p~ and a proton of momentum

1629
  E
p, ~k in Hilbert space and the final state
~k, which corresponds the basis vector ~
may then contain an electron and proton of  momenta E p~ 0 and a proton of mo-
mentum ~k 0 , corresponding to the vector ~ p 0 , ~k 0 in Hilbert space. It seems

like we might think of this as if the physical system corresponds
 E  to aEvector in
 ~  0 ~0
Hilbert space that rotates as a function of time from ~ p, k to ~ p ,k .

We cannot, however, describe things in this way. The reason is that scattering
clearly involves interaction between particles, and we have set up our Hilbert
space for free particles only. We must rethink our procedures if we want to
introduce interactions.

Right now it suffices to say that physical quantities will correspond to the el-
ements of a certain matrix defined in Hilbert space. What we need is basic
building blocks, in some sense matrices like the Pauli spin matrices that can be
used to describe the full set of 2 2 matrices. The matrices that we need will
fulfill certain basic requirements, in particular the requirement of locality. This
is the requirement that physical processes cannot influence each other if they
are outside each others light cone, i.e., if speeds larger than that of light are
needed to connect the events.

This we hope will be achieved by insisting that the operator (matrix) describing
a process at the space-time point x will commute with a similar operator for
the space-time point y if x and y are outside each others light cone.

Locality is a touchy point of quantum theory, and to a large extent it remains,

in its formulation, an article of faith. There is however, at this time, no reason
to suspect anything wrong with the procedure adopted above.

To explain what we are going to do let us first consider three-dimensional space.

What would be the most elementary building blocks that can be used to build
up any 3 3 matrix? We can do this with two matrices:

0 1 0 0 0 0
a= 0 0 2 , a = 1 0 0 (21.25)
0 0 0 0 2 0

This can be seen by working out the various products involving a and a and
showing that any 3 3 matrix can be obtained as
a linear combination of these
matrices and associated products. The factor 2 was introduced for reasons
that will become transparent later.

We now turn to Hilbert space.

The above example can be used for the subspace referring to a definite momen-

1630
tum p~ . A matrix a as above can be constructed in the subspace of states:

1 0 0
|0i = 0 , |~ pi = 1 , |~ p, p~,i = 0 (21.26)
0 0 1

These matrices represent a universe with n particles all with the same momen-
tum p~ where 0 n 2 as can be seen from the relations

a |0i = 0 , a |~pi = |0i , a |~ p, p~i = 2 |~ pi

a |0i = |~
pi , a |~
pi = 2 |~
p, p~i , a |~
p, p~i = 0
aa |0i = 0 , aa |~
pi = 1 |~
pi , aa |~
p, p~i = 2 |~
p, p~i

We may extend the example to more dimensions. In a large dimensional space

we still can do everything with two matrices a and a with

0 1 0 0 0 . . .

0 0 2 0 0 . . .


0 0 0 3 0 . . .

0 0 0 0 4 . . .
a= (21.27)

. . . . . . . .


. . . . . . . .

. . . . . . . .
. . . . . . . .

The above example can be used for the subspace referring to a definite momen-
tum p~. A matrix a as above can be constructed in the subspace of states

|0i , |~
pi , |~
p, p~,i , etc (21.28)

The matrix a when applied to a state vector with n particles of momentum p~

the vector for a state with n 1 particles of momentum p~ with a
then gives
factor n as we saw in the special case 0 n 2 above.

Examples
Consider

a |~
p, p~, p~i = 3 |~
p, p~i , a |0i = 0 (21.29)
This particular matrix operating in the p~ subspace will be denoted by a(~ p).
Similarly, we will have matrices for any other subspace, and we then have an
p) (with a+ = aT which is the same as aT
p) and a+ (~
infinite set of matrices a(~
since a is real here).

Having done this, we do not need to introduce new matrices for that part of
Hilbert space where we have particles of different momenta, such as the state

1631
|~
p, p~, ~q, ~q, ~qi with p~ 6= ~q. The matrices a(~
p) and a(~q) will by definition act on
these states as if the other particles are not there. Thus (with p~ 6= ~q):
a(~p) |~q, ~q, ~q, ~q, ~qi = 0

a(~p) |~p, p~, ~q, ~q, ~qi = 2 |~
p, ~q, ~q, ~qi

a(~q) |~
p, p~, ~q, ~q, ~qi = 3 |~p, p~, ~q, ~qi
and so on.

We now have a set of matrices a(~ p) and a+ (~ p), called annihilation and creation
operators respectively, defined over the whole Hilbert space, that can be used
to build up any other matrix(operator). Note that by construction a(~ p)a(~q) =
p) if p~ 6= ~q and of course also if p~ = ~q. They do not interfere with each
a(~q)a(~
other.

The matrix X p
H= p0 a+ (~
p)a(~
p) , p0 = p~2 + m2 (21.30)
p
~

is diagonal(as seen from the relations in the above example) and it is the energy
operator, that is:
H |i = E |i (21.31)
where E is the total energy of the state |i. Here may be any number of
particles of any momentum. Explicitly, we have
H |0i = 0 , H |~qi = q0 |~qi ,
H |~
p, p~i = 2p0 |~
p, p~i , H |~
p, ~qi = (p0 + q0 ) |~
p, ~qi
and so on.

21.1.2 Fields
The Fourier transforms of the matrices a and a+ are called fields. To be precise
we have the field A(x):
X 1
A(x) = p)eipx
a(~ (21.32)
2V p0
p
~

Similarly
X 1
A+ (x) = p)eipx
a+ (~ (21.33)
2V p0
p
~

Notice that A(x) is no longer a real matrix due to the complex factors exp(ipx).
The matrices A(x) and A+ (x) can be taken as basic building blocks also since
a can be recovered from A. For example, a(~q) can be obtained from A(x) by
another Fourier transformation:
Z
1
d3 x eiqx A(x) = a(~q) (21.34)
2V p0

1632
The Hermitian combination

(x) = A(x) + A+ (x) (21.35)

is called the field corresponding to the particles considered. This field has a
number of properties that make it very useful for the construction of physical
quantities. The main property is that it is local.

The commutator
[(x), (y)] = (x)(y) (y)(x) (21.36)
is zero if x and y are outside of each others light cone. To see this we compute
things step by step:

[A(x), A(y)] = 0 This follows because a(~

p)a(~q) = a(~q)a(~
p)
 + + +
p)a (~q) = a+ (~q)a+ (~
+

A (x), A (y) = 0 This follows because a (~ p)
X 1
A(x), A+ (y) = eip(xy)
 
(21.37)
2V p0
p
~
 +  X 1 ip(xy)
A (x), A(y) = e (21.38)
2V p0
p
~

These last two results follow from the equations

p), a+ (~q) = p~q~ = a+ (~

   
a(~ p), a(~q) (21.39)

Altogether we find:
X 1  ip(xy) 
[(x), (y)] = e eip(xy) (21.40)
2V p0
p
~

We now show that the right hand side is zero if x and y are outside each others
light cone. First we take the continuum limit, V :
X Z V
d3 p
(2)3
p
~

Calling the right hand side of the above commutator the function c (x y) we
have, Z
1 3 1  ip(xy) ip(xy)

c (x y) = d p e e (21.41)
(2)3 2p0
To prove that this is zero if the four vector z = x y is outside the light cone
(which means (zz > 0) we proceed in two steps. First we will show that c is
Lorentz invariant. Then we will show that c (z) is zero for any z with z0 = 0
by a Lorentz transformation (that leaves c unchanged) we will have proven
the required result. The argument is illustrated in Figure 21.1 below.

1633
 Figure 21.1: Light Cone Arguments

If the function is zero at the location of the cross it is zero in the whole
shell(formed from curve shown) going through that point, because all the points
in the shell can be obtained from the cross point by means of a Lorentz transfor-
mation, and the function is Lorentz invariant. If the function is zero along the
whole horizontal axis (the equal time line) then the function is zero everywhere
outside the light cone.

To show the Lorentz invariance we first rewrite the three-dimensional integral

as a four-dimensional integral. This may be done as follows.
p
Note that in the above p0 = p~2 + m2 . We write

Zq
1
= d( 2 p~2 m2 )() (21.42)
2p0
q

where (
1 if > 0
() = (21.43)
0 if < 0

Indeed, the only points contributing to the integral are

p
= p~2 + m2 (21.44)

We may write
p p
( 2 p~2 m2 ) = (( p~2 + m2 )( + p~2 + m2 )) (21.45)

Now
1 1
(ab) = (b) + (a) (21.46)
|a| |b|

1634
 p
Furthermore, the solution = p~2 + m2 will not give anything because the
function restricts us to positive values for . Thus, we get
Zq
1 1 p 1
= d p ( p~2 + m2 )() = p (21.47)
2p0 2
+ p~ + m2 2 p~ + m2
2
q

which is in agreement with what we stated above.

Although it might cause some confusion we are going to use for the name p0 ,
and the above result is
Zq
1
= dp0 (p20 p~2 m2 )(p0 ) (21.48)
2p0
q

Using this result, the function c becomes

Z Zq
1  
c (z) = 3
d p dp0 (p20 p~2 m2 )(p0 ) eip(z) eip(z)
(2)3
q
Z
1
d4 p(p2 m2 ) eipz eipz (p0 )


= (21.49)
(2)3

where we have used (a) = (a).

We now show that this integral is Lorentz invariant. Thus, we will let z = Lz 0 ,
where L is some Lorentz transformation, and we will show that

c (z) = c (z 0 ) (21.50)

We have
Z
01  0 0

c (z) = c (Lz ) = d4 p(p2 m2 ) eipLz eipLz (p0 ) (21.51)
(2)3
Now introduce four new variables q1 , q2 , q3 , q4 related to the p by p = Lq. This is
as if we did a Lorentz transformation on p, but it is really a change of integration
variables. The integration volume element d4 p becomes d4 q times the Jacobian
of transformation
d4 p = det(L)d4 q (21.52)
Since det(L) = 1 that gives no change. Furthermore:

(p, Lz 0 ) = (Lq, Lz 0 ) = (q, z 0 )

p2 = (p, p) = (Lq, Lq) = (q, q) = q 2

Finally, what happens with (p0 )? This is more subtle and requires a detailed
investigation of the action of a Lorentz transformation on the vector p. This

1635
vector is restricted to values inside the upper light cone, because we must have
p2 = m2 (p is said to be on the mass-shell ) and the function restricts us to
the upper light cone. Since any four-vector in the upper light cone transforms
into another vector in the upper light cone, q will also be in the upper light cone.
Therefore, (q0 ) will also be non-zero if (p0 ) was nonzero and zero if (p0 ) was
zero. In other words, (p0 ) = (q0 ) if p = Lq, but for this the function is
crucial, because otherwise there would not be the restriction to the upper light
cone. The result of all this is:
Z
1 
iqz 0 iqz 0

c (z) = d4
q(q 2
m 2
) e e (q0 ) = c (z 0 ) (21.53)
(2)3

The last step follows since this differs from the original expression only by a
different notation for the integration variables.

Now step 2. Let us suppose z0 = 0. Then

Z
1 4 2 2

iqz 0 iqz 0

c (z)z0 =0 = d q(q m ) e e (q0 ) (21.54)
(2)3

The actor restricts the q0 integration but not the ~q integration. The ~q integral
goes from to + for every component. Therefore
Z Z
d3 qf (~q) = d3 qf (~q) (21.55)

and we see that the two terms cancel and thus

c (z)z0 =0 = 0 (21.56)

This concludes the proof.

Another important property of the field is that it obeys the Klein-Gordon

equation. The complete expression for the field (x) is:
X 1 p
(x) = p)eipx + a+ (~
(a(~ p)eipx ) , p0 = p~2 + m2 (21.57)
2V p0
p
~

It is easy to see that

2
 
2 2
+ m (x) = 0 (21.58)
t2

due to the mass-shell relation for p0 . Finally, operating with (x) on the vacuum
state we obtain the state for a particle located at the point x at time x0 = 0 we
have
X 1 X 1
(x) |0i = (a(~p)eipx + a+ (~p)eipx ) |0i = eipx |pi (21.59)
2V p0 2V p0
p
~ p
~

1636
We can also derive commutation rules for equal times.

[(x), (y)]x0 =y0 = c (z)z0 =0 = 0 , z = x y (21.60)

   
(x) c (z)
, (y) =i = i (3) (~x ~y ) (21.61)
x0 x0 =y0 x 0 z0 =0

21.1.3 Structure of Hilbert Space

The connection between vectors in Hilbert space and measurements in quantum
mechanics is through probability. Physical states correspond to vectors(of unit
length) in Hilbert space, where length is defined by the usual scalar-product in
complex space. If a vector |i has the components 1 , 2 , ..., that is

|i = 1 |1i + 2 |2i + ...... (21.62)

where |1i , |2i , ..., etc, represent the first, second, etc, basis vectors. Then the
scalar-product of |i with itself is given by

h | i = 1 1 + 2 2 + ...... (21.63)

Similarly, X
h | i = 1 1 + 2 2 + ...... = i i (21.64)
i

The connection with physical measurements is now as follows. Suppose we have

a vector |ci corresponding to some physical system. Now, let there be two other
mutually orthogonal vectors |ai and |bi, also corresponding to physical systems.
Suppose now |ci = |ai + |bi. Remember that |ai, |bi and |ci must all be of
unit length, because they represent physical states.

Now, if the system is in the state described by |ci and one tries to measure
whether the system is in state |ai or |bi one will find:
2
probability to find system in state |ai = || (21.65)
2
probability to find system in state |bi = || (21.66)

Here in means that the system is measured to have the same properties as state
a or state b.

More generally, if a system is in a state |ci, then the probability to observe the
state |ai is given by
2
|ha | ci| (21.67)
This is the fundamental connection between Hilbert space and physical mea-
surements. Since hc | ci = 1 we must have
2 2
|| + || = 1 (21.68)

1637
The two probabilities add up to 1, as should, of course, be the case.

There is some potential trouble here in connection with Lorentz invariance,

quite different from the trouble arising from dealing with a finite volume. As
noted before, states in Hilbert space are generally not invariant under Lorentz
transformations, but they transform according to some representation of these
Lorentz transformations. The trouble now is that we certainly want probabil-
ities to be invariant under Lorentz transformations, in other words, the above
defined scalar-product must be invariant.

In addition, since the scalar-products are no longer necessarily positive, we must

then insist that states with negative probability are unphysical.

By itself this is not new: even in ordinary space there are restrictions, and mo-
menta of particles are physical only if their scalar-product with themselves is
positive (corresponding to real mass). Particles with momenta such that (pp) is
negative have not been seen.

So, much the same as in ordinary space where some domains seem to be ex-
cluded for physics, the same might happen in Hilbert space. In Hilbert space, we
would be truly in difficulty if we had to allow for physical systems with negative
probability. No consistent theory can then be constructed, because probability
is positive by its very definition.

For scalar or pseudoscalar particles there is not yet any problem, the transfor-
mations in Hilbert space are rather trivial (as for example the Hilbert space
vector |~
pi transforming to the vector |~qi where ~q is the Lorentz transform of p~.
In dealing with particles with spin, such as electrons and photons, the transfor-
mation of the states in Hilbert space becomes more complicated and we leave
that for a more advanced text.

21.2 Interacting Fields - Part 1

21.2.1 Physical System
We will now discuss systems of interacting particles.

We use the Lehmann-Symanzik-Zimmermann formalism, and Kallens method

to derive the S-matrix from the equations of motion.

For definiteness, we assume the existence of two kinds of particles called and .
Both are spinless, since including that complication is not needed to understand
the derivation of Feynman rules. We assume masses M and m for the and
respectively. This model is a simplified version of electrodynamics of electrons
and photons. The will play the role of the photon.

1638
To begin we will focus on a specific problem, namely scattering. The physical
process is shown in Figure 21.2 below.

Figure 21.2: scattering

Two pions with momenta p and q (with p2 = q 2 = M 2 ) meet and scatter, and
we are interested in the probability that a final configuration of two pions with
momenta p0 and q 0 (with p 02 = q 02 = M 2 ) is produced. This probability, when
multiplied with the appropriate flux factors, will be the differential cross section
for this process.

21.2.2 Hilbert Space

This aspect needs very careful consideration. A vector in Hilbert space repre-
sents a physical state.

What is a physical state?

A physical state is simply a possible physical system, with particles moving here
and there, with collisions, with dogs chasing cats, with people living and dying,
with all kinds of things happening.

Often people make the mistake of identifying a physical state with the system
at a given moment, one picture from a movie, but that is not what we call the
physical state. The system at some moment may be seen as a boundary condi-
tion, that is, if one knows the whole system at some moment, and one knows
the laws of nature, then in principle we can deduce the rest!

Thus a physical state may be characterized by the system at a definite moment,

but the state itself refers to the whole world including its progress in time.

Conveniently, especially for scattering processes one may use the time points
. Thus the above process corresponds to a vector in Hilbert space, and we
can denote that vector by
|p, qiin (21.69)
By this we mean: that physical system that has two pions of momenta p and q at
time t = (the in configuration). It must be understood that |p, qiin contains
everything, including how the system looks at other times. For example, we

1639
could define the state |p, qi0 as that physical system which has at time t = 0
exactly two pions of momenta p and q. The above described state, |p, qiin ,
has two such pions at t = , but it may well be that they scatter before
t = 0, and, thus, the probability that we have still two pions with that same
momentum at t = 0 is smaller than one.

The above discussion leads up to the following identification.

Let |p, qiin and |p, qi0 be the systems as described above. They are different
systems. Let the system be in the state |p, qiin (two pions at t = ). The
probability of having two pions with momenta p and q at t = 0 is the square of
the absolute value of the scalar-product between the states:
2
|0 hp, q | p, qiin | (21.70)

If some collision took place before t = 0 we may actually still have two pions
but with different momenta, say k and r. A state with two pions with momenta
k and r at time t = 0 is denoted by |k, ri0 . If the system is in the state |p, qiin
then the probability of observing two pions of momenta k and r at t = 0 is given
by:
2
|0 hk, r | p, qiin | (21.71)
Thus the state |p, qiin when viewed at time t = contains two pions of
momenta p and q, but if we look to it at time t = 0 we see with some probability
two pions that may or may not have the momenta p and q. More generally, new
particles may be produced in a collision, so we may also see three, four, etc.
pion configurations. For example,
2
|0 hk, r, s | p, qiin | (21.72)

gives the probability of observing three pions of momenta k, r and s at time

t = 0 if we know that the system is in a state characterized by the fact that at
time t = there were two pions of momenta p and q.

Magnitude of Hilbert Space

It follows from the above that we can have as many states in Hilbert space as
possible systems at t = .

We now assume the following asymptotic condition: if particles are sufficiently

far apart in space they do not interact and behave as free particles. If we go
sufficiently far back in time, particles will be separated. Therefore, we may
assume that the possible systems at t = are precisely the systems of non-
interacting particles.

Thus the Hilbert space of interacting systems is by this assumption equally as

large as the Hilbert space of free particles.

1640
Clearly, we will have to modify this if we want to consider stable bound states.
No matter how far back we go in time, the electron and proton in a hydrogen
atom do not separate. To describe such systems properly we must enlarge
Hilbert space and allow states containing hydrogen atoms. Of course, such
atoms again can just be considered as a new kind of particle, and the Hilbert
space becomes then effectively the free Hilbert space of three (in this case)
particles (electrons, protons and hydrogen atoms). We will consider, however,
only simple particle states.

Now that we are clear about the meaning of states and their representation in
Hilbert space we can proceed and postulate equations that will describe particles
in interactions. Experiment must then decide which equations describe nature.
Of course, whatever we postulate, it will be within the framework of Lorentz
invariant quantum mechanics. Only a limited degree of freedom is left.

U-matrix, S-matrix
In describing scattering problems, where one typically considers initial and final
configurations of widely separated particles, the description in terms of in and
out states is advantageous. A vector |aiin in Hilbert space corresponds to a
physical system characterized by the configuration a at time t = . Similarly
a vector |biout corresponds to a system characterized by the configuration b at
time t = +. The states |aiin can be counted in the same way as the free
particle states:
|0iin = vacuum at t =
|~
p1 iin = one particle (pion) with momentum p~1 at t =
..............................................................
|~
p1 , p~2 , ..., ~q1 , ~q2 , ......iin = pions with momenta p~1 , p~2 , ... and
particles with momenta ~q1 , ~q2 , ...... at t =
A remark needs to be made here.

Since for free particles the energy is known if the three-momentum is known
the state is characterized by the three-momenta only. That is why we used the
three-vector as argument in |~p1 i rather than |p1 i. In the following we will often
drop the arrow, assuming that the reader is aware that the particles indicated
are on mass shell(p2 = m2 ). Note that for finite times, when the particles are
not necessarily far apart, the energy is not simply given by the usual mass shell
relation.

Similarly, for out states.

Of course, in general if there is interaction the in states are different from the
out states, although both are in the same Hilbert space. Thus
|~
p1 , p~2 iin 6= |~
p1 , p~2 iout (21.73)

1641
because a system that at t = has two pions with momenta p~1 and p~2 is
unlikely to still have two pions with momenta p~1 and p~2 at t = +. There is
some probability that the pions do not scatter; it is given by the absolute value
squared of the scalar-product between the two states, that is,
2
|out h~
p1 , p~2 | p~1 , p~2 iin | (21.74)

is the probability that, starting with two pions with momenta p~1 and p~2 at
t = we will still find two pions with momenta p~1 and p~2 at t = +.

Similarly,
2
p 0 , ~q 0 | p~, ~qiin |
|out h~ (21.75)
is the probability that when measuring on a system characterized by there being
two particles of momenta p~ and ~q at t = we will find two particles of
momenta p~ 0 and ~q 0 at t = +.

We thus have two sets of basis vectors in the same Hilbert space, namely the in
basis (|0iin , |~
p1 iin , .....) and the out basis (|0iout , |~
p1 iout , .....). Since a system
without any particles at t = will still not have any particles at t = +, we
have |0iin = |0iout . Similarly for one particle states, |~ p1 iin = |~ p1 iout . But for
two or more particle states this is not true if there is any interaction.

Since physical states correspond to vectors of unit length both the in and out
bases are orthonormal. Therefore, there must exist a matrix that transforms
the in basis into the out basis:

|aiout = S + |aiin (21.76)

for any configuration a. We have used S + (complex conjugation + transpose =

Hermitian) here purely by convention. The S-matrix clearly must contain all
physical information for any scattering process. For instance, the probability to
have configuration a at t = and to find configuration b at t = + is given
by
2 2 2 2
|out hb | aiin | = |in ha | biout | = in ha| S + |biout  = |in hb| S |aiin |

(21.77)

The first step is because ha | bi = hb | ai , and the last step is nothing but the
very definition of S + relative to S. Since both in and out states are orthonormal
sets, the S-matrix is unitary (scalar-products are unchanged), so that S + S = I.

Later we will interpret this relation as conservation of probability. If there is no

interaction the S-matrix is I(the identity). One therefore often writes

S = I + iT (21.78)

and the relation S + S = I becomes

i(T T + ) = T + T (21.79)

1642
Exactly the same as in the case of free particle states, we may define matrices
a and a+ in Hilbert space. We can do that on both in and out bases. Thus,
ain (~
p) is a matrix such that
 
 +  +
 
p) p~, p~, ..., ~q, , ......
ain (~ = n  p~, p~, ... , ~q, , ...... (21.80)

| {z }  | {z }
 n p~0 s n1 p~0 s
in in

Similarly aout (~
p) is defined by its action on the unit vectors of the out basis.

Note that at this point we have no idea what happens when we apply ain (~
p) on
some out basis vector.

From the ain (~

p) and aout (~
p) we may construct fields in (x) (and out (x)) as
before: X 1
p)eipx + a+ p)eipx


in (x) = ain (~ in (~ (21.81)
2V p0
The momentum p refers here to a momentum that can be found with the par-
ticles at t = , when they are presumably far apart, and thus their energy
will be purely kinetic(no potential energy).
p p
Thus, we then have p0 = p~2 + M 2 for pions and p0 = p~2 + m2 for particles,
which have the field in (x).

Since the S-matrix transformsthe in basis into out basis it must also transform
ain (~
p) into aout (~
p)

p)S + |~
Saout (~ p) |~
p, p~, .....iin = Saout (~ p, p~, .....iout

= nS |~ p, .....iout = n |~ p, .....iin
p) |~
= ain (~ p, p~, .....iin

We used S 1 = S + to transform out into in state. We thus see that

ain (~ p)S +
p) = Saout (~ (21.82)

Similarly
in (x) = Sout (x)S + (21.83)
Both fields, in and out satisfy the Klein-Gordon equation.

It must be understood that in and out are well defined for all space and time.
Thus in (x) for example is perfectly well defined and non-zero for x0 = +.

The essence of the above may be put as follows.

The assumption that for t = any physical system becomes a system of free

1643
particles allows a mapping of all possible physical systems (by how they are
at ) on all possible free particle systems. Then we can use the formalism
developed for those systems, and build fields. The fields so constructed are the
in fields. Similarly out fields, related to labeling physical systems by how they
are at +. This then exhibits the role of assumptions on asymptotic behavior,
which are clearly of fundamental importance.

Let us formulate this once more in different terms.

Consider a physical system as a movie, showing a lot of action as it is projected.

This movie is in a can, and the can may be labeled in different ways. The
labeling may be on the basis of the opening scene of the movie, or alternatively,
on the basis of the final scene.

Clearly, these are very different ways of labeling.

Only empty movies (the vacuum) or movies containing just one actor (one par-
ticle of some momentum) are likely to have identical in and out scenes.

A physical state in Hilbert space is like a movie in a can. It is the whole movie,
not just the opening scene, even if the can is labeled that way. Seeing things
this way it hopefully becomes clear that a progressing physical system is not a
vector in Hilbert space (such as |~p, ~qiin rotating to another state (|~
p, ~qiout ) in
the course of time.

A vector in Hilbert space has no time dependence, but, like in a movie, all action
is contained in that state.

In that sense the S-matrix is a cross-index register, showing the relation between
two labeling systems. Given the beginning scene of a movie, the S-matrix tells
us what the final scene is.

21.3 Interacting Fields - Part 2

21.3.1 Interpolating Fields
We are now in a position to formulate equations of motion that describe an
interaction.

We assume the existence of a field (x) (no in or out index) that is equal to
in (x) if x0 (the time) is and is equal to out (x) at x0 = +. Again,
in (x) is well defined at x0 = +, but then it will be very different from (x).
Thus, we have

lim (x) = in (x) , lim (x) = out (x) (21.84)

x0 x0 +

1644
Figure 21.3 below shows some attempt to visualize the system.

Figure 21.3: Visualizing in to out

Note that in (x) and out (x) are well defined for all times. Now, (x) will also
satisfy an equation of motion, but it will not be a simple Klein-Gordon equation.
We therefore write:
2
 
2 2
2 + M (x) = j(x) (21.85)
t

The minus sign is part of the definition. The quantity j(x) is called a current,
and j(x) is such that if we solve this equation with the boundary condition

(x) = in (x) at x0 = (21.86)

then for
x0 = + we will have (x) = out (x) (21.87)
If we know j(x), and can solve this equation then we can find out from in ,
and thus also determine the S-matrix since S relates in- and out-fields.

Unfortunately, the equation can be solved only in successive approximations(perturbation

theory). We start by assuming that j(x) can be constructed from the them-
selves. This assumption has its basis in the fact that the a and a+ . Since (x) is
to be local (nothing moving with speeds exceeding the speed of light), it makes
sense to build up j(x) from the (x).

Basically all this is one big assumption. The system is really very complicated,
even for the simplest cases as we will see. Only for those simple cases can the
above equation be solved, and even then only in terms of successive iterations
(perturbation theory). Thus the scheme developed below is to a large extent
determined simply by the requirement that we can solve it. Fortunately these
methods give rise to results that agree very well with experimental observations.
One truly may be thankful to nature for this, that is, limiting itself to something
that we can compute!

Let us now write down a simple expression for j(x) and solve the equation.

1645
Since we want the pions to interact with the particles we will also include
fields in j(x). We assume
j(x) = 2g(x)(x) (21.88)
The constant g is called the coupling constant. We will now write (x) instead
of (x) to exhibit more clearly the fact that this particular field is associated
with pions.

The above choice is really the simplest non-trivial form for j(x) since if we chose
only (x) or (x) we could, with some reshuffling of the equation, make j(x)
zero and, therefore, this would not correspond to any interactions.

Of course, there is also an equation of motion for the fields:

2
 
2 + 2 m2 (x) = j(x) (21.89)
t
with some other current j(x) instead of j(x). As we will see one cannot freely
choose j(x) if j(x) is already fixed. There is a strong interconnection.

Intuitively, this is simple to see. If j(x) determines how pions interact with s
then evidently this fixes also the interaction of s with s. In fact, one can
determine the S-matrix from either j(x) or j(x) above, and this better be the
same S-matrix! We will see later that the choice j = 2g implies j = g 2 .

It should be stressed here that the field (x) in j(x) is not in or out but also
an interpolating field (interpolating between in and out ) just as (x) is an
interpolating field (interpolating between in and out ).

To find the S-matrix we assume the existence of a matrix U (x0 ), time-dependent,

but not space-dependent, such that:
(x) = U + (x0 )in (x)U (x0 ) , (x) = U + (x0 )in (x)U (x0 ) (21.90)
Clearly U () = I and U (+) = S. Since we will explicitly find U (x0 ), there
is no point in discussing whether this assumption makes sense.

To be complete, it should be noted that from a strict mathematical point of

view one can raise many questions and objections, relating to the fact that we
are dealing with infinite matrices. But being physicists, we, as usual, ignore
these potential problems.

From the fact that (x) satisfies an equation of motion we should be able to
deduce an equation of motion for U (x0 ). Here we need to be careful because all
the objects that we are dealing with are big, generally non-commuting, matrices.
Basic equations, derived earlier, that we will use are:
 
(x)
[(x), (y)]x0 =y0 = 0 , , (y) = i (3) (~x ~y ) (21.91)
x0 x0 =y0

1646
We will now show that, if U (x0 ) satisfies the following differential equation:
Z
U (x0 )
= ig d3 y in
2
(y)in (y)U (x0 ) with y0 = x0 (21.92)
x0
then (x) as defined above satisfies

2
 
2 + 2 M 2 (x) = 2g(x)(x) (21.93)
t
The proof of this statement is not particularly difficult, just a little cumbersome.
It is important to remember that U (x0 ) is time dependent, not space dependent.
In other words
U (x0 )
= 0 , = 1, 2, 3 (21.94)
x
First we introduce a notation:
Z
H(x0 ) = g d3 y 2 (y)(y) with y0 = x0 (21.95)

and Z
Hin (x0 ) = g d3 y in
2
(y)in (y) with y0 = x0 (21.96)

It follows that
H(x0 ) = U 1 Hin (x0 )U (21.97)
The equation to be solved is now:
U (x0 )
= iHin (x0 )U (x0 ) (21.98)
x0
The time derivative of (x) and (x) can be computed. Remember, in general:

(I) = 0 = (U 1 U ) = (U 1 )U + U 1 (U )
(U 1 ) = U 1 (U )U 1 (21.99)

We then find:
= U 1 in U (21.100)
and
(U 1 in U ) (U 1 ) (in ) (U )
= = in U + U 1 U + U 1 in
x0 x0 x0 x0 x0
(U ) ( in )
= U 1 U 1 in U + U 1 U + U 1 in iHin U
x0 x0
(in )
= U 1 iHin U U 1 in U + U 1 U + U 1 in iHin U
x0
(in )
= U 1 i [in , Hin ] U + U 1 U (21.101)
x0

1647
We can see a general rule here. The second time derivative of the field
becomes:
2
2 = U 1 [[in , Hin ] , Hin ] U + U 1 i [in , Hin ] U
x0 x0
2 in
 
in
+ U 1 i , Hin U + U 1 U (21.102)
x0 x20
As noted before the spatial derivatives of U vanish. Considering now the Klein-
Gordon equation for (x) we find

2 2
   
2 + 2 M 2 (x) = 2 + 2 M 2 U 1 in U


t t
2
  
= 2 M 2 U in U
t
2
  
+ U 1 2 + 2 M 2 in U
t
2
  
+ U 1 in 2 M2 U
t
2
  
2 1
= 2 M U in U
t
2
  
1 2
+ U in 2 M U (21.103)
t
and finally
2
 
2 + M (x) = U 1 [[in , Hin ] , Hin ] U
2 2
t

+ U 1 i [in , Hin ] U
x0
 
in
+ U 1 i , Hin U (21.104)
x0
Since the field in satisfies the free Klein-Gordon equation, the first term on the
right hand side vanishes. Also, the second and third terms vanish because in (x)
and in (y) commute for x0 = y0 (as far as in (x) and in (x) are concerned, they
commute always because the matrices a and a+ for the and fields commute).
The last term gives
  Z  
in in 2
, Hin = ig d3 y(y) , in
x0 x0 x0 =y0
Z
3 (3)
= 2ig d yin (y) (~x ~y )in (y)
x0 =y0

= 2igin (y)in (y) (21.105)

1648
The final result is:
2
 
2 + 2 M 2 (x) = 2gU 1 in (x)in (x)U
t
= 2g(x)(x) (21.106)

because

U 1 in (x)in (x)U = U 1 in (x)U U 1 in (x)U = 2g(x)(x) (21.107)

which is precisely what we set out to prove.

It is then straightforward(same techniques) to show that

2
 
2 2
2 + m (x) = g 2 (x) (21.108)
t
This is the moment to consider the question of the connection between j and
j. It is clear from the above derivation that they follow from the same U , i.e.,
from the same H. As a matter of fact one notes the formal rules:
H(x) H(x)
j(x) = and j(x) = (21.109)

with Z
H(x) = d3 yH(y) , H(y) = g 2 (y)(y) (21.110)

The quantities H and H are called the interaction Hamiltonian and the interac-
tion Hamiltonian density respectively. It is clear that it is better to start from
a Hamiltonian and then to derive the equations of motion to avoid inconsisten-
cies. This is what we will do in general. In fact, at this point we do not really
need the equations of motion for the fields any more. We will simply take some
Hamiltonian, and we then know that the U matrix satisfies the equation

U (x0 ) = Hin (x0 )U (x0 ) (21.111)
x0
and solve U from that. Once we have U we have the S-matrix, namely, S =
U ().

It should be clearly understood what we have here. The solution of the equation
for U will give us the matrix U as a function of in and in . Thus we will obtain
S as a function of in and in . This is precisely what we need. As noted before,
the probability to find the configuration c at time t = + if at time t =
we have the configuration b, is given by
2 2
|out hc | biin | = |in hc| S |biin | (21.112)

We need in hc| S |biin since if S is given in terms of in fields we know exactly

what S gives when operating on an in basis vector.

1649
Thus, in principle there is no problem here. However, the actual calculation of
matrix elements in hc| S |biin remains a complicated matter.

The diagram technique introduced by Stuckelberg and Feynman is a very pow-

erful tool to analyze this system. This technique is generally useful for solving
equations of the type
X
= AX (21.113)
t
where A is a given, time dependent, quantity. If A s a constant the solution is
simple:
X = eAt (21.114)
If A is time-dependent, but not a matrix, just some function, then the solution
is: t
Z
X = exp A(t 0 )dt 0 (21.115)
c

where c is an arbitrary constant. If A(t) is a matrix such that A(t1 ) and A(t2 ) do
not necessarily commute, then the solution is more complicated and essentially
can be given only in terms of a series expansion that looks very much like an
expansion of the above exponential, but not exactly (see Chapter 11).

To be precise, it is easy to verify(Chapter 11) that a solution in this case is

Zt Zt Zt
X 1
X =1+ dt1 dt2 ......... dtn T (A(t1 )A(t2 ).....) (21.116)
n=1
n!

We have taken the constant c as , to avoid a number of irrelevant compli-

cations. The time ordered product T is defined as follows. For the product of
two As (
A(t1 )A(t2 ) if t1 > t2
T (A(t1 )A(t2 )) = (21.117)
A(t2 )A(t1 ) if t2 > t1
For any number of As it works similarly, i.e., the As must be arranged in order
of decreasing time. Obviously, if A(t1 ) and A(t2 ) commute for any t1 and t2
then the above expression reduces to an exponential function.

We now proceed to show the correctness of the above solution. It can of course
be verified directly by putting the above solution into the equation, but we will
use iteration instead. Suppose we want to find X as a power series in A. Let us
find the lowest order term. We write X = 1 + 1 , where 1 is of first order in
A. Neglecting terms of order A2 such as A1 the equation for X becomes
Zt
d1
= A 1 = A(t1 )dt1 (21.118)
dt

1650
To obtain the next iteration we write:
Zt
X =1+ A(t1 )dt1 + 2 (21.119)

where 2 is of second order in A. The equation for X becomes (to second order
in A)
Zt
d2
= A(t) A(t1 )dt1 (21.120)
dt

The solution is
Zt Zt2
2 = dt2 A(t2 ) dt1 A(t1 ) (21.121)

Notice that t2 > t1 , that is, the matrices A appear in descending order of time.
To further proceed with this integral we claim that

Zt Zt
2 = dt2 dt1 A(t1 )A(t2 ) (21.122)
t2

also. Note the order of the A and the integration limits of the second integral
where again the A appear in descending order. This may be verified either
by direct insertion into the equation for d/dt or by transforming the integral.
This becomes very easy by considering Figure 21.4 below showing the integra-
tion domains.

Figure 21.4: Integration Domains

Taking c as lower integration limit, with c some number, the first integral corre-
sponds to domain I in the figure and the second to domain II. It is clear that the
two domains can be obtained from each other by exchanging t1 and t2 . Since
indeed the integrands have t1 and t2 interchanged we see that the two integrals

1651
are equal. We may therefore take also for 2 half the sum of both expressions
and thus obtain
Zt Zt
1
2 = dt1 dt2 T (A(t1 )A(t2 )) (21.123)
2

For the sake of clarity we will show directly that the above is the correct solution.
First,

T (A(t1 )A(t2 )) = (t t2 )A(t1 )A(t2 ) + (t2 t1 )A(t2 )A(t1 ) (21.124)

This shows explicitly the meaning of the T product. Now,

Zt
d2 1
= dt2 ((t t2 )A(t)A(t2 ) + (t2 t)A(t2 )A(t))
dt 2

Zt
1
+ dt1 ((t1 t)A(t1 )A(t) + (t t1 )A(t)A(t1 )) (21.125)
2

Essentially the first part is obtained by setting t1 = t, the second by setting

t2 = t. The very first term contains (t t2 ) and is zero unless t2 < t which is
always true since t2 runs from to t. We can therefore omit the function
in that term, to get
Zt
1
A(t) dt2 A(t2 ) (21.126)
2

The second term is zero unless t2 > t, which is never true, and that term is zero.
Similarly the last two terms, which really differ from the first two only in that
the integration variable is called t1 instead of t2 . Together we get the desired
result.

Similarly, one finds in general:

Zt Zt Zt
1
n = dt1 dt2 ....... dtn T (A(t1 )A(t2 )........A(tn )) (21.127)
n!

In this case one must consider n! domains of integration, all obtained from each
other by some permutation of t1 t2 ......tn , but there is no essential difference from
the case of just two variables.

1652
21.4 Interacting Fields - Part 3
21.4.1 Feynman Rules
We will now work out the lowest non-vanishing order of the Smatrix for the
case of the and fields given before. We have
U (x0 )
= iHin (x0 )U (x0 ) (21.128)
x0
or Z
U (x0 )
=i d3 yHin (y)U (x0 ) (21.129)
x0
x0 =y0

In first approximation we find:

Zx0 Z
U (x0 ) = 1 + i dt1 d3 yHin (y) (21.130)
y0 =t1

where we used y0 directly instead of t1 . Note that d4 y = d3 ydy0 . The Smatrix

follows by taking the limit x0 = +.
Z
S = 1 + i d4 yHin (y) + ....... (21.131)

The general expression for S is:

n Z
X i
S =1+ d4 y1 d4 y2 ......d4 yn T (Hin (y1 ).......Hin (yn )) (21.132)
n=1
n!

Let us now concentrate on a specific process. Consider the scattering of two

pions with momenta p and q giving rise to two pions with momenta p0 and q 0 .
We must calculate:

out hp 0 , q 0 | p, qiin = out hp 0 , q 0 | S |p, qiin (21.133)

with S as above, and
2
H(y) = gin (y)in (y) (21.134)
Remember also that
X 1
in (y) = (ain (k)eiky + a+
in (k)e
iky
) (21.135)
k
2V p0

where ain (k) transforms a state with m pions into a state with m 1 pions, and
gives zero if no pions are present, while a+
in (k) gives the opposite result.

Let us first consider the lowest order term of S. As we have now exclusively in
type objects we will drop this subscript. We have:
Z
hp , q | S |p , qi = hp , q | p , qi + i d4 y hp 0 , q 0 | H(y) |p , qi
0 0 0 0
(21.136)

1653
plus terms of higher order in H. Now, if p0 , q 0 is different from p,q then the
first term is zero (orthogonal vectors). The second term contains one H and
therefore only one field. This applied to a state without particles gives zero
(for the a(k) part) or a state containing a particle. But the dot product of
such a state with the state |p0 , q 0 i, containing no particle is zero. Therefore
also the second term is zero.

Generally, any product of an odd number of Hs gives zero between states with-
out particles, by similar arguments.

Let us now consider the second order term. It is given by:

ig 2
Z
d4 yd4 y 0 hp 0 , q 0 |T 2 (y)(y) 2 (y 0 )(y 0 ) |p , qi


(21.137)
2
Now we will have non-zero terms. For instance, some a+ term in the field
can transform the state |p, qi into a state containing in addition a particle of,
say, momentum k, i.e.,
|p, qi = |p, q, ki (21.138)
Then two atype terms in the following two fields may transform this state
into a state containing no pions, i.e.,

2 |p, qi = |ki (21.139)

Next the appropriate a(k) term in the field transforms this state into the state

2 |p, qi = |0i (21.140)

Finally, selecting the terms with a+ (p0 ) and a+ (q 0 ) in the last two pion fields
transforms the state |0i into the state |p0 , q 0 i. The scalar product of this state
with |p0 , q 0 iis non-zero; in fact it is one.

This may be graphically depicted in the following way. Particles are described
by lines, and the action of and fields is to either end or start a line. The
action of H is thus to start or end two lines and one line. The above exam-
ple, drawn in the opposite direction (i.e. with y 0 left of y) is shown in Figure
21.5 below.

Figure 21.5: Feynman diagram for scattering

H(y 0 ) ends two lines and starts a line corresponding to 2 |p, qi = |ki and

1654
H(y) ends a line and starts two lines.

We now can draw pictures corresponding to all possibilities. They are shown in
Figure 21.6 below.

+ further permutations

Figure 21.6: All Possible Feynman diagrams for scattering

We have drawn the vertices as visible dots, to avoid confusion with crossing
lines. The last case shown differs from the first only by the interchange of y and
y 0 . Since the whole is symmetric in y and y 0 it follows that both cases give the
same result. Also the first four diagrams all correspond to the same expression.

All together we get the contributions shown in Figure 21.7 below:

1655
 Figure 21.7: Different Feynman diagrams for scattering

Generally, in higher orders one gets the same result for all permutations of
y,y 0 ,y 00 , ...... , which gives a factor n! for the nth order. This cancels against the
factor 1/n! in front.

So far we have not worried about the various factors going with the a and a+ .
This is not very difficult.

Let us take the first diagram shown in Figure 21.8 below.

Figure 21.8: First Feynman diagrams for scattering

We find:
  
1 0 1 0 1 0 1
eipy eiqy eiky eiky
2V p0 2V q0 2V k0 2V k0
!
1 0 0 1 0 0
p eip y p eiq y (21.141)
2V p00 2V q00

This is for y0 > y00 . For y00 > y0 the order of the H is reversed, which is of no
consequence to the part, but now the starts in the point y and ends in y 0 .

1656
All together we get:
ig 2
Z
0 0 0
d4 yd4 y 0 p 0 0
ei(p+q)y ei(p +q )y
4V 2 p0 q0 p0 q0
!
X 1 ik(yy0 ) X 1 0
(y0 y00 ) e + (y00 y0 ) eik(y y) (21.142)
2V k0 2V k0
k k

Let us now first work out the expression in brackets. The sum over k may be
written as an integral over d3 k, and by methods as described before we may
rewrite the whole in terms of a 4-dimensional integral.
X 1 Z
0 1 0
eik(yy ) 3
d4 keik(yy ) (k0 )(k 2 + m2 ) (21.143)
2V k0 (2)
k

This function is denoted by + (y y 0 ). Similarly the second term, differing

only by the interchange of y and y 0 .
X 1 Z
0 1 0
eik(y y) 3
d4 keik(y y) (k0 )(k 2 + m2 ) (21.144)
2V k0 (2)
k
Z
1 0
= d4 keik(yy ) (k0 )(k 2 + m2 )
(2)3
= (y y 0 ) (21.145)

In the last step we replaced k by k. One has evidently

+ (z) = (z) (21.146)

The combination

F (y y 0 ) = (y0 y00 )+ (y y 0 ) + (y00 y0 ) (y y 0 ) (21.147)

is called the propagator of the field. It can be worked out easily using a
Fourier expression for the function. On has
Z
1 ei z
(z) = d , lim 0 , > 0 (21.148)
2i i

You can confirm this equation by considering the poles of the integrand in the
complex plane. Add an integral over a large half circle to make a closed
contour; take this circle either in the upper or lower plane depending on the
sign of z such that the exponential becomes very small on the circle.

With this expression we have:

(k0 )(k 2 + m2 ) (k0 )(k 2 + m2 )
Z Z  
1 4 ikz+i z0
F (z) = d k d e +
(2)4 i i i
(21.149)

1657
The trick is to get out of the exponential. This may be achieved by a change
of variable for the k0 integration. We take

k0 = k00 + (21.150)

Note that kz = k0 z0 ~k ~z. We then find

(k0 + )(~
k2 (k0 + )2 +m2 )
Z Z !
1 4 ikz i
F (z) = d k d e ~
k2 (k0 + )2 +m2 )
(2)4 i + (k0 )( i
(21.151)
where we renamed k00 to k0 . Next we do the integral. The argument of the
functions is zero if
q

k0 + = ~k 2 + m2 = (21.152)

The functions select the + root for the first term and the root for the
second. The argument of the function can be rewritten
~k 2 (k0 + )2 + m2 = (k0 + ) + (k0 + )



(21.153)

Remember again
1 1
(ab) = (b) + (a) (21.154)
|a| |b|
We find
Z !
1 4 ikz 1 1 1
F (z) = d ke +
(2)4 i
p
2 ~k 2 + m2 k0 + i k0 + i
(21.155)
The complete expression for the diagram considered is
i(p+q)y 0 i(p0 +q 0 )y ik(yy 0 )
ig 2
Z Z
4 4 4 0e e e
d k d yd y
k 2 m2 i
p
(2)4 iV 2 16p0 q0 p00 q00
(21.156)
Both y and y 0 occur only in the exponents, and the integrals can be done using
Z
1
(a) = dx eiax (21.157)
2

and we find
g 2
Z
1
(2)8 d4 k (4) (p + q k) (4) (k p0 q 0 )
k 2 m2 i
p
(2)4 iV 2 16p0 q0 p00 q00
(21.158)
The integral over k can be done

g 2 (2)8 (4) (p + q p0 q 0 )
(21.159)
(2)4 iV 2 16p0 q0 p00 q00 (p + q)2 m2 i
p

1658
From the above calculation we can see how things go in general. Write down
all possible diagrams, and then for any diagram write down the correct factors.
As much as possible factors relating to permutations should be absorbed into
some easy rules. This is not always possible, but in most cases that one meets
there is really not much of a problem.

First, the combinatorial factor relating to there being two pion lines in a vertex
that can be interchanged, is easily taken care of by including a factor of 2 in
the vertex. The factor of two relating to the symmetry in y,y 0 interchange
cancels against the factor 1/2! in front of the second order term of the S-matrix
expansion. Now we have three essentially different diagrams left:

Figure 21.9: The Essentially Different Feynman diagrams for scattering

The contribution due to the first diagram has been computed rules for the
theory that we are considering here. Here are the Feynman rules for the case
H(x) = g 2 (x)(x):

1. Toevery incoming or outgoing or of momentum p corresponds a factor

1/ 2V p0

2. To every vertex corresponds a factor 2i(2)4 g (4) (.....). Note: 2 for two
pion lines, i from the original equation for the Smatrix, (2)4 from the
integral giving the function, and g as found in the interaction Hamil-
tonian

3. To every propagator corresponds a factor

Z
1 1
d4 k 2 (21.160)
(2)4 i k m2 i

For the pion field one has M 2 instead of m2

Many of the propagator integrals can usually be done, thereby getting rid of
the functions due to the vertices. The general rule is that one (4) (.....)

1659
remains, assuring that the sum of incoming momenta equals the total of the
outgoing momenta, thus guaranteeing conservation of energy and momentum
in any process. In the first non-trivial order (as we are considering here) no
momentum integral remains. In the next order one four-dimensional integral
remains non-trivial, and in every next order there is one more four-integral.
This is what makes it so hard to do higher order calculations.

21.5 Interacting Fields - Part 4

21.5.1 Feynman Propagator
This is the moment to reflect on the most important central quantity, the Feyn-
man propagator F . Its definition was
F (y y 0 ) = (y0 y00 )+ (y y 0 ) + (y00 y0 ) (y y 0 ) (21.161)
The functions + and are
Z
0 1 0
(y y ) = 3
d4 k eik(yy ) (k0 )(k 2 + m2 ) (21.162)
(2)
In words, one may understand, the Feynman propagator as follows:

If the time y0 is larger than the time y00 then this propagator equals a function
containing plane waves for a particle of positive energy on mass shell. We can
literally say that if the time y0 > y00 then the Feynman propagator represents
a physical particle moving from the space-time point y 0 to the space-time point
y. In fact, the exponential is nothing else but the wave function for a plane
wave for a particle leaving y 0 multiplied by the wave function for a particle of
the same mass and momentum arriving at y. This product is the overlap of
these functions, something that relates to the probability for this to happen. The
total propagator is obtained when integrating over all possible physical momenta
(positive energy, on mass shell). If the time y00 > y0 , then the particle moves in
the opposite direction.

There is a causality idea in there: energy moves from the earlier point to the
later. There is another feature: the probability for this to happen must not be
negative, which is embodied in the sign of the . Indeed, having a theory with
as above but with a sign in front would give rise to negative probabilities.
This then is the physical content of the Feynman propagator.

The appearance of the functions in F and thus relates closely to physical

concepts. It turns out that these same functions are crucial for the study of
unitarity of the Smatrix, i.e., conservation of probability. And the sign of
the relates to the sign of probability: if it happens to be minus for some
particle, then that particle better be a ghost, meaning that the sum total of its
effects must somehow cancel. Things like that happen in gauge theories, where
then the symmetry of the theory guarantees the necessary cancellations.

1660
21.5.2 Scattering Cross Section
We now introduce a new particle in addition to the and , and we will call it
P . Apart from the spin, which one usually neglects in first approximation, this
P is to play the role of the proton.

It interacts with the in the same way as the (the electron). Thus the inter-
action Hamiltonian becomes:

H = g 2 gP 2 (21.163)

The minus sign reflects the fact that the proton charge is opposite to the electron
charge.

We then obtain the following equations of motion:

2
 
2 + 2 M 2 (x) = 2g (21.164)
t
2
 
2 + 2 m2 (x) = g 2 + gP 2 (21.165)
t
2
 
2 2
2 + MP P (x) = 2gP (21.166)
t

These follow from previous results, as well as the fact that the P field commutes
with all and fields, including time derivatives of these fields. The Feynman
rules are as before, except we now have an extra particle, the P , to be denoted
by a broken line. There is also a new kind of vertex, showing the P coupling
as shown in Figure 21.10 below.

Figure 21.10: New Particle Additions

Next we consider P scattering as shown in Figure 21.11 below.

1661
 Figure 21.11: P Scattering

In this case, to second order, we wind up with only one diagram as shown in
Figure 21.12 below.

Figure 21.12: Only 2nd order diagram in P Scattering

and the corresponding expression for the Smatrix element, or amplitude, is

(2g)(2g)i2 (2)8 (4) (p + q p0 q 0 )
hSi = (21.167)
(2)4 iV 2 16p0 q0 p00 q00 (p p0 )2 m2 i
p

where m is the mass.

To obtain a cross section we must take the absolute value squared of this expres-
sion, which is not immediately clear because of the function. To get around
this we first go back to finite volume V , which amounts to the replacement
V
p + ~q p~0 ~q0 )
(3) (~ p~+~q,~p0 +~q0 (21.168)
(2)3
Thus,
V2
( (3) (....))2 p~+~q,~p0 +~q0 (21.169)
(2)6

1662
since there is nothing difficult about squaring a Kronecker, but we now have
V 2 instead of V . Recombining one V factor with the we have
V
( (3) (....))2 (3) (....) (21.170)
(2)3
Now what about the fourth function (relating to energy conservation)? Here
we must introduce a time interval T . Since with plane waves as we consider
here there is really no beginning and end to the scattering process we limit our
observations to a time interval T , and will compute the transition probability per
unit of time. Essentially, now things are entirely the same for time and space,
and squaring the fourth function gives us a factor T /2. The transition
probability is therefore
 2
 4ig 2 (2)4 1  VT
2
|hSi| =  p (4) (....) (21.171)
 
 V 2 16p0 q0 p00 q00 (p p0 )2 m2 i  (2)4


We want to compare our results to the classical Rutherford scattering cross

section formula, thus we must work this over to a cross section. Imagine the
(proton) P to be at rest (q = (MP , 0, 0, 0)) and the (electron) comes in along
the zaxis (p = (p0 , 0, 0, pz )). We thus have a stream of coming along the
zaxis. Since we have one particle in the whole universe the flux (= number of
particles per unit surface per unit of time) is v/V where v is the velocity of the
. This velocity is given by
|~p|
v= (21.172)
p0
The cross section is the probability per unit of time, for unit flux, summed over
all possible final states. Thus,
Z Z
V 3 0 V V p0 2 1
tot = 3
d p 3
d3 q 0 |hSi| (21.173)
(2) (2) |~
p| T
The integrals with the factors in front are simply the continuum limit of sum-
mation over all p0 and q 0 . Thus, we arrive at the equation
Z
V V
tot = d3 p 0
(2)3 (2)3
 2
Z
V p  4ig 2 (2)4 1  VT 1
0
d3 q 0 (4) (....)
 
|~
p| V
 2p 0 2 2 4
16p0 q0 p0 q0 (p p ) m i (2)
0 0 T

 
(21.174)
or
16g 2 1
Z Z
3 01
tot = d p 0 d3 q 0
(2)2 4p0 q0 2p0
 2
1 p0  1  (p + q p0 q 0 )
 (4)
0 0 2 2
(21.175)
2q0 |~
p| (p p ) m i 


1663
The integral over q 0 can be done, using up three of the functions. We get

16g 2 1
Z
tot = d3 p0
(2)2 4p0 q0
 2
1 p0  1  (p0 + q0 p00 q00 )

0 0 (21.176)
p|  (p p0 )2 m2 i 
4p0 q0 |~
p
where q00 = ~q2 + M 2 with ~q 0 = p~ + ~q p~ 0 .

We now make the non-relativistic approximation and also the no-recoil approx-
imation, which is the approximation that the P mass MP is much heavier than
the mass M of the .

We then introduce polar coordinates for p~ 0 , the outgoing momentum as shown

in Figure 21.13 below.

Figure 21.13: momentum polar coordinates

Then we can write Z Z Z

d3 p0 = d y 2 d3 y (21.177)

where y = |~p|. Conservation of momentum tells us that q00 , the energy of the
outgoing P , is given by
q
q00 = Q~2 + M2 (21.178)
P

with Q = p + q p0 , and thus

~ = p~ + ~q p~ 0 = p~ p~ 0
Q (21.179)

~ is called the
because ~q = 0, as the initial proton is at rest. The quantity Q
momentum transfer. It is the amount of momentum given by the to the P .

If MP is very large we may approximate

~2
Q
q00 = MP + + ..... (21.180)
2MP

1664
The no-recoil approximation is to neglect the term

~2
Q
(21.181)
2MP
with respect to MP . Thus, to this approximation the proton remains at rest,
q00 = MP . The expression for tot now becomes
2
g4 1
Z Z 
1 p0  1  (p0 p00 ) (21.182)
y 2 d3 y

tot = d 0 0
4 2 p0 q0 4p0 q0 |~ 2 2
p| Q m i 

p
p|, thus p00 =
Now y = |~ y 2 + M 2 . It then follows that

dp00 1 1 y
= p 2y = 0 (21.183)
dy 2 y +M
2 2 p0

or ydy = p00 dp00 or |~

p 0 | d |~
p 0 | = p00 dp00 . Furthermore, the function assures us
that p0 = p00 . Thus p p
p~2 + M 2 = p~ 02 + M 2 (21.184)
so that |~ p 0 |. We therefore arrive at
p| = |~
2
g4 1 1 yp00 p0
Z Z 
1
dp00 0
 
tot = d  Q2 m2 i  (p0 p0 ) (21.185)
 
4 2 p0 q0 4p00 q00 |~
p|

Using the fact that y = |~ p 0 | and doing the now trivial p00 integration we
p| = |~
get
2
g4 1
Z 
p0  1 
tot = 2
d 0  2 2
 (21.186)
4 p0 q0 q0 Q m i 
 
~
It should be noted that in the no-recoil approximation Q0 << Q . This follows
because Q0 is the difference between the initial and final P energy

~2
~ 2 + M 2 MP Q
q
Q0 = q00 q0 = Q P (21.187)
2MP
~ 2 to a good approximation. Replacing, nonrelativistically, q0
Therefore Q2 = Q
by MP and p0 by M we have the final result
2
g4
Z 
 1 
tot = d   (21.188)
4 MP2
2 2 2
Q m 

We have omitted the i in the propagator, because both Q ~ 2 and m2 are positive,
so the infinitesimal is of no relevance here. If we take the mass to be zero
we have
g4
Z
1
tot = 2 2 d (21.189)
4 MP ~4
Q

1665
 ~ = p~ p~ 0 and
Now Q

~ 2 = 2 |~ 2 2
Q p| sin2
p| (1 cos ) = 4 |~ (21.190)
2
where we used |~ p 0 | and is the angle that the outgoing makes with the
p| = |~
zaxis (the direction of the incoming ), Thus

g4
Z
1
tot = 2 2 d 4
4 MP p| sin4 2
16 |~
g4 2g 4
Z Z
1 sin d
= d 4 4 = (21.191)
64 2 MP2 M 4 v sin 2 64 2M 2 M 4
P v 4 sin4 2

which is indeed the angular distribution as given by the Rutherford scattering

formula. In here v is the velocity of the initial , with |~
p| = M v as usual and
d = sin dd.

21.5.3 Lifetime
In this last section we will consider another application of the theory developed
so far, namely the calculation of a decay rate, or a lifetime for an unstable
particle.

In principle we have here a contradiction. An unstable particle lives a finite time,

and therefore it is impossible to be present at either minus or plus infinite time.
In other words, an unstable particle will not occur in in-states or out-states.
We will do as if it can be in an in-state and then calculate its decay probability.
The full justification for that requires a complete treatment of unstable particles,
which we will not do here.

Going back to the interaction Hamiltonian described before, H = g 2 , we will

now consider the case that the particle is heavier than two pions, so that
it normally will decay. The decay probability per unit time is called the decay
rate, and the inverse of the decay rate is the lifetime.

The process of interest has initially a and finally two pions. We therefore
must consider
hp, q| S |ki (21.192)
where k denotes the momentum of the initial and p and q are the momenta
of the final pions.

In lowest order there is one non-vanishing Feynman diagram is shown in Figure

21.14 below.

1666
 Figure 21.14: Feynman diagram for decay

The corresponding expression is

4
i (2) 2g
hSi = p 4 (k p q) (21.193)
8p0 q0 k0 V 3
The transition probability is
8
2 (2) 4g 2 V T
|hSi| = 4 (k p q) (21.194)
8p0 q0 k0 V 3 (2)3 2
The transition probability per unit of time follows by dividing by T . We must
also sum over all final states if we want the total decay rate. This rate is
therefore
Z Z 8 2
1 V 3 V 3 (2) 4g V 1 (4)
( 2) = d p d q (k p q)
2 (2)3 (2)3 8p0 q0 k0 V 3 (2)3 2
g2
Z Z
1 1 (4)
= d3 p d3 q (k p q) (21.195)
(2)2 k0 2p0 2q0
There is a subtlety here, we have divided by 2. This is because the two pions in
the final state are identical, and if we integrate over all momenta we will count
double, because the final state with the two pions interchanged is the same state
(Bose-Einstein statistics). To work out the above expression we go to the a rest
system. In this system k0 = m and ~k = 0. This makes the ~q integral trivial,
giving ~q = ~p and therefore q0 = p0 . We get
g2
Z
1
( 2) = 2
p2 dp 2 (m 2p0 ) (21.196)
4 m 4p0
We go to polar coordinates. The integral over angles is also trivial giving a
factor 4. We have
g2
Z
1
( 2) = p2 dp 2 (m 2p0 ) (21.197)
4m p0
Now p = |~
p|. Using the relation pdp = p0 dp0 this last integral is trivial and we
obtain
g2 p
( 2) = (21.198)
8m p0

1667
 p
with p0 = m/2 and p = p20 M 2 . Thus

g2 p 2
( 2) = m 4M 2 (21.199)
8m
The lifetime is the inverse of this, = 1/.

Numerical Evaluation
Generally one wants a cross section in terms of cm2 and a lifetime in seconds.
We have used ~ = c = 1 and will express everything else in M eV . The cross
section will have the dimension of (M eV )2 , the decay rate is of dimension
M eV and lifetime (M eV )1 . To go to cm2 the cross section must be multiplied
by (hc)2 = 1.97327 1011 (M eV cm)2 . To go from M eV to sec1 the decay
rate must be divided by h = 6.582122 1022 M eV sec and the lifetime is
thus h/. Note that in the examples above the coupling constant g has the
dimension M eV .

1668
Index

Adiabatic Approximation, 950 GHZ state, 1339

Hidden Variables Model, 1269
Dirac Equation Last Try, 1322
Currents and Continuity, 1575 Simple Introduction, 1299
Free Particles, 1577
Hole Theory, 1604 Identical Particles, 971
Hydrogen Atom, 1586 Identical Particles
Lamb Shift, 1600 Bosons with Spin=0, 976
Non-relativistic limit, 1573 Helium Atom, 990
Dirac Equation for Spin=1/2, 1563 Multielectron Atoms, 998
Multielectron Atoms
Electromagnetic Radiation and Matter, Angular Momentum Coupling,
1606 1005
Absorption/Emission, 1610 Screening, 1000
Gauge Invariance, 1608 Shell Structure, 1001
Interactions, 1606 N-electron atom, 984
Electromagnetic Radiation and MatterQuan- Spin=1/2 Fermions, 978
tization and Spontaneous Emis- Interacting Fields, 1625
sion, 1613 Feynman Propagator, 1660
EPR, Hidden Variables and Bell, 1265 Feynman Rules, 1653
1st Try, 1265 Fields, 1632
Bayesian Thoughts, 1328 Free particles, 1625
Changing Beliefs, 1329 Hilbert Space, 1625, 1639
BCHSH Inequality, 1302, 1305 Examples, 1631
Bells Inequality, 1275, 1326 Structure, 1637
Bells Inequality:Proof, 1304 Interpolating Fields, 1644
Bells Theorem, 1271 Lifetime, 1666
Experiments, 1271 Physical System, 1638
Details, 1278 Scattering Cross Section, 1661
Aspect Experiments, 1309 U-Matrix,S-Matrix, 1641
Example using Photons, 1318 Interacting Fields Feynman Diagrams,
Formalism, 1293 1625
History, 1296
Non-separability, 1316 Measurement
Philosophy, 1296 Comparing Approaches, 1192
EPR Argument - Quick Overview, Composite Systems, 1157
1273 Coupling to Environment, 1208

1669
 Density Matrix, 1145, 1194 Quadratic Lagrangian Propagator,
Actual Process, 1205 1368
Kochen-Specker Theorem, 1212 Quantum Mechanics, 1346
Polarization of Spin=1/2 Systems, Aharonov-Bohm Effect, 1357
1204 Topology, 1357
Projection Operators, 1198 Quantum Mechanics
Pure and Mixed Ensembles, 1194 Examples, 1351
Spin=1/2 Systems, 1200
Stern-Gerlach Example, 1206 Quantum Logic, 1226
von Neuman Equation, 1199 Meaning of True, 1226
Filtering, 1185 FALSE and NOT TRUE, 1230
General Theorem, 1165 Logical Connectives, 1231
Interpretation of State Vector, 1167 Quantum System Examples
Interpretation of States, 1163 Coherent and Squeezed States, 1093
Joint and Conditional Probabili- Quantum System Examples
ties, 1180 Electron in a circular wire, 1098
Mechanisms, 1169 Neutron Interferometry, 1107
Observation and Time Evolution, Penning Trap, 1112
1210 Quantum Eraser, 1133
Probability Distributions, 1181 Schrodinger Cat, 1116
Pure State Factor Theorem, 1159 Spin-Orbit Coupling in Complex
Spin Example, 1163 Atoms, 1102
Spin Recombination Experiment, Zeeman Effect in Complex Atoms,
1174 1105
State Determination, 1150 Quantum World View, 1237
State Preparation, 1145
Relativistic Wave Equation
Wave Function, 1171
Klein-Gordon Equation, 1541
Molecular Physics, 1059
Bound States, 1559
Born-Oppenheimer Approximation,
Free Particles, 1555
1065
Non-relativistic limit, 1562
General Properties, 1059
Physics, 1549
Molecular Physics
Hydrogen Molecular Ion, 1073 Scattering Theory, 1033
Hydrogen Molecule, 1077 Born Approximation, 1047
Vibrational/rotational levels, 1081 Partial Waves
Optical Theorem, 1047
Path Integrals, 1345 Phase Shift Property, 1050
Classical Limit, 1354 Scattering Theory
Least Action Principle, 1354 Born Approximation Examples, 1055
Evaluation Details, 1360 Cross Sections, 1041
Free Particle Propagator, 1361 Greens Functions, 1035
Harmonic Oscillator Propagator, Partial Waves, 1042
1372 Phase Shift Examples, 1051
History, 1345 Second Quantization, 1483
Motivation, 1345 Bogoliubov Transformation, 1518

1670
 Hanbury, Brown and Twiss Exper-
iment, 1511
Identical Particles, 1483
N-Particle Non-Interacting Gas, 1503
Occupation Number Space, 1485
Operators, 1499
Pair Correlation Function, 1506
Spontaneous Symmetry Breaking,
1523
The Hamiltonian, 1515
Solid State Physics, 1377
2D Lattices
Fermi Surface, 1446
Bloch Theorem, 1391
Alternative Form, 1394
Derivation, 1393
The Wave Vector, 1394
Translation Operators, 1391
Crystal Structure, 1377
Crystal Symmetry, 1378
3D Crystals, 1388
Classifications of 2D Lattices, 1384
Group Theory, 1382
Point Symmetry, 1381
Translation Symmetry, 1379
Kronig-Penney Model, 1427
Weak Binding Limit, 1433
Reciprocal Lattice, 1395
1st Brillouin Zone, 1399
Brillouin Zone Properties, 1403
Brillouin Zones, 1399
Higher Brillouin Zones, 1401
Square Lattice, 1397
Reciprocal Lattice Vectors, 1396
Spherical Harmonics Addition Theorem,
1018
Sudden Approximation, 950

Time-dependent perturbation theory, 921

Time-dependent perturbation theory
Electric dipole approximation, 941
induced emission/absorption, 943
Ionization, 946
physical meaning, 927
Selection rules, 940

1671