P. 1
Introduction to Quantum Mechanics Schrodinger Equation and Path Integral~Tqw~_darksiderg

Introduction to Quantum Mechanics Schrodinger Equation and Path Integral~Tqw~_darksiderg

|Views: 1,973|Likes:
Published by Bitan Chakraborty

More info:

Published by: Bitan Chakraborty on Sep 10, 2010
Copyright:Attribution Non-commercial

Availability:

Read on Scribd mobile: iPhone, iPad and Android.
download as PDF, TXT or read online from Scribd
See more
See less

03/19/2014

pdf

text

original

Sections

  • Introduction
  • 1.1 Origin and Discovery of Quantum Mechanics
  • 1.2 Contradicting Discretization: Uncertainties
  • 1.3 Particle-Wave Dualism
  • 1.4 Particle-Wave Dualism and Uncertainties
  • 1.4.1 Further thought experiments
  • 1.5 Bohr's Complementarity Principle
  • 1.6 Further Examples
  • Hamiltonian Mechanics
  • 2.1 Introductory Remarks
  • 2.2 The Hamilton Formalism
  • 2.3 Liouville Equation, Probabilities
  • 2.3.1 Single particle consideration
  • 2.3.2 Ensemble consideration
  • 2.4 Expectation Values of Observables
  • 2.5 Extension beyond Classical Mechanics
  • Mathematical Foundations of Quantum Mechanics
  • 3.1 Introductory Remarks
  • 3.2 Hilbert Spaces
  • 3.3 Operators in Hilbert Space
  • 3.4 Linear Functionals and Distributions
  • 3.4.1 Interpretation of distributions in physics
  • 3.4.2 Properties of functionals and the delta distribution
  • Dirac's Ket- and Bra-Formalism
  • 4.1 Introductory Remarks
  • 4.2 Ket and Bra States
  • 4.3 Linear Operators, Hermitian Operators
  • 4.4 Observables
  • 4.5 Representation Spaces and Basis Vectors
  • Schrodinger Equation and Liouville Equation
  • 5.1 Introductory Remarks
  • 5.2 The Density Matrix
  • 5.3 The Probability Density p(x, t)
  • 5.4 Schrodinger Equation and Liouville Equation
  • 5.4.1 Evaluation of the density matrix
  • Quantum Mechanics of the Harmonic Oscillator
  • 6.1 Introductory Remarks
  • 6.2 The One-Dimensional Linear Oscillator
  • 6.3 The Energy Representation of the Oscillator
  • 6.4 The Configuration Space Representation
  • 6.5 The Harmonic Oscillator Equation
  • 6.5.1 Derivation of the generating function
  • Green's Functions
  • 7.1 Introductory Remarks
  • 7.2 Time-dependent and Time-independent Cases
  • 7.4 Green's Function of the Harmonic Oscillator
  • 7.5 The Inverted Harmonic Oscillator
  • 7.5.1 Wave packets
  • 7.5.2 A particle's sojourn time T at the maximum
  • Time-Independent Perturbation Theory
  • 8.1 Introductory Remarks
  • 8.2 Asymptotic Series versus Convergent Series
  • 8.2.1 The error function and Stokes discontinuities
  • 8.2.2 Stokes discontinuities of oscillator functions
  • 8.3 Asymptotic Series from Differential Equations
  • 8.4 Formal Definition of Asymptotic Expansions
  • 8.5 Rayleigh-Schrodinger Perturbation Theory
  • 8.6 Degenerate Perturbation Theory
  • 8.7 Dingle-Muller Perturbation Method
  • The Density Matrix and Polarization Phenomena
  • 9.1 Introductory Remarks
  • 9.2 Reconsideration of Electrodynamics
  • 9.3 Schrodinger and Heisenberg Pictures
  • 9.4 The Liouville Equation
  • Quantum Theory: The General Formalism
  • 10.1 Introductory Remarks
  • 10.2 States and Observables
  • 10.2.1 Uncertainty relation for observables A, B
  • 10.3 One-Dimensional Systems
  • 10.3.1 The translation operator U(a)
  • 10.4 Equations of Motion
  • 10.5 States of Finite Lifetime
  • 10.6 The Interaction Picture
  • 10.7 Time-Dependent Perturbation Theory
  • 10.8 Transitions into the Continuum
  • 10.9 General Time-Dependent Method
  • The Coulomb Interaction
  • 11.1 Introductory Remarks
  • 11.2 Separation of Variables, Angular Momentum
  • 11.2.1 Separation of variables
  • 11.3 Representation of Rotation Group
  • 11.4 Angular Momentum:Angular Representation
  • 11.5 Radial Equation for Hydrogen-like Atoms
  • 11.6 Discrete Spectrum of the Coulomb Potential
  • 11.6.1 The eigenvalues
  • 11.6.2 Laguerre polynomials: Various definitions in use!
  • 11.6.3 The eigenfunctions
  • 11.6.4 Hydrogen-like atoms in parabolic coordinates
  • 11.7 Continuous Spectrum of Coulomb Potential
  • 11.7.1 The Rutherford formula
  • 11.8 Scattering of a Wave Packet
  • 11.9 Scattering Phase and Partial Waves
  • Quantum Mechanical Tunneling
  • 12.1 Introductory Remarks
  • 12.2 Continuity Equation and Conditions
  • 12.3 The Short-Range Delta Potential
  • 12.4 Scattering from a Potential Well
  • 12.5 Degenerate Potentials and Tunneling
  • Linear Potentials
  • 13.1 Introductory Remarks
  • 13.2 The Freely Falling Particle: Quantization
  • 13.2.1 Superposition of de Broglie waves
  • 13.2.2 Probability distribution at large times
  • 13.3 Stationary States
  • 13.4 The Saddle Point or Stationary Phase Method
  • Classical Limit and WKB Method
  • 14.1 Introductory Remarks
  • 14.2 Classical Limit and Hydrodynamics Analogy
  • 14.3 The WKB Method
  • 14.3.1 The approximate WKB solutions
  • 14.3.2 Turning points and matching of WKB solutions
  • 14.3.3 Linear approximation and matching
  • 14.4 Bohr—Sommerfeld—Wilson Quantization
  • 14.5 Further Examples
  • Power Potentials
  • 15.1 Introductory Remarks
  • 15.2 The Power Potential
  • 15.3 The Three-Dimensional Wave Function
  • Screened Coulomb Potentials
  • 16.1 Introductory Remarks
  • 16.2 Regge Trajectories
  • 16.3 The 5-Matrix
  • 16.4 The Energy Expansion
  • 16.5 The Sommerfeld—Watson Transform
  • 16.6 Concluding Remarks
  • Periodic Potentials
  • 17.1 Introductory Remarks
  • 17.2 Cosine Potential: Weak Coupling Solutions
  • 17.2.1 The Floquet exponent
  • 17.2.2 Four types of periodic solutions
  • 17.3 Cosine Potential: Strong Coupling Solutions
  • 17.3.1 Preliminary remarks
  • 17.3.2 The solutions
  • 17.3.3 The eigenvalues
  • 17.3.4 The level splitting
  • 17.4 Elliptic and Ellipsoidal Potentials
  • 17.4.1 Introduction
  • 17.4.2 Solutions and eigenvalues
  • 17.4.3 The level splitting
  • 17.4.4 Reduction to Mathieu functions
  • 17.5 Concluding Remarks
  • Anharmonic Oscillator Potentials
  • 18.1 Introductory Remarks
  • 18.2 The Inverted Double Well Potential
  • 18.2.1 Defining the problem
  • 18.2.2 Three pairs of solutions
  • 18.2.3 Matching of solutions
  • 18.2.6 The complex eigenvalues
  • 18.3 The Double Well Potential
  • 18.3.1 Defining the problem
  • 18.3.2 Three pairs of solutions
  • 18.3.3 Matching of solutions
  • 18.3.6 Eigenvalues and level splitting
  • 18.3.7 General Remarks
  • Singular Potentials
  • 19.1 Introductory Remarks
  • 19.2.1 Preliminary considerations
  • 19.2.4 Notation and properties of solutions
  • 19.2.5 Derivation of the S-matrix
  • 19.2.6 Evaluation of the S-matrix
  • 19.2.7 Calculation of the absorptivity
  • 19.3.1 Preliminary remarks
  • solutions
  • 19.3.4 The connection formulas
  • 19.3.5 Derivation of the ^-matrix
  • 19.4 Concluding Remarks
  • Large Order Behaviour of Perturbation Expansions
  • 20.1 Introductory Remarks
  • 20.2 Cosine Potential: Large Order Behaviour
  • 20.3 Cosine Potential: Complex Eigenvalues
  • 20.3.1 The decaying ground state
  • 20.3.2 Decaying excited states
  • 20.3.3 Relating the level splitting to imaginary E
  • 20.3.4 Recalculation of large order behaviour
  • 20.4 Cosine Potential: A Different Calculation
  • 20.5 Anharmonic Oscillators
  • 20.5.1 The inverted double well
  • 20.5.2 The double well
  • 20.6 General Remarks
  • The Path Integral Formalism
  • 21.1 Introductory Remarks
  • 21.2 Path Integrals and Green's Functions
  • 21.3 The Green's Function for Potential V=0
  • 21.3.1 Configuration space representation
  • 21.3.2 Momentum space represenation
  • 21.4 Including V in First Order Perturbation
  • 21.5 Rederivation of the Rutherford Formula
  • 21.6 Path Integrals in Dirac's Notation
  • 21.7 Canonical Quantization from Path Integrals
  • Classical Field Configurations
  • 22.1 Introductory Remarks
  • 22.2 The Constant Classical Field
  • 22.3 Soliton Theories in One Spatial Dimension
  • 22.4 Stability of Classical Configurations
  • 22.5 Bogomol'nyi Equations and Bounds
  • 22.6 The Small Fluctuation Equation
  • 22.7 Existence of Finite-Energy Solutions
  • 22.8 Ginzburg—Landau Vortices
  • 22.9 Introduction to Homotopy Classes
  • 22.10 The Fundamental Homotopy Group
  • Path Integrals and Instantons
  • 23.1 Introductory Remarks
  • 23.2 Instantons and Anti-Instantons
  • 23.3 The Level Difference
  • 23.4 Field Fluctuations
  • 23.4.1 The fluctuation equation
  • 23.4.2 Evaluation of the functional integral
  • 23.4.3 The Faddeev—Popov constraint insertion
  • 23.4.4 The single instanton contribution
  • 23.4.5 Instanton-anti-instanton contributions
  • 23.5 Concluding Remarks
  • Path Integrals and Bounces on a Line
  • 24.1 Introductory Remarks
  • 24.2 The Bounce in a Simple Example
  • 24.3 The Inverted Double Well: The Bounce and Complex Energy
  • 24.3.1 The bounce solution
  • 24.3.2 The single bounce contribution
  • 24.3.3 Evaluation of the single bounce kernel
  • 24.3.4 Sum over an infinite number of bounces
  • 24.3.5 Comments
  • 24.4 Inverted Double Well: Constant Solutions
  • 24.5 The Cubic Potential and its Complex Energy
  • Periodic Classical Configurations
  • 25.1 Introductory Remarks
  • 25.2 The Double Well Theory on a Circle
  • 25.2.1 Periodic configurations
  • 25.2.2 The fluctuation equation
  • 25.2.3 The limit of infinite period
  • 25.3 The Inverted Double Well on a Circle
  • 25.3.1 Periodic configurations
  • 25.3.2 The fluctuation equation
  • 25.3.3 The limit of infinite period
  • 25.4 The Sine—Gordon Theory on a Circle
  • 25.4.1 Periodic configurations
  • 25.4.2 The fluctuation equation
  • 25.5 Conclusions
  • Path Integrals and Periodic Classical Configurations
  • 26.1 Introductory Remarks
  • 26.2 The Double Well and Periodic Instantons
  • 26.2.1 Periodic configurations and the double well
  • 26.2.2 Transition amplitude and Feynman kernel
  • 26.2.3 Fluctuations about the periodic instanton
  • 26.2.4 The single periodic instanton contribution
  • 26.2.5 Sum over instanton-anti-instanton pairs
  • 26.3 The Cosine Potential and Periodic Instantons
  • 26.3.1 Periodic configurations and the cosine potential
  • 26.3.2 Transition amplitude and Feynman kernel
  • 26.3.3 The fluctuation equation and its eigenmodes
  • 26.3.4 The single periodic instanton contribution
  • 26.3.5 Sum over instanton—anti-instanton pairs
  • 26.4 The Inverted Double Well and Periodic In- stantons
  • 26.4.1 Periodic configurations and the inverted double well
  • 26.4.2 Transition amplitude and Feynman kernel
  • 26.4.3 The fluctuation equation and its eigenmodes
  • 26.4.4 The single periodic bounce contribution
  • 26.4.5 Summing over the infinite number of bounces
  • 26.5 Concluding Remarks
  • Quantization of Systems with Constraints
  • 27.1 Introductory Remarks
  • 27.2 Constraints: How they arise
  • 27.2.1 Singular Lagrangians
  • 27.3 The Hamiltonian of Singular Systems
  • 27A Persistence of Constraints in Course of Time
  • 27.5 Constraints as Generators of a Gauge Group
  • 27.6 Gauge Fixing and Dirac Quantization
  • 27.7 The Formalism of Dirac Quantization
  • 27.7.1 Poisson and Dirac brackets in field theory
  • 27.8 Dirac Quantization of Free Electrodynamics
  • 27.9 Faddeev-Jackiw Canonical Quantization
  • 27.9.1 The method of Faddeev and Jackiw
  • The Quantum-Classical Crossover as Phase Transition
  • 28.1 Introductory Remarks
  • 28.2 Relating Period to Temperature
  • 28.3 Crossover in Previous Cases
  • 28.3.1 The double well and phase transitions

INTRODUCTION TO OUHNTUM MECHANICS

Schrddinger Equation and Path Integral

INTRODUCTION TO QUANTUM MECHANICS
Schrodinger Equation and Path Integral

This page is intentionally left blank

*»«*?-:#•

m
m

-'t

Schrodinger Equation and Path Integn

University of Kaiserslautern, Germany

\(P World Scientific
N E W J E R S E Y • L O N D O N • S I N G A P O R E • B E I J I N G • S H A N G H A I • HONG KONG • T A I P E I • CHENNAI

Published by World Scientific Publishing Co. Pte. Ltd. 5 Toh Tuck Link, Singapore 596224 USA office: 27 Warren Street, Suite 401-402, Hackensack, NJ 07601 UK office: 57 Shelton Street, Covent Garden, London WC2H 9HE

British Library Cataloguing-in-Publication Data A catalogue record for this book is available from the British Library.

INTRODUCTION TO QUANTUM MECHANICS: Schrodinger Equation and Path Integral Copyright © 2006 by World Scientific Publishing Co. Pte. Ltd. All rights reserved. This book, or parts thereof, may not be reproduced in any form or by any means, electronic or mechanical, including photocopying, recording or any information storage and retrieval system now known or to be invented, without written permission from the Publisher.

For photocopying of material in this volume, please pay a copying fee through the Copyright Clearance Center, Inc., 222 Rosewood Drive, Danvers, MA 01923, USA. In this case permission to photocopy is not required from the publisher.

ISBN 981-256-691-0 ISBN 981-256-692-9 (pbk)

Printed in Singapore by World Scientific Printers (S) Pte Ltd

Contents
Preface 1 Introduction 1.1 Origin and Discovery of Quantum Mechanics 1.2 Contradicting Discretization: Uncertainties 1.3 Particle-Wave Dualism 1.4 Particle-Wave Dualism and Uncertainties 1.4.1 Further thought experiments 1.5 Bohr's Complementarity Principle 1.6 Further Examples xv 1 1 7 12 14 17 19 20 23 23 23 29 29 31 34 38 41 41 41 49 53 54 55 59 59 60

2 Hamiltonian Mechanics 2.1 Introductory Remarks 2.2 The Hamilton Formalism 2.3 Liouville Equation, Probabilities 2.3.1 Single particle consideration 2.3.2 Ensemble consideration 2.4 Expectation Values of Observables 2.5 Extension beyond Classical Mechanics 3 Mathematical Foundations of Quantum Mechanics 3.1 Introductory Remarks 3.2 Hilbert Spaces 3.3 Operators in Hilbert Space 3.4 Linear Functionals and Distributions 3.4.1 Interpretation of distributions in physics 3.4.2 Properties of functionals and the delta distribution . . Dirac's Ket- and Bra-Formalism 4.1 Introductory Remarks 4.2 Ket and Bra States v

4

VI

4.3 4.4 4.5 5

Linear Operators, Hermitian Operators Observables Representation Spaces and Basis Vectors

62 68 71 73 73 73 77 78 80 83 83 84 90 91 98 98 105 105 105 Ill 113 118 118 123 129 129 130 133 139 143 146 147 152 155

Schrodinger Equation and Liouville Equation 5.1 Introductory Remarks 5.2 The Density Matrix 5.3 The Probability Density p(x, t) 5.4 Schrodinger Equation and Liouville Equation 5.4.1 Evaluation of the density matrix Quantum Mechanics of the Harmonic Oscillator 6.1 Introductory Remarks 6.2 The One-Dimensional Linear Oscillator 6.3 The Energy Representation of the Oscillator 6.4 The Configuration Space Representation 6.5 The Harmonic Oscillator Equation 6.5.1 Derivation of the generating function Green's Functions 7.1 Introductory Remarks 7.2 Time-dependent and Time-independent Cases 7.3 The Green's Function of a Free Particle 7.4 Green's Function of the Harmonic Oscillator 7.5 The Inverted Harmonic Oscillator 7.5.1 Wave packets 7.5.2 A particle's sojourn time T at the maximum Time-Independent Perturbation Theory 8.1 Introductory Remarks 8.2 Asymptotic Series versus Convergent Series 8.2.1 The error function and Stokes discontinuities 8.2.2 Stokes discontinuities of oscillator functions 8.3 Asymptotic Series from Differential Equations 8.4 Formal Definition of Asymptotic Expansions 8.5 Rayleigh-Schrodinger Perturbation Theory 8.6 Degenerate Perturbation Theory 8.7 Dingle-Miiller Perturbation Method

6

7

8

Vll

9

The 9.1 9.2 9.3 9.4

Density Matrix and Polarization Phenomena Introductory Remarks Reconsideration of Electrodynamics Schrodinger and Heisenberg Pictures The Liouville Equation

161 161 161 166 167 169 169 169 170 173 176 178 184 185 189 191 195 199 199 199 205 206 210 213 215 215 . . 219 223 227 234 237 239 243 249 249 250 251 254

10 Quantum Theory: The General Formalism 10.1 Introductory Remarks 10.2 States and Observables 10.2.1 Uncertainty relation for observables A, B 10.3 One-Dimensional Systems 10.3.1 The translation operator U(a) 10.4 Equations of Motion 10.5 States of Finite Lifetime 10.6 The Interaction Picture 10.7 Time-Dependent Perturbation Theory 10.8 Transitions into the Continuum 10.9 General Time-Dependent Method 11 The Coulomb Interaction 11.1 Introductory Remarks 11.2 Separation of Variables, Angular Momentum 11.2.1 Separation of variables 11.3 Representation of Rotation Group 11.4 Angular Momentum:Angular Representation 11.5 Radial Equation for Hydrogen-like Atoms 11.6 Discrete Spectrum of the Coulomb Potential 11.6.1 The eigenvalues 11.6.2 Laguerre polynomials: Various definitions in use! 11.6.3 The eigenfunctions 11.6.4 Hydrogen-like atoms in parabolic coordinates 11.7 Continuous Spectrum of Coulomb Potential 11.7.1 The Rutherford formula 11.8 Scattering of a Wave Packet 11.9 Scattering Phase and Partial Waves 12 Quantum Mechanical Tunneling 12.1 Introductory Remarks 12.2 Continuity Equation and Conditions 12.3 The Short-Range Delta Potential 12.4 Scattering from a Potential Well

vm 12.5 Degenerate Potentials and Tunneling 13 Linear Potentials 13.1 Introductory Remarks 13.2 The Freely Falling Particle: Quantization 13.2.1 Superposition of de Broglie waves 13.2.2 Probability distribution at large times 13.3 Stationary States 13.4 The Saddle Point or Stationary Phase Method 14 Classical Limit and W K B Method 14.1 Introductory Remarks 14.2 Classical Limit and Hydrodynamics Analogy 14.3 The WKB Method 14.3.1 The approximate WKB solutions 14.3.2 Turning points and matching of WKB solutions . . . . 14.3.3 Linear approximation and matching 14.4 Bohr-Sommerfeld-Wilson Quantization 14.5 Further Examples 15 Power Potentials 15.1 Introductory Remarks 15.2 The Power Potential 15.3 The Three-Dimensional Wave Function 16 Screened Coulomb Potentials 16.1 Introductory Remarks 16.2 Regge Trajectories 16.3 The S-Matrix 16.4 The Energy Expansion 16.5 The Sommerfeld-Watson Transform 16.6 Concluding Remarks 17 Periodic Potentials 17.1 Introductory Remarks 17.2 Cosine Potential: Weak Coupling Solutions 17.2.1 The Floquet exponent 17.2.2 Four types of periodic solutions 17.3 Cosine Potential: Strong Coupling Solutions 17.3.1 Preliminary remarks 17.3.2 The solutions 259 265 265 265 266 270 272 276 281 281 282 286 286 290 293 297 301 307 307 308 315 319 319 322 328 329 330 336 339 339 341 341 350 353 353 354

ix

17.3.3 The eigenvalues 17.3.4 The level splitting 17.4 Elliptic and Ellipsoidal Potentials 17.4.1 Introduction 17.4.2 Solutions and eigenvalues 17.4.3 The level splitting 17.4.4 Reduction to Mathieu functions 17.5 Concluding Remarks 18 Anharmonic Oscillator Potentials 18.1 Introductory Remarks 18.2 The Inverted Double Well Potential 18.2.1 Defining the problem 18.2.2 Three pairs of solutions 18.2.3 Matching of solutions 18.2.4 Boundary conditions at the origin 18.2.5 Boundary conditions at infinity 18.2.6 The complex eigenvalues 18.3 The Double Well Potential 18.3.1 Defining the problem 18.3.2 Three pairs of solutions 18.3.3 Matching of solutions 18.3.4 Boundary conditions at the minima 18.3.5 Boundary conditions at the origin 18.3.6 Eigenvalues and level splitting 18.3.7 General Remarks 19 Singular Potentials 19.1 Introductory Remarks 19.2 The Potential 1/r 4 — Case of Small h2 19.2.1 Preliminary considerations 19.2.2 Small h solutions in terms of Bessel functions . . . . 19.2.3 Small h solutions in terms of hyperbolic functions . . 19.2.4 Notation and properties of solutions 19.2.5 Derivation of the S-matrix 19.2.6 Evaluation of the S-matrix 19.2.7 Calculation of the absorptivity 19.3 The Potential 1/r 4 — Case of Large h2 19.3.1 Preliminary remarks 19.3.2 The Floquet exponent for large h2 19.3.3 Construction of large-h 2 solutions

361 363 371 371 373 375 377 378 379 379 382 382 384 391 393 396 402 405 405 407 412 414 417 424 427 435 435 436 436 438 441 442 446 455 458 460 460 461 464

X

19.3.4 The connection formulas 19.3.5 Derivation of the S-matrix 19.4 Concluding Remarks 20 Large Order Behaviour of Perturbation Expansions 20.1 Introductory Remarks 20.2 Cosine Potential: Large Order Behaviour 20.3 Cosine Potential: Complex Eigenvalues 20.3.1 The decaying ground state 20.3.2 Decaying excited states 20.3.3 Relating the level splitting to imaginary E 20.3.4 Recalculation of large order behaviour 20.4 Cosine Potential: A Different Calculation 20.5 Anharmonic Oscillators 20.5.1 The inverted double well 20.5.2 The double well 20.6 General Remarks 21 The 21.1 21.2 21.3 Path Integral Formalism Introductory Remarks Path Integrals and Green's Functions The Green's Function for Potential V=0 21.3.1 Configuration space representation 21.3.2 Momentum space represenation Including V in First Order Perturbation Rederivation of the Rutherford Formula Path Integrals in Dirac's Notation Canonical Quantization from Path Integrals

466 468 470 471 471 476 479 479 486 493 494 495 500 500 501 502 503 503 504 510 510 513 514 518 524 533 537 537 539 544 549 554 557 564 570 574 579

21.4 21.5 21.6 21.7

22 Classical Field Configurations 22.1 Introductory Remarks 22.2 The Constant Classical Field 22.3 Soliton Theories in One Spatial Dimension 22.4 Stability of Classical Configurations 22.5 Bogomol'nyi Equations and Bounds 22.6 The Small Fluctuation Equation 22.7 Existence of Finite-Energy Solutions 22.8 Ginzburg-Landau Vortices 22.9 Introduction to Homotopy Classes 22.10The Fundamental Homotopy Group

XI

23 Path Integrals and Instantons 23.1 Introductory Remarks 23.2 Instantons and Anti-Instantons 23.3 The Level Difference 23.4 Field Fluctuations 23.4.1 The fluctuation equation 23.4.2 Evaluation of the functional integral 23.4.3 The Faddeev-Popov constraint insertion 23.4.4 The single instanton contribution 23.4.5 Instanton-anti-instanton contributions 23.5 Concluding Remarks

583 583 583 592 596 596 603 609 613 614 618

24 Path Integrals and Bounces on a Line 619 24.1 Introductory Remarks 619 24.2 The Bounce in a Simple Example 625 24.3 The Inverted Double Well: The Bounce and Complex Energy 631 24.3.1 The bounce solution 631 24.3.2 The single bounce contribution 635 24.3.3 Evaluation of the single bounce kernel 637 24.3.4 Sum over an infinite number of bounces 641 24.3.5 Comments 644 24.4 Inverted Double Well: Constant Solutions 644 24.5 The Cubic Potential and its Complex Energy 645 25 Periodic Classical Configurations 25.1 Introductory Remarks 25.2 The Double Well Theory on a Circle 25.2.1 Periodic configurations 25.2.2 The fluctuation equation 25.2.3 The limit of infinite period 25.3 The Inverted Double Well on a Circle 25.3.1 Periodic configurations 25.3.2 The fluctuation equation 25.3.3 The limit of infinite period 25.4 The Sine-Gordon Theory on a Circle 25.4.1 Periodic configurations 25.4.2 The fluctuation equation 25.5 Conclusions 649 649 650 650 659 663 664 664 667 669 670 670 671 673

Xll

26 Path Integrals and Periodic Classical Configurations 675 26.1 Introductory Remarks 675 26.2 The Double Well and Periodic Instantons 676 26.2.1 Periodic configurations and the double well 676 26.2.2 Transition amplitude and Feynman kernel 678 26.2.3 Fluctuations about the periodic instanton 679 26.2.4 The single periodic instanton contribution 684 26.2.5 Sum over instanton-anti-instanton pairs 688 26.3 The Cosine Potential and Periodic Instantons 690 26.3.1 Periodic configurations and the cosine potential . . . . 690 26.3.2 Transition amplitude and Feynman kernel 693 26.3.3 The fluctuation equation and its eigenmodes 694 26.3.4 The single periodic instanton contribution 696 26.3.5 Sum over instanton-anti-instanton pairs 700 26.4 The Inverted Double Well and Periodic Instantons 702 26.4.1 Periodic configurations and the inverted double well . 702 26.4.2 Transition amplitude and Feynman kernel 705 26.4.3 The fluctuation equation and its eigenmodes 706 26.4.4 The single periodic bounce contribution 708 26.4.5 Summing over the infinite number of bounces 710 26.5 Concluding Remarks 714 27 Quantization of Systems with Constraints 27.1 Introductory Remarks 27.2 Constraints: How they arise 27.2.1 Singular Lagrangians 27.3 The Hamiltonian of Singular Systems 27.4 Persistence of Constraints in Course of Time 27.5 Constraints as Generators of a Gauge Group 27.6 Gauge Fixing and Dirac Quantization 27.7 The Formalism of Dirac Quantization 27.7.1 Poisson and Dirac brackets in field theory 27.8 Dirac Quantization of Free Electrodynamics 27.9 Faddeev-Jackiw Canonical Quantization 27.9.1 The method of Faddeev and Jackiw 28 The 28.1 28.2 28.3 Quantum-Classical Crossover as Phase Transition Introductory Remarks Relating Period to Temperature Crossover in Previous Cases 28.3.1 The double well and phase transitions 715 715 717 720 723 726 727 734 736 740 740 745 745 753 753 755 756 757

Xlll

28.3.2 The cosine potential and phase transitions 28.4 Crossover in a Simple Spin Model 28.5 Concluding Remarks 29 Summarizing Remarks A Properties of Jacobian Elliptic Functions Bibliography Index

759 760 771 773 775 779 797

This page is intentionally left blank

Preface
With the discovery of quantization by Planck in 1900, quantum mechanics is now more than a hundred years old. However, a proper understanding of the phenomenon was gained only later in 1925 with the fundamental Heisenberg commutation relation or phase space algebra and the associated uncertainty principle. The resulting Schrodinger equation has ever since been the theoretical basis of atomic physics. The alternative formulation by Feynman in terms of path integrals appeared two to three decades later. Although the two approaches are basically equivalent, the Schrodinger equation has found much wider usefulness, particularly in applications, presumably, in view of its simpler mathematics. However, the realization that solutions of classical equations, notably in field theory, play an important role in our understanding of a large number of physical phenomena, intensified the interest in Feynman's formulation of quantum mechanics, so that today this method must be considered of equal basic significance. Thus there are two basic approaches to the solution of a quantum mechanical problem, and an understanding of both and their usefulness in respective domains calls for their application to exemplary problems and their comparison. This is our aim here on an introductory level. Throughout the development of theoretical physics two types of forces played an exceptional role: That of the restoring force of simple harmonic motion proportional to the displacement, and that in the Kepler problem proportional to the inverse square of the distance, i.e. Newton's gravitational force like that of the Coulomb potential. In the early development of quantum mechanics again oscillators appeared (though not really those of harmonic type) in Planck's quantization and the Coulomb potential in the Bohr model of the hydrogen atom. Again after the full and proper formulation of quantum mechanics with Heisenberg's phase space algebra and Born's wave function interpretation the oscillator and the Coulomb potentials provided the dominant and fully solvable models with a large number of at least approximate applications. To this day these two cases of interaction with nonresonant spectra feature as the standard and most important xv

Diverse and more detailed quantum mechanical investigations in the second half of the last century revealed that perturbation theory frequently does permit systematic procedures (as is evident e. any problem more complicated is frequently dispensed with by referring to cumbersome perturbation methods. like periodic potentials. be treated to a considerable degree of satisfaction perturbatively.g. the aforementioned exactly solvable cases. With various techniques and deeper studies. even though the expansions are mostly asymptotic. one problem being that there are few nontrivial models which permit a deeper insight into their connection. In the first part. the path integral method of Feynman was soon recognized to offer frequently a more general procedure of enforcing first quantization instead of the Schrodinger equation.e. and it will be seen in the final chapter that potentials with degenerate vacua are not exclusively of general interest. Then the Schrodinger equation is introduced and the two main exactly solvable cases of harmonic oscillator and Coulomb potentials are treated in detail since these form the basis of much of what follows. again point the way: For scattering problems the path integral seems particularly convenient. With the growing importance of models in statistical mechanics and in field theory. The introduction to quantum mechanics we attempt here could be subdivided into essentially four consecutive parts. numerous problems could. However. Thus this first part . has not always been understood well. its mathematical foundations. Thus important level splitting formulas for periodic and anharmonic oscillator potentials (i. basic postulates and standard applications. These basic cases will be dealt with in detail by both methods in this text.XVI illustrative examples in any treatise on quantum mechanics and — excepting various kinds of square well and rectangular barrier potentials — leave the student sometimes puzzled about other potentials that he encounters soon thereafter. with degenerate vacua) were first and more easily derived from the Schrodinger equation. in fact. Our approach to quantum mechanics is through a passage from the Poisson algebra of classical Hamiltonian mechanics to the canonical commutator algebra of quantum mechanics which permits the introduction of Heisenberg and Schrodinger pictures already on the classical level with the help of canonical transformations. screened Coulomb potentials and maybe singular potentials. we recapitulate the origin of quantum mechanics. in Feynman diagrams in quantum electrodynamics). Excluding spin. whereas for the calculation of discrete eigenvalues the Schrodinger equation. but arise also in recently studied models of large spins. Chapters 1 to 14. To what extent the two methods are actually equivalent. but also about complex energies that he encounters in a parallel course on nuclear physics. that is the Coulomb potential and the harmonic oscillator.

and the elliptic or Lame potential — here introduced earlier as a generaliza- . The solution of this case — however in nonperiodic form — turns out to be a prerequisite for the complete solution of the Schrodinger equation for the singular potential 1/r 4 in Chapter 19. This is followed by the important case of the cosine or Mathieu potential for which the perturbation method was originally developed. we derive respectively the level-splitting formula and the imaginary energy part for these cases for arbitrary states. Using perturbation theory. After a treatment of power potentials. The most prominent examples here are the double well potential and its inverted form. we deal mostly with applications depending on perturbation theory. and their eigenvalues. In the majority of the cases that we treat we do not use the standard Rayleigh-Schrodinger perturbation method but the systematic perturbation procedure of Dingle and Muller which is introduced in Chapter 8. Thereafter the concepts of instantons. The earlier Chapter 17 also contains a brief description of a similar treatment of the elliptic or Lame potential. We also consider inverted double wells and calculate with the path integral the imaginary part of the energy (or decay width). the method of matched asymptotic expansions with boundary conditions (the latter providing the so-called nonperturbative effects). and the behaviour of the eigenvalues is discussed in both weak and strong coupling domains with formation of bands and their asymptotic limits. which is presumably the only such singular case permitting complete solution and was achieved only recently. In the second part. The following Chapter then deals with Schrodinger potentials which represent essentially anharmonic oscillators. the chapter thereafter deals with Yukawa potentials. The potentials with degenerate minima will be seen to re-appear throughout the text.e.XVII deals mainly with standard quantum mechanics although we do not dwell here on a large number of other aspects which are treated in detail in the long-established and wellknown textbooks. i. Chapters 15 to 20. The following chapters deal with the derivation of level splitting formulas (including excited states) for periodic potentials and anharmonic oscillators and — in the one-loop approximation considered — are shown to agree with those obtained by perturbation theory with associated boundary conditions. In the final chapter of this part we discuss the large order behaviour of the perturbation expansion with particular reference to the cosine and double well potentials. periodic instantons. In part three the path integral method is introduced and its use is illustrated by application to the Coulomb potential and to the derivation of the Rutherford scattering formula. bounces and sphalerons are introduced and their relevance in quantum mechanical problems is discussed (admittedly in also trespassing the sharp dividing line between quantum mechanics and simple scalar field theory).

and whenever available also with the results of WKB calculations. however. for instance. The introduction of collective coordinates of classical configurations and the fluctuations about these leads to constraints. In addition this part considers in more detail the region near the top of a potential barrier around the configuration there which is known as a sphaleron. the Mathieu equation.g. Employing anharmonic oscillator and periodic potentials and re-obtaining these in the context of a simple spin model. All results are compared with those obtained by perturbation theory. Our fourth and final part therefore deals with elementary aspects of the quantization of systems with constraints as introduced by Dirac. We then illustrate the relevance of this in the method of collective coordinates.xvm tion of the Mathieu potential — re-appears as the potential in the equations of small fluctuations about the classical configurations in each of the basic cases (cosine. except that these are no longer of hypergeometric type. in compiling this text it was not possible to transcribe anything from the highly condensed (and frequently unsystematic) original literature on applications of path integrals (as the reader can see. Comparing the Schrodinger equation method with that of the path integral as applied to identical or similar problems. we can make the following observations. anharmonic oscillator potentials on a comparable level with. This puts Schrodinger equations with e. the Schrodinger equation can be solved for practically any potential in complete analogy to wellknown differential equations of mathematical physics. An expected observation is that — ignoring a minor deficiency — the WKB approximation is and remains the most immediate way to obtain the dominant contribution of an eigenenergy. This method is therefore more complicated. for instance. cubic). The physical behaviour there (in the transition region between quantum and thermal physics) is no longer controlled by the Schrodinger equation. it is. The application of path integrals to the same problems with the same aims is seen to involve a number of subtle steps. The particular solutions and eigenvalues of interest in physics are — as a rule — those which are asymptotic expansions. In fact. we consider the topic of transitions between the quantum and thermal regimes at the top of the barrier and show that these may be classified in analogy to phase transitions in statistical mechanics. quartic. These considerations demonstrate (also with reference to the topic of spin-tunneling and large-spin behaviour) the basic nature also of the classical configurations in a vast area of applications. such as limiting procedures. an approximation whose higher . this comparison on a transparent level being one of the main aims of this text. from our precise reference to unavoidable elliptic integrals taken from Tables). With a fully systematic perturbation method and with applied boundary conditions.

Throughout the text some calculations which require special attention. Andrews). are relegated to separate subsections which — lacking a better name — we refer to as Examples. Park (Masan). B. e. at the end of a chapter (after troublesome turning of pages). we also consider at various points of the text comparisons with WKB approximations. H. whose research into asymptotic expansions laid the ground for detailed explorations into perturbation theory and large order behaviour. Nonetheless. Tchrakian (Dublin) and Jianzu Zhang (Shanghai). Thus when instantons became a familiar topic it was natural to venture into this with the intent to compare the results with those of perturbation theory. Whittaker and G. D. like E. so that the reader is spared difficult and considerably time-consuming searches in a source (and besides. For ease of reading. As a rule. Miiller-Kirsten *In the running text references are cited like e. Dingle for paving him the way into this field which — though not always at the forefront of current research (including the author's) — repeatedly triggered recurring interest to return to it. shows him that each such formula here has been properly looked up). W. Liang (Taiyuan). In writing this text the author considered it of interest to demonstrate the parallel application of both the Schrodinger equation and the path integral to a selection of basic problems. also for the verification of results. T.g. with the source given in the bibliography at the end. K. Instead a glance at a nearby footnote provides the reader immediately the names of authors. The line of thinking underlying this text grew out of the author's association with Professor R. formulas taken from Tables or elsewhere are referred to by number and/or page number in the source. as well as applications and illustrations. not the least for permitting a more detailed and hopefully comprehensible presentation here. in particular Professors J. B. The author is deeply indebted to his one-time supervisor Professor R. the references referred to are never cited by mere numbers which have to be identified e.* H. which is particularly important in the case of elliptic integrals which require a relative ordering of integration limits and parameter domains. The author has to thank several of his colleagues for their highly devoted collaboration in this latter part of the work over many years. Whittaker and Watson [283].g.XIX order contributions are difficult to obtain. an additional motivation was that a sufficient understanding of the more complicated of these problems had been achieved only in recent years. Their deep involvement in the attempt described here is evident from the cited bibliography.-Q. Since this comparison was the guide-line in writing the text.g. N. D. J. This endeavour developed into an unforeseen task leading to periodic instantons and the exploration of quantum-classical transitions. Watson [283]. Dingle (then University of Western Australia. thereafter University of St. . other topics have been left out which are usually found in books on quantum mechanics (and can be looked up there).

that the formula he had established for the energy distribution of electromagnetic black body radiation was in agreement with the experimentally confirmed Wien.and Rayleigh-Jeans laws for the limiting cases of small and large values of the wave-length A (or AT) respectively is generally considered as the discovery of quantum mechanics.1 Origin and Discovery of Quantum Mechanics The observation made by Planck towards the end of 1900. The best approximation to such a body is a cavity with a tiny opening (of solid angle d£l) and whose inside walls provide a diffuse distribution of the radiation entering through the hole with the intensity of the incoming ray decreasing rapidly after a few reflections from the walls.Chapter 1 Introduction 1.2 cm at moderate temperatures T) is the radiation emitted by a body (consisting of a large number of atoms) as a result of the temperature (as we know today as a result of transitions between a large number of very closely lying energy levels). which involved also thermodynamics and statistical mechanics (in the sense of Boltzmann's statistical interpretation of entropy). We do not enter here into detailed considerations of Planck. very few explain in this context what he really did in view of involvement with statistical mechanics. the amount of radiation absorbed by a body is equal to the amount the body 1 . Kirchhoff's law in thermodynamics says that in the case of equilibrium. Instead. we want to single out the vital aspect which can be considered as the discovery of quantum mechanics. Planck had arrived at his formula with the assumption of a distribution of a countable number of infinitely many oscillators. Thermal radiation (with wave-lengths A ~ 10~ 5 to 10 . Although practically every book on quantum mechanics refers at the beginning to Planck's discovery. A "perfectly black body" is defined to be one that absorbs all (thermal) radiation incident on it.

. kT kXT (1. Fig. and the other in the region of large A (or AT).1) containing the constant h by treating the radiation in the cavity as something like a gas? By 1900 two theoretically-motivated (but from today's point of view incorrectly derived) expressions for u(u. It was found that one expression agreed well with observations in the region of small A (or AT).2) and the .e. y J c 3 \ex . How did Planck arrive at the expression (1. i. A being the wave-length of the radiation. Let us look at the final result of Planck. of the photons or photon gas) in the cavity with both possible directions of polarization (hence the factor "2") in the frequency domain v. u(u. h = 6.3 4 J s.T) = dv3e-C2U/T. radiators. In Eq. T) were known and tested experimentally.1) c is the velocity of light with c = u\. v + dv in equilibrium with the black body at temperature T.e. (1. energy per unit volume) of the radiation (i.626 x 10 .e.1 Absorption in a cavity.38 x 1(T 23 J K'1. and then measuring the increase in temperature of the heat bath. The parameters k and h are the constants of Boltzmann and Planck: k = 1. The (equilibrium) radiation of the black body can be determined experimentally by sending radiation into a cavity surrounded by a heat bath at temperature T.l ) where x = ^ = ^ .2 CHAPTER 1. (1.e.1) ' y Here u(v.T) = 2*?£(-?-)kT. the formula (to be explained) u(u. Black bodies as good absorbers are therefore also good emitters. Introduction emits. i. These expressions are: (1) Wien's law. T)du is the mean energy density (i. 1.

Here W is a number which determines the distribution of the energy among a discrete number of objects. and he had succeeded in finding such an expression of the form u(v.T) 2^^kT. Indeed. This is the point. 1. exp(x) ~ 1+x) and "large" (exp(—x) < 1). C3 being constants. Planck now imagined a number T of oscillators V or iV oscillating degrees of freedom. To this end he considered Boltzmann's formula S — klnW for the entropy S. discretization begins to enter. . but — here the discreteness appears properly — only in elements (quanta) e. and thus over a discrete number of admissible states. every oscillator corresponding to an eigenmode or eigenvibration or standing wave in the cavity and with mean energy U. (x large).3) are contained in Eq. in the first place Planck had tried to find an expression linking both. Moreover Planck assumed that these oscillators do not absorb or emit energy continuously. e xhv. Considering Eq. C2. he searched for a derivation. 3 (1. which are indistinguishable) among the N indistinguishable oscillators at . so that W represents the number of possible ways of distributing the number P := NU/e of energy-quanta ("photons". "small" (i. We see.1) in regions of a. that the formulas (1.1. (1.1) as approximations.e.2) and (1. we obtain: u(i/.3) Ci.T) = av e6"/T-i' where a and b are constants. where the Fig. (1.1 Origin and Discovery of Quantum Mechanics (2) Rayleigh-Jeans law: u(i>. T) u(i/. When Planck had found this expression.T) = 2^C3T.2 Distributing quanta (dots) among oscillators (boxes).47TZ/ 2 {x small).

p. i. We visualize the iV oscillators as boxes separated by N — 1 walls. there not called Stirling's formula. formula 8.3 Comparing the polarization modes with those of a 2-dimensional oscillator.2) requires that e ex is. with riydu — 2 x —w—dv. N -* oo. U{T) being the average energy emitted by one oscillator.g. u(i>. S. v + dv. (1. i. The Stirling formula or approximation will appear frequently in later chapters. 940. Magnus and F. one obtains (cf. Then W is given by w = (N With the help of Stirling's formula* {N + p-iy. W. Agreement with Eq. with the quanta represented schematically by dots as indicated in Fig.g.T)dv.2.7) *See e. We now obtain the energy density of the radiation. e. Oberhettinger [181]. Ryzhik [122].343(2). and the second law of thermodynamics ((dS/dU)v Example 1. (1. 1.e.4 CHAPTER 1.e. e = his.4) IniV! ~ JVlniV-iV + O(0).3.1)!P! (1. . p. as for small values of e). . Introduction temperature T. = 1/T). 1.6) Fig. (1. I. as in most other Tables. Gradshteyn and I. h = const. by multiplying U with the number nvdv of modes or oscillators per unit volume with frequency v in the interval v. M.1) u = vmrri (L5) as the mean energy emitted or absorbed by an oscillator (corresponding to the classical expression of 2 x kT/2.

'''From the equation I -\ JW .V 2 ) E = 0. KT = I J . (1..T)dX = ^ehc/*kT_idX.UJ = 2-KV. as in electrodynamics.1 Origin and Discovery of Quantum Mechanics 5 where the factor 2 takes the two possible mutually orthogonal linear directions of polarization of the electromagnetic radiation into account.1. We observe that u(v.2. is given by . rii = 1.3.T) has a maximum which follows from du/dX = 0 (with c = vX).3 (as for ideal conductors). 2 2 2 r2 L K — 7T n . so that (lvL\A 0 I I = rr. Then L^j = nrii..1)) The solutions of this equation are ^max = 4.1). where we have for the electric field E oc elwt \ J eK sin KI^I sin K2X2 sin K3X3 K with the boundary condition that at the walls E = 0 at Xi = 0... i.7).3.1. (1. nvdv per unit volume.i = 1.. where^ 2 [2-KUY . d ™4*±\IL> — -—dv = nvdv dv dv |_8 3 \ c / 14 8 2 4TTV2 = 3 dv 83 ^ ^ = ^ ^ ' as claimed in Eq. In terms of A we have u(X. The number of possible modes (states) is equal to the volume of the spherical octant (where n^ > 0) in the space of n^. so that .2.3. (1.965 and xmin = 0. .UJ2/C2 + K? = 0. so that the derivative of u implies (x as in Eq.e. . We obtain the expression (1. L for i = 1. as indicated in Fig.7) for instance. dj\l dM .. v + dv. The number with frequency v in the interval v. We obtain therefore u^T) = Unv = 2^-fJ^—i.8) This is Planck's formula (1.2.

we then have at T = 0 independent oscillators. that of an oscillation system at absolute temperature T ^ 0.2 (1.e. the behaviour of the system at zero absolute temperature. (1. . We shall see later that in the case of this linear harmonic oscillator the energies En are given by En= (n + -jhu= U + I W h=—.6 The first value yields Amax-T = CHAPTER 1. This is Wien's displacement law. We are not dealing with the linear harmonic oscillator familiar from mechanics here. but one can expect an analogy. One might suppose now. which can not be eliminated without a different approach. We therefore examine such contradictions next. that we arrive at quantum mechanics simply by discretizing the energy and thus by postulating — following Planck — for the harmonic oscillator the expression (1. Lorentz and Planck that Eq. .l. ra = 0. Thus we have a rather complicated system here. n = 0. Since temperature originates through contact with other oscillators. However. If an oscillator with thermal weight or occupation probability exp(—nx) can assume only discrete energies en = nhu.10). . and from which the constant h can be determined from the known value of k. x — oo) the mean energy vanishes (0 < U < * > oo). which did not arise in Planck's consideration of 1900.1. One expects. then (with x = hv/kT) its mean energy is En=0e = nX dx ^0 — /ii/— In = hu-f r%e dx 1 — e_x (1 — e~x)z hv .—v <L9) We observe that for T — 0 (i. of course. Introduction he 4. . that it is easier to consider first the case of T = 0.965K = Const. i. such a procedure leads to contradictions. Later it was realized by H. which can assume the discrete energies en — nhv.e.10) Thus here the so-called zero point energy appears.8) could be derived much more easily in the context of statistical mechanics. A. which had also been known before Planck's discovery. . 2 .

. where k is Boltzmann's constant and W is the number of times P indistinguishable elements of energy e can be distributed among T V indistinguishable oscillators.kT This means U is then the classical expression resulting from the mean kinetic energy per degree of freedom.InP!] ~ kN The second law of thermodynamics says 1+ 7 ln 1 + ( 7)~7ln7 \au)v For a single oscillator the entropy is s = S/N.e. around 1926.ln(TV . so that 1 f ds\ T — \dUjy . for 2 degrees of freedom.= . with Heisenberg's discovery of the uncertainty relation.1)! . i. We "This is what was effectively done before 1925 in Bohr's and Sommerfeld's atomic models and is today referred to as "old quantum theory". {NW •• 1)! (TV-1)!P! and P = UN Show with the help of Stirling's formula that the mean energy U of an oscillator is given by U •• exp(e/fcT) .In ..1. kT/2.2 Contradicting Discretization: Uncertainties 7 Example 1.2 Contradicting Discretization: Uncertainties The far-reaching consequences of Planck's quantization hypothesis were recognized only later. we obtain T V S =fc[ln(TV+ P .1: Mean energy of an oscillator In Boltzmann's statistical mechanics the entropy S is given by the following expression (which we cite here with no further explanation) S = fcln W.+ 1 e \U e u = exp(e/fcT) which for e/kT — 0 becomes > -.k dU 1+U\ f ln(l T + U\ - U U k ( e . d .1 ' u~ e . 1.1)! .1 Solution: Inserting W into Boltzmann's formula and using In TV! ~ A In T — TV.i n . In the following we attempt to incorporate the above discretizations into classical considerations* and consider for this reason socalled thought experiments (from German "Gedankenexperimente").

8

CHAPTER 1. Introduction

shall see that we arrive at contradictions. As an example^ we consider the linear harmonic oscillator with energy E = -mx2 + - w V .
ZJ

(1.11)

Zi

The classical equation of motion dE n — = x(mx + mco x) = 0 permits solutions x = Acos(u>t + S), so that E = -mco2 A2, where A is the maximum displacement of the oscillation, i.e. at x — 0. We consider first this case of velocity and hence momentum precisely zero, and investigate the possibility to fix the amplitude. If we replace E by the discretized expression (1.10), i.e. by En — (n + 1/2)HUJ, we obtain for the amplitude A

A A

^ -=\[Ef+l-

(i i2)

-

Thus the amplitude can assume only these definite values. We now perform the following thought experiment. We give the oscillator initially an amplitude which is not contained in the set (1.12), i.e. for instance an amplitude A with An <A<An+l. Energy conservation then requires that the oscillator has to oscillate all the time with this (according to Eq. (1.12) nonpermissible) amplitude. In order to be able to perform this experiment, the difference AA = An+1 - An must not be too small, i.e. the difference AA =

V mu>

n~V tfAA

n+

£HV

1 2h 2 [l + 0 ( l / n ) ] . mui 4^/n

For m = 2kg, h = i x 10-- 3 4 J s , u = I s - 1 , we obtain lO" 17 [l + 0 ( l / n ) ] meter. 2^

+ H. Koppe [152].

1.2 Contradicting Discretization:

Uncertainties

9

This distance is even less than what one would consider as a certain "diameter" of the electron (~ 10~ 15 meter). Thus it is even experimentally impossible to fix the amplitude A of the oscillator with the required precision. Since A is the largest value of x, where x = 0, we have the problem that for a given definite value of mx, i.e. zero, the value of x = A can not be determined, i.e. given the energy of Eq. (1.10), it is not possible to give the oscillator at the same time at a definite position a definite momentum. The above expression (1.10) for the energy of the harmonic oscillator, which we have not established so far, has the further characteristic of possessing the "zero-point energy" Hu>/2, the smallest energy the oscillator can assume, according to the formula. Let us now consider the oscillator as a pendulum with frequency u in the gravitational field of the Earth. * Then

"2 = f,

(1-13)

where I is the length of the pendulum. Thus we can vary the frequency cv by varying the length I. This can be achieved with the help of a pivot, attached to a movable frame as indicated in Fig. 1.4. The resultant of the tension in the string of the pendulum, R, always has a nonnegative vertical component. If the pivot is moved downward, work is done against this vertical component of R; in other words, the system receives additional energy. However, there is one case, in which for a very short interval of time, 8t, the pendulum is at angle 0 = 0. Reducing in this short interval of time the length of the pendulum (by an appropriately quick shift of the pivot) by a factor of 4, the frequency of the oscillator is doubled, without supplying it with additional energy. Thus the energy En= ( n + - ) fojj becomes I n + - IH2co,

without giving it additional energy. This is a self-evident contradiction. This means — if the quantum mechanical expression (1.10) is valid — we cannot simultaneously fix the energy (with energy conservation), as well as time t to an interval 8t —• 0.§ The source of our difficulties in the considerations of these two examples is that in both cases we try to incorporate the discrete energies (1.10) into the framework of classical mechanics without any changes in the latter. Thus the theory with discrete energies must be very different from classical mechanics with its continuously variable energies.
H. Koppe [152]. See also Example 1.3.

10

CHAPTER 1. Introduction

It is illuminating in this context to consider the linear oscillator in phase space (q,p) with
P2
1

29

E —

1—mco q = const. 2m 2 *

(1.14)

Fig. 1.4 The pendulum with variable length. This equation is that of an ellipse as a comparison with the Cartesian form a2
+

'" b2

reveals immediately. Evidently the ellipses in the (g,p)-plane have semi-axes of lengths 2E b= V2mE. (1.15) a = mw' Inserting here (1.10), we obtain 2(n + l/2)fr^ (1.16) hn = ^2m{n + l/2)^. mar We see that for n — 0,1, 2 , . . . only certain ellipses are allowed. The area enclosed by such an ellipse is (note A earlier amplitude, now means area) An = nanbri or ,( pdq — 2irh I n + '- ]. (1.17b) 2irEn
UJ

2Tih{n+

-

(1.17a)

In the first of the examples discussed above the contradiction arose as a consequence of our assumption that we could put the oscillator initially at

1.2 Contradicting Discretization:

Uncertainties

11

any point in phase space, i.e. at some point which does not belong to one of the allowed ellipses. In the second example we chose n = 0 and thus restricted ourselves to the innermost orbit. However, we also assumed we would know at which point of the orbit the pendulum could be found. Thus in attempting to incorporate the discrete quantization condition into the context of classical mechanics we see, that a system cannot be localized with arbitrary precision in phase space, in other words the area AA, in which a system can be localized, is not nought. We can write this area AA > An+1 -Any'=
(1.17a) 2TT/L

since the system cannot be "between" An+i and An. Since A A represents an element of area of the (q, p)-plane, we can write more precisely ApAq > 2irh. (1.18)

This relation, called the Heisenberg uncertainty relation, implies that if we wish to make q very precise by arranging Aq to be very small, the complementary uncertainty in momentum, Ap, becomes correspondingly large and extends over a large number of quantum states, as — for instance — in the second example considered above and illustrated in Fig. 1.5.

Fig. 1.5 Precise q implying large uncertainty in p. Thus we face the problem of formulating classical mechanics in such a way that by some kind of extension or generalization we can find a way to quantum mechanics. Instead of the deterministic Newtonian mechanics — which for a given precise initial position and initial momentum of a system yields the precise values of these for any later time — we require a formulation answering the question: If the system is at time t = 0 in the area defined by

12 the limits 0 < q < q + Aq, 0<p<p

CHAPTER 1. Introduction

+ Ap,

what can be said about its position at some later time t = T? The appropriate formulation does not yet have anything to do with quantum mechanics; however, it permits the transition to quantum mechanics, as we shall see. Before we continue in this direction, we return once again briefly to the historical development, and there to the ideas leading to particle-wave duality.^

1.3

Particle-Wave Dualism

The wave nature of light can be deduced from the phenomenon of interference, as in a double-slit experiment, as illustrated in Fig. 1.6.

Fig. 1.6 Schematic arrangement of the double-slit experiment. Light of wave-length A from a source point 0 can reach point P on the observation screen C either through slit A or through slit B in the diaphragm placed somewhere in between. If the difference of the path lengths OBP, OAP is n\,n € Z, the wave at P is re-inforced by superposition and one observes a bright spot; if the difference is n\/2, the waves annul each other and one observes a dark spot. Both observations can be understood by a wave propagation of light. The photoelectric effect, however, seems to suggest a corpuscular nature of light. In this effect* light of frequency v is sent onto a metal plate in a vacuum, and the electrons ejected by the light from the plate are observed by applying a potential difference between this plate and another one. The energy of the observed electrons depends only on v and
"See also M.-C. Combourieu and H. Rauch [58]. "This is explained in experimental physics; we therefore do not enter into a deeper explanation here.

1.3 Particle-Wave Dualism

13

the number of such photo-electrons on the intensity of the incoming light. This is true even for very weak light. Einstein concluded from this effect, that the energy in a light ray is transported in the form of localized packets, called wave packets, which are also described as photons or quanta. Indeed the Compton effect, i.e. the elastic scattering of light, demonstrates that photons can be scattered off electrons like particles. Thus whereas Planck postulated that an oscillator emits or absorbs radiation in units of hv = hu>, Einstein went further and postulated that radiation consists of discrete quanta. Thus light can be attributed a wave nature but also a corpuscular, i.e. particle-like, nature. In the interference experiment light behaves like a wave, but in the photoelectric effect like a stream of particles. One could try to play a trick, and use radiation which is so weak that it can transport only very few photons. What does the interference pattern then look like? Instead of bands one observes a few point-like spots. With an increasing number of photons these spots become denser and produce bands. Thus the interference experiment is always indicative of the wave nature of light, whereas the photoelectric effect is indicative of its particle-like nature. Without going into further historical details we add here, that it was Einstein in 1905 who attributed a momentum p to the light quantum with energy E = hv, and both he and Planck attributed to this the momentum

The hypothesis that every freely moving nonrelativistic microscopic particle with energy E and momentum p can be attributed a plane harmonic matter wave ip(r,t) was put forward much later, i.e. in 1924, by de Broglie.t This wave can be written as a complex function ij)(T,t) =Aeik-r-iut,

where r is the position vector, and to and k are given by E — hio, p = /ik. Thus particles also possess a wave-like nature. It is wellknown that this was experimentally verified by Davisson and Germer [64], who demonstrated the existence of electron waves by the observation of diffraction fringes instead of intensity distributions in appropriate experiments.
f

L. de Broglie [39].

14

CHAPTER

1.

Introduction

1.4

Particle-Wave Dualism and Uncertainties

We saw above t h a t we can observe the wave nature of light in one type of experiment, and its particle-like nature in another. We cannot observe both types simultaneously, i.e. the wave-like nature together with the particle-like nature. Thus these wave and particle aspects are complementary, and show up only under specific experimental situations. In fact, they exclude each other. Every a t t e m p t to single out either of these aspects, requires a modification of the experiment which rules out every possibility to observe the other aspect.* This becomes particularly clear, if in a double-slit experiment the detectors which register outcoming photons are placed immediately behind the diaphragm with the two slits: A photon is registered only in one detector, not in b o t h — hence it cannot split itself. Applying the above uncertainty principle to this situation, we identify the attempt to determine which slit the photon passes through with the observation of its position coordinate q. On the other hand the observation of the interference fringes corresponds to the observation of its momentum p.§ Since the reader will ask himself what happens in the case of a single slit, we consider this case in Example 1.2. Example 1.2: The Single-Slit Experiment
Discuss the uncertainties of the canonical variables in relation to the diffraction fringes observed in a single-slit experiment. Solution: Let light of wave-length A fall vertically on a diaphragm Si with slit AB as shown schematicaly in Fig. 1.7.

^y

Ax

Fig. 1.7 Schematic arrangement of the single-slit experiment.
On the screen S2 one then observes a diffraction pattern of alternately bright and dark fringes, in the See, for instance, the discussion in A. Messiah [195], Vol. I, Sec. 4.4.4. Considerable discussion can be found in A. Rae [234].

1.4 Particle-Wave Dualism and Uncertainties

15

figure indicated by maxima and minima of the light intensity I. As remarked earlier, the fringes are formed by interference of rays traversing different paths from the source to the observation screen. Before we enter into a discussion of uncertainties, we derive an expression for the intensity I. Since the derivation is not of primary interest here, we resort to a (still somewhat cunbersome) trick justification, which however can also be obtained in a rigorous way." We subdivide the distance AB = Ax into N equal pieces AP\, P1P2,..., as indicated in Fig. 1.8.

Si

t
Ax
\

»v
"^^J
B ^ ^

*

w

p3

Q

Fig. 1.8 The wave-front

WW.

We consider rays deflected by an angle 9 with wave-front WW' and bundled by a lense L and focussed at a point Q on the screen S2. Since WW1 is a wave-front, all points on it have the same phase, so that light sent out from a source at Q reaches every point on WW1 at the same time and across equal distances. Hence a phase difference at Q can be attributed to different path lengths from Pi,P2,... to WW'. Considering two paths from neighbouring points Pi,Pj along AB, the difference in their lengths is Axsva6/N. In the case of a wave having the shape of the function 2?r sin kr = sin this implies a phase difference given by < ;v = 5 2n Ax sin 6 \ N (1.20)

Just as we can represent an amplitude r having phase 6 by a vector r, i.e. r — |r|exp(iS), we > can similarly imagine the wave at Q, and this means its amplitude and phase, as represented by a vector, and similarly the wave of any component of the ray passing through AP\, P1P2, • • •• If we represent their effects at Q by vectors of equal moduli but different directions, their sum is the resultant OPN as indicated in Fig. 1.9. In the limit N — 00 the N vectors produce the arc of a > circle. The angle 5 between the tangents at the two ends is the phase difference of the rays from the edges of the slit:
27T

5 = 2a =

lim NSN = — A a ; s i n 0 .

(1.21)

If all rays were in phase, the amplitude, given by the length of the arc OQ, would be given by the chord OQ. Hence we obtain for the amplitude A at Q if AQ is the amplitude of the beam at the slit: . length of chord OQ , 2a sin a , sin a A0 , ,,—; =A0 . (1.22) 7^-=A0 length of arc OQ a2a "S. G. Starling and A. J. Woodall [260], p. 664. For other derivations see e.g. A. Brachner and R. Fichtner [32], p. 52.

16
The intensity at the point Q is therefore

CHAPTER

1. Introduction

h = h
where from Eq. (1.21)
•K ,
.

a = -flisinB = -Aisint A 2

.

k

Fig. 1.9 The resultant OPM of N equal vectors with varying inclination.
Thus the intensity at the point Q is

Ie=Io

sin 2 (fcAx sin6(/2) (fcAx sin 0/2) 2

(1.23)

The maxima of this distribution are obtained for fcAxsinfl = (2n + 1 ) - , i.e. for A x sin0 = (2n + 1 ) - = (2n + 1) and minima for 1 fcAx sin # = 7171", i.e. for A x i
: TlA.

A

(1.24a)

(1.24b)

The maxima are not exactly where only the numerator assumes extremal values, since the variable also occurs in the denominator, but nearby. We return to the single-slit experiment. Let the light incident on the diaphragm S i have a sharp momentum p = h/\. When the ray passes through the slit the position of the photon is fixed by the width of the slit A x , and afterwards the photon's position is even less precisely known. We have a situation which — for the observation on the screen S2 is a past (the uncertainty relation does not refer to this past with px = 0, rather to the position and momentum later; for the situation of the past A x A p is less than h). The above formula (1.23) gives the probability that after passing through the slit the photon appears at some point on the screen 52. This probability says, that the photon's momentum component px after passing through the slit is no longer zero, but indeterminate. It is not possible to predict at which point on S2 the photon will appear (if we knew this, we could derive px from this). The momentum uncertainty in the direction x can be estimated from the geometry of Fig. 1.10, where 6 is the angle in the direction to the first minimum: Apx = 2px =2psin6 = — sing.
A

(1-25)

From Eq. (1.24b) we obtain for the angle 9 in the direction of the first minimum Ax sin 6 = A,

1.4 Particle-Wave Dualism and Uncertainties

17

Fig. 1.10 The components of momentum p.
so that Ax Apx = 2h. If we take the higher order minima into account, we obtain AxApx A x Apx > h. We see that as a consequence of the indeterminacy of position and momentum, one has to introduce probability considerations. The limiting value of the uncertainty relation does not depend on how we try to measure position and momentum. It does also not depend on the type of particle (what applies to electromagnetic waves, applies also to particle waves). = 2nh, or

1.4.1

Further thought experiments

Another experiment very similar to that described above is the attempt to localize a particle by means of an idealized microscope consisting of a single lense. This is depicted schematically in Fig. 1.11. light

Fig. 1.11 Light incident as shown. The resolving power of a lense L is determined by the separation Aa; of the first two neighbouring interference fringes, i.e. the position of a particle is at best determinable only up to an uncertainty Ax. Let 9 be one half of the angle as shown in Fig. 1.11, where P is the particle. We allow light to fall in the direction of —x on the particle, from which it is scattered. We assume a quantum of light is scattered from P through the lense L to S where it

18

CHAPTER 1. Introduction

is focussed and registered on a photographic plate. For the resolving power Ax of the lense one can derive a formula like Eqs. (1.24a), (1.24b) . This is derived in books on optics, and hence will not be verified here, i.eJ Ax~-±-. (1.26a) 2 sm 0 The precise direction in which the photon with momentum p = h/X is scattered is not known. However, after scattering of the photon, for instance along PA in Fig. 1.11, the uncertainty in its x-component is 1h Apx = 2psin0 = — sine A (1.26b)

(prior to scattering the x-components of the momenta of the particle and the photon may be known precisely). From Eqs. (1.26a), (1.26b) we obtain again Ax Apx ~ h. The above considerations lead to the question of what kind of physical quantities obey an uncertainty relation. For instance, how about momentum and kinetic energy T? Apparently there are "compatible!'1 and "incompatible" quantities, the latter being those subjected to an uncertainty relation. If the momentump x is "sharp", meaning Apx = 0, then also T = px2/2m is sharp, i.e. T and px are compatible. In the case of angular momentum L = r x p, we have |L| = |r||p'| = rp', where p' = p sin 0. As one can see, r and p' are perpendicular to each other and thus can be sharp simultaneously. If p' lies in the direction of x, we have Ax Ap' > h, where now Ax = rAip, ip being the azimuthal angle, i.e. rAipAp'>h, i.e. ALA<p>h.

Thus the angular momentum L is not simultaneously exactly determinable with the angle </?. This means, when L is known exactly, the position of the object in the plane perpendicular to L is totally indeterminate. Finally we mention an uncertainty relation which has a meaning different from that of the relations considered thus far. In the relation Ax Apx > 0 the
"See, for instance, N. F. Mott, [199], p. 111. In some books the factor of "2" is missing; see, for instance, S. Simons [251], p. 12.

1.5 The Complementarity

Primciple

19

quantities Ax, Apx are uncertainties at one and the same instant of time, and x and px cannot assume simultaneously precisely determined values. If, however, we consider a wave packet, such as we consider later, which spreads over a distance Ax and has group velocity VQ = p/m, the situation is different. The energy E of this wave packet (as also its momentum) has an uncertainty given by

AE « -T^Ap = vGAp. op
The instant of time t at which the wave packet passes a certain point x is not unique in view of the wave packet's spread Ax. Thus this time t is uncertain by an amount
At w Ax

vG

.

It follows that AtAE^AxAp>h. (1.27) Thus if a particle does not remain in some state of a number of states for a period of time longer than At, the energy values in this state have an indeterminacy of |Ai£|.

1.5

Bohr's Complementarity Principle

Vaguely expressed the complementarity principle says that two canonically conjugate variables like position coordinate x and the the associated canonical momentum p of a particle are related in such a way that the measurement of one (with uncertainty Ax) has consequences for the measurement of the other. But this is essentially what the uncertainty relation expresses. Bohr's complementarity principle goes further. Every measurement we are interested in is performed with a macroscopic apparatus at a microscopic object. In the course of the measurement the apparatus interferes with the state of the microscopic object. Thus really one has to consider the combined system of both, not a selected part alone. The uncertainty relation shows: If we try to determine the position coordinate with utmost precision all information about the object's momentum is lost — precisely as a consequence of the disturbance of the microscopic system by the measuring instrument. The so-called Kopenhagen view, i.e. that of Bohr, is expressed in the thesis that the microscopic object together with the apparatus determine the result of a measurement. This implies that if a beam of light or electrons is passed through a double-slit (this being the apparatus in this case) the photons or

20

CHAPTER 1. Introduction

electrons behave like waves precisely because under these observation conditions they are waves, and that on the other hand, when observed in a counter, they behave like a stream of particles because under these conditions they are particles. In fact, without performance of some measurement (e.g. at some electron) we cannot say anything about the object's existence. The Kopenhagen view can also be expressed by saying that a quantity is real, i.e. physical, only when it is measured, or — put differently — the properties of a quantum system (e.g. whether wave-like or corpuscular) depend on the method of observation. This is the domain of conceptual difficulties which we do not enter into in more detail here.*

1.6

Further Examples

Example 1.3: The oscillator with variable frequency
Consider an harmonic oscillator (i.e. simple pendulum) with time-dependent frequency w(t). (a) Considering the case of a monotonically increasing frequency w(t), i.e. dui/dt > 0, from LUQ to u>', show that the energy E' satisfies the following inequality Eo < E' < —y-Eo, w o (1.28)

where Eo is its energy at deflection angle 6 = 0Q. Compare the inequality with the quantum mechanical zero point energy of an oscillator. (b) Considering the energy of the oscillator averaged over one period of oscillation (for slow, i.e. adiabatic, variation of the frequency) show that the energy becomes proportional to ur. What is the quantum mechanical interpretation of the result? Solution: (a) The equation of motion of the oscillator of mass m and with variable frequency co(t) is mx + mui (t)x = 0, where, according to the given conditions, — > 0, dt
dui

u> = u>o a,t t = 0, w = ui at t = T,

.

_

i.e. io{t) grows monotonically. Multiplying the equation of motion by x we can rewrite it as
1 , 1 w -mx • 2-\—mui 2 (t)x 2 W 2 2
2 1 n —mx 2 ^ = 0. 2 dt

dt The energy of the oscillator is l „ E — -mx1 2

l 0 , z l 29 + -mu} (t)x , y 2 '

so that

dE 1 —- = -mxz dt 2

9

dJ1 > 0, dt ~

(1.29) v '

where we used the given conditions in the last step. On the other hand, dividing the equation of motion by UJ2 and proceeding as before, we obtain - [mx + mur (t)x\ = 0, i.e.
1 1 1 — —--mx -2 H—mx 2 dt u22 2

1 mx 2

2d

— . 2 dt\u)

"See e.g. A. Rae [234]; P. C. W. Davies and J. R. Brown [65].

1.6 Farther

Examples

21

d ( E\ 1 , d / 1 \ — — ) = -mx2 — ( — = dt\uJ2J 2 dt\u>2)

mx2 dw < 0, UJ3 dt ~

(1.30) v

where the inequality again follows as before. We deduce from the last relation that 1 dE —2 u} dt Integrating we obtain fE' dE ^ f"'2 / < /
2 dw2 . _ E , _, n • n 2 ,a,' —7T, i-e. [[InE f„ [ lna; 2J u 22 , i.e. 2 E < E

E dw2 < 0, UJ4 dt ~

i.e.

1 dE 1 dw2 < —2 . E dt ~ u) dt

(1.31)

JEo
or

E - Jui

u, '

> °-

"o

E0 ~ UJ22

' ^ u'2 — < —-

E'

<

-^-EQ.

Next we consider the case of the harmonic oscillator as a simple pendulum in the gravitational field of the Earth with

e + wge-o, ^o = f.
and we assume that — as explained in the foregoing — the length of the pendulum is reduced by one half so that J2 = 2 - =2u;2. Then the preceding inequality becomes E' < 2E0. In shortening the length of the pendulum we apply energy (work against the tension in the string), maximally however EQ . Only in the case of the instantaneous reduction of the length at 6 = 0 (the pivot does not touch the string!) no energy is added, so that in this case E' = EQ, i.e. E0 < E' < 2E0. We can therefore rewrite the earlier inequality as , u'2 < -5-Bo.

E0<E'

Just as the equality on the left applies in the case of an instantaneous increase of the frequency (shortening of pendulum string), so the equality on the right applies to d = # m a x . In classical physics we have 1 -2 1 —mx H—? 2 2 If no energy is added, but u> is replaced by 2a; 2 , then x changes, and also x, i.e. x becomes shorter and x becomes faster. The quantum mechanical expression for the energy of the oscillator in its ground state is the zero point energy E = Hu>/2. Here in quantum physics we cannot change UJ without changing E. This means if we double tj instantaneously (i.e. in a time interval A t — 0) > without addition of energy (to fojj/2), then the result E' = Tiw is incorrect by A E = HUJ/2. We cannot have simultaneously A t — 0 and error A E = 0. > (b) The classical expression for E contains u> quadratically, the quantum mechanical expression is linear in OJ. We argue now that we can obtain an expression for E c i a s s i c a l by assuming that w(t) varies very little (i.e. "adiabatically") within a period of oscillation of the oscillator, T. Classical mechanics is deterministic (i.e. the behaviour at time t follows from the equation of motion and

22

CHAPTER

1.

Introduction

the initial conditions); hence for the consideration of a single mass point there is no reason for an averaging over a period, unless we are not interested in an exact value but, e.g. in the average

(lmX/ = ^I0 \mx2{P>dtx2(t) = UJ2X2 and hence

(L32)
-E i.e. in

If u> is the frequency of x(t), i.e. x(t) oc cosujt or sinu>t depending on the initial condition, then l-mw2x2\ (as follows also from the virial theorem). Eq. (1.29), for mx2 / 2 the mean value = (-mx2\ =

If we now insert in the equation for dE/dt,

/I 1£ 2\ ( - mx ) = , \ 2 / 2u2' we obtain dE_/l

~dt ~ \2mX
and hence

2\dw

2

_Edw2

dE _ 1 dw2 _ du

/ ~dT ~ 2w2~dT'
E — = const. w

°r

~E ~ ~iU> ~~ ~u7'

In quantum mechanics with E = hw{n + 1/2) this implies H(n + 1/2) = const., i.e. n = const. This means, with slow variation of the frequency the system remains in state n. This is an example of the so-called adiabatic theorem of Ehrenfest, which formulates this in a general formJ

Example 1.4: Angular spread of a beam
A dish-like aerial of radius R is to be designed which can send a microwave beam of wave-length A = 2irh/p from the Earth to a satellite. Estimate the angular spread 6 of the beam. Solution: Initially the photons are restricted to a transverse spread of length A x = 2R. From the uncertainty relation we obtain the uncertainty /\px of the transverse momentum px as Apx ^ h/2R. Hence the angle 0 is given by

~~ p

2R\2nh)

~ A-KR'

See e.g. L. Schiff [243], pp. 25 - 27.

Chapter 2

Hamiltonian Mechanics
2.1 Introductory Remarks

In this chapter we first recapitulate significant aspects of the Hamiltonian formulation of classical mechanics. In particular we recapitulate the concept of Poisson brackets and re-express Hamilton's equations of motion in terms of these. We shall then make the extremely important observation that these equations can be solved on the basis of very general properties of the Poisson bracket, i.e. without reference to the original definition of the latter. This observation reveals that classical mechanics can be formulated in a framework which permits a generalization by replacing the c-number valued functions appearing in the Poisson brackets by a larger class of quantities, such as matrices and operators. Thus in this chapter we attempt to approach quantum mechanics as far as possible within the framework of classical mechanics. We shall see that we can even define such concepts as Schrodinger and Heisenberg pictures in the purely classical context.

2.2

The Hamilton Formalism

In courses on classical mechanics it is shown that Hamilton's equations can be derived in a number of ways, e.g. from the Lagrangian with a Legendre transform or with a variational principle from the Hamiltonian H(qi,Pi), i.e.
rt2 r
6

/

^2PiQi-H(qi,Pi) dt = 0,

where now (different from the derivation of the Euler-Lagrange equations) the momenta pi and coordinates qi are treated as independent variables. As
23

chapter VII. which all together describe the state of the system. All functions u(qi.g. Hamiltonian Mechanics is wellknown.1) as d . H. (2. The total time derivative of u can therefore be rewritten with the help of Eqs.n n.e. (2. whereas the velocity requires observations of space coordinates at different times. Pi has to be considered as an independent quantity. Goldstein [114]. Goldstein [114] remarks at the end of his chapter VIII. A. (2. du \ x^/dudH du dH\ . M. dH In this Hamilton formalism it is wrong to consider the momentum pi as mqi. One can verify readily that Eq.Pi. chapter VIII. Rather.4) contains as special cases the Hamilton Eqs. x^fdu. . one obtains the Hamilton equations* • OH . We can therefore consider Eq.2) as This equation is. as mass times velocity.4) as the generalization of Eqs. (2. qi(t + 6t) .Pi) is contained implicitly in the canonical variables q^ and pi.t). in analogy with Eqs. Compared with an arbitrary function f(qi.pi) of qi. (2. S »(«. since . which can be observed directly at time t. i. ^As H. It suggests itself therefore to consider more closely the properties of the symbols (2. (2. The following properties can be verified: *See e.) = £ [wm + WiK) = £ [WiWi . the entire time-dependence of observables u(qi.1).. Dirac [75]. the equation of motion of the observable u.3).1). (2. It was only with the development of quantum mechanics by Heisenberg and Dirac that Poisson brackets gained widespread interest in modern physics. the standard reference for the application of Poisson brackets is the book of P. 6t->0 5t Real quantities which are directly observable are called observables.p% are therefore again observables. . (2.P.2) If we have only one degree of freedom (i = 1). this expression is simply a functional determinant.qi(t) qi = hm f . One now defines as (nonrelativistic) Poisson bracket the expression^ With this definition we can rewrite Eq.^ ^ j .24 CHAPTER 2.1). A system consisting of several mass points is therefore described by a number of such variables.

the Poisson bracket {A.5d) above. in other words without any reference to the original definition (2. Since Eqs. Later we shall consider noncommuting quantities. Property (2.5b) (3) complex conjugation (note: observables are real.B}C or C{A.6).A}. B}. {qi.B*}. {C. but could be multiplied by a complex number): {A. (2. B2}. a i S i + a2B2} = ax{A. (2. {A.4). Bx} + a2{A. 25 (2. can be solved solely with the help of the properties of Poisson brackets and the fundamental Poisson brackets (2.BC} (5) Jacobi identity: {A. (2. that the very general Eq.5d) is useful in calculations. As long as we are concerned with commuting quantities.5d) = {A*. (2) linearity: {A.e.3) of the Poisson bracket. C}} + {B. If we evaluate the Poisson brackets for qi.Qk} = 0.B} = -{B. where A and B are arbitrary observables.6) We can now show. like here. i. B}} = 0. The original definition of the Poisson bracket will not . which combines the Hamilton equations.5e) = {A.B}* (4) product formation: {A.6) give the values of these. for example. (2. If. then the ordering is taken as in (2. As an example we consider a case we shall encounter again and again. that of the linear harmonic oscillator.Pk} = 0. A}} + {C. (2. we wish to evaluate {A. (2.2 The Hamilton Formalism (1) Antisymmetry: {A.C}.B}. {B.B}C + B{A. {pi.Pi. it is irrelevant whether we write {A. B} is completely evaluated.5a) (2.2.Pk} = 5ik. we expand A and B in powers of qi and pi and apply the above rules until only the fundamental brackets remain.5c) The first three properties are readily seen to hold. we obtain the fundamental Poisson brackets. These are {Qi.

These are transformations qi—>Qi = Qi(q.10) In classical mechanics one studies also canonical transformations. (2.26 CHAPTER 2. from which we infer that q(t) — qocost + posint. Then we obtain: (2.8a) (2.4) we have for u = q. q + q = o. (2. From Eqs.p) is an ordinary function is also irrelevant.12) . "4' = q. (2.H}.7) q = [q.7) into (2.p. (2.q2}) = = 2i{q. and Pi are ordinary real number variables and that H(q. or q(t) = qo+ Pot . q = {q. (2. (2.9) Similarly we obtain from Eq. and P={p.yPot3 + •••• (2.11a) 'q' = -q.8b) We insert (2.8b) p=-q. In the evaluation one should also note that the fact that g.H}.11b) (2. Since constants are also irrelevant in this context.p.t)..P2} + {q. (2. and so q = -q..6). Pi—> Pi = Pi(q.-qot2 .p}p + p{q.\(p2 + q2)} = l({q..p.. According to Eq.p\) P.t).10) we deduce q = p = -q. we consider as Hamiltonian the function H(q.8a) and use the properties of the Poisson bracket and Eqs.9) and (2.p) = ±(p2 + q2). Hamiltonian Mechanics be used at all. (2.

(2. verify that the Poisson bracket of two observables A and B is invariant. Example 2.2.15a). {Qi.3) we can now express the Poisson bracket {^4. {PhPk} = 0.13) We write the reversal of the transformation (2. i.e.qk} = 0.P is canonical in the sense defined above. which means that a Hamilton function K(Q.P. *H.15a) °raS {AiB}Q.B}Q>P. The proof requires the invariance of the fundamental Poisson brackets under canonical transformations. chapter VIII.P One can then show that {A. _ax p__dK_ (2. In classical mechanics we learned yet another important aspect of the Hamilton formalism: We can inquire about that particular canonical transformation.3 deals with the relativistic extension. .15b) The proof of the latter invariance is too long to be reproduced in detail here but can be found in the book of Goldstein.Qk} = 0.e. those at a time t — 0. Of course.e.pk} = 5ik. P) exists.14) With the help of the definition (2. {qi. (2.p = {A. this transformation is described precisely by the equations of motion but we shall not consider this in more detail here. {qi. Example 2. (2. i. i. .12) Qi—>qi = qi{Q.* Hence in Example 2.2 The Hamilton Formalism 27 for which the new coordinates are also canonical. (2.P- provided the transformation q.B}q.15a). Eq. as {A. {QhPk} = Sik.2 below contains a further illustration of the use of Poisson brackets. Pi^Pi=Pi(Q. i.e.1 we verify only Eq.P.t). which transforms qi.p <-> Q. i.Pi back to their constant initial values. and Example 2.e.t).Pk} = 0. for which Hamilton's equations hold.B}q. B} of two observables A and B in terms of either set of canonical variables.1: Canonical invariance of Poisson bracket Assuming the invariance of the fundamental Poisson brackets under canonical transformations Qj = Qj(Q'P)>Pj = Pj{Q>P). (2. that (dropping the subscripts therefore) {Pi. Goldstein [114].

236.P-^pr.A}q.2: Solution of Galilei p r o b l e m w i t h Poisson brackets Consider the Hamiltonian for the free fall of a mass point mo in the gravitational field (linear potential). § See P.p v ^ (9A dB ^ | _ _ *-? V dqj dpj dA 8B —— dpj dqj 8B_dI\\ dPk dpj J _ 8A_f^B_dQk dpj \ 8Qk dqj dB 8Pk dPk dqj = E E k dAL/'^B_dQk dqj \ dQk dpj {A.( 9A = {Pk. (2. V V °Qk dPk dPk dQk J E x a m p l e 2.Pj}q. Mittelstaedt [197]. 1 „ H = p + m0gq and solve the canonical equations with Poisson brackets for initial conditions q(0) = qo. Hamiltonian Mechanics S o l u t i o n : U s i n g t h e definition of t h e P o i s s o n b r a c k e t a p p l i e d t o ^4 a n d B w e h a v e r „ „•. . § E x a m p l e 2.B}QIP.s F 5 7 r ={A. OQk Pk}q. we have —Et. Solution: Relativistically we have to treat space and time on an equal footing.q/c) and (E. Replacing here A by Q and B by A.p-^v dP.p = J2 d~Fk Replacing in the above A by P and B by A.Qj}q.p-^ "Pk. (2.15a) r„ ™ v .pc). dA {Qk.p = dB dA dB \ rA 1 {A'B^ E b r s r .28 CHAPTER 2.A}q. {A.F}r ~di d~E ~ d~E~8t Consider F = H(q. The relativistic Poisson bracket (subscript r) therefore becomes du dF dq dp du dF dp dq du &F _ du dF ' {u.p^— + {A. we obtain analogously dA dQk' Inserting both of these results into the first equation. Qk}q. Thus we extend q and p to space-time vectors (t.p(0) = poSolution: The solution can be looked up in the literature.p) . we obtain as claimed by Eq.3: Relativistic Poisson brackets By extending qi. p.B}q. we obtain dA {Qk.pt to four-vectors (in a (1 + 3)-dimensional space) define relativistic Poisson brackets.15b) dA {Qk. their product Et — qp being relativistically invariant.E(t). Thus whenever q and p are multiplied.+ dQ.

.e.. numerically zero. We distinguish in the following between two kinds of probabilities.to. Probabilities Single particle consideration We continue to consider classical mechanics in which the canonical coordinates qi.y. This probability is evidently proportional to AqAp. Probabilities 29 (This is. Instead we assume a case in which we know only that the system is located in some particular domain of phase space.Pi are the coordinates of some mass point m. g oc AqAp.p) to G\{q.Po.t). Let us assume that at some initial time to the system may be found in a domain Go(q.)•'=(*)*-*£. i. of course. (1. it is the equations of motion which lead from Go(q.po.1 Liouville Equation._. so that Go(qo. and E as a function of t). we have A = (p apaq — J . in the case of the linear oscillator with energy E given by Eq. one obtains Gi with q = q((lo.t). and at time t > to in a domain G\(q. H(q. du . since H is expressed as a function of q and p. which is the probability of a particle having a coordinate q between q and q + Aq and a momentum p between p and p + Ap. (*.po is a point on Go.po) = Gi(q. Of course. (2. We consider first the a priori weighting or a priori probability. .3 Liouville Equation. but partial derivatives of F do not vanish.p). dE .e. (1..2..14) and area A of the phase space ellipse given by Eq. if qo. But now we consider a system whose phase space coordinates are not known precisely. I .E(t)}r = Hence at Relativistically we really should have clu/dr. p= p(qo.3. c dt 2 2 w^ du du dr .p) . rrr . to .3 2. i.16) For example. du q-\ P = —.t0.p) around some point qo.p)..17a). also boundary points of one domain are mapped into boundary domains of the other. and the space spanned by the entire set of canonical coordinates is described as its phase space. where dr is the difference of proper time given by dudH 1 du dE(t) du — = 1 du .Po.p). dudH s l u . Then . g.. Since Hamilton's equations give a continuous map of one domain onto another.' ±=j1V c \dtj' -u^.dt y/l - qZ/c2' 2. 2TTE UJ and hence g oc — .

/. Pi 2 / \. dt dq and with with Hamilton's equations (2.4: Liouville's theorem Show that A q A p is independent of time i. dq Aq dq 2 dq Aq dq 2 qH to the right and q to the left.. dpdq and similarly dt = — Ap. — dp A A — In(AqAp)s = — + — = Example 2. Example 2.1): d . = 2TrIEsm6. and hence g oc %ir2IdE. dt dt Aq d ( A p ) dt ' Ap Here d(Aq)/dt is the rate at which the q-walls of the phase space element move away from the centre of the element.30 CHAPTER 2. this has the same value at a time to.p^.eded<t> = 8TT2IE.14) t o (1. as at a time t'0 ^ to • Solution: We consider -dln(A9AP)-rf(A<Z) ' . Show that the total volume of phase space covered for constant E is 8n2IE.4. in view of this independence it can be expressed in terms of the conserved energy E. Example 2.— = — Aq.5: A priori weighting of a molecule If the rotational energy of a diatomic molecule with moment of inertia / is 1 / 2 . Hamiltonian Mechanics If g depended on time t it would be dynamical and would involve known information about the particle. the (pg. as is demonstrated by Liouville's theorem in Example 2. dt dq dq dp dp d2H dqdp d2H = 0.15) — an ellipse of area § dpgdp.5 thereafter provides an illustration of the a priori weighting expressed in terms of energy E.)-curve for constant E and $ is — as may be seen by comparison with Eqs. ? « + •sin 2 6>/' 21E 2IEsin2e' in spherical polar coordinates. Thus g must be independent of t. (1. Hence from the difference: — . Je=0 . Solution: Integrating over the angles we have =2TT fe=7r 2frEsin. J0=0 Hence g oc 8n2IdE. which means.

e. i. (2.2 E n s e m b l e consideration 31 We now assume a large number of identical systems — the entire collection is called an ensemble — all of whose initial locations are possible locations of our system in the neighbourhood of the point qo.p.t).1 The system moving from domain Go to domain G\.3. The total number of systems N is obtained by integrating over the whole of phase space.Thus we assume a large number of identical sytems. Since W has the dimension of a reciprocal action.t)dqdp= 1.3 Liouville Equation.p + dp.p.17) F ^GT^^ 0 Fig.19) Thus W is the probability to find the system at time t at q.18) With a suitable normalization we can write this / W(q. (2.2. (2. We consider the totality of these systems which is described by a density of points p (number dn of points per infinitesimal volume) in phase space.*).Po. by -j-r = P(9»P.q + dq.p.p.p.20) . Probabilities 2.t)dqdp = N. dn — p(q. i. it is suggestive to introduce a factor 2-KK with every pair dpdq without. (2. 2.e. W=-p(q. whose positions in phase space are characterized by points. Thus dn is that number of systems which at time t are contained in the domain q. however. d d 1 P = JJdqidpi. leaving the basis of classical mechanics! Hence we set / «w)^ =.

p) and vp(q. t + dt)dqdp. 2.e. We are now interested in how n ov W changes in time.3 The region G.e. we consider the domain G in Fig.p) denote the velocities in directions q and p — p{q. t + dt)dqdp — p(q.p. i.3 and establish an equation for the change of the number of points or systems in G in the time interval dt.p.e.a t dt + q+dq. is equal to the number in G at time t plus the number that went into G in the time interval dt minus the number that left G in the time interval dt.p. Hamiltonian Mechanics We can consider 2irh as a unit of area in (here the (1 + l)-dimensional) phase space.p. p + dp G P O q qn-dq Fig.p. i.t)dp( — I . In order to derive this equation.2 The ensemble in phase space. 2.P dt p(q + dq. p(q.t)dp{ -jt Q. — if vq(q. The number of points at time t + dt in domain G. The equation of motion for n or W is the so-called Liouville equation.p / . 2. i. In doing this. that in our consideration no additional points are created or destroyed.32 CHAPTER 2. t)dqdp p(q. how the system moves about in phase space. we take into account. -* q Fig.

Comparison of Eq. p.t) dt p(q.t)vq{q. (2. Probabilities and thus p(q. t)vq(q + dq. p.p) .p.p.3 Liouville Equation.19).p.t)^- = 1.t + dt)dqdp — p(q.p.P)]-|K(Q. vp(q.p).p)dtdp +p(q.p(q.p. so that dt ~~dq\P Hence dp) + dp\P ~dq)~ ~~dq~~dp~ + dp~~dq~ ~ { . (2.p + dp) dp = or | However.p.t)vq(q.p.21) with Eq.p) . the probable motion of the system under consideration.P *' % = iH>P}- ( 2 .p(q. = -|K(9.p) = q = -g^.p(q + dq.20) and (2. .t + dt) .p.t)vq(q + dq.W(q. 33 Dividing both sides by dqdpdt this becomes p(q.t)vp(q. (2.t)} with JW(q.2.t) dt {H(q.p(q. (2.p + dp)dtdq. put differently.p)=p = 3H -—.p)dtdq .P.P)].p + dp. t)dqdp = p(q. With Eqs.p)dtdp .22) The generalization to n degrees of freedom is evident: The volume element of phase space is .p) dq p(q.p(q + dq.21 ) This is the Liouville equation which describes the motion of the ensemble or. .p + dp.4) shows that p and u satisfy very similar equations.p. . (2. .t)vp(q. dH Mq.t)vp(q.21) we can also write dW(q.t)vp(q.p.

1 the area Go is equal to the area G\. and hence that equal phase space volumes contain the same number of systems.e.4). With Eq. 2. The first and most immediate possibility is — as indicated .24) since the total derivative is made up of precisely the partial derivatives contained in Eq. i. p.e.t)(^-J.p (cf.4). that dtj\y v if qo. (2.p. t) depends explicitly on time t (if determined at a fixed . We shall see that we have two possibilities for this.p)W(q. and this means — since these systems are contained in a finite part V of phase space — that dt We have in particular. since no systems are created or destroyed.24) implies that p is a constant in time.34 CHAPTER 2. We now inquire about the time variation of the expectation value (it) of u. (2. (-.24).p. Equation (2. Hamiltonian Mechanics where is the probability for the system to be at time t in the volume q. 2. for i<«>-!/«<*p>"w>(^)". We deduce from the Liouville equation the important consequence that ^ M = 0. Example 2.p).that the density or probability W(q.po are the initial values of q.p) be an observable. (2.p) the following expression: (u)=Ju(q.4 Expectation Values of Observables Let u = u(q. p+dp. We define as expectation value of u(q. Thus in Fig. (2. i.26) we described the time variation of the observable u(q. q+dq.

Po. i-e.22).24): W(q. Then Eq. we have Qo = g(q. at t = 0.p) = uo{qo. where we used Eq. that (u) = (u)0.p.p.po.27) becomes !<»> - /«(*P>!"W>(^)" (2.t).29) The distribution of the canonical variables is given by W(q. Thus we can write. since W oc p is constant in time according to Eq.p. and hence u(q.Po.t).t).2.p.p).Po.t).po. Goldstein [114].t). We verify this claim as follows.Po) at time t = 0. (2. However.32) In this expression the time t is contained explicitly in the observable u(q. (2.0) = W0(q0. so that Q = g(qo. (2. (2.po.t). W is the density in the neighbourhood of a given point in phase space and has an implicit dependence on time t.t) W(q0.t) = = W(g(q0.t)}(^y. of course.W(q. . With these expressions we obtain for the expectation value (U)Q: (u)o = Ju0(qo.po.28) = Ju(q. 8.t). and the time variation d(u)/dt is attributed to the fact that it is this probability (that u(q.e.t).31) i.Po.0) =u(g(qo.t).^ Solving the equations of motion for q.8.t). Sec. (2.f(qo.p){H(q.4 Expectation Values.33) po = f(q. (2.34) S e e also H.f(q0. we can express these in terms of their initial values qo. (2.p) assumes certain values) that depends explicitly on time. Observables 35 point in phase space). 1 (2.Po.29). Reversing Eq.p. we can also employ a more complicated consideration.po.p0)(^^-J. We expect.0) = u0{qo.i).t)W0(q0. (2.p. (2.p.p) = u(q.p0.30) p = f(qo.

t) 0H_ 0 q = - {{H(q. p = f(<j(q.t).38) we obtain -£<«)„ = .Po)(^^)n.t).t) .po.30) Moreover.P.t).PO.\„. / ^fdq0dpc . .0)=u(q.p).t) d t _ (du\ \ d<i dq / P=f(qo.0) u(q.p.P)}).p.39) Substituting this into Eq.) Wo(qo.p.p).f(q.f({H(q. Hamiltonian Mechanics so that on the other hand with Eq.36) and (2.p.t). i.p0) = W(q. (2.(dq0dp0\n We deal with the partial derivative with the help of Eq.Po.p.) f / u (q0.t) dp M du\ ° q 'P=f(qo.P0.e.t) = (2 30) (2. (2.t) M f 9u\ V °P / P=f(q0.p)})q=g(QO.p)W{q.u{q. (2. we obtain an expression which is different from that in Eq.t).Po.29) q = g(g(q. (2.po)(~ . Inserting Eqs.t).p). Eq.po. [ du0{q0.(2.32).f(q.t) dB__ (du\ °P \°PS P=f(qo. we obtain («)o = / u{q.t).36 CHAPTER 2.u(q.37) and (2. Taking now the total time derivative of (it)o.p.t)W "(q0.p. (2.30) u0(g(q.=9(.35) Inserting these expressions into no we obtain uo(qo.t).31)) W0(q0.p.f(g(q.p.t).Po.32). (2-40) .t) (2. (2.t)J(q. d dt {U)0 = dt. (2. (2.P0.t). (cf. (2.30): duo(qQ.37) as had to be shown.f(q.t).p.t). Eq.36) = (2.35).28).t)\—^\ = (u)." V " " ™0' -> • • » « » 0 .P.POit)i p=f(qo. (2.po.t). (2.25) into Eq.t) v{g(<j(q.Po.f(q.M 27Th i .o.t).p.

Observables 37 Here we perform the transformation (2..p)}W(q. W}.44) iji^ioo dpi (which is reasonable since the density vanishes at infinity).p. (2. dt <«)o v f u(q. u}W + u{H. This timedependence is described by the Liouville equation dW(q.p.p.p..p)) is attributed to the probability W(q.t)} . The phase-space integral of a Poisson bracket like I (2.p). from the properties of the Poisson bracket. / d q d p \ n 2irh J (2.. Consider {H.31).42) w-xtf)" ~dHd(uW) dqi dpi vanishes under certain conditions.34) and use (2. lim ——uW — 0. The considerations we just performed demonstrate that we have two ways of treating the time-dependence: The explicit time-dependence can either be contained in the probability W or in the (transformed) observables.25) and (2. the observable u is treated as a function u(q.p){H(q. we obtain zero after partial integration of I and hence from Eqs.45) in agreement with Eq.t) dqdp\n 2-irhJ (2.t)}.28) and (2.W(q.28). (2.t) of u assuming certain values q and p. ^ „.p).41) This expression contains {H. and the time variation of (u(q. uW} = {H.W(q. .42) the relation d . (2. described as "Schrodinger picture". Alternatively we could deduce from Eqs.(2.45) the Liouville equation.2.41) and (2.u(q.W} in Eq.4 Expectation Values.p.43) ——uW = 0. x r „. we obtain {H. (2.28).t) dt {H(q. In the first case.p). u}W instead of u{H.uW} = J2 i OHd(uW) dpi dqi =£ If for all i: lim Pi->±oo Oqi _9_(dEuW dpi V dqi d fdH uW dqi V dpi (2. . However. so that ~d~t <«>0 = -J{H(q.p).

does not have to be commutative.F(«.5 Extension beyond Classical Mechanics With the above considerations we achieved a general formulation of classical mechanics.4) of an observable on the other.** These considerations point the way to a generalization which results if we permit u and W to belong to a more general class of mathematical quantities. as we observed.po are constant initial values — we have du0(q0. 2. which. Koppe [152] at the university of Munich around 1964." and the explicit time-dependence is transferred into the correspondingly transformed observables Uo(qo. Hamiltonian Mechanics In the other case. This formulation deals with observables u representing physical quantities.46) We thus also recognize the connection between the Liouville equation. (b) As usual a muliplication by a complex number is defined. (c) A multiplication of the quantities among themselves is defined..p)}. which are described as "Schrodinger picture" and "Heisenberg picture"— all this on a purely classical level but with the use of canonical transformations. and the equation of motion (2. The time dependence of the expectation values can be dealt with in two different ways.po) is assumed. .rt. the reason being that — since qo. Thus we arrive at a more general theory if we define u and W with the following properties: (a) An addition is defined between the quantities.t).t) dt du(q. called u Heisenberg picture". for which the axioms of a commutative group apply. as the equation of motion of an ensemble or of a probability distribution on the one hand. (2.p)}. the probability of the initial values Wo(qo. (cf. " This means: Only in the Heisenberg picture dW/dt = 0.p) It = {u(q.38 CHAPTER 2. and probabilities W.t) dt __ ~ (2J30) ( 4) du0(qo.po. i.po. which describe the state of a system.4)) ^ = {«(. (2.p). Eq. " T h e author learned this approach from lectures of H.H(q.p0.e. The equation of motion is then that of an observable. but does satisfy the usual associative and distributive laws.

Introducing it here assumes.2. can be deduced from the following characteristic properties of a trace.48): (a) TV u* = (Tr «)* = TV^.26)) (u) = Tr(wW). We shall then interpret as "canonical quantization" the procedure which allocates to each of the fundamental Poisson brackets (2.49) With these considerations we have reviewed aspects of classical particle mechanics in as close an approach to quantum mechanics as seems possible. (b) TV(cm + j3u) = oTV u + /3Tr v.6) a so-called "commutator''' [A.5 Extension beyond Classical Mechanics 39 Quantities satisfying these properties define a linear algebra.p)(^y. also called "density matrix". Eq.48) In matrix theory the symbol "trace" has a well-defined meaning. (2. ox In view of our later considerations.p} = l {q. y:\p. That this is the case. it is helpful to introduce already at this stage some additional terminology. ~[q. The quantity corresponding to W in quantum mechanics is the so-called "statistical operator'''. In Chapter 9 we attempt a corresponding approach for classical systems with . (2. px -> -in—. and to be able to correlate these with the above classical considerations. that its use here implies the essential properties it has in matrix theory. Moreover.q} = 0 {p. for a phase-space integral we define the word or symbol "trace". by Traceu:=Ju(q.) = TV (vu). therefore.p] = l. if u ^ 0. also written "TV". (2.p]=0(2-47) One verifies readily that the commutator relations are satisfied by the differential operator representations qx -> x. Thus we can write the expectation value of an observable u (cf.p} = 0 • — -^ -~[q. which all apply to (2. B] := AB — BA in the following way: 'i {q. so that we consider this next.q}=0. (c) TV (u*u) > 0. (d) Tr(ut.

40 CHAPTER 2.e. These aspects will be considered in Chapter 27. it will be shown that in electrodynamics. obtain corresponding results — as one would envisage in view of the expected particle-wave duality in quantum mechanics. and. the Poisson brackets require modification to Dirac brackets. electrodynamics. since gauge fixing (i. a constraint) has to be taken into account.e. . Hamiltonian Mechanics a wave-like nature. i. Thus we can now proceed to prepare the ground for the extension of classical mechanics into an operator formulation. However. excepting the Poisson brackets.

We found that the Poisson algebra permits extensions to non-cnumber formulations. i.1 Introductory Remarks In Chapter 2 we investigated the algebraic structure of classical Hamiltonian mechanics.e. of measurable quantities. = {ipi} is called a linear vector space on the set of numbers I 6 {C}. quantum mechanics: The Hilbert space as the space of state vectors representing the states of a physical system.Chapter 3 Mathematical Foundations of Q u a n t u m Mechanics 3. we can define the Hilbert space as the space of states of a physical system. i. with the canonical commutation relations or Heisenberg algebra defining the basic product relations. which turn out to be those of the theory today known as quantum mechanics.e. if the elements ipi of M satisfy the usual axioms of addition and K 41 . A set M. In this chapter we therefore introduce important basic mathematical concepts of this non-c-number mechanics.2 Hilbert Spaces We first recapitulate some fundamental concepts of linear algebra and begin with the axioms defining a linear vector space. and selfadjoint operators in this space as representatives of observables. These somewhat abstract considerations — although later in many cases not referred back to — are a necessary prerequisite for the formulation of a mechanics which is not of the c-number type as classical mechanics. We also introduce in this chapter the concepts of linear functionals and distributions so that we can make free use of the delta distribution and similar objects in later chapters. 3. Building upon this.

the vectors il)i. (ipi + ipj) + ijjk = ipi + (ipj + ipk).M is said to be a metric vector space or a pre-Hilbert space. i. . . A + tpj = tpj + i'i. .ipn are said to be linearly dependent. ( n (3. E (3./?eK).^i) . x M. i=l (3.i = 1.2) Vectors ip\. (aptyi.ip2).^) called inner product. . 2 .ip2. if numbers eti. (3 5c) + a2{ip. .M.tp2.aiV>i + a 2 ^ 2 ) = ai(ip. fa. .^1) = (-01. (V>. . n exist (not all zero).---. (a. - . f > 0 if V T ^ O . if any two elements ip\. (0 : null element and with complex numbers a and j3: a(tpi + tpj) = a{Wi) lipi = = < M). so that n 5 > ^ = 0. ip2 of this space can be associated with a complex number (V'l. (3.5a) (3. (ipi. if every vector ip E M can be associated with numbers Q . . n.ipn are said to be linearly independent. are linearly dependent.42 CHAPTER 3.3) If all on = 0. with the properties (a* G IK): (^2. .-02)* (hermiticity).~ip2) '• M. ^ i = 0 if ^ = 0. Mathematical Foundations of Quantum Mechanics multiplication by complex numbers. such that n ^ = YJCi^i.5b) .1) aipi + aipj. If n + 1 elements ^ € M. • • • . . i = 1. and n is the smallest such number. (3. 2 .e. —»• IK. n is called the dimension of .4) The vector space . In each case n linearly independent vectors are said to form a basis.

3.^) = ai*(V'i. (3. The norm of the vector ip (pre-Hilbert space norm) is defined as |H|:=(^)1/2.9a) (3. (3. In order to verify Eq.9c) HV'i + tp2\\2 + \\1p1 .*h)--=\\A-H\(3-8) (3. which we can write 0 < ( ^ i .rh) (3. meaning one-and-a-halffold linearity).9b) = 0 (Pythagoras theorem).2.)> (3-6) i. (3. ^ 2 ) | . if Wi.e.V') +"2*(V. linearity in the second component.tp2 £ M..is defined by d(ipi.V. for ip\. WA+Ml2 = ll^il 2 + ll^2|| 2 .9d) | | V i | | = sup | ( ^ i .i>2\\2 = 2||'i/'i||2 + 2||'i/'2||2 (parallelogram equation). ^ ) + |A| 2 (V2.7) In addition the following relations hold in a metric space M. for arbitrary A and ip2 ^ 0: (V>i + AV>2.9e) 11^11=1 We restrict ourselves here to some remarks on the verification of these wellknown relations. »2|| 2 . The distance between two vectors if>i. tp2 £ -M: 1(^1. (3. \\1p1 + ^211 < HV'ill + IIV^II (triangle inequality). ^ i ) + A * ( ^ .^1) 2 h 11 0 <-%«e + M .^2)1 < H^ill • 11^211 (Schwarz inequality)..^ 2 ) = | | ^ | | 2 + 2 ^ i .2 Hilbert Spaces 43 where the asterix * means complex conjugation. A V 2 ) + |A|2||V2||2 2 \m\\ llwl For if)2 7^ 0 we set A so that |(^2^l)|2 2 IWI + ll^i||. The first two properties imply ( a i ^ i + a2ip2. (3-10) (^2.V>i + A</>2) > 0 . ^ i ) + A ( ^ i . antilinearity in the first component (also described as sesquilinearity.9a) we start from if) = ipi + \if>2 6 M.

^ e M are said to be orthogonal if (</>!. beginning with A = 1 in the second line of Eq.9d).ii) thus verifying Eq.* Two vectors i f i . so that | | ^ l + ^ 2 | | < | | ^ l | | + ||^2||. (3. (3. a £ K : ll^ill HVill 11^1 +^211 |M| If for a vector tp e M: > = < = 0. We also omit the verification of the following properties of the norm of a vector tpi € M with ^ € X . o ^ V i = o.V2) IIV'l||2 + ||V'2||2 + 21(^2. be the set of all column vectors v\ V = (Vi) = I 2 V with complex numbers Vi. (3.| | ^ l l = (HlMI + | | « 2 . Mathematical Foundations of Quantum Mechanics 1(^2.^1)1 ||Vi|| 2 + ||^2|| 2 + 2 | | ^ | | .13) the vector is said to be normalized. ll^ill + llVdl. for which CO |2 := V^l^l 2 < 00.12) We omit here the verification of the remaining relations (3.44 or CHAPTER 3. (3. V 2 ) = 0 . Examples of metric vector spaces: (1) Let M. In verifying the triangle inequality we use this result (3.^)1 <IIV>il|. (3. i=l * Not all the wave functions we consider in the following and in later chapters are automatically normalized to 1.9e). H-IHI (3. .\\ih\\.10): ll^+^H2 = < < | | ^ | 2 + ||V2||2 + 2K(Vl. (3. hence verification in each case is necessary.11).9a).9c).

w) :=J2v*Wi. But this applies also in the case of any function which is nonzero only on a set of measure zero. with inner product oo 45 (v. Elements of the classes are then called representatives of these classes.5c)) / ( x ) = 0 = > ( / . = C? be the set of all complex-valued integrable functions / ( x ) on 5 C IR3 (in the sense of Lebesgue) for which / VSCR 3 |/(x)| 2 d 3 x < oo. Js the space C? is not yet a metric vector space although for (cf.g)= [ /(x)^(x)d3x. Then L2 is the space of all these equivalence classes. i. for which the scalar product. (2) Let M. •'^ \ 0 otherwise. (3. With the scalar product (f. (3. In order to avoid this difficulty. i. and one defines addition and multiplication by complex numbers with respect to these classes. are combined to an equivalence class [/] (with space L2). Eq.5a).2 Hilbert Spaces Then we define v + w := (v{) + (wi) := (vi + Wi).e. / ) = 0. (3. is defined by ([/]. The Schwarz inequality is then oo oo \M\<* ^2\Vi\ ^2\Wj\2. all square-integrable functions / which are "almost everywhere equal". which satisfies relations (3.3. 2 and so on. . which differ solely on a set of measure zero.5c). etc.e.[$]):= / JM /(x)* 5 (x)d 3 *.5b). */ x } = / /o for x = 0.

46 and

CHAPTER 3. Mathematical Foundations of Quantum Mechanics

ll[/]||=0=»[/] = [0],
where [0] is defined as the class of all functions which are almost everywhere zero. This means that functions that differ only on a pointset of Lebesgue measure zero are looked at as identical. Unless necessary, we ignore in the following mostly for simplicity the distinction between C2 and L2. Convergence of sequences in Hilbert space is then called convergence "almost everywhere". With the help of the concept of a norm we can introduce the concepts of convergence and point of accumulation. Definition: A sequence {tpn} € M. is said to converge (strongly) towards tp E M., if the distance \\tp — ipn\\ tends towards zero, i.e. lim | | ^ - V n | | = 0.
n—•oo

(3.14)

The vector ip is then called point of accumulation. The point of accumulation does not have to be an element of A4. If M. contains all of its points of accumulation, the set M is said to be closed. A normalized vector space M. which with every convergent sequence contains a vector towards which the sequence converges, is said to be complete, i.e. if ipn e M. with lim \\ipn - i/jm\\ = 0
m,n—>oo

(called Cauchy sequence), there is a ip 6 M. with ip = lim ipn,
n—>oo

i.e.

lim \\ip — ipn\\ = 0.
n—>oo

(3.15)

Every finite-dimensional vector space (on IK) is complete in the sense of the concept of convergence defined above (so that completeness does not have to be demanded separately). In order to see this, we consider the convergent sequence
n

<Pa = ^2CmiPi,
where i/ji,...,ipn€.A>i ras)

CaieK,

(3.16)

constitute a basis in M- Then (according to Pythagon

nv« - M = i sec™ - c0i)A\\ =YI i c - -c^2'

2

2

n

(3-i?)

3.2 Hilbert Spaces

47

a relation also known as Parseval equation. The convergence of the sequence i/ja implies the convergence of the sequence {Cai} towards a number Cj. Then for the vector
i=l

we have

ll^a-V'II^ElC^-Cil2,
i=i

(3.18)

i.e. that the sequence of the vectors tpa converges towards tp. We thus arrive at the definition of a Hilbert space. Definition: An infinitely dimensional, metric vector space, which is also complete with regard to (strong) convergence, is called a Hilbert space "K. The given definition of a Hilbert space is that usually given in mathematics.t In physics this is generally supplemented by the requirement that the space be separable, i.e. of a countably infinite dimensionality. Naturally Hilbert spaces with a countable basis are the simplest. We supplement the above by referring to the concept of a dense set or subset M of "K. A subset M of "K is said to be dense in "K, if to every / 6 "K there exists a sequence of vectors fn, fn < M, so that fn — f, S > i.e. fn converges strongly to / , implying that every vector / e Ji can be approximated arbitrarily precisely. We consider next some examples. Examples of Hilbert spaces: (1) The hyperspherical functions Yitm(6,(p) define a complete set of basis functions on the unit sphere. Any function f(9,(p) with

[\f(e,tp)\2dn<™
can be written as a convergent series
oo 1=0 /

f{9, ip) = Y, E
m=-l

C

l,mYl,m(0, ip).

(3.19)

For completeness we recall here the definition

(o<p)-twi r(2*+iw-m)!i x v„ sin ™ 0 (±y +m (cos2e _ lV Yl
*i,mW<P)2in

[

47r(Z + m ) !

J

e

sm

\

d 9

)

^cos V

^'

See e.g. N. I. Achieser and L. M. Glasman [3].

48 so that

CHAPTER 3. Mathematical Foundations of Quantum Mechanics

Y0fl = —=,
J Air

1

Yito = \ —cosO,
V 47T

/ 3

Yit±i =

I o

^\—e±tvswt

A Hilbert space contains a complete set of orthonormal basis vectors or a corresponding sequence precisely then if it is separable. (2) On the space L2(0,2n), i.e. on the space of square-integrable functions on the interval [0, 2TT], an orthonormal system, called the trigonometric system, is defined by the functions
1

einx,

±n = 0 , 1 , 2 , 3 , . . . .

(3.20)

/2TT

(3) On the space L2(a,b), where (a, b) is an arbitrary but finite interval, a complete but not orthonormal system is given by

In order to obtain the orthonormalized system, one employs the orthogonalization procedure of E. Schmidt.* The sequence of polynomials thus obtained consists of the Legendre polynomials which are defined on the interval — 1 < x < 1. These polynomials are defined as follows: Po(*) = l, *>„(*) = _ 1 dn —(x2-ir, n = l,2,.... (3.21)

These polynomials satisfy the following normalization conditions, i.e. are orthogonal but are not normalized to 1:
l

/_

Pm(x)Pn{x)dx
l

= 0, (m ^ n),

(3.22a)

and

J' [Pn{x)\2dx = 1 ^

l

.

(3.22b)

(4) By orthogonalization of the following functions
-x2/2
xe~x 2

/2

x2p-x

2

/2

one obtains the following functions defined on the space L2{—oo,oo): ^[x) = ( - l ) n e * 2 / 2 f ^ e - * 2 = Hn{x)e-x2'2 dxn (3.23a)

''This procedure is wellknown in analysis, and hence will not be elaborated on here.

3.3 Operators in Hilbert Space with
/*C5tJ

49

j —(

bn(x)<t>m(x) = 2nn\yft6nm,

(3.23b)

where Hn (x) is the Hermite polynomial of the n-th degree (which we do not go into in more detail at this stage).§

3.3

Operators in Hilbert Space

Having defined the admissible states of a physical system as the vectors which span a Hilbert space, the next step is to introduce quantities representing operations in this space. This is our objective in this section. We begin with a number of definitions. Definition: Let T>A, the domain of definition, be a (dense) subspace of the Hilbert space "K. Then one defines as a linear operator A on "K the mapping A : VA -> H with (a,/3 G C, ^ , ^ e ^ c M ) A(cnl)i + 0rfo) = a(A^i) + /3(At/>2). (3.24b) (3.24a)

Definition: One defines as norm (i.e. operator norm) of the operator A the quantity sup l i M . (3.25) ipevA\{o} WW Definition: An operator A is said to be bounded, if its norm is finite, i.e. \\A\\ < oo. Definition: Two operators A : T>A — "K and B : T>B —>"Kare said to be > equal if and only if Atp = Btp, for every ip G VA and Z>A = T>BExample: An example of a linear operator is given by the differential operator D:= —
d

||A||:=

_.:.LU ^ f./. _ T1 ^ rwith VD = iil>eL*,j-eL*>.

T2

(3.26)

Definition: We define the operations of addition and multiplication of operators by the relations (A + B)tp:=Aip
s

+ Bip, \/i/j<EVA+B = VAnVB,

(3.27a)

Hermite polynomials are dealt with in detail in Chapter 6.

50

CHAPTER 3. Mathematical Foundations of Quantum Mechanics (AB)ip := A(Bip),
V^GPB,

B^eVA.

(3.27b)

Definition: We define operators called commutators as follows: Let A : T>A —> "K, and B : T>B — "K be linear operators in the Hilbert space "K. Then we > define as commutator the expression [A, B] := AB - BA with V[A,B\ = D A n £>B -» ft. (3.28)

Definition: If .A : X^ — IK for i/> 6 T>A\{0}, then ^ is called eigenvector with • respect to the eigenvalue A € C if and only if AV> = \i/>. (3.29)

Very important for our purposes are the concepts of adjoint and selfadjoint operators. Definition: Let A : T>A — IK and ip € P A C IK. Then A^ is called adjoint » operator of A if for A^4> := <jj, <f> € T>Aj, the following relation holds: {A* 4,il>) = (<!>, Ax/,). (3.30)

Definition: The operator A : X>A — ft, is said to be symmetric if and only if >

0 M ^ ) = (MV), V ^ i e P

A

c%

(3.31)

Definition: The operator A, A : X^ — IK, is said to be hermitian or selfad> joint if and only if A = Af, One can verify that: (A*)* = A, (A + S j ^ A t + (AA)+ = A*A , (AB)t = S U * , (A" 1 )* = (A*)" 1 .
f

i.e. D A = D A t

and A ^ = A</>.

(3.32)

(3.33a) fit, (3.33b) (3.33c) (3.33d) (3.33e)

We are now in a position to construct relations between operators A and B in K, which correspond to the relations (2.5a) to (2.5e) of Poisson brackets, however, we omit their verification here: [A,B] = -[B,A], (3.34a)

3.3 Operators in Hilbert Space [A, ai-Bi + a2B2] = ax[A, B{\ + a2[A, B2], [A,B]* = -[A\B\ [A,BC] = [A,B}C + B[A,C], [A, [5, C]] + [B, [C, A}} + [C, [A, B}} = 0.

51 (3.34b) (3.34c) (3.34d) (3.34e)

The last relation is again called a Jacobi identity. Comparison of Eq. (3.34c) with Eq. (2.5c) requires here in the definition of the corresponding quantity the introduction of a factor "i" (see Eq. (2.47)). As important examples we consider the following operators. (1) q3 : Vqj - L 2 (R 3 ). We write sometimes the application of the operator qj to <f> G L2(IR3): (qj4>){x), and we define M)(x) :=arj-0(x). (3.35) We can read this equation as an eigenvalue equation: Operator qj applied to the vector <f> yields the eigenvalue Xj multiplied by <> in the present case on /, 3 R . Since Vqj = {<£ G L 2 (R 3 ) : qj* G L 2 (R 3 )}, we have
^ = ^ t -

Furthermore, for instance for ip, <p G L2(IR1), we have (ip,q<p) = / ip*(x)(q(f))(x)dx = / ip* (x)x<p(x)dx

=

(#,«•

(3.36)

Since for the adjoint operator A^ of A:

it follows (with Pg = V t from above) that qj — q3^. (2) P j : VPj -> L 2 (R 3 ) defined by (p 3 »(x) := -ih^-<t>{y). In this case we have VPj = <<t>€ L 2 (R 3 ) : c continuous, -~- G L 2 (R 3 ) I = / > Vpj]. (3.37)

52

CHAPTER 3. Mathematical Foundations of Quantum Mechanics

Here for ip,(f> e L2(Ul): (V,W>) = /V(aO(p0)dx
(3

=7)-ifi

f^*(x)-^(f>(x)dx

= -inr(x)<i>(x)\f00-J^r(x)<i>(x)dx
— — I [ ih—i/}(x) \ (f)(x)dx = J (pip)*(x)<f>(x)dx = (p^,0) = (pty,0), (3.38)

so that p* = p. Something similar applies in the case of the following operator which represents classically the kinetic energy T: (3) T:VT^ L 2 (R 3 ) and

™ ( x ) := "£ A<Kx) = ( ^ & 2 V ( x ) '
As a further example we consider the commutator. (4) Let the commutator be the mapping

(3-39)

Then for <p E L 2 (R 3 ): \Pj,Qk]<l>(x) = i.e. formally The following commutators which define the Heisenberg algebra \Pj, Qk] = -ihSjk, \pj ,Pk] = 0, [qj ,qk] = 0 = (pjqk ~ 9fePj)^(x) = (pjqk(t>)(x) - (qkPj(t>)(x) - i ^ f ^ X f c 0 ( x ) - xfe—</>(x) J = -ih6jk(j)(x.),

are called canonical quantization conditions with respect to a theory whose classical version possesses the fundamental Poisson brackets {Pi, ?*} = Sjk, {Pj,Pk} = 0, {qj,qk} = 0.

The simplest example to consider is the harmonic oscillator. We postpone this till later (Chapter 6). We add, that the quantization must always be

3.4 Linear Functionals and Distributions

53

carried out on Cartesian coordinates. Moreover, the above relations assume that the three degrees of freedom are independent, i.e. there are no constraints linking them. Systems with constraints can be handled, but require a separate treatment.^

3.4

Linear Functionals and Distributions

We now introduce the concept of a continuous linear functional on a so-called test function spaced Our aim is here, to provide a frame in which the formal Dirac bra- and ket-formalism to be developed later finds its mathematical justification. We require in particular the delta distribution and Fourier transformations. A subset of a Hilbert space "K is called a linear manifold D, if along with any two elements (f>i,4>2 G ^ C 'K this also contains the linear combination of these, i.e. « i ^ i + a2<^2 £ ^ , ai,a2€C. (3-41) A linear functional [/] in the Hilbert space "K is a mapping of the manifold T> into the set of complex numbers, i.e. [/] : V - C and is written [/]<<£> = / < 0 ) : = / with the property of linearity, i.e. f(<l>i + <h) = f(<l>i) + f{<h), (i.e. the expression (3.42) is antilinear in the first component). If / ( x ) is for instance a locally integrable function (i.e. for a compact set e R n ) like exp(ik • x), then it is clear that the function </>(x) has to decrease sufficiently fast at infinity, so that the expression in Eq. (3.42) exists. Functions which provide this are called test functions (see also later). Instead of the Hilbert space we therefore consider now a space of test functions (vector space of test functions), and on this space linear functionals. Definition: The compact support of a continuous function <fi : Rn — C is > defined to be the compact (i.e. closed and bounded) set of points outside
f

/(x)0(r)dx,

(V<P€V)

(3.42)

S e e Chapter 27. "We follow here to some extent W. Giittinger [127].

54

CHAPTER 3. Mathematical Foundations of Quantum Mechanics

that of (ft = 0. Test functions cf> with compact support are exactly zero outside their support; they define the space D(Rn). A different class of test functions <j> consists of those which together with all of their derivatives \Dn(f>\ fall off at infinity faster than any inverse power of |x|. These test functions are called "rapidly decreasing test functions" and constitute the space S(Rn): D(Rn) S(Rn) := {(f> G C°°(DRn -> C) : support of </> compact}, := {4> £ C°°(Rn -> C) : \x\m\Dn(j>\ bounded, m,n,... e I > 0}. N (3.43) Definition: Distributions f{<f>) are defined to be the linear functionals on D(Rn) and tempered distributions the linear functionals on S'(IRn). A subset of distributions can obviously be identified with ordinary functions, which is the reason why distributions are also called "generalized functions".

3.4.1

Interpretation of distributions in physics

It is possible to attribute a physical meaning to a functional of the form

f (</>):= J dxf(x)<j>(x).

(3.44)

In order to perform a measurement at some object, one observes and hence measures the reaction of this object to some tests. If we describe the object by its density distribution f(x), like e.g. mass, and that of its testing object by <t>(x), then the product f(x)(p(x) describes the result of the testing procedure at a point x, since f(x)4>(x) = 0, provided f(x) ^ 0 and <p(x) ^ 0, i.e. if object and testing object do not meet. The expression then describes the result of the testing procedure in the entire space. If we perform the testing procedure with different testing objects and hence with different test functions <pi(x),i = 1,2,..., we obtain as a result for the entire space a set of different numbers which correspond to the individual 4>i{x). These ^-dependent numbers are written as in Eq. (3.44):

m = j f{x)<t>{x)dx.
If f(<j>) = 0 for every continuously differentiable function <f>{x), then f{x) — 0. In general one expects that a knowledge of the numbers f{<j>) and the test

3.4 Linear Functionals and Distributions

55

functions <p(x) permits one to characterize the function f(x) itself, provided the set of test functions is complete. In this way one arrives at a new concept of functions: Instead of its values y = f{x) the function / is now determined by its action on all the test functions <f>(x). One refers to this as a functional: The functional / associates a number /(</>) with every test function (f>. Thus the functional is the mapping of a space of functions into the space of numbers. f{4>) is the value of the functional at the "point" <f>. With this concept of a functional we can define quantities which are not functions in the sense of classical analysis. As an example we consider in the following the so-called "delta distribution".

3.4.2

Properties of functionals and the delta distribution

The delta distribution is defined as the functional 5(<j)) which associates with every test function <fi(x) a number, in this case the value of the test function at x = 0, i.e. 6(4>) = </>(0), (3.45a) where according to Eq. (3.44) 6((j)) = f 8(x)(/)(x)dx. (3.45b)

The result of the action of the "delta function" 5(x) on the test function 4>(x) is the number 0(0). The notation J 5{x)(j>(x)dx is to be understood only symbolically. The example of the delta function shows that a function does not have to be given in order to allow the definition of a functional. In order to insure that in the transition from a function f(x) defined in the classical sense to its corresponding functional /(</>) no information about / is lost, i.e. to insure that f(4>) is equivalent to f(x), the class of test functions must be sufficiently large. Thus, if the integral J f(x)4>(x)dx is to exist also for a function f(x) which grows with x beyond all bounds, the test functions must decrease to zero sufficiently fast for large x, exactly how fast depending on the given physical situation. In any case the space of test functions must contain those functions <f>(x) which vanish outside a closed and bounded domain, since these correspond to the possibility to measure mass distributions which are necessarily restricted to a finite domain. Furthermore, for the integral (3.44) to exist, the test functions must also possess a sufficiently regular behaviour. For these reasons one demands continuous differentiability of any arbitrary order as in the case of 5(0?") above. Certain continuity properties of the function f(x) should also be reflected in the associated functional. The reaction of a mass distribution f(x) on a test

56

CHAPTER 3. Mathematical Foundations of Quantum Mechanics

object <p(x) is the weaker, the weaker cp(x) is. Thus it makes sense to demand that if a sequence {(pi(x)} of test functions and the sequences resulting from the latter's derivatives of arbitrary order, (p^(x), converge uniformly towards zero, that then also the sequence of numbers ((pi) converges towards zero. We refrain, however, from entering here into a deeper discussion of convergence properties in the space of test functions. The derivative of a distribution f((p), indicated by a prime, is defined by the following equation f((P):=-f(cP>). (3.46) This definition suggests itself for the following reason. If we associate with the function f(x) the distribution
oo

dxf(x)</>(x),
/
-oo

then the derivative f'(x) becomes the functional
oo

/

dxf(x)cP(x)
-oo

= -/($'),

(3.47)

as in Eq. (3.44). Partial integration of this integral then yields, in view of the conditions <p(±oo) = 0,
oo

dxf(x)<P'(x) = -/(</>'),
/
-oo

as in Eq. (3.47). Equation (3.46) defines the derivative of the functional f(<p) even if there is no function f(x) which defines the functional. For instance in the case of the delta distribution we have 6'(cP) = -S((P') = -<j>'(0) according to Eq. (3.45a). Formally one writes, of course,
oo

(3.48)

/•oo

/
-oo

dx5'(x)<p(x) =

[<J(a;)0(a;)]?foo - /
J—oo

dx5(x)(P'(x) (3.49) (3-50) (3.51)

= -5(<P>) = -<P'(Q). For an infinitely often differentiable function g(x) one has apparently (5 •/)<</>> = / ( # ) , so that (x6)((P) = S(x(P) = [x(P}x=0 = 0 x (P(0) = 0,

3.4 Linear Functional and Distributions or ' x6(x)</>(x)dx = 0,
/ '

57

x5{x) = 0.

(3.52)

Thus formally one can operate with the delta distribution or delta function in much the same way as with a function of classical analysis. As a further example we consider the relation f(x)S{x) = f(0)S(x). According to Eq. (3.50) we have f{x)5{x)4>{x)dx = «J(/0) = [f(x)4>(x)]x=0 = f(0)<f>(0) (3.53)

/

=

f(0)5(<f>) = J

f(0)6(x)<j>{x)dx,

as claimed in Eq. (3.53). Formal differentiation of the relation (3.53) yields f(x)6\x) = f(0)S'(x) - f'(x)S(x). (3.54)

One can convince oneself that this formal relation follows also from the defining equation (3.46). In particular for f(x) = x we obtain the useful relation x5'{x) = -S(x). (3.55)

A very important relation for applications is the Heaviside or step function 9{x) which is defined as follows:

From Eq. (3.56) we can deduce the relation 6'{x) = 6{x). (3.57)

For the verification we associate with the step function the following functional
oo

roo

/
-oo

0(x)<f>(x)dx = / <f>(x)dx.
JO
/•oo

For the derivative we have according to Eq. (3.46) e'(x)(<p) = -8(x){4f) = dxc/)'(x) = (f)(0) - 0(oo) = 0(0)

= 8{x)(<i>),

58

CHAPTER 3. Mathematical Foundations of Quantum Mechanics

or symbolically 0'(x) = 5(x), i.e. Eq. (3.57). After the introduction of the delta function it is customary to consider briefly Fourier transforms, i.e. in the case of one dimension the integral relations f{x) g(k) = = - =
V Alt
I
1

\ /

f°°

dkg(k)e-ikx dxf(x)eikx

=

F~lg{k), (3.58)

J-oo
roo

- =

= Ff(x).

\Z2TT J-OO

We assume here some familiarity with these integral relations. It is clear that the existence of these integrals assumes significant restrictions on the functions f(x),g(k). As a formal relation in the sense of the theory of distributions we deduce from Eq. (3.58) the important formal integral representation of the delta function, i.e. the relation
I
/•OO

S(x) = — /

dkeikx = S(-x).

(3.59)

One can see this as follows. According to Eq. (3.58) /(0) = - - L fdkg(k) yzir J and g(k) = - L y lis J fdxf(x)eikx.

Inserting the second relation into the first, we obtain

/(0) =

Jdxf(x)^jdke^,

and comparison with Eqs. (3.45a) and (3.45b) yields (3.59). We close this topic with a comment. The singular delta distribution was introduced by Dirac in 1930. The rigorous mathematical theory which justifies the formal use of Dirac's delta function was only later developed by mathematicians, in particular L. Schwartz. Thus today the singular delta distribution is written as a regular distribution, i.e. as an integral operator, by writing, for instance, 5a= f <5(x - a)^(x)dx = ^(a) V 0 e S ( [ R n ) , (3.60)

and one derives from this that the delta function <J(x) has the (impossible) properties of being (1) zero everywhere except at x = 0 where it increases so enormously that (2) the integral over <5(x) is unity:

f S(x)dx = 1.

Chapter 4

Dirac's Ket- and Bra-Formalism
4.1 Introductory Remarks

In this chapter we introduce the (initially position or momentum space representation-independent) notation of Dirac* which is of considerable practicability. We also introduce those properties of the notation which make the calculations with states and operators amenable to simple manipulations. The notation will be used extensively in later chapters. The integral representation of the delta function discussed in Chapter 3 can be considered as a formal orthogonality relation for harmonic waves exp(ikx) which finds its rigorous justification in the context of distribution theory. Thus we have — again for simplicity here for the one-dimensional case —
1
/"OO

S(x -x') or 6(k -k')

= — /
27T
-i

dkeikxe~ikx'

(4.1a)

J_00
/-oo

= — /
27T J_00

dxeikxe-ik'x.

(4.1b)

Since k and similarly x here assume a continuum of values, Eqs.(4.1a) and (4.1b) are described as normalization conditions of the continuum functions exp(ikx) as distinct from orthogonality conditions for functions depending on integers m,n as, for instance, the trigonometric functions um(x) = cos(mx), vm{x) = sin(mx),
'See P. A. M. Dirac [75].

59

60 for which

CHAPTER 4. Dirac's Ket- and Bra-Formalism

1 fn - / dxum(x)un(x)
^

= 6mn,

(4.2a)

J-K

i r
- /
^ J-7T

dxum(a;)i;ri(x) = Smn,

(4.2b) (4.2c)

i
- \
^

r
dxum(x)vn(x) = 0.
J-K

In the following we shall therefore "orthogonalize" continuum states represented by vectors of a Hilbert space which is no longer separable (i.e. which has at least a subset whose vectors are characterized by a continuous parameter) in the sense of the relation (4.1a) to a delta function instead to a Kronecker delta as in Eq. (4.2a). With this formal use of the delta function we can manipulate continuum states easily in much the same way as discrete states which implies an enormous simplification of numerous calculations, t The Fourier transforms introduced in Chapter 3 permit an additional important observation: We have several possibilities to represent vectors in Hilbert space, since the Fourier transform describes the transformation from one representation to another (" configuration" or "position space representation" <-> "momentum space representation"). A position space Schrodinger wave function ?/>(x), which we shall consider in detail in numerous examples later, corresponds to the representation of the vector ip of a vector space (as representative of a state of the system under consideration) in the position space representation, i.e. as a function of coordinates x. In the momentum space represenation the vector tp is the Fourier-transformed V>(k) of ip(x); this representation will be used in particular in Chapter 13.

4.2

Ket and Bra States

In the following we introduce the notation of Dirac.* We define ket-vectors as elements of a linear vector space and bra-vectors as those of an associated dual vector space. The syllables "bra" and "kef are those of the word "brackef. The spaces are linear vector spaces, but not necessarily separable Hilbert spaces, unless we are dealing with an entirely discrete system. More as an excercise than as a matter of necessity we recall in the following some considerations of Sec. 3.1, expressed, however, in the notation of Dirac. Hopefully this partial overlap is instructive.
'Formally this means that we have to go to an extended Hilbert space, which contains also states with infinite norm, see e.g. A. Messiah [195], Vol. I, Sec. 7.1.3. *P. A. M. Dirac [75].

4.2 Ket and Bra States

61

In order to achieve a representation-independent formulation we assign to every state of the system under consideration a vector, called ket-vector and designated |), in a vector space V. In the symbol \u), for example, u is an index of the spectrum, i.e. a discrete index in the case of the discrete spectrum, or a continuous index in the case of a continuous spectrum. The linearity of the vector space implies, that if £ is a continuous index, and A(£) an arbitrary complex function, then with |£),

\w) = J\(0\Odt

(4-3)

is also an element of V. As mentioned, the space can be finitely or infinitely dimensional. In the vector space V of ket-vectors a set of basis vectors can be defined, so that every vector can be expressed as a linear combination of these basis vectors. With the vector space V of ket-vectors we can associate a dual space V of bra-vectors, which are written (x|. The bra-vector (%| is defined by the linear function <X|{|«» of ket-vectors \u). For a certain ket-vector \u) the quantity x has the value (x|it), which is, in general, a complex number. For a better understanding of the concept of the dual space, we recall first the difference between a linear operator and a linear functional. According to definition, the linear operator which acts on a ket-vector \u) G V yields again a ket-vector \u') G V. On the other hand, a linear functional x i s a n operation, which assigns every ket-vector \u) G V linearly a complex number {x\u}, i.e. X= with x ( A i K ) + A 2 |« 2 )) = A i x O i ) ) + A 2 x ( M ) The functionals defined on the ket-vectors \u) G V define the vector space V called dual space of V. An element of this vector space, i.e. a functional x is symbolized by the bra-vector (x|.§ Then (x\u) is the number that results by allowing the linear functional (x| G V to act on the ket-vector \u), i.e. X(\u}) = <xl«>One should compare this with our earlier definition of the functional / on the space of test functions <> We wrote this functional /. |«)eV->x(l«»

m = w».
By construction every ket-vector is assigned an appropriate bra-vector. The reverse is not true in general. See e.g. C. Cohen-Tannoudji, B. Diu and F. Laloa [53], p. 112.

if (u\v) — 0. if x\u) = 0 V kets \u) E V.and Bra-Formalism Furthermore we have: ( x | = 0. (4.4c) (4-4d) (v\ = J\*(0(m- One defines as the scalar product or inner product of \u) E V and \v) E V the c-number (v\u). This expression is linear with respect to \u) and antilinear with respect to \v). the norm squared. It follows that we can now write the Schwarzian inequality (3.5) 4.e.(t. We consider next the important case that a unique antilinear relation called conjugation exists between the kets \u) E V and the bras (x| E V. Because of the anti-linearity the conjugate bra-vector associated with the ket-vector |u) = Ai|l> + A2|2> is (v\ = Xl(l\ + X*2(2\.3 Linear Operators.5c)). which we write (u\. As in our earlier considerations we demand for (v\u) the following properties: (1) (u\v) = (v\u)*. The dimension of V is the same as the dimension of V. Also: (xi| = (X2I) if (xi\u) = (X2\u) for all \u) E V.4b) (4. if \u) = 0 (see (3.\u) = (u\0*.4a) \v) = J\(m<% and (4. (2) (u\u) > 0. (3) (u\u) = 0. Two ket-vectors \u). V \u) (in the case of |£) with infinite norm!). i.9a): \{u\v)\2 < (u\u)(v\v). \v) E V are said to be orthogonal. or (4. that the vector \u) E V is associated with a vector in V. Dirac's Ket. Hermitian Operators A linear operator A in the space of ket-vectors V acts such that A\u) = \v) EV .62 CHAPTER 4.

e. Furthermore. Two operators A and B are said to be inverse to each other.4. The product vectors l?^ 1 )^ 2 )) span a new vector space. the space Vi <S> V2 has the dimension . as we can see as follows. The operator A. Hermitian Operators 63 with A = 0 if A\u) — 0 for every vector \u) £ V. The inverse of A is written A . It follows that (r. Then the following rules apply: A + B = B + A. In this way the operations of operators A. -1 and \u) = B\v). Then one says: (771 results from the action of A on (x|.or direct product of vector spaces.\u) = ((X\A)\u) = (X\(A\u)) = (x\A\u). Let \u)(l> be vectors of the space Vi and similarly |?j)(2) the vectors of the space V2. B.. if AB = 1. we have \u^u^) = X1\v^u^) + \2\w^u^) and so on. This does not always exist. if \v) — A\u) i.e.. defines the unit or identity operator. also acts on the dual space V. the space Vi <S> V2. This product is commutative. on bra-vectors are similar to those on ket-vectors. l|tt) = \u).. Then (x|(^4|it)) is a linear functional of \u) £ V (since A is linear).e. Also A = B it for every vector \u) £ V : (u\A\u) = (u\B\u). let (771 6 V be the bra-vector defined by this functional. and one writes: (V\ = (X\A. i. If Vi has the dimension A/i. BA = 1. The inverse of a product is (AB)'1 =B-1A~1.or Kronecker. Multiplication by unity. Let the following bra-vector be given: (x\ £ V. Furthermore linearity applies. denned as a linear operator on V. AB^BA. The product of such vectors is written: \u^u^) = \u)^\u)^. Finally we define the tensor. with |n)(1) = Ai|w)( 1 )+A 2 k) ( 1 ) . i.3 Linear Operators.

we write \v) = A*\u). we obtain the conjugate relation <t|At|«> = (u\A\t)* (4.7b) (4. i.e. also Sec. Every operator A^> commutes with every operator A(2\ i. by replacing in the above the bra-vector (t| by <t|flt) (AB)* = B*A\ (cA)* = c*A\ (A + 5) t = A* + B\ (4. and (v\ = (u\A. [A^. here designated with the same symbol.A^] = 0. then \v) is a linear function of \u). Every such operator is then also associated with an operator. Examples of operator relations: (1) (AB\u)(v\Cy = C^\v)(u\B^Al (4.e. If (v\ = (u\A.e.3).6) for all \u). AW\uW)\uW) = \vWUW). in the product space. Dirac's Ket. i.7a) (4.and Bra-Formalism an operator in the Mi A/2. Let A^> be an operator in the space Vi and A^ space V2.e.7C) (|«)(u|)t = \v){u\. (v\t) = (u\A\t). since AWAW\UW)\UW) = \vWuW) = AWAW\UW)\UW). As a consequence we arrive at the following properties (e. AW\U)W = \V)W. Assuming now that \v) G V and (u\A G V are conjugate vectors as explained above. Next we construct the following scalar products: (t\v) = (t\Ai\u). 3.g.7d) .64 CHAPTER 4. Since we demanded the scalar products to have the property (t\v) = (v\ty. the operator A^ is called hermitian conjugate or adjoint operator of A (cf. i. \t) G V.

-H^K] The separation of HK into hermitian and anti-hermitian parts is therefore HK=±{HK + KH)+l-[H. i. so that (n|^4|n) is real.e. if H — H' and anti-hermitian.3 Linear Operators. and (b) that the eigenvalues with respect to \u) are equal to those with respect to (u\ and vice versa. An important theorem is the following. if H is hermitian and (u\H\u) > 0 for all ket-vectors \u). Obviously the quantity \a)(a\ is an hermitian operator. but only if these commutators commute. for we had (\u){v\)i = \v){u\. 65 The linear operator H is called hermitian. An hermitian operator A has the properties (a) that all its eigenvalues are real. Verification: Since A — A^ and A\u) = a\u).4. [H. It follows that a is real.e. The commutator of two hermitian operators is anti-hermitian: [H. i. it follows that (u\A\u) = a(u\u). Analogous considerations apply in the case of bra-vectors (u|.K]^ = = [HK-KH]* KH-HK = = K^H^ -[H. Hermitian Operators (2) The conjugate bra-vector of AB\u)(v\C\w) is {w\C^\v){u\B]A]. The operator H is said to be positive definite. hermitian The product of two hermitian operators is not necessarily hermitian.K\. . as is also (u\u). Every operator A can be written as the sum of two such parts: A=\(A + A*) + \(A-tf) anti—hermitian . K] = 0. so that (|a)(a|)t = |a)(a|. or (u\A\u)* — {u\A^\u) = (u\A\u).K}. if H = H' and K = K\ then (HK)* = HK only if tfrf = KH.

(u\u') = 5{v . (v\A = b(v\. Dirac's Ket. This follows from observing that (u\Ps\v) = (u\vs) = (us\v3) = (us\v).e. A continuous spectrum can immediately be included by passing to the extended Hilbert space. i. Next we introduce the very useful concept of projection operators. but the continuum vectors are normalized to a delta function. then the hermitian operator defined on this space is an observable. {us\us*) = 0. i.e. a ^ b. we have 0 = (u|A|rt) — (u|. (4.and Bra-Formalism Moreover. Ps\ua. since a ^ b. Orthogonality: Two eigenvectors of some hermitian operator A belonging to different eigenvalues are orthogonal.) = 0.8) The projection operator has the following important properties: Ps is hermitian.A|u) = a(v\u) — b(v\u) — (a — b)(v\u). it follows from A\u) — a\u) that (u\A = a(u\ or the other way round. The entire set of ket-vectors spans the extended Hilbert space. (u\Ps = {us\. i. The above theorem remains unchanged.9a) . and let S* be its complement. so that (u\Ps = (us\. The operator Ps which projects onto S is defined by the properties: Ps\u) = \us) = Ps\us). Let S be a subspace of the Hilbert space !K. Ps2 = Ps(4.u'). i. Ps is idempotent. ( n | z / ) = 0 .9b) (4. so that (v\u) = 0. (n\n')=5nnl. If these form a complete system. since a is real. which includes vectors of infinite norm. Then every vector \u) £ "K can be written \u) — |its) + \us*) with \us) e S. \u8*) E S*. Assuming A\u)=a\u). as we can see as follows.e.e.66 CHAPTER 4.

i. so that \ua) = (a\u)\a) = \a)(a\u). = (p2-p)\p). i. .. Let p be an eigenvalue of the projection operator P. Ps2 = PsThe only eigenvalues of Ps are : 0 and 1.e. \ua) = c\a).9d) Thus the vectors P\u). (4. i.4. |iV) is a set of orthonormalized states. Hermitian Operators This property follows from the observation that Ps2\u) = Ps\us) = \us) = Ps\u). \u) arbitrary. (1 — Ps) is projector onto S*. Ps is projector onto S.1.3 Linear Operators. (1 — P)\u) are orthogonal. 67 (4. |2).e. Set \u) = \ua) + \ua*) with (a\ua*) = 0. Then 0 = (P2-P)\p) and hence p = 0.e. Example: The vector \a) normalized to 1 with (a\a) — 1 spans a 1-dimensional subspace. If for these (by construction) the projector P2 onto a subspace S2 is P2= J*I f2\odm. c = (a\u).p = 0. if the set of vectors |1). .. i. as can be seen as follows. where £ is a continuous index. p 2 .e. P\p) =p\p). Then (a\u) = (a|u a ) + (a\ua*) = (a\ua) and so {a\u) — {a\ua) and hence {a\u} = c(a\a) = c. The projection \ua) of an arbitrary vector \u) onto this subspace is |ita) = |a)(a|w). Let a continuum state be written |£).9c) This property can be seen as follows. The quantity \a)(a\ is called elementary projector. Very similarly we can construct operators which project onto the subspace spanned by the continuum states. Obviously N PN = y]ln)(re| n=l is the projection operator onto the A^-dimensional subspace SN.. so that (Ps2 ~ Ps)\u) = 0.

4 Observables Operators which play a particular role in quantum mechanics are those called observables. in this case the spectrum also has a continuous part. see Eq. which we introduced earlier. In the case of the infinite-dimensional Hilbert space with a countable number of orthonormalized basis vectors.. and Eq. which are orthogonal to each other (i. ' The projector onto the subspace with eigenvalue Aj can then be written Pi = ^2\ui. Let us assume that A is an hermitian operator with a completely discrete spectrum (as for instance in the case of the harmonic oscillator). be a system of basis vectors in this space. are linearly independent).6. The eigenvalues Aj then form a discrete sequence with associated eigenvectors \ui) € !K.114b).1 and 11. P = £|n)(n|.e. 4..and Bra-Formalism and hence '6 Numerous properties of the projection operators Pi are self-evident. However. Degeneracy will be discussed at various points in this text. one says the degree of degeneracy is r . Examples 8. (11. . Observables are representatives of measurable quantities. See e.r)(ui. n=l In the case of continuum states (which are no longer countable) one has correspondingly as projector of all states in a domain £ e [£2>£i] or in the case of the differential domain d£ of £: dp = j '£ de'ion This latter operator is called differential projection operator. that one and the same eigenvalue Aj is associated with several eigenfunctions. 8. i.e. Dime's Ket. the operator P is correspondingly P2\u)= f2\0m\u).6. Sec. (11. If there are r such vectors..r\ "An example is provided by the case of the hydrogen atom.68 CHAPTER 4. Let \u\). Coulomb potential.g. In general it is possible.114c). |u2).

i.e. 69 i If A is an observable and if the spectrum is purely discrete.4 Observables The dimension of this subspace is that of the degree of degeneracy.r\u)\2.r (4.r \(ui.r) (the convergence.10). It follows that the norm squared is given by (u\u) = {u\PA\u) = ^2(u\ui. PA = y£pi i = Yt\ui.r\ui/. (4. i.e. If Aj ^ Aj'.r)(ui. and the eigenvectors \ui. the projection of this operator is the entire space.r\u).11) i.r\ = l.12) .r\u) i.r =^ i.\i)Pi = 0. Let us set {A .r \ui. ^ A i P i = A = ^Ai|ui. i. the operator A is completely determined by specification of the eigenvalues A.10) gives the linear combination |it) = ^ i. is always implied). and APi = XiPi. The operators Pi are linearly independent.10) expresses the completeness of the orthonormal system. in the case of degeneracy with (ui.r)(ui. The uniqueness of the expression (4. Applied to an arbitrary vector \u) G "K Eq.r'} = Sii>6rr>. we obtain i i i i i. which we do not enter into here. (4.r|. then PiPi' = 0.4. the relation (4. i i (4.r)(ui.e.10) This expression is known as completeness relation or closure relation.e.10) follows from the fact that i Applying the operator A to the projector (4.r>(ui. or also as subdivision of unity or of the unit operator. Together with the orthogonality condition.

as explained earlier. the eigenvector \ui): f(A)\ui) = f{\i)\ui).e. now however. i.17) previously. Then we denote by \uv) the eigenvector of A whose eigenvalue is \{v). i.and Bra-Formalism This is the expression called Parseval equation which we encountered with Eq. In a similar way we can handle functions f(A) of an observable A. (4.v').14) . for instance. (note: {uv\A\uvi) is the matrix representation of A) A\uv) = \(v)\uv). (4. (3. 1/2)) PA = 22 \ui)(u*\ + / dv\uv){uv\ = 1. (u„|zv) = 8{v . for instance exponential functions. written in Dirac's notation. Then f(A) = f(A)PA =^/(AOk)^! + / f(\{v))\uv)(uv\dv. Let v be the continuous parameter which characterizes the continuum. If the spectrum of an observable A consists of discrete as well as continuous parts. we have the corresponding generalizations.e. One defines as action of f(A) on. Dime's Ket. > rV2 + / dis\(uu\u)f \uj)(ui\Xi + / \(v)\uv){uv\dv. The ket-vectors \uv) are orthonormalized to a delta function. Then the operator PA which expresses the completeness of the entire system of eigenvectors is (all continuum states assumed to be in the interval (yi.13) For an arbitrary vector \u) of the appropriately extended Hilbert space we then have v—<• fV2 \u) = PA\U) = 22 \ui)(ui\u) + / and hence i) = (u\PA\u) = J2 \(ui\u)\2 and A = APA = ) J dv\uv){uv\u).70 CHAPTER 4.

In the one-dimensional case we write {x\x') = S(x-x').4. if (a) they all commute pairwise. (4.and bra-vectors.58) in terms of ket.. Extending this we can say: A sequence of observables A.e. [A. (4. |x) € Fx. Finaly we recapitulate the following theorem from linear algebra: Two observables A and B commute. here presented without proof. 4. We shall return to this theorem. which is discrete in the case of discrete energies and continuous in the case of scattering states (with continuous energies). For many practical purposes the use of the Fourier transform is unavoidable.16) represent subdi- . with energy E. Therefore we want to re-express the Fourier transform (3. First of all we can rewrite the integral representation of the delta function. In the following we consider more generally ket-vectors \ip) as representatives of the physical states.B. if and only if they possess at least one common system of basis vectors. as a formal orthonormality condition. for which the completeness relation is f dk\k)(k\=l.16) Correspondingly we also have a complete set of basis vectors {\k)} of an associated vector space F^. In the differential operator representation the problem to establish such a relation is known as the Sturm-Liouville problem.59). form a complete set of commuting observables. i. i. Since both expressions (4.. and (b) if they possess a uniquely determined system of basis vectors. again later. where the symbol ip is already indicative of the Schrodinger wave function ij. which all commute with one another.15) where the vectors {\x)} are to be a complete set of basis vectors of a linear vector space in its position space representation Fx.15) and (4.5 Representation Spaces and Basis Vectors 71 This relation expresses an arbitrary function f(A) of an observable A in terms of the eigenfunctions of this operator.17) The Fourier transform provides the transition from one representation and basis to the other. (3. \k)€Fk. / dx\x){x\ = 1... Eq. their commutator vanishes.C. B].e.5 Representation Spaces and Basis Vectors We began by considering ket-vectors \u) in which u is a parameter of the (energy) spectrum. (4.

Rather \x) and \k) serve as basis vectors in the representation spaces Fx. The representation of a state vector \ip) G 'K in position space Fx is the mapping of the vector \ip) into the complex numbers (x\ip).15): 5(x . The vectors \x) and \k) G F are not to be confused with the vectors |u) or \ijj) G !K. i.L e " * * ' = {x'\k)\ V27T (4. i. . Obviously we obtain this by inserting a complete system of basis vectors of Fk. (4. (4.and Bra-Formalism visions of the unit operator.-space representation. called wave function. The Fourier representation ij){x) = -)= V 27T J Ieikx^{k)dk (4.1a).20) The representation of the corresponding bra-vector (^1 G IK in the position space Fx is correspondingly written (il>\x) = {x\ipy. we can rewrite Eq.18) According to Eq. this expression has to be identified with — / dkeikxe-ikx\ i. (x\k) = -^=eikx. (3. shows that these expressions are the corresponding continuum functions (the continuous parameter k replaces the discrete index n). V 27T (k\x') = . ${k) := (k\if>).Fk.e. Dime's Ket.22) All of these expressions can readily be transcribed into cases with a higher dimensional position space.59) or Eq. which are representatives of the states of our physical system. (4. <a#) = [ dk{x\k)(k\ip).72 CHAPTER 4.21) provides the ket-vector \ij)) in the A.19) Comparison with the orthonormalized system of trigonometric functions (4. (4.2a) etc.e. il>{x) := (x\i/}) : "K -* C.x') = (x\x') = (x\t\x') = (x\ f dk\k)(k\x'). (4.e.

Schrodinger [244]. We mentioned earlier (in Sec. if we do not take into account the complementary part of the universe.p. Finally we introduce briefly the canonical distribution of statistical mechanics and show that with this the density matrix can be calculated like the Green's function of a Schrodinger equation (inverse temperature replacing time). This will then also clarify the role of p as an operator in the quantum mechanical analogy with the Liouville equation.1 Introductory Remarks In this chapter we remind ourselves of the measuring process briefly alluded to in Chapter 1 and of the necessity to recognize the abstraction of a physical situation that we perform.t).e. i. 73 .Chapter 5 Schrodinger Equation and Liouville Equation 5. the calculation of which for specific cases is the subject of Chapter 7. We postulate the (time-dependent) Schrodinger equation* (with the Hamiltonian as the time-development operator) and obtain the quantum mechanical analogue of the Liouville equation. with the complementary system or *E. Thus we distinguish between so-called pure states and mixed states of a system and thereby introduce the concept of density matrix and that of the statistical operator.5) that in general and in reality we ought to consider the system under consideration together or in interaction with the rest of the universe. 1. which is the quantum mechanical analogue of the classical probability density p(q. 5.2 The Density Matrix We shall establish a matrix p.

For an exact treatment we would have to express the actual state \ip) of the system as a supersposition not only of the states \i). where i represents collectively all quantum numbers of the state of the system under consideration. which in Dirac's notation we can write as = J]C a »|i) a. The states \a) £"K'on the other hand are the corresponding states of the complementary part of the universe. The states \i) £ "K. are called "pure states".b. of an operator that acts only on the states of the system we are interested in. The difficulty we have to face is the impossibility to specify precisely the stationary state of our limited system.e. i. is then given by (A) := {MA\^) = Yl U\{b\A\a)\i)CaiCbj* a. T .j ij (5-4) Plj = (i\p\j)^YC-iC-J*a (5-5) Here p is an hermitian matrix. \a). as well as 2~] |«)(*| = 1 in the subspaceof pure states i (i\j) = 5ij. \a). p" ( C C * ) t = CTC* = p.i.^ Since p is hermitian..b. where T means transpose: p = CTC*.i.. (a\i)=0 = {j\b)... (5-2) (5.e. we have (a\b) = 5ab. These are the quantum mechanical states which are also referred to as micro-states.j = where J2 (J\A\i)5baCatCbj* = X>'l4*>/0« a. a (5. the matrix can be diagonalized by a transition to a new set of states ^The hermiticity can be demonstrated as follows. that is. i eH ®^' (5J) We assume that the states are orthonormal and in their respective subspaces also complete. \i)... called the density matrix.3b) The quantum mechanical expectation value of an observable A.3a) and N J | a ) ( o | = l in the complementary subspace. we have to deal with a so-called "mixed state". The actual and real state of a system is then a superposition of two types of states. but also of the states of the complementary set of the universe.74 CHAPTER 5.. i. Schrodinger Equation and Liouville Equation the measuring instruments.

3a). seen macroscopically of Z copies of the system). can be achieved with the help of the completeness relation of the vectors \i): \i') = J2\i)(i\i') i.j(x'|i)(i|a. This transition to a new set of basis vectors in which the matrix becomes diagonal.) = YJc t . as we discussed previously. If the ensemble corresponding to the system under consideration consists of Z elements (i.g. (5.10) and is therefore.. where x represents collectively the entire set of position coordinates of the (particle) system. Thus the pure states \i) form in the subspace of the space of states which is of interest to us. or (5.ji(x / )i*(x). but occupy in general different microscopic..2) and (5. P ^ ) = |i)(#) = (#)|i) (5.5.6) (i'\p\i") = Wi'Si'vi = real. i. Gibbs introduced the concept of ensemble.3 with Pi^YtCaiCaj*. Thus the real system is in the pure state \i) with probability uii/ J2i Ui.9) is diagonal. then (cui/ '^Zi^i)Z is the number of ensemble elements in the pure state |i).\p\i")(i"\ = Y. a projection operator. (see below) eM. This ensemble is interpreted as a large set of identical systems.^'\i'^i'\> (5-8) since then (f\p\j") = ^Ui'ij'li^ii'lj") v = ^Ui'Si'fSi'j" v = UjiSj/j.e. which today is a fundamental term in statistical mechanics. a complete orthonormal system with properties (5.7) p = ^2\ii)(i. In the following we write simply \i) instead of \i'). This projection operator represents a quantum mechanical probability as compared with the numbers ui{ which represent classical probabilities. .2 The Density Matrix 75 \i') 6 "K. The operator Pi = \i)(i\ projects a state \<p) G "K onto \i). a (5-12) *In his reformulation of the statistical mechanics developed by Boltzmann. as we saw earlier.. In this sense we have p(x'. The ensemble can be represented as a set of points in phase space.e.e. quantum states.e. e.4) we had the expression (A) = Y/Pij(j\A\i) i. i.* We can specify the state of our system. in the position state representation or xrepresentation.x) := (x'\p\x) = yju. i i (5-11) In Eq. in a specific representation. (5. (5. which are all subjected to the same macroscopic boundary and subsidiary conditions.

On the other hand. We can also show that Ui > 0. (5.4)) (5. (5. (5.) := (mm = (#') W> = l(#')|2 > 0. Then the expectation value of the observable A is (cf. In particular for A = 1. 3 (5. (5.14) which we can write (A) = Tr(pA). Schrodinger Equation and Liouville Equation We now define the operator p by the relation (cf. (A. Thus Tr(j>i|i)<i|)=l.76 CHAPTER 5.18) Substituting this into Eq. we have Tr(p) = l. Eq.13) {A) = ^(j\A\i)(i\p\j) ij = 52{j\Ap\j). (5.3 = pvv = uv. To this end we set A —>• Ai' := \i')(i'\ (no summation over i').15) 5^0'l5^w«l*X*b') j i = 1 » or J^wi(i|t> = ^ W i = l. We recall that the elements of p originate from the coefficients Cai in w = Y^cai\a)\i) = Yl ( E ^ M V ^5-19) .17) Hence the sum of the classical probabilities is 1.16) (5. Hence Wj/ > 0. (5. Eq. according to Eq. i t (5. as claimed by Eq.4). (5.5) above) (i\p\j)=Pij.14) we obtain (Ai') = ^2(j\i')(i'\i)(i\p\j) ij = ^Sji'Su'Pij i.18).

but we see here.3 The Probability Density 77 We see therefore that p defines the mixed states. the system is in a pure state. so that p{x.e.3 The Probability Density p(x. i. .t) and instead of (x\Pi\x) we now write p(x.22) In the following (x\i) is always to be understood as (x\i(t)}. (5. not in a mixture of states.t)\2.5. The effect of the interaction of the system under consideration with the surroundings. since they do not take into account the (interaction with the) rest of the universe.e. We observed above. Equation (5. that — different from u>i — P.t) = \^{x. that these are actually not sufficiently general.x'). Then {x\Pi\x) = (x\i)(i\x) = \{x\i)\2.§ {\x)} is a complete system of basis vectors in the position representation space Fx with dx\x){x\=l. which is contained in the coefficients Cai. (5. in which the numbers Wj represent classical probabilities and the operator \i)(i\ quantum mechanical probabilities or weights. In the following we are mostly concerned with considerations of systems in pure states. one says.t). represents a quantum mechanical probability. and in the other states with probability zero. defines the statistical operator. i. i.t) we now write tp(x. We now consider the latter expression.e.20) expresses that the system can be found with probability 1 in the state \i). ^2ui\i){i\. by representing the vector \i) by the wave function i(x. t) p= i The expression (5. the only remaining one is 1. in all other cases the state of the system is a mixed state. (5. (Question: Is the state of the universe a pure state?) 5.t) = {x\i).21) p=\i){i\=Pi. For the case n we have TJ=J- (5-20) (5. If all ui but one are nought.24) s (x\x') = 5{x . Thus we have a system in a pure state.8).23) Instead of i(x. thus enters these states. We see this more clearly by going to the position space representation.

8) and multiplying by ih. and hence the integral over all space dxp(x. t h e equation of motion with states represented in a system for which the q's or here x's are diagonal). and the time-independent equation should correspondingly be called Schrodinger wave equation (i.^ The equation which is the conjugate of Eq. with respect to time. Equation (5.3. 2. it is not "derived"). which is chosen in such a way that the desired analogy is obtained.e. i.e. To this end we have to differentiate the density matrix p. (5.27) is known as the Schrodinger equation.27) is -iHJ\ jt= m . This differentiation requires the time derivative of a state vector \i) E "K.78 CHAPTER 5. (5. Hence we postulate: The time development of a state \j) E "K is given by the equation ih^\j)=H\j). the Schrodinger equation is always postulated in some way. We therefore postulate first an equation for the time dependence of a state vector. (5. We now want to obtain the quantum mechanical analogue.8). or the wave function tp(x. can be found with probability 1 in the space R1.t) = / dx{i\x)(x\i) = (i\i) = 1. (5. Schrodinger Equation and Liouville Equation The relation (5. (5. .27) where H is the Hamilton operator of the system. t). the expression (5.t)\2.26) 5. The generalization to a three-dimensional position space is self-evident: JK 1 / p(x.25) J Thus the particle whose state is described by \i).4 Schrodinger E q u a t i o n a n d Liouville E q u a t i o n We encountered the classical form of the Liouville equation in Sec.28) Differentiating Eq. we obtain ih % = ^ E ^ w o i = ^E^ij)o-i+^Ewi(^iJ))oi dt +ihYJ^\J){jt{J\)- (5-29) "More precisely this equation should be called Schrodinger equation of motion (since H is fixed. (5.22) can also be written |(x|i)| 2 = (i\x)(x\i). but the linear operator states are moving — even if the system is observably stationary).t) = \rp(x. We shall convince ourselves later that the postulated equation is sensible (in fact.

p also contains "quantum mechanical probabilities" (i. (5.27).28) we obtain (see also Eq. the relation ih^ = [H. (5. (2.31) below) 3 j 3 = y £u>J{HPj-pjH)+thyE^m\ dt 3 = [H. We thus obtain the operator form of the Liouville equation. not the partial derivative. (5.27) in the position space representation: ih(x\-\j) = (x\H\j) = {x\H(pi.31) The eigenvalues u>i of p determine the fraction of the total number of systems of the ensemble describing the actual system which occupy the state \i).28) the states are considered as time-dependent.Xi)\j).5. With Eqs.34) .24)). Eq. i.4 Schrodinger Equation and Liouville Equation 79 Inserting here Eqs. dp/dt ^ 0. The reason for this is that in addition to the "classical probabilities". We can now consider the Schrodinger equation (5.e. both uji and |i)(i|). such a formulation is described as Schrodinger picture.27) and (5. (5.32) Comparing Eq.e.e. (5.32) with its classical counterpart (2. (5. M\/n\ (5-30) where in statistical equilibrium dt £^">01=03 (5.p} + ihJ2^\J)(J\.p}. (3. The correspondence to the classical case is obtained with the substitution h J-< > <5'33> and dp/dt = 0 (cf. we observe that here on the left hand side we have the total time derivative. Eq.21). (cf. The Schrodinger picture and the alternative Heisenberg picture will later be considered separately. i.37)) (5.

since the solution of Eq.x 1 ^ ( x . P. with respect to time t.36) however.0): where H(-ih—.35) Here we can look at the Hamilton operator as a "time generator3''. In expression (5. since the time-dependence cancels out (the exponential functions involve the same operator H but with signs reversed. 0) can be expanded in terms of a complete set of eigenvectors \En) of the Hamilton operator H. t ) .4. in Sec. we can replace these states by corresponding timeindependent states. this is an application of the Baker-Campbell-Hausdorff formula which we deal with in "Later.t) = Hld . Schrodinger Equation and Liouville Equation or with (x\j) — ip(x. however.4. which we shall need later.. to) = exp[—iH(t — to)/H\. it is plausible to refer to this equation at this stage. „ / ith—. T meaning temperature. i. we shall describe as "time development ator given by U(t. (5.t) = exp ip(x. With the help of Eq.36) in in The initial wave function or time-independent wave function ^(x.Xi){x\En) = En(x\En). (5. we write V>(x. but with respect to the parameter /3 = 1/kT. namely that the density matrix satisfies an equation analogous to the time-dependent Schrodinger equation — not.80 CHAPTER 5.e. n (5-37) This equation is described as the time-independent Schrodinger equation.38) (x\j)t=o = Yt(x\E^(En\J)t=o. For a mixture of states \i) of the system under consideration caused by the rest of the universe we have Ui ^ 1.and temperature-dependent Green's functions.1 Evaluation of t h e d e n s i t y m a t r i x As a side-remark with regard to statistical mechanics we can make an interesting observation at this point. .35) can be written'! ip(x. 0) or \il))t = \il>)t=o exp (5. a (5..8) for p. Feynman [94].. 5. 10.t): > ih—</>(x. the state \i) is still time-dependent. (5.** In view of the close connection between time. appearing in the Boltzmann distribution.e. operator" the exponentiated oper- . i. in p = ^2i^i\i}(i\. **R. or H\En) = En\En).

9) with En -> nhu) L0icxe-PE\ so that Yliui becomes = f3 = l/kT. (5.x')^p(x. (5.44) r (Tre-PH)' (5. or ^ = ^ e -0Ei _m. (5.1). Eq. i (5. Hence we have (with (Ei\Ej) = 5^) in what may be called the energy representation p = YJ"i\Ei)(Ei\.45) since on the one hand •PEi i i i and on the other hand .e. (543) Since H(f>i(x) = Ei<f>i(x). (1. we can rewrite p.x'-(3) = Si e ~^W*V).4 Schrodinger Equation and Liouville Equation 81 Example 5.41 ) or with 4>i(x) := (x\Ei) this is p(x. i ( 5 .40).39) Without proof we recall here that in the so-called canonical distribution the weight factors uji (similar to those of the Boltzmann distribution) are such that (cf. the expression _Zie-^\Ei)(E>\ also as (see below) e-0H (5.5.39) therefore (x\p\x') = J2"i(x\^)(Ei\x').40) 1. (5.^ n tne position space representation Eq.42) Inserting here Eq. x') = ^ i LUiMxWix'). i. we obtain p(x.

Equation (5. We now rewrite the factor exp(—(3H) in Eq.x'. (5. (5. M.-L [B. (5.35).1: Baker—Campbell—Hausdorff formula Verify that if A and B are operators. [B.48) = 1. (5. [A. (7.tt tt For a more extensive discussion see R.(3).4> = ^ g ^ .46) is pN(x. (3). (5. (5. B] + -L [A. whose calculation for specific cases is a topic of Chapter 7 (see Eq. A}} + • Solution: The relation can be verified by expansion of the left hand side.X'. PN(0) := (EilpNiP^Ej) = (Ei\e-PH\Ej) =e"^^ (5. Schrodinger Equation and Liouville Equation so that p is given by Eq.45) and (5. Wilcox [284].51) where the subscript x indicates that H acts on x of PN(X./?). (5.45).55)). Tr(pN) = / dxpN(x. With Eqs.The quantity pjv is obtained from the solution of the Schrodinger-like equation (5. we obtain = -HxPN{x.45) without normalization as PN(J3) := e~pH. we obtain 9pN gpP) = SijEie-^ = -EiPNij(J3).82 CHAPTER 5.49) In the position or configuration space representation Eq. (5. B}} . (5.46) In the energy representation this expression is pNij(P) with PNij(P) = 6ij.51) is seen to be very similar to the Schrodinger equation (5.50) Differentiating this equation with respect to j3. (5.x'. This solution is the Green's function or kernel K(x. Example 5.46) we can write the expectation value of an observable <^Tr<„.47) with respect to (3.51).47) (5. .x./3). Differentiating Eq. the following relation holds: exp(A) exp(S) = exp ( A + B + i [A.(3) := (x\pN(P)\x') = (x\e~0H\x').x'.52) Hence this expectation value can now be evaluated with a knowledge of pN.

(8. The importance of the harmonic oscillator can also be seen from comments in the literature such as:* ". * Comment of Y. However." Quite apart from this basic significance of the harmonic oscillator.Chapter 6 Q u a n t u m Mechanics of t h e Harmonic Oscillator 6. An alternative method of quantization of the harmonic oscillator — by consideration of the Schrodinger equation as the equation of Weber or parabolic cylinder functions — is discussed in Chapter 8 (see Eqs. its associated space of state vectors also provides the best illustration of a properly separable Hilbert space. like the Coulomb potential. Kim and M.51) to (8.53)). S..1 Introductory Remarks In the following we consider in detail the quantization of the linear harmonic oscillator. can be treated exactly and therefore plays an important role in numerous quantum mechanical problems which can be solved only with the help of perturbation theory. E. This is a fundamental topic since the harmonic oscillator. Noz [149]. its quantization here in terms of quasi-particle creation and annihilation operators also points the direction to the quantization of field theories with creation and annihilation of particles. Since the spectrum of the harmonic oscillator is entirely discrete.the present form of quantum mechanics is largely a physics of harmonic oscillators. that the oscillator. 83 . we shall see on the way.. which is often described as "second quantization" as distinct from the quantization of quantum mechanics which is correspondingly described as "first quantization" (thus one could visualize the free electromagnetic field as consisting of two mutually orthogonal oscillators at every point in space). can — in fact — be quantized in quite a few different ways.

in the present case + [E--mooJ2x2)ip = 0.' ^mgLOh by setting and *~.MvT« 7=r)^(VT-vir>' + («> 'Note that we assume it is obvious from the context whether q a n d p are operators or c-numbers. hence we do not use a distinguishing notation. 2mo dx \ 2 2 (6. like taking half and half. .84 CHAPTER 6.p]=ih (6. which is — in the first place — representationindependent. This equation would then have to be solved as a second order differential equation. the boundary conditions on the solutions ip(x) would be square integrability over the entire domain of x from —oo to oo. However. of which the only non-trivial one here is [q. To this end we introduce first of all the dimensionless quantities rriQUi fl q.1) for the hermitian operators q. For instance we can go to the special configuration or position space representation given by q^x.^ The first problem is to determine the eigenvalues and eigenfunctions of H. so that there is no ambiguity due to commutation. p^-ih—.2 The One-Dimensional Linear Oscillator In the case of the one-dimensional. we can also proceed in a different way. h = l. This can be done in a number of ways.p. ox and then to the time-independent Schrodinger equation Hip(x) = Eip(x). which is called Weyl ordering. When such a term arises one has to resort to some definition.2) which for mo = 1/2.to = 1 reduces to the parabolic cylinder equation. Also observe that H does not contain terms like pq or qp. Quantum Mechanics of the Harmonic Oscillator 6. 1 2 2 The quantization of this oscillator is achieved with the three postulated canonical quantization conditions. linear harmonic oscillator the Hamilton operator is given by the expression XJ 1 2 .

+ 1). A^] = A\ (6. We now use the relation (3.5) Re-expressing q and p in terms of A. (6.2 The One-Dimensional Linear Oscillator 85 (p.6. C}B. (6.14b) [A* A.e. (6. i. Prom these relations we obtain ( ^ = ^ . With the help of Eq.1) we obtain immediately the commutation relation [A.B}C + B[A. C] + [A.7) (6.14a) (6. (6.C]. (6. q are hermitian).6) and p = y/motoh -A-Ai -=^. for ^ A | a ) = a\a). C] = A[B.14a) we deduce for an eigenvector \a) of A^A.9) We observe first that if \a) is a normalized eigenvector of N with eigenvalue a.e. i. if A*A\a) = a\a).12) in order to obtain the following expressions: [tfA. (6. A] = [A\ A]A = -A. A*} = A^A.34d): [A.1 ) .13) ( ^ A ) i t = A\A^A From Eq.10) then a = (a\A^A\a) = \\A\a)\\2>0. and [AB.A^] = 1.BC] = [A. i\/2 Inserting these expressions for p and q into H we obtain H = hiu{A^A + AA*) = hi* (A^A + ^ .8) The eigenstates of H are therefore essentially those of N := ^ f A (6. (6. A\ we obtain h A + A^ rriQLU y^ (6. (6.11) Thus the eigenvalues are real and non-negative. .

(6.2A)\a) 2 (a . (6. unless A^\a) = 0. This would mean that for sufficiently large values of n the eigenvalue would be negative. (6. However. we must have (a) Let \a-n):= An|a)^0.l)\a) = (a . (6.2)A |a). or \\A\a)\\ = Va. in view of Eq. A 2 |a) is eigenvector of A^A with eigenvalue (a — 2). but (b) An\n>\ llA An+1|a)=0. | a) is eigenvector of A^A with eigenvalue (a + 1). we find that A n |a) is eigenvector of A^A with eigenvalue (ct — n). unless A\a) = 0.2A)\a) = A(Aa . unless A 2 |a) = 0.86 the relation CHAPTER 6.11) this is not possible.16) This means.15) Thus A\a) is eigenvector of A^A with eigenvalue (a — 1). Next we consider the vector A 2 |a).19) i.A)\a) A{AAU .l)\a) = A(a . In this case we have (A*A)A2\a) (6 (6. (6.18) = '= a ) A(A^A .l)A\a). Thus for a certain value of n > 0.17) (6.e. Quantum Mechanics of the Harmonic Oscillator (A*A)A\a) = A{A*A .14b) : (AiA)A*\a) = A\A^A + l)|a) = A\a + 1)| a) = (a + l)A^\a).21) .l)A\a) = A(A^AA . since this equation implies that the eigenvalues cannot be negative. If we continue like this and consider the vectors An\a) ^ 0 for all n. (6. or ||At|a)|| = v ^ T T . The norm of A la) is ||A|a)|| 2 = {a\A^A\a) = a ( a | a ) = a.20) . (6. Similarly we obtain from Eq.. Similarly \\A^\a)\\2 = (a|AAt|a) = (a|l + A^A\a) = (a + l)(a|a) = a + 1.

22) With relation (b) of Eq (6. v. or a = n > 0. so that A^\n)^0 In particular we have A^\0) ^ 0 and II^IO))!2 = (OIAA+IO) = (0\l + AU\0) = (0|0) = 1.18) we obtain for a = n: p t | n ) | | 2 = n + l.n | a .28) .' " /.5) to shift operators A to the right which then acting on |0) give zero. For a — n = 0 we deduce from (b) of (6.23) Hence the eigenvalues of the operator iV := A^A are nonnegative integers. (6.6. that a-n = 0.n ) = " ||A»|a>||=» ' = ||A«|a)|F = 1 . Moreover.20) we obtain from this that the right hand side vanishes.2 The One-Dimensional Linear Oscillator be a normalized eigenvector of A^A with eigenvalue (a — n).e..I n U .. \ I 2 2 x 87 Replacing in Eq. i.26) <0|A4 t AA t + A4 f |0) = (Q\2(A]A + 1)|0) = 2. The state |0) is called ground state or vacuum state. I „ .25) (6. so that (a|(A n )tA n \a) __ \\An\a)\\2 ( a . \ 9 \ I / II 4 T ) .20) we have for a = n = 1: A2\1) = 0. From relation (6.20) that A|0) = 0.24) This is a very important relation which we can use as definition of the state vector |0). (6. But A 2 A t |0) = A(AA^)\0) = A(A^A + l)|0) = 0. In the following manipulations we always use the commutator (6. (6. (6. HAU+IO)!!2 = = (O\AAAWI\O) = {O\A{A^A + I)A^\O) for all n. (6.17) a by (a — n).27) According to (b) of (6. we obtain a-n = \\A\a-n)\\2(6=) An+l\a) \\An\a) (6. (6.

11) we have [A. (A*)2} = [A. A^A* + A*[A.29).Al] = 2A^A^ + (A^)2 x 1 = 3(A^)2.^ ( O l A ^ n o ) .(6.34) (6. Similarly we find |2)oc^Ut|0).(A*)n] = n(Ai)n-\ (6.32) we have (n|m) = .88 and CHAPTER 6. (6. and in view of Eq.30) (arbitrary phase factors which are not excluded by the normalization have been put equal to 1). (6. i.5) and again (6. |l>ocAt|0).31) (6. A^} = 2A\ and [A.11).32) (6. The states \n) thus defined are orthonormal. (6. (6. we have |2> = ^=dAi\0). (0|^ 2 (At) 2 |0) = 2. In general we have \n) = -^(A^n\0) (6. and in general [A.33) .26) the equality |l) = A t |0).e. (A^)3] = [A. as we can see as follows. According to Eq.29) Hence we obtain and in view of Eq. (6. But using Eqs. Quantum Mechanics of the Harmonic Oscillator A2AtA*\0) = A(A^A + 1)4*10) = (AA^AA^ + AA*)\0) = = = {AA\A* A+1) + A* A+ 1}\0) (AA^+ 1)\0) = (A^ A+ 2)\0) 2|0) ^ 0. (At)2]A+ + (A*)2[A.

the eigenvalues of the Hamiltonian H are En = hw\n+-\+ i ) .37) n^)71-1}^) = v n | n .1 ) . l . in quantum mechanics. •(» (6.6. The terminology is. . i. ldnm = n\5nm.L ( A t ) n + 1 | 0 ) = VnTT\n Vn! and with Eq. In view of the same properties A is also called annihilation operator and A^ creation operator of so-called u quasi-particlesv.35) the operators A^ and A are called respectively raising and lowering operators or also shift operators.1 |0). (6. however.38) (6.'n\ and = -v n ![ ( A t ) M i= + (6. In view of the properties (6.1) (6. we do not have creation or annihilation of any real particles as in field theory (hence the word "quasi-particle").33). chosen in close analogy to the postulated "second quantization relations" of field theory.39) A^A\n) = y/nA^\n . (6. n = 0 .40) (6.J \n) = hw (n + i J \n). whose number is given by the integer eigenvalue of the number operator N = A^A.34) and (6. (6.41) The contribution hu/2 is called zero point energy. 2 . (6. .1 (A + ) m .8) we obtain therefore H\n) = hu (A^A + ]. .34) A\n) = -±=A{Al)n\0) .35) Inserting this result into Eq.32): rf\n) = . With Eq.2 The One-Dimensional Linear Oscillator so that {0\An(A^)m\0) = = {0\An-1A^mA\0) + (0\An-1m{A^)m-l\0} 89 0 + m(0|A n . By repeated application of this relation on itself it follows that (since this is nonzero only for n = m) (0\An(A^)m\0) = n{n . (6. . .e. .36) + 1) (6. we obtain (n\m) = 5nm We also deduce from Eq. .1) = n\n). Here.

The representation of the Hamiltonian in this energy representation is {n\H\n') = En5nn. A^ from the relations (6. and — in fact — we have used this already previously (see e.38) and (6. Eq.e.2..39)).90 CHAPTER 6. also called Heisenberg representation.6) and (6.7) the energy representation of the operators q and p. Quantum Mechanics of the Harmonic Oscillator 6. (6.42) We can deduce the energy representation of the operators A. n = 0...'\n) i'\A*\n) (n + l\A'\n) and similarly (n'|A|n) (n — l\A\n) In matrix form this means 0 0 0 0 0 0 0 0 0 A/4 = = Vn + l(n'\n + 1) = \/n + 15 n / iTl+1 .39): {n\A. i. (5. other elements zero.3 The Energy Representation of the Oscillator The energy representation.46) v^ . other elements zero. En = hu(n + -).1.45) Correspondingly we obtain with Eqs. (6.43) = = y/n(n'\n — 1) = y/n5n^n-i. (6. ( At = Vi 0 0 0 ° \ f° 0 Vi 0 0 0 0 0 0 0 0 V2 0 0 0 0 0 V2 0 0 v^ 0 0 0 0 0 V3 0 V5 0 v / ( ° VI 0 V2 0 0 v / (6.g. is defined by projection of operators on eigenfunctions of energy. Vi 0 0 0 2mQUJ 0 V2 0 y/3 0 0 0 V3 0 y/l 0 -\/2 0 V3 0 \ V • / P = mohu) VT 0 0 0 ° -VT 0 V2 0 0 / 0 0 -V3 0 (6...44) yfn. (6. Vn + 1.

6.1) and (6. This is a simple differential equation of the first order with solution (x|0) = The normalization constant C is determined by the condition* OO /-C 1 = (0|0) = / / dx(0\x)(x\0) = \C\2 I dx(0\x)(x\Q) = \C\2 / -OO J —< e-mou}x2/hdx = \C\2 mow' so that We choose the arbitrary phase 6 to be zero. Correspondingly the position space representation is given by the wave function <t>n(x) := (x\n). ™ou(x 2h \ + J _ d ] m = 0 (649) mQUJ dx Ce-mouJx2l2h.4 The Position Space Representation 91 It is an instructive excercise to check by direct calculation that Eqs. (6.5) are also satisfied as matrix equations.47) -L-P |0) = 0 . (6. (6.32).4 The Configuration Space Representation We saw that the eigenstates are given by Eq.48) Applying from the left the bra-vector (x\ and remembering that (x\p\<f>) = we obtain -ih—(x\(j)}. 2 . mow / (6. 6. Hence ( * Recall J™ dxe-™ * ? 2 2 2 1/4 W \ i/4 \ e-m^x*/2h_ (g 5 Q ) = yfln/w .3)) ^ (q 2n \ + (6. by (cf. i. (6. A\Q) = 0.24). The ground state wave function (/>o(x) is defined by Eq. Eq.e.

e.92 CHAPTER 6.56) < ^ = We have as an operator identity the relation . it suffices according to Eq. Vn! Now {XlA (6. These polynomials are real for real arguments and have the symmetry property # n ( . (6. so that <t>n(-x) = ( . In order to obtain the wave functions of higher states. i.l ) > n ( x ) .l ) n f f „ ( 0 . they are called Hermite polynomials.L ( x | ( A t ) » | 0 ) .54) (6 55) i/4 r-zri7r 1 / 4 vo.32) to apply the appropriate number of creation operators A^ to the vacuum state |0).dx'{X] so that <*'->^(irrv^i« mQU dx Setting a we have 7T 1 /4 v ^!2«/2 \a Setting a and ff„(0 we have = ^ / 2 u^n-^m^-''^ h mouj dx) K ' ( f .2 n n! The functions Hn(£) are obviously polynomials of the n-th degree.| ) e"^2. (6.0 = ( .51) \l 2h {X q \ m0uP)-\l 2h \X m0u. Quantum Mechanics of the Harmonic Oscillator This is therefore the ground state wave function of the one-dimensional harmonic oscillator. c/>n(x) := (x\n) = . (6.

6. . ( 2 0 4 . Magnus and F. (203-6(20.58b) by W.57b) • Generalizing this result and inserting it into Eq.e. -e. 93 . gain as an operator relation we have de = een± = een da e-e/2±_^e-e/2 d£ 1 u ^ d? dt. 2£.4 The Position Space Representation i.12(20 .58a) ' T h i s definition — apart from the usual one — is also cited e. (202"2. (6. 81.54) we find that (observe that there is no factor 2 in the arguments of the exponentials!)§ Hn{i) = ( .It follows that 0en(Pe/2^Le-z 2 n\P-?i2 — c2 d d£ -? = 2f.l ) n e ^ In particular we have #o(0 #i(0 #2(0 #3(0 tf4(0 = = = = = 1. — = n+ d ~^ -i 2 + d 2 de) ' 1+e ~ (6. d£" (6. p.2)m = (±-t)m for some function VKO.g. Oberhettinger [181]. (6.12.en_ dt.

59c) The recurrence relations satisfied by these Hermite polynomials (the second will be derived later in Example 6.59a) The differential equation obeyed by the Hermite polynomials Hen(£) defined in this way is (^2 . 81.59d) The normalization of Hermite polynomials is given by the relations: oo / d£ Hn(OHm(Oe~e -oo = 2 n n\^5nm. £2-l.He'n{£). (6. 80.94 CHAPTER 6.3) are (as given in the cited literature and with a prime meaning derivative) Hen+1(0 Hen(0 He'n(0 = ZHenifi) . £3-3£. £ 4 . (6. Mathematicians often define Hermite polynomials Hen(^) as^ Hen(0 Then He0(O Hex{0 He2(0 He3(H) fle4(£) = = = = = 1.^ + n) He n(0 = 0- (6. (6. Magnus and F. = ffen+itO+ntfen-iCO. . Oberhettinger [181]. (6.58c) Here special care is advisable since the Hermite polynomials defined above were motivated by the harmonic oscillator and hence are those frequently used by physicists.59b) = (-l)^2/2^. Quantum Mechanics of the Harmonic Oscillator The differential equation of the Hermite polynomials Hn(£) defined in this way is ( ^ 2 ^ + 2n )ff"(O=0. = nHe^iO.60b) J—< "See for instance W. £. pp.6 £ 2 + 3.60a) and I d£ Hen{Z)Hem{Z)e-?l2 = n! V & „ m - (6.(e-^/2) or Hn(0 = 2n/2Hen(V2ti). (6.

63) The meaning is that if we expand F(£. Oberhettinger [181].61) e-?l*2-nl2Hn{Z).64) n\ See e..62) J-oo The following function is generally described as generating function of Hermite polynomials Hn{£): F(S.6.s) = J2 n=0 (6. . (6.s) = eM?-(Z-s)2}(6.0) we obtain + s fdFs 1! V ds s=0 + OF ds 2 dF ds2 = s=0 2(f d S )^-«-)a s=0 = 2£ = ffi(0.l ) ^ e-^HeniO 2 / 4 | . 93. Magnus and F.s)=F(£. p. F(S..g. and so on. 2(£ - s)£2-& s=0 s=0 ~ ds = m -sf-2]e?= •«-«) : S=0 2(2e2-l) = (2e)2-2 = ^2(0. W. Their normalization is therefore given by OO /"OO / OO J —OO z (6.^ = e-^/42~n/2Hn(aV2).4 The Position Space Representation 95 The relation between Hermite polynomials and parabolic cylinder functions Dn(0 — which we use frequently here — is given by" Dn(0 = = Dn(V2£) = ( . Thus the generating function can be written F(Z. s) as a Taylor series.

and we see t h a t even and odd states alternate. has parity + 1 . But this would mean t h a t b o t h x{t) and p{t) would assume simultaneously the "sharp" values zero. 2 . i. W i t h Apx = moAvx we have = moLoAx Ax and thus h/2 Apx h/2 mQuAx Ax I h/2 . i. ground) state is fko/2. We see t h a t the lowest state is symmetric. T h e comparative behaviour of t h e wave functions of the three lowest states of the harmonic oscillator is shown in Fig.e. E0 ~ -—{Apx)2 + IrriQ l -mQuj2{Ax)2 moLoh/2 2TUQ 1 moU!2h/2 2 TTIQU = -Ku). for x(t) = 0 and p(t) = mox(t) — 0.96 CHAPTER 6. We also see from the eigenvalue t h a t the energy of the lowest (i. Quantum Mechanics of the Harmonic Oscillator Fig. 6.e. In fact we can use the uncertainty relation AxApx ~ H/2. mow' Apx = ^m0Luh/2. 6.1. in order to estimate the lowest energy of the harmonic oscillator.1 The comparative behaviour of the lowest three wave functions of the harmonic oscillator. Furthermore we observe t h a t the wave function of the n-th state has n zeros in finite domains of the variable x.e. In classical mechanics the energy of the oscillator can be zero.

But it has a pole at infinity. Example 6.2 contains only the answers since the evaluation of the integrals proceeds as in Example 6. to —a.. and with potential V(q) = 2-K2mov2q2 and total energy E = p2/2mo + V(q).: 27 -T mov2a2.2 The original path and the transformed path at infinity.6. E .-1/a O l/a Rez all at infinity Fig. and is directly observable.1 we demonstrate how the eigenvalues may be obtained by the method of "poles at infinity". original path q . classical orbit i. to 0. Solution: Consider the simple harmonic oscillator of natural frequency of vibration u.4 The Position Space Representation 97 The zero point energy plays an important role in atomic oscillators. Finally we point out that the Hamiltonian of the harmonic oscillator can be diagonalized in many different ways. like two-atomic molecules. 6. According to "old quantum theory" with integer n. In the plane of complex q the integral has no poles for finite values of q.plane pole at z=0 ^ . back to 0. mass mo. 27rmoi/ <f> V'a 2 — q2dq = n*h. This observability can be taken as direct proof of the uncertainty relation. setting q = 1/z. we obtain the integral . In Example 6. The analogous Example 6. Sec. 14.plane Imz z .4) to obtain by contour integration the eigenvalues of the harmonic oscillator. n*h. Thus. in corrected version instead with n* = n + 1/2.e. -/classical orbit Here q goes from 0 to a. It is clear that the uncertainties are not due to deficiencies of measuring instruments (we are concerned here with a system in the pure state |0)).1. pdq.1: Eigenvalues obtained by contour integration Use the corrected Bohr-Sommerfeld-Wilson quantization condition (cf. which is the complete orbit. The kinetic energy is always positive or zero.

e. 1 . With a2 = -E. We can evaluate the integral with the help of the Cauchy formula (the superscript meaning derivative) Jc Hence we obtain (z . Quantum Mechanics of the Harmonic Oscillator This integral has a triple pole at z = 0 (i. Example 6. £ = n*/w = ( n + . .65) f -^Va2z2 .) hv. 2 .V62 -a2). (m . .5 The Harmonic Oscillator Equation In the above we considered mainly the energy representation of the harmonic oscillator and encountered the Hermite polynomials. After removal of a Gaussian or W K B exponential. mo = 1/2) the integral I2 can be checked to give the eigenvalues for the Coulomb potential -Ze2/q (2 turning points).2TTI a ^ = .e.1.c2 = (I + 1/2) 2 (natural units. 6.2: Contour integration along a classical orbit Setting q = 1/z. are the most basic and exactly solvable cases in quantum mechanics.98 CHAPTER 6.5. In the following we investigate in more detail the important differential equation of the harmonic oscillator. 6. i. z = 0 V-i 2-Kmov(-Ka2) = n*h = E/v. a J \-d- Solution: The solution proceeds as in Example 6. in q at infinity). we start with a Laplace transform ansatz for the remaining part of the solution.5 . verify that II = d> dq yi-92 aq + ft -2TTI 9 /y/F^I 9z V a + bz z=o ..e.1 D e r i v a t i o n of t h e g e n e r a t i n g function From the above we know t h a t the energy is quantized. and obtain with this all properties of the solution.1. -E = Z2e4/4(l + n+ l ) 2 (cf. . for instance — without deriving all properties of the solutions.a) m = 2 . Eq. but here we leave this .114a)). In Chapter 11 we perform similar calculations for the radial equation of the Coulomb potential — as in Example 11.It follows that 1 = s2 i ^V»a*a-i ~2f z=0 2-wi r m '„2 = 7ria 2 La*Va 2 2 2 . harmonic oscillator and Coulomb interaction. 2ft a2 + 9 c 2 ! / 2 27T In f j — (ac — b) and I3 = <t \-a2 r 2b _ c^n/2 = ^( a 3 6 _a2c2).1)! (6.( 6 . hence we consider the following differential equation in which n = 0 . i. / " ^ . « qdq + I2 = f dq . since these two cases. (11.2b = Ze2.3.

E. / dpf(p)e~px the Laplace transform of f(p). (11. (6. We demand that C(p) := — \ps(p)e~pt] = 0.66) First we have to remove the £2-term. Thus — ignoring details — / dpf(p)erpx is the Fourier transform of / ( p ) . W.g.69) **See e.68) § = J p2s(p)e-*dp.p t dp = \ps(p)e-Pt} . we obtain -P2/2 ln(ps(p)) = — -p — nlnp or s(p) <x |Tl+l P (6. i. Magnus and F. Eq. This has to be true for all values of t.g. Hence we set (with the chosen sign in the exponential) m=y{t)e-t2l* with *0. p. since this can give rise to exponentially increasing behaviour (at most a factor t can be tolerated). fp{ps{p)) = -{v2 + n)s{p). (this determines the limits).67) we obtain -\ps(p)e -pt] + I P2s(p) + —(ps(p)) + ns{p) e~ptdp = 0.J ±(ps(p))e-*dp.i ^ | (ps(p)) -P n P Integrating this expression.** We make the Laplace transform ansatz^ (with limits to be decided later) y(t) = [ s(p)e~ptdp. (6. . (6.6. 80. Oberhettinger [181].67) the latter being the differential equation of Hermite polynomials Hen{t) for integral values of n.5 The Harmonic Oscillator Equation latter property open for determination later: 99 ~dP + 1 1 2 V> = 0 . . so that the integrand of the remaining integral vanishes. (6.e. Substituting these expressions into Eq. Then (with partial integration) *dt = ~*/Ps(p)e. (cf.111)) the Mellin transform of e~p is n!. and / dpf(p)pn the Mellin transform of / ( p ) .-t*ij& + ny{t) = O.

(1) This will be satisfied if \p\ — ±oo. (3) We can choose a contour once around p = 0 in the complex p-plane. y(t) oc 2vri J pn x -dp for n integral and ept p 2 = coefficient of pn in ~^ (6. then for large t we have y(t) ~ exp(i 2 /2). Hence a possible path of integration > is from — oo to oo. To see this. This exponential assumes a maximum value when pt -\—p2 = minimal. The value of the exponential at this point is ~ exp(£ 2 /2).100 and hence CHAPTER 6. (2) If n < 0. we can choose a path from zero to infinity (the condition will vanish at p = 0). since otherwise (with p = \p\ exp(i8)) we will get part of the solution from 0 to infinity. when p = —t (which is possible. so that y(t) ~ exp(i 2 /2). so that p cannot go to infinity. Quantum Mechanics of the Harmonic Oscillator -pt-p2/2 » ( « ) -pt-p2/2 -dp with C{p) = (6. n must then be an integer. = / • Im p Rep n not an integer Fig.-.e. Thus the only solution left which does not involve large values of \p\ is that with § Q for n an integer.3 The path when n is not an integer.70) = 0 P. as indicated in Fig. i.67)) M) * e"* /4y(t) ~ et +2/ 2 /4 But this is not permissible. We now investigate possibilities to satisfy the boundary condition C(p) = 0.3. Eq. This can be attained if |p| is allowed to be large (t can be of either sign). For single-valuedness. consider the exponential in the integrand. We can show that if \p\ is allowed to be large.n+l between the two limits of integration. (6. 6. of course). 6.71) . The quantum mechanical wave function is then (cf. Therefore we take pt-p2/2 y(t) .

The Hermite polynomial Hen (t) is now defined such that the coefficient of tn is unity.0) p p»+i v ' Hen(t) = (-l)^ e * 2 /2 / 2vri /(9=t) dg " ^(g-i)"+i e ?2/2 = e i 2 / 2 (-|)ne"i2/2. The Hermite polynomial is therefore given by H ^{t)-{-^—f—^r~dp. We can also obtain the derivative form of this Hermite polynomial. The exponential function here is described as the generating function of the Hermite polynomials Hen(t).„y2 v l { ' ^ r (^) \ 2 ' where v = 0. . 4 .3.5...5 The Harmonic Oscillator Equation 101 with Cauchy's residue theorem.if n is odd. . .P 2 / 2 00 C—tY ( .l ) n n ! x coefficient of pn in = = (~l)nn\ e~ ~ x coefficient of pn in V l z ^ e .. 2 . . pt p2/2 (6. the term of highest degree in t in the exponential corresponds to taking all p's from exp(— pt). In y(t) above. Therefore the coefficient of (pt)n in exp(—pt — p2/2) is (—l) n /n\. We have 27T! Setting q — p + t.6. (6-76) . if n is even and v = 1. n—v even ^ 2 ' Hence we obtain the polynomial of degree n Hen(t) = nl £ see below ( _ 2 ) ( n .l ) n n ! x coefficient of pn~v in V LJ_L e -p 2 /2 *-^ v\ u=0.72) We now obtain the polynomial form with Cauchy's residue theorem as Hen(t) = ( . this becomes jfj.

P-pt-p 2 / -00 /2 -dp. n + l) = l.77) = t. Quantum Mechanics of the Harmonic Oscillator which we recognize as agreeing with Eq. (n. one can construct the tower of polynomials Example 6. Re-interpreting the remaining integrals with Eq. and elsewhere.e. as in the normalization of asymptotic expansions of Mathieu functions). Solution: We multiply the generating function by a factor t and perform a partial integration on the factor exp(— pt) contained in the following integral (the differentiation of the other factor leading to two contributions). (6. Hen(t)Hem(t)e-t ™ It follows that y ' (27Ti)2 fp=0Jq=0 P "V+l^+l J ^ ^ The integral with respect to t is a Gauss integral^ and yields ePde(p 2 +12)/2 f°° J —OO e-{t+(p+<.3: Recurrence relation of Hermite polynomials Show that tHen(t) = (n.7 that we use throughout this text) and their orthogonality and normalization to Examples 6. and we obtain then: -pt-p2/2 —^T-i -pt-p 2 /2-| d : P r e-pte~p2/2 •(n + l) <t —5 f dp - e-pte-p 2 /2 Here the bracketed expression vanishes in view of our condition (6.)} 2 /2dt = ePde(P 2 W)/2^^} J —CO . in radiation problems.102 CHAPTER 6. Hei(t) Hen--j.70). Starting from Heo(t) = 1.n— 1) = n.4: Orthonormality of Hermite polynomials Establish the orthogonality and normalization of Hermite polynomials. n — l)Hen-i(t). (6.75) as Hermite polynomials. Hen(t). we obtain tHe„{t) (-l)"n! ~ i.3 and 6.5 is a further application of the method for the evaluation of integrals that occur frequently in practice (e. Example 6. Example 6.it) _ (n + l)Hen+1{t) (-l)""1^-1)! (—l)"+ 1 (n + 1)! ' tHe„(t)=nHen-1{t) + Hen+1(t). ( n .g. (6.1 '2dt using Hen(t) = (-l)n27 r . Solution: We have to evaluate °° 2 . We leave the consideration of the recurrence relation of Hermite polynomials (of paramount importance in the perturbation method of Sec.4.n + l ) f f e n ^ i ( £ ) + (n. for the evaluation of expectation values. 8.59a).

80) f(p.q) = = e . the coefficient of qn+2r-n the coefficient of p"qn+2r m exp(. (6. Hence the normalized wave function of Eq.n)\ i s an+2r-ij.5 nm .5 The Harmonic and hence Oscillator Equation 103 / = (-!)" But (2iri)2 { J Jp=0 J q-0„ pn+ i0m+l P I e d d P 1- 1 / epq p" <k rrdq = coefficient of qm in epq = — 27ri / qm+1 m Therefore (-IT ^n+m ^ n l I Jp=o ' v d P ' 27U pn-m+1 =2ni if Ti=?7i. g) is n! Next.78) \/27rn!.66) is \Zn\V2w Example 6. and 0 otherwise \/2n(—l)n+mn! if n = TO.82) ^ .q). It follows therefore that the following integral (with an exponential exp(—at) from whose expansion we pick later the power of i) is given by °° 2 / where -OO e-atHen(t)Hen+2r(t)e-t l2dt = n\(n + 2r)\ X coefficient of pnqn+2r in f(p. zero otherwise (6.5: Generalized power integrals with Hermite polynomials Establish the following general formula for Hermite polynomials: / Jo t2sHen(t)Hen+2r{t)e-t ir n\(2sy.6. g) is the product V^Tre" ^oM!(n-/i)!(n + 2r-M)! (6. (6./(n _|_ 2r — /i)!.(n + 2r)\ ' 2 ~ s r '2dt r 2* ^ i / ! ( 2 r + i/)!(Ti-i/)!(s-r-i/)!' Solution: From the above we know that He„(t) is the coefficient of p " in (—l) n n! exp(—pi—p 2 /2).^ + s 2 ) / 2 f°° 2 e-(P+9+«)*~t /"GO 2 /2di 2 J — oo o = e " /2eP9e"(p+9) / e-(t+p+9+a!) /2d4 J —oo ^/^Tga /2e9Qep(q+«) ^ gl^ The coefficient of p n in / ( p .jQ) n n!I ^—' ^Q fi\(n . so that /2 Q2(n+r-rf in / ( p .

n + 2){n + 2.ra + 3)(n + 3. the coefficient of T 2s -»{s-r -u)\ As we observed. n + 2) + (n.(t)e-«2/2dt Now." ) a2^-) r coefficient of (2s)! in f^ (s . n + l ) ( n + 1.n + 2) + (n. -.104 CHAPTER 6.' .. n + l ) ( n + 1.78).n+ l)(n + 1.r . n + 2)(n + 2. Quantum Mechanics of the Harmonic Oscillator It now follows t h a t — in later equations with v := (n — /i). n + 4)_ffe n +4(t) + [(ra. i. n + 2)]ife„ + 2 (t) + ••• . This result can be verified by inserting in the integral for t Hen(t) the linearized expression obtained with the help of the recurrence relation (6. n + l)(n + l . so that now with integration limits from 0 to oo K = coefficient of a2s (2s)! in J a = 2. n + l ) ( n + 1.l ) ( n .83) = (n + 2r)!e a / 2 V „ „.r . It follows that the coefficient of a 2 s / ( 2 s ) ! in this expression J is the quantity K given by f^ot2sge„(t)Jfen+2t.i / ) ! ( 2 r + i/)! > . n ) ( n . n . n + l)(n + 1.2s _oo ( a 2 / 2 ) i . n + l ) ( n + 1.A o?(. the relation (n.A i ) ! (6. n — 1) are given by the first index.^ / 2 * . Hence t4 He„ (t) = = Hen+i Hen+4{t) (t) + [(n + 3) + (n + 2) + (n + 1) + n] H e n + 2 (t) + • • • + 2(2n + 3)Hen+2(t) + --.v)\(2r + v)\ > Q (2s)!(n + 2r)! AT 2~ s r 2V v\(2r + u)\{n .r+v) a2s coefficient of in (n + 2r)\ V (2s)! h . Prom Example 6.° i i f e T 1 ( t ) i f e „ + 2 r ( t ) e .t • 2l/2dt _ J : Q2 » a2(n+r-^) ~ f ° ° f f „ t2/2 „ !ZoHen(t)e-2 Wdt .— — ^ . n + 2) + ( n .v)\ ^ Multiplying both sides by the value of the normalization integral.Y^ i[ ^0vKn-v)\C2r + ») cc2s A (n + 2r)! a 2 ^ .e.e '' ^ o^M ! ( n . .' '2dt= / Jo t4 Dn(t)Dn+2(t)dt = V2^(2n + 3)(n + 2)!. i. A « 2 ( r '+ I/ ) 2 coefficient of (2s)! in (n + 2r)\e ' ^ . The result then follows with half the normalization and orthogonality integrals (6.79).( n + Jr).e.e.3 we know that up-going coefficients are 1 and down-going coefficients (n.l . we obtain the result (6. n + 3)(n + 3.77). by n!(7r/2) 1 / 2 . K is the coefficient of a / ( 2 s ) ! in the expression J . n. n + 2)(n + 2.^ ) ! ( n + 2 r . i.u)\ 2s-<—" v\(n .v)\(s . As an example we obtain for the integral from 0 to oo (also re-expressing the integral in terms of parabolic cylinder functions Dn (t)): / Jo t 4 f l e „ ( t ) H e n + 2 ( t ) e . i / ! ( n . n)(n. /i = (n — v) _ / r o o e .

Chapter 7 Green's Functions 7. which is a point of unstable equilibrium for a classical particle. which we write G{x.t). in Feynman's path integral as the kernel or free particle propagator.1) be the time-independent Schrodinger equation which has to be solved. This example also offers a convenient context to introduce the inverted oscillator potential.2 Time-dependent and Time-independent Cases In the case of Green's functions we distinguish between those which are time-dependent. we consider the sojourn time of a quantum mechanical particle at a point. The first of these cases will re-appear later. of a differential equation of second order.x'. In particular we derive the time-dependent Green's functions for the case of a free particle and for that of a particle in an harmonic potential. We consider first time-independent Green's functions. These are the Green's functions of the time-independent Schrodinger equation or. Let H^ = E^> (7.x'). more generally.1 Introductory Remarks In this chapter we consider Green's functions of Schrodinger equations in both time-dependent and time-independent forms. 7. and those which are timeindependent. in Chapter 21 (with a different calculation). With reference to this case. which we write here K(x. which is normally not discussed. and is thus an important point to be noted here. We 105 . The other example of the oscillator enables us to evaluate corresponding expectation values of observables in the canonical distribution introduced in Chapter 5.

The boundary conditions which G(x.x') or ^2^(x)^*(x') i = S(x .5) M = Z*W=EW i l for Ei0) E *- (7 6) - This holds since *For simplicity we consider the one-dimensional case. and Hi is some other contribution.x').x'). for instance.EW)xG{x.e. cf. i. (7. We can obtain an expression for G(x.g.e. (7. The time-independent Green's function G(x.3) i which has the specific representation ^2{x\i)(i\x') i = 5(x . since the generalization to higher dimensions is self-evident. the relation X') = 5{x . (7. aq2. Green's Functions H = H0 + Hi.x') by recalling the completeness relation. . 'The case of £(°) equal to some Ei is considered later.g. a perturbation part like (3q4.106 assume a Hamiltonian of the form CHAPTER 7.x') is defined by the equation* (Ho . of the kinetic part p2/2mo and some part of the potential.2) where Ho consists. e. (7. appropriately decreasing behaviour at infinity). e.x') can be written t G (7. i.13).g.4) where the functions ipi(x) are solutions of HQij)i{x) = E^iix). x') has to satisfy correspond to those of * (to insure e. square integrability. Eq. where £ ( 0 ) = lim E. We can readily see that G(x. although the explicit calculations can be much more involved.

8) is satisfied as a consequence of the completeness relation of the wave functions ij)i{x).x'. (JL x'\ t) = H0K(x.8) This Green's function can be seen to be given by the following expansion in terms of a complete set of states (see verification below) K(x.Hi{x'))^(x'). i (7.x')(E . We rewrite the complete time-independent Schrodinger equation H^ = E^ as (#0-£(0))* = (£-£(0) -#/)*. (7.#.E<®)xG(x.(*'))*fr') Hi{x'))^(x') f dx'5{x .t).EM)X*(X) ( . (7.t) = ^Eie^ihM^i(x') i = Y^eE^ihUx)^(x').x'. . x'\ t) (7.t) since ih^K(x.E® (E-E{0) -Hi{x))m{x).2 Green's Functions: Time-dependent and Time-independent 107 We note that the Green's function possesses simple poles in the plane of complex E(°\ and that the residue at a pole is determined by the wave function belonging to the eigenvalues Ei.x'){E .9) = i H0YieE^i%(x)^(x') = H0K(x..E^ of the eigenvalue.x. which consists of the perturbation part Hi of the Hamiltonian and the corresponding correction E .7) with the initial condition K(x.E^ We can check this by considering (H0 . (7. The time-dependent Green's function K(x. We see that the initial condition (7.x'.11) = = 3) = fdx'(H0 .7.) = f dx'G(x.£(0) . For the solution V we can immediately write down the homogeneous integral equation (again to be verified below) *(a.x'. .0) = S(x-x').10) The contribution on the right hand side is the so-called perturbation contribution. We can see the significance of the time-independent Green's function for the complete problem as follows.t) (K for "kernel") is defined as solution of the equation ih—K(x. x'){E .

Bleecker [33]. (7.11) an inhomogeneous term which is a solution of the non-perturbed part of the Schrodinger equation? If we keep in mind the anharmonic oscillator with square integrable wave functions ^(x). This will be different. H. integral equations are more difficult to solve than the corresponding differential equations.15) ''This problem and its circumvention can be formulated as a theorem. (7. Instead of Eq.11) are called (homogeneous or inhomogeneous) Fredholm integral equations. see B. All such methods of solution are based on the analogy of Eq.2 (not contained in the first edition!). it is possible to solve the integral equation by an iterative perturbation procedure (see later: Born approximation). if we assume that E^ = Ej. the function ipi(x) would not be a solution of and it is not possible to add an inhomogeneous contribution. Maharana. 2nd ed. Merzbacher [194].108 CHAPTER 7. then at best such a contribution would be something like cipi(x). In cases where an inhomogeneous contribution is given. (7. W.x'.14) We can now see the significance of the time-dependent Green's function K(x. Sec. Wiedemann [183].Ej ffj(x'))*(a:') is a vector in TL which is orthogonal to ipj. This difficulty* can be circumvented by demanding from the beginning that 0= f dx'tf(x'){E-Ej-Hi(x?))V{x'). The theorem is also known as Fredholm alternative. But then the Green's function (7. however. In general. (7.11) with a system of linear equations of the form Vi = Mijyj. J.e. In this case the perturbation is restricted to that subspace of the Hilbert space which is orthogonal to ipj. The time-dependent Schrodinger equation ih-\><l>)t = H\il>)t (7.. that (E . Booss and D.12) Equations of the form of Eq. Examples in various contexts have been investigated by L.6) is not defined.t) for the complete problem as follows. . But if we assume that E^ ^ Ei for all i. (7. i. 17.6) we would have (7. see E. Miiller-Kirsten and A. D. Green's Functions Can we add on the right hand side of Eq.

We can write this expression also as§ ip(x. e.Q) = 6(x-x').The solution of Eq.51) shows that we obtain a very analogous expression for the density matrix PN(P) a s for the Green's function K(x.0). E.42)) Wi=0 = E l ^ ) ^ l ^ = o n (7-17) Thus at time t = 0 the state \ip)t is a linear superposition of the vectors \En) with coefficients (En\i/})t=o. x'.t) = where if(x^'.2 for the computation of the sojourn time. <a#) t = ]T n J [dx'(x\En)(En\x'){x'\^)t=0eE^\ (7. 1 See e.22) According to Eq. p.7.x'. .16) As usual.t) describes the evolution of the wave function from its initial value ^(a^O). this relation provides the probability density \ip(x.x'.7) with Eq. (7.19) i/>(x.17) is the initial condition of \4>)t.t) = ^ e £ " f / « V n W < ( ^ ) . we assume that the set of states \En) is a complete set of eigenvectors of the Hamiltonian H (in the energy representation or energy basis) so that (cf.21) dx'K(x.. H\En) = En\En).0)eEnt/in.20) is the time-dependent Green's function. i. 2nd ed. 7.' which obviously satisfies the initial condition K(x. x'. (7.e. (7.t) is known. 109 (7. Merzbacher [194]. \En)(En\ip)t=oeE^\ n (7.2 Green's Functions: Time-dependent and Time-independent permits stationary states \En) defined by \1>)t = \En)eE^ih. (7. Comparison of Eq.15) which contains (7.g. (7. *>0. for the oscillator potential.20) the Green's function K(x. (7. Eq. t)\2.158.t) = J2 fdx/^n(x)^*n(x')^(x'.17) as initial condition is obviously W)t = Y. in different formulations.x'.g. We use this in Sec. (7. (6. (5. t) with the Note that when K(x.t)il>(x'. t > 0.5.18) or.

x') and K(x. t) = J2 e £ " f M i ( x ) < ( x ' ) . we obtain / W : = . between (with E^0' = E) G(x. x')6(t) (7. x'. As a consequence of the above considerations one wants to know the connection between the time-dependent and the time-independent Green's functions.E . 7.t) K(x.x'.ie (7 .t). 24 ) - ReE Fig.23) along the contour C in the plane of complex E as shown in Fig. Inserting for GE+ie the expression above.x'.1 The contour of integration. = J2 ^„(s)C(*') En — E We see that G = GE depends on E. we obtain then also the corresponding density matrix.1.x') = GE(x.e. (7.«^|«2 W . 7.i /f e «/.110 CHAPTER 7. With Cauchy's residue theorem we obtain I(t) := J2eEnt/ihMxWn(x'Mt) n = K(x. t > 0. In the following we shall derive the Green's function for the case of the harmonic oscillator. Green's Functions difference that (5 = 1/kT plays the role of it.25) in agreement with the time-dependent Green's function . Jc 27T „ hn . i. We therefore consider the following integral with e > 0: I(t) := -i J ^eEt/*hGE^e(x.

0) = 5(x-x').3 The Green's Function of a Free Particle The time-dependent Green's function of a free particle which we now derive is an important quantity and will reappear later in Feynman's path integral method. which is moving in one space dimension.X>. v ' 2m 0 v h i. and identifying coefficients of the same powers of t on both sides.x'.26) This is the case of a free particle with mass mo.— d2 —2K(x.7).e.27) An equation of this type — called of the type of a diffusion equation — can be solved with an ansatz.x'. . A and B being constants.x>. i.30) into Eq.29) tJJ 2 2B(x .0) = 1. (7.27). (7. (7. B m0 n _ 2ih 2iK (7. K(x.t) = ^e-B^-X In this case we have dK ~dt and dK dx 2 8K dx2 A A 2t3/2 A tV2 '^'K (7. We therefore consider first the simplest case with Hn = P 2mo h2 a2 2mo dx2 (7.x1) t -B(x-x')2/t -B(x-x')2/t 2AB 4AB2(x-x')21 ~¥l 2+ W 2 (7.29) and (7. (7.30) Inserting Eqs. In this case the Green's function is the solution of the equation d h-K(x. 2mQK h h2 ih(AB) = --—{4B2A).31) The constant A has to be chosen such that K(x. / dxK(x.t).28) B(x-x')2 t2 D-B(x-x') 2 /t (7.e. It is clear that it is nontrivial to solve an equation like Eq.7.x'. Thus we try the ansatz.3 The Green's Function of a Free Particle 111 7.t) t h2 = . we obtain *I-£I = h2 -—(-2AB).

x'.zEnt. (7.e.36) |^ W Q ^ J e-mo(i-i') 2 = —eP2lia 27T dke~^k-^2^2 J„00 /2«tS .t)= f^em2t/2m0iheik(x~X')_ J 27T (735b) We set a = i-—. t) = /jJo_ e -mo(«-«') 2 /««.(7.36) 2mo Then — provided that the parameters assume values such that the intergral exists — K{x.0)oce"Qx2+ifcox."737) \27rii/i in agreement with Eq.x'-t) = Y.x'.38) . we have This is 1 provided A=JB= ^ (7 . we > have to make the replacements V ^ fdk. *>0. (7. p.t) — for instance for a wave packet given at time t = 0 of the form ^(x. n For a free particle moving in the one-dimensional domain \x\ < L — oo. from K{x.H We can insert the expression for K(x.33). (7. 1st ed. Merzbacher [194].. (7. "See the excercise in E. 32) It follows that x'. i.oo 27T (7.21). Green's Functions For parameter values such that the following integral exists. Mx)^M^) „ n so that J =^ .ihMxWn{x').112 CHAPTER 7.35a) K(xy. 2m0 (7. (7. v27r En^^-.34) K(x.33) V Lirvnt Can we demonstrate that this expression can also be obtained from Eq. j3 = i(x-x').20) and can then obtain ip(x. 158.t) = — / dke-ak2+Pk 2vr J.t) into Eq. (7.

t).25) — in the context of Feynman's path integral method. x'. mQLu .42) . . x'. x'. We consider the one-dimensional harmonic oscillator with Hamilton operator HQ. 1 2mo u>n ox n with the initial condition (7.e. (7.40) Then Eq. z . -§jK(x. —ih—K(x.t) 1 + -m0u2x2K(x. i.t). (7.t) = h2 2 d2 TW .x'..e.x'. I (7.33).8).7.e.7) is the solution of d ih—K(x. the relation* 6(x) = |o|<y(os).d r.x'.t) + -~-x K(x.t) We now set h2 d2 = .4 Green's Function of the Harmonic Oscillator The next most obvious case to consider is that of a particle subjected to the harmonic oscillator potential. .x'.39) In this case the time-dependent Green's function K of Eq. ..3 The Green's Function of the Harmonic Oscillator 113 The result (7. K(x. (21./ ) + fK(x. (7.x'.X . . 7. / ) (7.——K(x. HQ = ^—p2 ZrriQ + \m0uj2q2. We rewrite this initial condition in terms of £ and use for a = const. i. -—-KK(X.37) will later be obtained by a different method — see Eq. (7. (7. . / ) = ~^K(x..43) so that In the domain of small values of x (near the minimum of the potential) the Hamiltonian HQ is dominated by the kinetic energy.e.0) = 6(x-x') at / = 0.40) becomes 2 . i. 5{ax) = ± J eikaxdk = ArS(x).x usn ot i. . in this domain *Recall that S(x) = ± f eikxdk.

Integrating Eq. (/ = 2a.114 CHAPTER 7. b' = -Aab. as the limiting case / — 0.Z'..^ .50a) we obtain (7. so that for / — 0: > a^—. W ) .x'.e. we obttain a' = l .4a 2 )£ 2 . Feynman [94]. (7. (7. c(0)-|^.b2. 50. Identifying coefficients on both sides.4 a 2 . Green's Functions to the particle behaves almost like a free particle.50a) (7. (7.e.c o t h 2 / .48).33). (7. (7.47) We insert this ansatz for K into Eq.48) ocexp[-{a(/)C 2 + 6(/)e + c(/)}] (7. we must have /o = 0. i. K { ^ f ) ^ ^ e ~ { ^ ' ) 2 / i f f r ° ^ ' (7 -45> The same approximation is also valid for large energies E and for t or f small (near zero) in view of the relation AEAt ~ h.b2.45) in accordance with (7.) a'£2 + b'i + c' = (1 .t) become similar to expression (7. . that is. f i.50b) (7. and we expect K(x. P. (7.42) and obtain the equation (with a1 = da/df etc.45) in this sense. it • is suggestive to attempt for K the following ansatz^ m.46) If we interpret Eq.51) See R.f) with a(0)-»-y.47) becomes (7.49) a=icoth2(/-/J0 ) w 2 "' 2tanh2(/-/0) * To ensure that the expression (7.4a&£ + 2a .50c) (7. a = . p.

(7. c(0) = £' / 4 / .27).5.In B. (7. (7.48). 2TT^ Inserting a(f).* With this result we have another important quantity at our disposal. x .50c) yields for c(/)./3) = J IJIQUI 2irhsmh(hLu/kT) 2hSinh(hw/kT) \{X +X j C x exp ° S n kT lXX J (7. as one can verify. x'. K(x. (5.0).e.x'. 7. In order to satisfy Eq. we can use K(x.£) _ | {(x2 + x /z ) cos tot — 2xx'} 2msm(o. / —————rexp y 27rmsin(u.55) 'For an alternative derivation and further discussion see also B.40) of the time-dependent Green's function with Eq. p. with A. and to ensure that we obtain the prefactor of Eq.7.e. Felsager [91].33) for the Green's > function of a free particle.X'.c(f) K = 5 =exp Vsinh 2 / into Eq.51) for the density matrix PN(X. 174.50b) Kf) = —r^77) smh 2 / A independent of / .53) or. x') of the density matrix p^ (with respect to the canonical distribution with (5 = 1/kT): pN(x. (7.b(f). . we must have (besides A = —£') = /m 0 u. Finally Eq.47) we obtain ^ c o t h 2 / + ^ + ^ c ° t h 2 / (7.54) For t —• 0 this expression goes over into the expression (7.t) = . c(/) = i ln(sinh 2/) + ^ 2 c o t h 2 / . 115 To ensure that in accordance with Eq.i) (7. we must have A = -('.3 The Green's Function of the Harmonic Oscillator Correspondingly we obtain from integration of Eq. B independent of / . i.2. if we return to x. (7.x'.48) 6(0) = — £'/2f. (7.45). t. in particular for the derivation of the sojourn time in Sec.t) to obtain this element (x. (7. (7. Comparing the Eqs. i. as we shall see in the following.

at temperature T): {A) = Tl{pA) = ^ > . Eq. Feynman [94].56a) {q2) = = Y. we verify the relation: (Q2) ~ dxdx'(x\p\x'){x'\q2\x) = / / dxdx'(x\p\x')(x'\x) x2 x){x\p\x')(x'\q2\x"}. )x Inserting into Eq. Thus the system is in a mixed state and the expectation value "Cf. P. (5. e-P E i = i + y.» l. 52.e-ft >-E*) E (7 57) ' of the number of systems of the ensemble occupies the quantum mechanical state i. = jdxp(x.p)dx W ) = —^r = —p .x.e. For instance we have with Eq. (5.uij>0 ->• 0..e.x. (7.56a) the expression (7. Green's Functions With this expression we can evaluate (cf. we skip the algebra here. we obtain § < Z> = ^ c o t h ^ ^ ° — .116 CHAPTER 7. (5.x.P) 2.. ^—• Irpiv J pN(x. Eq.40)) 1 e-pEi 1 Ui = y. .55). <2 (7.52)) the expectation value of an observable A in the canonical distribution (i.56b) w v ' 2mQu} 2kT 2m 0 u ' What is the meaning of this expression? At temperature T the fraction (cf. [ [ fdxdx'dx"(i\x}(x\p\x')(x'\q2\x")(x"\i) Idxdx'dx"{x"\i){i\x){x\p\x'){x'\q2\x") J2 [ [ i = fjf^»(x>)(xW)(xW) dxdx'dx"5{x" i.p)dx Thus for (q2) = Tr(pq2) = ^2i(i\pq2\i) we obtain: 2 (7. 11 For T ~+ 0: u>0 .52): Tr (PNq2) Jx2pN(x. p. R.

and we replace to by LO — ie. x'.3 The Green's Function of the Harmonic Oscillator 117 (7. This is the first and hence dominant term of the expression (7. which means in the oscillator state \i) with eigenenergy fku(i + 1/2).54) is K(x.54).x'.56b) is that with respect to this mixed state (whose cause is the finite temperature T). 2 + x'2)l-e^-2xx' to—>w—it t—>co 2 . We return to the Green's function (7. i.J (w — ie)< e -(n+l/2)Et e -i(n+l/2)u. We assume t > 0 and t —> oo. t) = J2 eEnt/ihct>n{x)<t>n(x')For t > 0 and En = (n + l/2)frui the factor exp(Ent/ih) exp is (7. <*2>o dxx a •K j e IT I da r dxe-ax *\1/2 da a J _ 1 _ _ h 2a 2mou j —( .t "With the normalized ground state wave function of the harmonic oscillator given by Eq. (6.t) w—>UJ—ie to—n t—>oo moco/irh ^ ' piuit g—iwt exp m ^\(x 0 ^ .50) we obtain. of K(x. /2N 2~7T: = eE°/iht(t>Q(x)(t)0(x') for £ > 0.e.59a) E 0 = ^fiw. 1 mo^ exp lmow / 2 2 ft X (7.21). e > 0. (7.58) Next we explore the connection between the explicit form of K and the latter's expansion in terms of a complete set of states. If we consider the system in the pure state \i). the expression for (q2) would be:'I (fti = J(i\x)x2{x\i)dx j(i\x){x\i)dx (Q2)o = h 2m. In this case the Green's function K(x.59b) n+ l \ ftwt here 2hf exp _ i ( n + . x'. setting a = mooj/h.7.QUJ (7. t) of Eq.

Considered classically. i. In the following we want to calculate (more precisely estimate) with the help of the Green's function the time interval T which a pointlike particle can stay at the maximum of the potential before it rolls down as a result of the quantum mechanical uncertainties.1 W a v e packets The simplest type of wave is the so-called plane wave or monochromatic wave of frequency UJ represented by the expression exp[i(k-r-wt)].5 The Inverted Harmonic Oscillator We encountered the inverted harmonic oscillator already in some examples. 7. (7. as in Eq.1 we estimate T semiclassically.e. (7. the velocity of planes of equal phase. a particle placed at the maximum of the inverted oscillator potential (which is classically a position of unstable equilibrium) will stay there indefinitely. and with this we estimate the sojourn time T. In Example 7. lie on surfaces.118 CHAPTER 7. We first introduce the concept of a wave packet and then use the particular form of a wave packet in order to describe the state of the particle at time t = 0. (7. to infinity) the contribution with n — 0 dominates. Green's Functions For t large (i.e. of dpN as pN(x. The wavelength A is given by The phase velocity v<p.x'. the particle will stay there only for a finite length of time T. |k| = k. i.5. In a very analogous manner we obtain the solution of the equation for the density matrix. which are planes. is defined as vv = .62) .e.59a).61) The word "plane" implies that the points of constant phase <p := k • r — uit at t = const. However quantum mechanically in view of the uncertainties in position and momentum..(3) = J2^EnMx)K(x') n „ ^ ° e-^MxWo^')- (7-60) 7.

(7.Q ( k . for which \ip\ assumes its largest value: OO .)-¥'(*:")) +ism(<p(k') . duo' da .)ll/(fc")l[cos(¥'(fe. i. e .4 The Inverted Harmonic Oscillator 119 Every frequency u belongs to a definite (particle) energy E (cf. i. x-t— + — = 0. or dco' da The centre of mass determines the particular phase. The wave packet describes a wave of limited extent. ' t) dk'. We now ask: How and under what conditions does the time variation of the function ^(r.e.65) Let f(k') = \f(k')\eia and <p := k'x . essentially again a Gauss curve. i.Jt + a. e. t) is the spatial Fourier transform of this Gauss distribution — as we know. to the function ip given by oo / fik'y^'x-^dk'.<p(k"))}dk'dk". / -oo J—oo . /-OO / / / oo /*oo \f{k')\\f(k")\ei^W-^k"»dk!dk" l/(fe. A wave packet is defined as a superposition of plane waves with almost equal wave vectors k. ^ ( r . -oo (7. a > 0 .7. then ^(r. (7.e. t) describe the motion of a classical particle? For reasons of simplicity we restrict ourselves here to the one-dimensional case. u = w(k).64) where / ( k ' ) differs substantially from zero only near k' = k. One defines as centre of mass of the wave packet that value of x for which d(p — = 0.k ') 2 . i ) = /*/(k')e i(k '" r -'' .63) The relation u = w(k) is known as dispersion or dispersion law.e. the fundamental postulate on matter waves in Chapter 2): E = hu. If we assume for / ( k ' ) a Gauss distribution.

i.69) In general one uses the theory of functions for the evaluation of this integral. (7. a > 0. The three-dimensional generalization is evidently du .1: Fourier transform of a Gauss function Calculate the Fourier transform of the spatial Gauss function e . Solution: The function to be calculated is the integral dxe-ax -OO ezkx. however. we can use a simpler method. (7. Differentiation of g(k) yields g'{k) = / dxe~ax -2a y_oo 7 f°° 2 {-2ax)eikx = 7 2a / f°° J-x dx With partial integration we obtain from this 7 f°° 2 k g'{k) = — / ikeikxe-a* dx = . as claimed for ||=0.120 CHAPTER 7.. dE v 9 = .^ g=r a rad.e. = — (7-68) It is instructive to consider at this point the following examples. p = hk the group velocity is equal to the particle velocity v. Green's Functions This expression assumes its maximal real value when <p(k') = ip(k") = const. we obtain the de-Broglie relation p — hk.67) The centre of mass moves with uniform velocity vg called group velocity of the waves exp[i(kx — cut)]. Example 7. Thus the expression g{k) is solution of the following first order differential equation g'(k) + Aff(fc) = 0.— g(k).fc w> vv = ~a^ = S r a d P E «i7 Sd u . . In the present case. . For E = hio. 9 cku dk dhw dhk dE dp dE dp' We can also argue the other way round and say: By identifying vg = v.

7 °) With the help of this example we can obtain some useful representations of the delta function or distribution as in the next example.3: The uncertainty relation for Gaussian wave packets For the specific Gauss wave packet of Example 7.75) ' E . Since /•c 121 9(0) = J—00 dxe J_ we obtain S(fc) = ^ e " f c 2 / 4 a ( 7 . J. or see H.43) one can verify the following important relation* 5[{x -a)(xb)] = T—!-— [S(x .70) to verify the following representations of the delta distribution: -X2/€2 5(x) = lim ?—=e ^ o eV7r and 5{x) = \\m-^r^. (7.70) we obtain 6(h) = lim — f°° dxe-ax2eikx = . g . 2 f + . .73) 27T 7 .Afc = 81n2. \a — b\ (7.69) and (7.a) + 6{x . Miiller-Kirsten [215]. XZ (e > 0).2: Representations of the delta distribution Use Eqs. (7.72) Solution: From Eqs.1 verify the uncertainty relation Aa.7. (7. We have — 1 / * ° ° dke~eWeikx = - I / " 0 0 dke~tk cos kx = - 1 . (7. Appendix A. —±=e-k*/ia2 / 4 " = lim 1 — — lim ^e-* = The second important example can be verified by immediate integration. W...69) and (7. with partial fraction decomposition. where c = 9(0) is a constant.0 0 7T 7 0 7T e From Eq. Example 7..71) (7.4 The Inverted Harmonic Oscillator Simple integration yields g(k) = c e " f c 2 / 4 a .&)] for a + b... (7.r „• With the help of Eq.74) Example 7. (7. (7.73) we obtain the requested representation of the delta distribution with 1 f°° 1 € Six) = lim — / dke~e^eikx = lim .

where | / ( x ) | = m a x | / ( x ) | / 2 . The breadth A/c of the curve g(k) around k = kg.77b) (7. The wave function corresponds to the function f(x) in the above considerations (e. i. i. In quantum mechanics the square of the modulus of the wave function ip(x. (7.2 The Gaussian curve.e. 7.76) 2ira The uncertainty Ax is defined to be the width of the curve at half the height of the maximum.e. requires according to Eq. Thus a sharp maximum of the function f{x).77a) small values of a.70) the Fourier transform of f(x) is (7. when Ax is very small. Physically . at time t = 0). the width Ax is a measure of the uncertainty of the probability. Green's Functions 1/(27i)1/2a — 1/2(2. Thus a slim maximum of the curve of |/(a. is Afe = 2 V 2 1 n 2 - (7.t) is a measure of the probability to find the particle at time t at the position x.CHAPTER 7.2 we sketch the behaviour of the Gaussian function p—x 11a ~—ikox (7.g. (7. where g(k) = max|<?(fc)|/2.* o ) 2 / 2 . According to Eq.78) It follows that the product of the uncertainties is AxAk = 8 In 2.r)1/2a Re f(x) Ax Fig. A simple calculation yields A i = 2v/2~In2a.« 2 ( * . 7.)| leads to a very broad maximum of the Fourier transform \g(k)\.77a) g(k) = e . and hence implies large values of Ak. Solution: In Fig.

7. which means all values of the momentum are equally probable. In the reverse case a sharp localization of the particle in momentum space implies a correspondingly large uncertainty of its spatial coordinate. Barton [15]. Estimate the maximum length of time compatible with quantum limitations before the pencil falls over. Barton recalls that W. Barton mentions at the beginning of his paper. <7*» We assume that initially.. (7. (7. The function g(k) is therefore described as momentum space representation of the wave function f(x). many supposedly elementary problems request the calculation of the sojourn time of a quantum system near a classically unstable equilibrium configuration.* We obtain the Green's function K{0 for the inverted harmonic oscillator from Eq.«> = / ^ C ' W . O ) . ^In Eq. As G.81) *G.54) with the substitution ui — iu. The result (7. the less precise is the determination of its associated momentum.79) According to Eq.75) implies therefore: The more precise the coordinate of a microscopic particle is determined.71) > and (7. Lamb at Oxford set the following problem in an examination in 1957: "A pencil is to be balanced so as to stand upright on its point on a horizontal surface. (7.20) the wave packet at time t > 0 is obtained from that at time t = 0 with the relation «-.76) e-x 2 /2a2 ^2na a->0 um 1/0*01 = lim a^O . (7. . (7. Thus G. In the limit a — 0 we obtain from Eqs. For reasons > of simplicity in the following we set in addition t mo = l.5.. i.2 A particle's sojourn t i m e T at t h e m a x i m u m The very instructive topic of this subsection has been explored in detail in a paper by Barton. This means the particle is localized at x = 0.". so that Ki0(x.54) these parameters appear in the combinations moui/h (dimension: l e n g t h .7.e.u> = l.t) = V 27Ti s i n h t exp 2 sinh t {(x2 + x'2)cosht-2xx'} . In the same limit the momentum uncertainty Afc grows beyond all bounds. at time t = 0.2 ) and U)t.h = 1. E.x'.4 The Inverted Harmonic Oscillator 123 p = hk is the canonical momentum associated with x. the wave packet has its centre of mass at the origin and has the following Gaussian shape as the pure initial state: iP{x^) = ^L=e-x2l2b\ (7. nr- = S(x). Drastic idealizations are then required in order to reformulate the classical situation into that of a quantum system.

so t h a t (this defining C(t) and the imaginary part ip on the right) -{B2 + icotht) sinh 2 1 + (i/26 2 ) sinh 2t ( l / 6 ) s i n h 1 .81) and (7.82) i/f(x.t) = J oo P .t) = J oo dx' Setting A2:=±we can rewrite ip as ex b2 and AB sinh t' (7.cosh t sinh t' Wisll2b ^ (7. A .124 CHAPTER 7..t) = A\/2msmh 2TT 2 /2C2+i<px2 tVir^b In t h e further calculations it is convenient to set t a n 29 = b2cotht Then ~A 4 and VA + iB := ReiB (7.84) dx' oo tvirl/2b With some algebra one can show that -(B2 2y and A2 sinh 2 t f7 s ^ sinn / + icotht) = . = . . (7. (7.79) into (7.84) we obtain* ip(x. ' 2\Aswht % cosh t (7. 9 (7.85) = —r= o2 7 2 sinh 2i.83) ip(x. + cosh 2 1.88) .-(Ax' + Bxf \/27risinh + -(B2 + ^cothi)a.86) Evaluating the Gaussian integral of Eq.(i/2) sinh 2t 2 2 2C 2 (i) + *c^..83) sinh 2 t + (i/2b2) sinh 2i 2 sinh £ 2A 2 sinh 2 1 (7.87) <9 2 „9 2 sinh21 i? = A + B f / ^ d x e " ^ / 2 " 2 = v^F<*.2 (7. Green's Functions Inserting (7.80) we obtain „/2 oo ex P 2 s i n h t { ( : C + x')^ht-2xx'} \/27usmh .

7..92) The sojourn time T is now defined as the mean time T given by T:= r*° i-oo 2 (M)s r*(-f4 d^e. We have therefore .90) with the expected limiting behaviour \im \ilj(x. C2(t) (7.4 The Inverted Harmonic Oscillator The real part of Eq.t)\' 2 125 (7. (7. •&(t)=//c(t) = / or.t)\' 2 exp[-x2/b2] rrVaft In the following we choose 6 = 1 .t)\2 J-i =-= 71 / V " Jo dO ? 6' (7.94) / dxf(x)=g'(y)f(g(y)). t) the phase ip drops out and one has (x. _2 1 exp C 2 (i) = l + 2sinh 2 i.t)= where Co(t) / dx\^(x.95) .91) C(t)' (7.89) = exp[-x2/C2(t)} 7rV2C(t) (7.86) has C2(t) := b2 cosh 2 t + 72 sinh 2 1. In taking the modulus of ip(x. The probability for the particle to be at time t still within the distance I away from the origin is plausibly given by Q(l.-£ 2 (7.93) where —{dQ/dt)5t is the probability that the particle leaves the vicinity of the origin in the interval St around t (partial integration leads from the second expression back to the first). since dy Jo the derivative dT_ ~dT For b = 1 we have f°° dt-= / (7.

+ 0(e-'2) (7.98) We insert this into Eq. Dwight [81].=o V(i -v W + v2) o-V "dT J L 1 f A -x 2 __H_I r'=s/21 Adx'-. we obtain dT _ 1 ~dJ~l 'See for instance H.97) and obtain dT ~dl or with £ = ?y/7: 2 2 d^e.2)(p_7?2)(7. (7. C(t) = .=0 V^ C'(t) I Jt=0 V^ (7.e ~' V^~i Jo Vn~i Jo ^ rx=VM _Jl 2 Using the following expansion of the integral* _ _ r e~*' dt L 2 « 1 2e~*2/2 7T X /2^7-.= \ / l + 2sinh 2 t.96) °° l Jr.(4 2 . B. 136.. £ 1 dr\ 2Z %_ rl 2 2 yfx Jr.99) . V and (cosh t = y 1 + sinh2 t) 1 r CHAPTER 7 Green's Functions dr} dt C'lt) ' C(t)' (7.From this we obtain C(t) C'(t) Z2 ^(r?2_.97) sinh 2 1 = . p..1 ) 2 \n J ^-"•""ff^-^-V^K?.126 We set I C(t)' so that dT ~d~l Now.

Solution: In the case of the inverted oscillator (representing the egg) we have in the usual notation the classical Hamiltonian Hwith time derivatives dH dp P mo ' max = p dH dx = mow x. p. 2 2 2mo -mow x The classical equation of motion is therefore' it — uj x = 0 (7.4 The Inverted Harmonic Oscillator 127 for I sufficiently large. Fig. 238.4. this equation is the equation of motion of this scalar field with negative mass-squared.4: The sojourn time calculated semiclassically A tiny ball of mass mo is placed at the apex of an upright egg. See e. 7. Calculate with the help of a semiclassical consideration the (quantum mechanically limited) maximal length of time which passes before the ball is observed to roll down. T h e following Example 7.g. this is T ~ -]n(ly/mou>/h). (7. Such a field arises in the spectrum of states in string field theories and is there called a "tachyon". It follows t h a t T~lnZ.100) Re-introducing the dimensional parameters we had set equal to 1. also motivated by this paper. Zwiebach [294]. U) (7.3 How stable is the particle on top of the eggshell in quantum mechanics? E x a m p l e 7. .102) 'As a matter of interest we add that in the theory of a free scalar field x.7.101) A more detailed evaluation of the constant can be found in the paper of Barton cited above. The equation there describes the classical "rolling down" of this tachyon. is instructive in revealing the basic quantum mechanics involved in the finiteness of the sojourn time. Explain the parameters entering the calculation. B.

(x 2 ) h 2mou 2mou> and so l^J^ULl). but instead are subject to an uncertainty relation of the form AxAp > -h. / 7 ^ 2 =2J(x }. i. A p are defined by the mean square deviations given by (see Eq. ^(x2)(p2). (10.g. Green's Functions x = Aco sinh cot + Bco cosh cot. Ap= v /(p 2 )-<p)2. For a symmetric state like the usual ground state of the harmonic oscillator we have (since the wave function is an even function. and solution x = A cosh Lot + B sinh cot. Eq. the expectation value of x is the integral with an odd integrand) <x) = 0 = <p). 2 10 x(0) = B u . 2 x(0) ^ where A i . For h —> 0 the time T —• oo in agreement with our classical expectation. We ask: At what time T > 0 does the ball reach the point with horizontal coordinate x = I. so that raoto _ We assume now that at time t = 0 the ball is placed at the point x(fi) with momentum p(0). so that x(0) + il^-L mow It follows that For a minimal uncertainty (hence the factor of 2 in the following) we then have -h= 2 Hence uT P ( ° ) _ O . x(0) = A. the smallest macroscopic length. Quantum mechanically x(0). pifi) have to be replaced by the positions of the spatial and momentum maxima of a wave packet.103) to VV h Here I is a largely arbitrary but macroscopic length like the length of the power of resolution of a microscope — so to speak. (7. Thus we set I = A cosh LOT + B sinh coT ~ -(A + B)e"T.e. (7.102)) p(0) = ^/(p 2 >.128 with x2 = LO2X2 + const. (p2):=mW(x2). 2V ' Let A and Bio be the values of x and x at time t = 0. i.p(0) cannot be determined with arbitrary precision. so that e. v A x A p = \ (x2)(p2) v = m0u(x2).e. CHAPTER 7. so that AxAp= In our semiclassical consideration we set therefore x(0) = y/{x*). Thus quantum mechanically x(0). and for simplicity we take (cf.5)) Ax = <J(x2)-{x)2. .

in closed form) only for very few potentials. EW+(3EW+(32EW + --- ((3 can also be thought of as a kind of "book-keeping" parameter in retaining corresponding powers of some kind of expansion). In general perturbation series do not converge.2h2 cos 2 a # = 0. An example permitting convergent perturbation series is provided by the trigonometric potential cos 2x with one-dimensional Schrodinger equation given by ip" + [E. For mathematical purists the question of convergence of the series is an even bigger challenge.Chapter 8 Time-Independent Perturbation Theory 8. one would set tt E = = ^(°)+/ty.e. The perturbation method generally described in textbooks — and frequently called Rayleigh-Schrodinger perturbation theory — consists in assuming power series expansions for the wave function * and the eigenvalue E in terms of a parameter like (3 which is assumed to be small.(i)+/^(2) + -".1 Introductory Remarks The Schrodinger equation can be solved exactly (i. i.e. but this is no longer the case for an anharmonic oscillator potential like ax2+(3x4.E^ is already a bigg problem. In the case of the harmonic oscillator potential ax2 the Schrodinger equation can be solved exactly with ease. Frequently even the calculation of the next to leading contributions ip^. It follows that in general one depends on some approximation procedure which is usually described as a perturbation method. 129 .

This convergence can be shown with D'Alembert's ratio test since''' lim . It will be seen later (e.2 Asymptotic Series versus Convergent Series Before we actually define asymptotic series we illustrate some of their characteristic properties by considering specific examples which demonstrate also how they differ from convergent series. Time-Independent Perturbation Theory In the case of this equation.3) *See e.130 CHAPTER 8. concerning also the uniform convergence of the exponential series. E. (8. the expansions in ascending powers of the parameter h2 can indeed be shown to have a definite radius of convergence. Meixner and F . N. (8. which is known as the Mathieu equation. p..• + (-ir~ X X* 1' 2' r?l Xn + Rn(x).. E = E^ + h2E^+hAE^ 7-1 72 7 + --. 8. Actually expansions of this type which are ubiquitous in physics are so-called asymptotic series which were originally also described as semi-convergent series in view of the decreasing behaviour of their first few terms. e. perturbation theory and path integral methods — that the light-minded way in which perturbation theory is sometimes discarded is not justified. J.g.. the expansions in descending powers of h2. Watson [283]. 581.g.g. W. Whittaker and G. at the end of Chapter 26) — and by comparing the results of WKB. In the following we explain the difference between convergent and asymptotic series. The series n=0 of the exponential function is well known to converge absolutely for all real and complex values of x. T.* i. explore the latter in somewhat more detail and finally consider methods for deriving perturbation solutions of the Schrodinger equation. 0. h hl are in the strict mathematical sense divergent.g. .+ .= 0 < 1.2 However.1) and its terms with the behaviour of the following series and its terms: f(x) = 1 . Schafke [193]. Cll E = a_ 2 n + a_i/i + a0 + — + —r -\ . . the expansions i> = v (0) + /*V (1) + /*V 2 ) + • • •.e.2) n—»oo n It is interesting to compare the behaviour of the series (8. ^For further details. see e..

we observe that the individual terms of the asymptotic series have the form of a factorial divided by the power of some parameter which is large. and in fact increase indefinitely.222. .. the better the approximation obtained. This type of behaviour is characteristic of the terms of an asymptotic expansion as we shall see in more detail in Chapter 20. .5) Cf.0.g. As a second example we consider two series expansions of the gamma function or factorial T(z) — {z — 1)!. Normally a divergent series is characterized by an ever increasing behaviour of its terms as in the case of x = 1.0. for larger values of x. / ( I ) = 1 — 1! + 2! — 3! H • However. the series (8. Watson [283]. Whittaker and G.1)...3) with the convergent series (8. + 0.. N. It is evident that the larger the value of \x\. .29629. Thus e. T.001 + 0. We observe in the first place that since lim n—>oo co (8..22222 .g.22222 .2 Comparison of Asymptotic Series with Convergent Series 131 where Rn(x) is the remainder sum.33333 . E.000 002 00 . the partial sum of terms up to and including the least term yields a reasonably precise value of the function at that point with an error of the order of the first term of the remainder.. in spite of this the series (8. /(1000) = 1 .2. and we see that the moduli of successive terms first decrease and then.8.4938. +0.3) can still be used to obtain almost correct values of f(x) for x sufficiently large.000 000 06 + • • • . for x = 3.9876. after reaching 6/27 = 0..0. e. . i. we have ^ ' ~ = 3 + 9 ~ 2 7 + 8 1 ~ 2 4 3 + 7 2 9 ~ 2187 + " ' ' 1 . From the Weierstrass product which defines this function* one obtains the series _ M*-l)! = -7(*-l) + E [ ^ . 235.. Comparing the asymptotic series (8. ..l n ( l + ^ ) n=l L ^ ' (8. However.e. .. begin to increase again. p..0. We can see this as follows..3) diverges for every value of x. The theory of asymptotic expansions claims that if the expansion is truncated at the least term.376 . + •••.4) for every value of x. + 0.0.

the leading term of an asymptotic expansion. It seems that Schiff tries to cling to the idea that a proper series has to be analytic. 152. Vol. that E and \ijj) (i. This is not appreciated by mathematical purists. 236. ip) can be represented by rapidly converging power series in e. Whittaker and G. ^E. This latter observation hints already at the importance of asymptotic series in applications. Merzbacher [194]. § However. **A. T. In fact. Thus Merzbacher^ says:"Simple perturbation theory applies when these eigenvalues and eigenfunctions can be expanded in powers of e (at least in the sense of an asymptotic expansion) in the hope that for practical calculations only the first few terms of the expansions need be considered. whereas applications of the convergent series (8. It is interesting to observe the comments of various authors on this point. . Schiff [243]. Watson [283]. E. Thus one can say that the vast majority of expansions of this type in physics is asymptotic and not convergent. On the other hand Schiff" says:" We assume that these two series (for \I/ and E) are analytic for e between zero and one.5) are practically unknown. in fact. In many respects Messiah aims at more rigour in his arguments. (16. meaning convergent.1.5). the dominant approximation is. p. Messiah [195]. N." These "rapidly converging power series" in physical contexts are most likely very rare cases.1 between Eqs. although this has not been investigated except for a few simple problem^.5772157— Here the series on the right is an absolutely and uniformly convergent series of the analytic function. 11 L. p. it is sensible to make the assumption. Here the series on the right is an asymptotic series which is particularly useful for large values of z but is readily checked to yield very good approximations for values as small as 2 or so.41) and (16.J In2. It is therefore not surprising that he** says: "7/ the perturbation eV is sufficiently small. Sec." This is a very clear statement which does not try to pretend that a perturbation expansion would have to be convergent. 16. p. Time-Independent Perturbation Theory where 7 = 0. 371.132 CHAPTER 8. I. for ln(z — 1)! one can also obtain the Stirling series ln(z-l)! = ln(2vr)5 -z+ (z.e. Applications of Stirling's series can be found in all areas of the physical sciences (particularly in statistical problems). It is inherent in the nature of an approximation in a physical problem that in deciding between dominant or primary effects and those of secondary importance. at a later point in his treatise Messiah admits that the expansions are mostly § Cf.

ft Jx du = 2tdt.7) This function has been considered in detail in t h e book of Dingle.10) Vn Changing t h e variable of integration to u = t2-x2. we can rewrite Eq.ft somewhat afraid t o say so himself a n d therefore with reference t o investigations of T .2 Comparison of Asymptotic Series with Convergent Series 133 asymptotic. He says. Messiah [195]. p. We know from books on Special Functions t h a t all of these functions. Dingle [70].7) as / e'* dte-* dt = 1 . Replacing t h e exponential in Eq. . (8. seems to be closest t o t h e t r u t h a n d does not attempt t o give t h e impression t h a t t h e series has t o converge. B. (8. (8.. there is no need for such bias.^=e~x2 / e-^-^dt.. (8. It is therefore suggestive to write iy poo <t>(x) = 1 . Merzbacher. 8.1 T h e error function a n d Stokes discontinuities Another extremely instructive example which illustrates t h e n a t u r e — a n d in addition t h e origin — of asymptotic expansions is t h e error function 4>{x) denned by t h e integral <f>(x) = -7= / V71" Jo 2 fx e _ t dt.* and we follow some of t h e considerations given there.9) JO Jx J V 71" Jx The integral is expected to b e dominated by t h e behaviour of the exponential at t h e lower limit.. II.11) A . *R.-= / e~t2dt. 198 (German edition). Vol.2. 2 (8. In fact." Of these three authors t h e first.8.7) by its power series and integrating term by term one obtains t h e absolutely convergent series Considering for t h e time being only real and positive values of x. which have been studied in great detail. possess asymptotic expansions which have for a long time been important a n d accepted standard results of mathematics. (8. Kato: "Indeed the perturbation expansion is in most cases an asymptotic expansion . solutions of second order differential equations.

(8. Then e (n —\)\( n\ 2du u \n Xy/n o du VK n = 0 e 2 + XypK ~"(1+ ^ °° /•oo —u (n-i)!/ n! (nn\ 2)! u du+ Jx : e u du 1 ^ ( 1 + n=0 oo /_u_ V x2 ^ ) -K ^ n=0 (8. (8.134 we obtain CHAPTER 8.14) sin7T2. Time-Independent Perturbation Theory <P(x) = 1 .i)[(_!)«' Then.13) presupposes t h a t < 1.15) < 1 we can insert this expansion into Eq. du. 7T (n-i)!sin{7r(n+^)} ( „ .e. 2 9 f00 / Jo du e-" 2(u + x2)^ u <£(*) = 1 T h e binomial expansion . since (—4)! = y/n.e.^ X e U ~ ( 1 + -* \ 2 .12).16) . Using the reflection formula (*-!)!(-*)! = for z = n + i .' we have 2 -nH 7T 7T (8.-^e~x \Ar i. we obtain 1+ i. « \~2 =E n=0 l 1 V5F(n-i)!(-l)n nlTT (M x2) / « r oo 1+ For the domain \u/x2\ <f>(x) U \-2 1 \/K n=0 2-< n! V (8.12) VK Jo n V oo 7 n=o x) 2 2-*Qn\(-±-n)\\x2) (8.

19) originates.r|<-7r. We now want to relax this condition and allow for phases argx ^ 0.19) where "~" means asymptotically equal to which in turn means that the right hand side of Eq. (8.15) into (8.19) approximates <f){x) the better the larger x is. the asymptotic expansion of the error function is. i.e.18) or that we insert the binomial expansion (8. We can actually understand the deeper reason for this in practice. As in the case of the gamma function.l)! = / Jo Then 2 c-Hn-xdt. Whichever way we look at the result. argx = 0. the fact that we effectively use a binomial expansion beyond its circle of convergence implies that the resulting series is divergent so that even if the first few terms decrease in magnitude the later terms will increase eventually as a reflection of this procedure. It is fairly clear that the above considerations remain unaffected for |argx| < -7T.—yS!L^LL-Z e x OO / _ 1\| -X2 /-CO due-u L \ / V n = Q -1 and we write 0(X)~1 V 7 n=0 v 2^.19) implies either that we ignore the remainder or the correction term given by the integral in Eq. ' |arg. convergent) only in the restricted domain u < x2. /•oo T(n) = (n .12) ignoring the fact that the latter is valid (i. whereas in a convergent expansion the terms first increase in magnitude. i.17) 0(X) = i. The expansion (8. We now see how the asymptotic expansion (8. In an asymptotic expansion the first few terms successively decrease in magnitude.8. and in fact so slowly.e. much more useful than the convergent expansion. (8. In the foregoing discussion of the error function we assumed that x was real and positive. . (8.2 Comparison of Asymptotic Series with Convergent Series 135 We can evaluate the first integral with the help of the integral representation of the gamma function.e. (8. Frequently "=" is written instead of "~". that a large number of terms has to be summed in order to obtain a reasonable approximation of the quantity concerned. in general.

VK Jo V^ Jo (8. (8.22) we obtain 2 4>{iy) = . V7T Jo Changing the variable of integration to v (8.20) where we set t = is. =dv.2 y V . and we can write (j>(iy): y^ Jo t^Q n] \y J If we want to proceed with the evaluation of the integral in order to arrive at an expansion of (f>{iy).7) and obtain 4>(iy) = —= j e~l dt = -= / es2ds.^ f V e ^ .^ = .24) We can read off the binomial expansion of the factor in the integrand from Eq. i. We can therefore rewrite the integral as (j)(iy) i 2 fy2 y —-=e / e~v 1 . when |argx| = 7r/2.26) or else return .23) Throughout the range of integration 0 < v < y2 the integrand is real. We therefore consider this case in detail. dv = -2sds = . (8. (8. (8. we can try to proceed with Eq.21) = y2-s2.136 CHAPTER 8. however. (8. Time-Independent Perturbation Theory since for this range of phases the decreasing nature of the exponential is maintained.e.vds. Proceeding along lines similar to those above we write <f>{iy) = ^=ey2 I* e^-^ds.15). thus with \v/y2\ < 1. we set x = iy in Eq. The situation becomes critical. (8.

.17): J£° e~vvndv = n\.2 Comparison of Asymptotic Series with Convergent Series to Eq.^ ^ p . We have therefore <j>(iy) = -^e^TL fV e~v ( l .26) we can write* 137 v* Jv2 ie « „2 OO / 1 n=0 M ~k n! . (8.28) we observe that the expansion (8. we obtain 4>(iy) = -^ey2TZ f°° e~v (1 . (8. y y 2 -( n _i)! y arg(* = iy) = i * . Returning to Eqs.28) ignores the integral contribution of Eq. For v > y2 the second integral is seen to be f imaginary and thus drops out in taking the real part.24). If we proceed with Eq.28) n=0 We observe that this expansion differs significantly from (8. and so the integral is real. and the larger y.27) Then . We know that for v < y2 the integrand is real. (8.30) yvn f Jo V y J Cf. e ^ ( i y ) ^ ^ . which will be studied in more detail below. (8.29) where 5t means "real parf. (8. (8. „2 \v J v^(n-i)! y(n-j)l ^—-' y2n ie"2 /-00 ^ ieir /•" W7r /„2 y ^—' _v^{n-\)\(v n! .28) because then the correction term oc f°£ becomes smaller and smaller. These integral contributions are of order 1/y since the integral behaves like e~y and is multiplied by a factor e+y in front.^ ) * dv JO V y /»00 /»' yV^ 1 i Vsfv 2 ey n oo J0 Jy' 1%) 'dv. for | arg x\ < 7r/2 and for arg x = -IT/2 the error function 4>(x) possesses different asymptotic expansions. for large values of y. the better the approximation expressed by Eq. Eq.e. (8.24). i (8.^] * dv. — (8.2 yir (8. (8.8.27) and (8. i.e.27). i.19). However. Suppose now we consider Eq. ey is much larger than something of order 1/y..

| y.* We observe that this Stokes discontinuity is a property of the asymptotic expansion but not of the function itself. Looking at Eqs. Since (cf. (8.33)) would have • a maximally increasing exponential e+x . is obtained for <f>(x) with arg x = —ir/2.3!) which is (8. We have therefore found that XTT n=0 ^ ' and . For the sake of completeness of the above example we continue the phase to | arg x\ > IT/2. Chapter 1. Eq. (8. 7r = -. (8.31) and (8.33) we see that the Stokes discontinuities occur at those phases for which these expansions (i. With a similar type of reasoning one can show that the same expansion. Time-Independent Perturbation Theory If we insert here the expansion (8. .26).138 CHAPTER 8. and • higher order terms of the associated series all have the same sign.25) and integrate from 0 to oo.8)) 4>(x) = -<f>(-x) we see that _x2 OO / ] \ | n=0 for 7r/2 < I arg x\ < 3TT/2. we use the expansion outside its circle of convergence and hence obtain a divergent expansion. B.30) can be written </>(iy) = -^ey2 fV e-v(l-^) yVn Jo V y J "dv = ^!rV"E^(4)V (8. This result is identical with that obtained previously.e.28).2 OO rz=0 V .33) The sudden disappearance of "1" at |argx| = 7r/2 hints at something like a discontinuity of (j>{x) at arg x = n/2 which was discovered by Stokes and is therefore known as a Stokes discontinuity.. (8. since (8. (8.31). e xZ ^ (n — £)! „ ^ ) . The corresponding phase lines are called Stokes rays." — E ^ ^ for|arg. *Cf.e. . Dingle [70]. i. R. (8.

-1/-1 ipv{x) -v-1 -v-l (±ix). —1 cancel out. Dingle [70]. e. We also observe that the expansion for | arg x\ = ir/2 is half the sum of the expansions on either side of the Stokes ray.36) which is of the type of a Schrodinger equation for an harmonic oscillator potential* In view of the considerable importance of the harmonic oscillator and the associated parabolic cylinder functions in later chapters we consider this case now in more detail.32) and (8. though. .2. Oberhettinger [181]. i.e.2).37) Cf.e. Equation (8. i. R.2 Comparison of Asymptotic Series with Convergent Series 139 It should be observed that the phase of the Stokes ray is the phase at which an exponential is not just increasing but maximally increasing. B.35) 8. Eq. '"Cf. (6. Later we shall make extensive use of parabolic cylinder functions Du(x) which are solutions of the equation dx2 + v+ :X y =o .g. in the sum of the right hand sides of Eqs. 9 . See also Tables of Special Functions. and so may not be universally applicable. and 4>(x) argx=7r/2 = 2 r2 ^ + <f>(x) a. 91.34) the contributions 1. (8.2 Stokes discontinuities of oscillator functions Dingle' has formulated rules which permit one to continue an asymptotic series across a Stokes discontinuity without the necessity of a separate calculation of the asymptotic series in the new domain. W. apply predominantly to asymptotic series of the solutions of second order differential equations.rgx<n/2 argx>7r/2 _ X1T •<—' v — ( n=0 £ (n-i)! X2)n ' oo (8. 5.36) has an exponentially decreasing solution and an exponentially increasing solution for zero phase of x. Chapter 1. for arg x = 7r/2. we have e -z2/2 _ 2 e-(x R-xj+2ixRxI)/2_ The exponential is maximally increasing for XR = 0. These rules. Thus if we set x = XR + ixj. Magnus and F.13. pp. which can be written (for v not restricted to integral values) D^(x) DM(x) f xu(f>u(x).8. p.

Chapter 1.38) are of the type (2i)! i\ i\(-2x Y 2 (-2x2Y and so have the form of a factorial divided by a power which (as discussed previously) is the behaviour typical of asymptotic series.x 2 oo «*) - — E^fi=0 v . R. x~l/~1ipu(x). 1 1 Cf.38)) has a Stokes ray at Dl '(X) the arg x = 7r/2.e. (8. 0 < argz < | (8.40) § Cf. Dingle [70]. the series of D ^ x ) develops an additive contribution (the discontinuity) which is IT/2 out of phase with xu4>v{x) and proportional to the associated function. R. We consider Dv (x).37). Chapter 17.39) onto the Stokes ray and from there into the neighbouring domain is determined by the following rules which will not be established here:" • Dingle's rule (1): On reaching arg a. .e.39) (as in the example of the error function). Hence Dil){x) = xv(j>v(x).e. Thus D{J-\x) (i. D^(x) = = xv4>v{x)Jr\^l'1^'Kl'1\A~v~X^^) xv4>v{x) + \a. with a real proportionality factor. v — — f — 1. ti asymptotic series (8. as can most easily be seen in the case of the Mathieu potential. i. We also observe that the late (large i) terms of the series in (8. Chapters I (in particular p. = 7r/2. (8-38) We observe that one solution follows from the other by making the replacements^ x — ±ix. > > The equation (8. Dingle [70]. B. ' S u c h symmetries are extensively exploited in the perturbation method of Sec. B. Time-Independent Perturbation Theory i.140 where^ CHAPTER 8. 9) and XXI. i. (8.36) is invariant under these replacements.ei<v+Vx-v-Hv{x\ argx = £ .7. (14). Eq. 8. At arg x = 7r/2 the exponential factor becomes an increasing exponential and late terms of the series have the same sign. The continuation of (8.

e. is given by • Dingle's rule (2): One half of the discontinuity appears on reaching the Stokes ray.40).41) to arg a. Since Dv (x) is real when x is real.41) the Stokes multiplier a is. sm(iri/) = —a/3/2 or a/? = -2sm(irv). It is determined by continuing the asymptotic series to argx = n and demanding that the result be real since Do (x) is real when x is real.44) + ^iap) = 0 . i.41) xuct)v{x) + a{-x)~u~lipu{x). beyond arg x — 7r/2.42) el™ + ±iafi\ {-x)v<t>v{x) + a{-x)-v-l^u{x). the extra phase factor u ei7r/2 a n c j t n e p h a s e ew7r/2 0 f x <j>u(x) (so that as the rule requires the added contribution is 7r/2 out of phase with the first).42) T ^(eiwu i. Applying Dingle's rule (1) (to the Stokes discontinuity of ^v{x)) we obtain on the ray D<P(x) = x^M^ + aU-xy^M^ + l^e^e^lxrMx)] (8.8. (8.e. < 7T. where in the first line on the right hand side the second term contains the real proportionality constant a/2. Applying rule (2) we obtain D$\x) = U™ + ^iap) (-xyMx) + a{-x)-v-lMx) (8-43) for IT < arg a.e. with \x\ = (—x) for arg a. i.e.2 Comparison of Asymptotic Series with Convergent Series 141 for argz = 7r/2. we must have at arg x = 7 in the dominant factor on the right of Eq. The value of Du (x) on the other side of the Stokes ray. for xj = 0) and therefore possesses a Stokes discontinuity there. (8. We therefore proceed to continue (8. Thus for arg x > ir/2 we have D$\x) = xp(t>u{x) + = 2]iaei<v+^x-1/-l^v{x) ^ < arga. In (8. (8. < 37r/2. = ir. (8. another half of the discontinuity appears on leaving the Stokes ray on the other side. = ir. as yet an unknown real constant. e +fx2 _ 2 2 e±(x R-x I+2ixRxI) is maximally exponentially increasing (i. On reaching ir the part containing ^ ( x ) .

e.42)) for argx = ir (in the second line of Eq.e. (8. a{y)P{y) = a(-u . (3{-v .42) we insert from Eq.50a ) (aMb) »W = f "M=(=^)i In order to decide which case is relevant we observe from Eq.48) we can set <8 49) ' ^ ) = {_V\)V PM = ^f> or ( 8 . (8.v . We can see that the equation is satisfied by (3{u) = ±a(-u .47) Equation (8. i.1)! TT. Eq.51) .45) is therefore given by a{v)a(-v . or sin-7r(2: + 1) sin irz' W ( = 7 ^ ) ! = -2sil"r2Comparing this with Eq.1) (8.42) that a multiplies ipv(x).1) = -2sin7ri/. (8.1) (8.e. Time-Independent Perturbation Theory Now a and (3 can still be functions of v.—^ ON • — ^ > + ^ ^ L . a(u)P(v) = -2sin(7r^) = -2sin7r(-i/ .48) We compare this with the reflection formula (8. (8. 8. argx = vr. i. . Hence the left hand side must have the same property. i.1) = ±a(y). (8.14). V i=0 \U0 v 2w . Dl \X) is (cf.e.38) = cos7ru(-x)u(f)1/(x) + (—x)ve~*x COS7T^— —r- a(u)(-x)-l'-1^u(x) (2i . (8. (8. i.46) For a given function a(y) this equation determines /3(u).l)0(-v .142 CHAPTER 8.45) that af3 = -2sin(7r^)) D^\x) = (8.1).45) Thus the right hand side of this equation remains unchanged if v is replaced by —v — 1. (z)\(-z-iy.

) (8. Then (in the second step using (—u — l)\v\ = — ir/sm(Trv)) DW(x) ~ c o s ^ ^ * f ^ .1. i. (8.2^1 v y 7T i=0 ' (2i + v)\ z!(2x ) v (8. (-2x2)1 = COSTTU(-X) v .e. i. In a similar way we can examine Dv (x) and its Stokes discontinuities (at argx = 0.3 Asymptotic Series from Differential Equations In the case of the error function we obtained the asymptotic expansion from the integral representation of the function. (8.7).3 Derivation of Asymptotic Series from Differential Equations 143 We choose (8.35).7r).. However.. For details we refer again to the book of Dingle.A / ± sin^C-x)-"-M* 2 V ^ . solutions of Eq.*"e-i°> f ( ^ _ | _ L _ = (-!)««(.8.53) the factor (—n — 1)! is to be understood in association with (2i — n — 1)! in the numerator.2. (8. (2i-n-l)! (-n-1)! n! {n-2i)\ So far we have been considering the asymptotic series expansion of Di.. Proceeding along similar lines we can continue the expansion into the domain (2) ir < arg x < 2ir.53) We have therefore obtained in a natural way the quantization of the harmonic oscillator...52) (argx = 7r). ' ^ (-i/-l)!z! ^ .1 ) ( . i.I / .36)) in the domain 0 < arg x < IT. Eq. Chapter I.e.e^ . ' (X) (cf. —-f-. the second contribution vanishes and we obtain £>«(*) . the error . It should be noted that in Eq. which vanish at x = ±oo.36). (8.„ _ix2^(2i-v-l)l e 2 > —.)(-. We observe that for v — n = 0.1 ) ! 4~L i\{-2x2Y i=0 V2^(-x).e. (8. u 1 y> {2i + v)\ 1 n 0.50a) because this enables us to define normalizable parabolic cylinder functions.. 8.

60) into (8. 0. % = {. and demonstrate. S\.36)) 0 where + x2Q(x)y = 0.58) Since x2 is the highest power of x appearing in x2Q(x) we set S(x) = ]-S0x2 + SlX + S2 In a.59) and equating to zero the coefficients of powers of x2. x°. previously we had ^. Solving these equations we obtain = = 0.60) 2 x x l Inserting (8. we obtain equations from which the coefficients So. (8. ^ + 2 .37). Instead of dealing with Eq (8. • • • can be determined. Thus 2S 0 Si 5o + 25 0 5 2 + ^ + ^ and so on.e.e.38) can be obtained. how series of the type (8. Eq. i. x.. (8. (8. -\ + U (8. i. ^ = 0 dxA dx with the boundary conditions lim 4>(x) = ± 1 . x—>±oo (8.144 CHAPTER 8.55) The large x asymptotic expansion can also be obtained from the differential equation. Eq.56) ~l^(8 57) Q{x) = - We then set y = es^ so that the equation becomes (8. S2. Time-Independent Perturbation Theory function can also be obtained as the solution of a second order differential equation. (8.54) (8. We write Eq. (8.. 5„=4 Sl=o. cf.^ + 1 ) .36)..36) in the form (coefficient of — x2 here chosen to be ~. + — + ^ + • • • .54) we consider again the important equation of the harmonic oscillator.

In particular we shall need the asymptotic expansions of parabolic cylinder functions in various domains. p.+ X M W = Q. In the following we are mostly concerned with asymptotic series in some parameter.37). (8. a z a a ( + 1) z2 which has the unit circle as its circle of convergence.35) and (8. i. function.19).61) (the dominant terms of (8. » 1.b. and it is clear t h a t higher order terms follow accordingly.z) is a confluent hypergeometric T-. T h e expansions (8. 2 ) + V2(-|-l)! l-t1! 2 1 — v 3x\ 1 2 2' 2 where iF\(a.3 Derivation Then of Asymptotic Series from Differential Equations 145 eS(x) = 2 e±^ xS2 1+ 0 (8. The derivation of the asymptotic series from integral representations (cf.63) **W.36) have correspondingly e±x / 2 ) . therefore. (8.8.38).38) in a variable for the solutions of approximated differential equations. Magnus and F. do not enter here into their derivation and simply quote the result: W i t h \x\ . T h e parabolic cylinder function normally written Dv(x) is frequently defined via the Whittaker function WK^{x) which is a solution of Whittaker's equation d2W dx2 Thus* Du(x) = 24 + 1 K 1 + T+ . 2 X -2Wi. etc.e. Whittaker and Watson [283]) is now somewhat elaborate (although in principle the same as before). n 22e"T JziL (8.62) 4 \ 2 XiFi 2'2 . Nonetheless we shall need asymptotic expansions like (8.91. / i N -.37) with (8. 4~l~2 ' i i _i. . \x\ ^> \v\ and (a) for | a r g x | < D„{x) 3/4TT: e 4 x x i/(i/-l) 2x2 | v(y-\){V-2){y-S) 2Ax2 | . We. are asymptotic series in a variable x. Oberhettinger [181].

N. top of p. 349. If Sn(x) is the sum of the first n + 1 terms of the divergent series expansion . Time-Independent Perturbation Theory (b) for 5/47T > arga? > n/4: Dv{x) ~ e 4 v x v{y-\) 2x 2 V{y-\){y-1){y-S) 2. Thus the series expansion (a) in this case is multiplied by an increasing exponential and the asymptotic equality " ~ " means an exponentially decreasing contribution has been ignored. Whittaker and G.+ — + ••• of a function f(x) for a given range of a r g x .67) Cf.4x2 (8.0. Thus there is no contradiction^ between (a) and (b). 8. Watson [283]. whereas e~x '4 increases exponentially.64) (c) for —7r/4 > a r g £ > —57r/4: Dv{x) e 4x !-*£=£2 2x -i/-i + +• (-I/-1)! + xe 4 x (i/+ !)(„ + 2 ) 2x 2 " ' (8.66) 7 =e|(^) 2 -i(3x)2ei(^)(?Jx) decreases exponentially.4x 2 27F (-*/-!)! | e^e^x-""1 1 + (v + l)(i/ + 2) 2x 2 | (i/ + !)(»/ + 2)(i/ + 3)(»/ + 4) 2. 4 (8. . then the series is said to be an asymptotic expansion of f(x) in t h a t range if for a fixed value of n lim \x\—>oo ft xn\f(x)-Sn{x)\=0 3.65) We observe t h a t the series of (a) and validity 1 1 -7r < a r g x < 4 In this domain 'Qx > |9ffcr| and so e (b) have the common domain of 3 3 -7r. E.146 CHAPTER 8.1 a2 an a0 + — + ^ + --.4 Formal Definition of Asymptotic Expansions Finally we introduce the formal definition of an asymptotic expansion. T.

8.5 Rayleigh-Schrodinger Perturbation although Theory 147 lim \xn [f(x) . In fact. (8. (8. the divergent nature of asymptotic expansions together with a vagueness of definition (which is avoidable as explained by Dingle) frequently tempt mathematically prejudiced purists to turn away from asymptotic expansions.Sn{x)) | = oo n—>oo (8.73) Vn + q/#> + e2T/i2) + • • • . in particular with regard to possible exponentially small contributions to the expansion for which e.68) The definition(8. which is assumed to be small. The definition is due to Poincare (1866). Chapter I. lim xne-W |x|—>oo = 0 (8.70) It is assumed that the spectrum {En are known with and the orthonormality Smn = (lpm\lpn) = } and the eigenfunctions {ipn} of HQ W n = 4°Vn (8-71) / dx%l)*m(x)lpn(x). The equation to be solved is the equation H<S> = EI/J with H = H0 + eV.72) For the eigenvalue E near En and the eigenfunction \l/ near ipn we assume power expansions in terms of the parameter e. 8. first published in 1902. (8. . possesses a whole chapter on asymptotic expansions. we set E^En * -> ^ n = En°1+eEnV = + e2EnV + ---.69) and the consequent nonunique definition of the expansion.g. A critical discussion of the definition can be found in the book by Dingle [70]. i.5 Rayleigh-Schrodinger Perturbation Theory Rayleigh-Schrodinger perturbation theory is the usual perturbation theory which one can describe as textbook perturbation theory as distinguished from other less common methods. It may be noted that Whittaker and Watson's internationally aclaimed text on "Modern Analysis".e. thereby (unknowingly) discarding the vast majority of expansions which have been used in applications to physics.68) is (effectively) simply a statement of the observations made at the beginning.

With the help of the orthonormality condition (8.As a consequence the state |vl/) is a vector in this space. Ho^+V^n = EW^+E^n. so that we choose / • dxrpM^O. i.76) The first equation is that of the unperturbed problem.148 CHAPTER 8. J f°° dxrn E(^ 0) .w » .70) and equate on both sides the coefficients of the same powers of e. (8-79) We multiply this equation from the left by ip^ and integrate over x.72) we obtain oo roo dxifc{H0 -OO / ~4 0 ) ) E • / °i ^ = / J— OO d x < i E n ] ~ V )^n. i.n = 40)V>n. ^ ) = 0.e.4 0 ) M = / ^C(4 1} . i.74) (8. .e.75) HoW + vW = E^+E^+E^.e. The second equation can be rewritten as {Ho . -°° i J or 0 = EW-Jdx{rnVil>n). Time-Independent Perturbation Theory We insert these expansions into Eq. (8. ( ^ . We obtain then the following set of equations: Hrj.n.i4 0) )V4 1} = {EM . (8.V)i. (8.78) into Eq.77) and obtain {Ho ~ 4 0 ) ) E "i^i = (Ein] ~ W n . we write in the position space representation ^ ^ E ^ - (8-78) A contribution to ipn can be combined with the unperturbed part of the wave function \P. as are also the states Wn )j Wn )i — We therefore write these vectors as superpositions of the basis vectors provided by the unperturbed problem. We insert the ansatz (8. (8. (8.77) The states \tpn) of the unperturbed problem span the Hilbert space"K.

i.^(EW ~ V ) ^ + E& / J—oo dx^miln- .82) *> = ^ + E . multplying Eq. (8.e.Viprt e (8.a\^(E:S0) „• -4 0 ) )^ = / J—oo dxrl?m(E$-V)1>n. S ' ti^+Q^)• We observe that Eq.V)^ Setting + 42tyn.e.4 0 ) M 2 ) = ( 4 1 } . (8. and hence It follows that to the first order in the parameter e: En = 4 ° ) + e ( ^ . (8-83) Thus the procedure forbids the equality of any E\ '. (8. the procedure forbids degeneracy and we can consider here only nondegenerate eigenstates. i. we return to Eq.i ^ 0. with En . ^ n ) + 0(e 2 ).m ^ n and integrate: oo /•oo / -oo dxrmY.84) (8-85) ^2) = £ a S 2 ) ^ (excluding i = n for reasons discussed above).8. (8.76) which we rewrite as (H0 . We therefore consider Eq.83) makes sense only if Ef] ^ EnV for all i + n. (il)i. In order to appreciate the structure of these expansions it is instructive to go one step further and to derive the next order contribution. We multiply the equation by ipm. the coefficients a\ .84) from the left by ip^ and integrating over the entire domain of x we obtain coo J —( oo />oo / -oo dxi.5 Rayleigh-Schrodinger Perturbation and so Theory 149 E& = {n\V\n) = Wn.79). It is clear that we can proceed similarly in the calculation of the higher order contributions to the expansions of En and \I/n.Vil>n)- (8-80) In order to obtain ipn . (8.

^j)(V .81) we obtain (2) _ _ {lt>m. FVn) + e 2 ^ ( V > n .90) + £ ( 4 0) -4 0) )(4 0) -^ 0) ) ^ +' . (8.85) together with (8.^w .^(8.150 CHAPTER 8. En0) (8.88) into xjjT we obtain similarly * r i + f ^ i^n (lpi. ~Ej E- E.88) (Q) Inserting Eq.87) (i>k.V'>pTl (E^-Eny V'fc (8.2 (V> m . ipn ) = 0 i.VV .4 0) ^+ e 2 E (lpk.Vlpn] M _ P (0K 2 ^r . Inserting the expression (8.V^n) . Time-Independent Perturbation Theory Setting m = n we obtain with (^„.Vlpn) E^ .V4>n){lpn.86) E^ iy^n J-Jn J J -i ' i^n For m ^ n w e obtain from Eq. ^ n ) j&iift + ^n . n) -£f) (8.89) • • • where we guessed the term of 0(e 3 ). V ^ i ) X (Q) (Vi.(1) (i&0) .79) •/—oo oo / Using (8. j.V^j)(^j.£i°») 2 Hence +^£o (£«? . (8.Vlpn)(lpn.B i 0 ) ) ( £ f .86) into the expression for En we obtain En = 4 ° ) + e ( ^ n .^ ) + $& w £ (^fc.Vlpn)E. (8.e.

in the form i Adopting this procedure for the perturbing potential V(x) we can set V(x)i..5 Rayleigh-Schrodinger Perturbation Theory 151 The expressions (8. Merzbacher[194]). . We can therefore rewrite tpn ': The perturbation theory described in Sec. These recurrence relations allow one to re-express expressions like Xm^n as linear contributions of ipi with constant coefficients. (8.90) is the term with only one summation J2 m the contribution of 0(e 2 ). 8. The functions ipn(x) are eigenfunctions of some second-order differential equation.7 is based on this procedure.n+i = Ck-n(2) We can now rewrite ipn ' as / (2) _ V"^ c ^ n k-nc0 ~ 2-j ~ TJfi)0 ) k^n (4 -^) 2 n(0).89).g. One would prefer to lump it together with the other contribution.8. An ugly feature of the expansion (8.n(x) = J2crtn+ii (8.(0 4>k- The sum J2j^n includes a term with j = k which is exactly cancelled by the other contribution.91) The coefficients Cj follow from the recurrence relations. + l^t /„(0) V~^ c k-jCj-n &(Efi»-Ei )(Ef-E&»)\ 0) „(0)w. These functions generally obey one or more recurrence relations which are effectively equivalent to the differential equation in the sense of difference equations. We then obtain (lpk.lpn+i) = ^ Cj5k.90) can be found in just about any book on quantum mechanics. sometimes expressed in terms of projection operators which one can construct from the states \ipm) (see e. i.Vlpn) = y^Ci(lPk.e.-.

—I. Equality of eigenvalues.70).97a) (8. (8.We now insert Eqs. i. (8. .6 Degenerate Perturbation Theory It is clear from the expressions obtained above for En and tyn that no two eigenvalues En '.e. Equations (8. Equations (8.96) are effectively the known equations of ipnitpm and so yield no new information. all n..95a) (8.94a) into Eq.94a). in a central force problem all have the same energy unless the system is placed in a magnetic field.97b) are .95b) H0(c2l^n and + c22ipm) = En°\c21^n + c22^m) (8.e. (8. Em = 4 ° ) +eE$ + e2E$ + . . (H0 + eV)^n and (HQ + eV)Vm = Emym.Hence we write * n = CllV'n + CisV'm + € ^ + d 2 ^ 2 +••• . In order to deal with the problem of degeneracy we consider the simplest case.97a) and (8. so that En = 4 ° ) + eE™ + e2EP + ••• . are allowed to be equal since otherwise the expansion becomes undefined.93b). For example the magnetic states enumerated by the magnetic quantum number m.ipn). i.. I — 1 .93a). Time-Independent Perturbation Theory 8.96) (Ho . degeneracy.94a) * m = c 2 i ^ n + c22tpm + e ^ + e ^ + ••• (8.93b) The lowest order eigenfunctions belonging to En. Proceeding as before we obtain Ho(cniln + Ci21pm) = En0) (cU1pn + C121pm) = EnVn. . (8. is not uncommon. . (8.93a) (8.ipm) = 0 = (^m. + c22^m). (8. that of double degeneracy En = Em .4 0) )</4 1} = (En1] ~ V)(cu^n (Ho .V)(c2l^n + ci 2 ^ TO ).97b) with ipn y£ ipm. Em can now be linear combinations of ipni'ipm.94b) with (ipn. (8. m = I. however. (8. i.4 0 ) ) ^ } = (Em] .152 CHAPTER 8. (8.e. .

y>m.(^0-4°))41)) = EM^CuTpn -(ipl. (8.6 Degenerate Perturbation Theory 153 inhomogeneous equations.H0^) .(Vn. (8. ^ ) = ( V . . and ( ^ m .H0il>W) = cn[EW . But (8.(lPn. (^m.98b) + Culpm) + cuV^m).Vil>m)] V^m).H0xl>$) . Wm. t f ) = a k 1 } 4 0 ) " Eg>aU = 0. ^ o ^ ) ) = 0. Vlf>m)] . c2iV^n + c22V7pm).^ ( ^ . Vi/>n)] . H0lP$) = C22[E$ ~ (Vm.ipn n) = 0 = (ipm./>£>) = C2l[E£> .£ < « = ag)^) . Equations (8.C 21 (^ m . H0lpV) .98a) Setting I = n.ipm ) the equations (iPn. ^ ) ) = CU[EW .ci 2 (V„.C22(Vn. V^n).m). £ aW^°Vi) . Vl/. Similarly Wn. (^n.99) and ( ^ ^ O ^ ) " ^ ^ . We now multiply these equations from the left by ip* and integrate.m.cnVipn and -tyi.EJ®(lPn.J#»a&> = 0j since ii4i = En (degeneracy). we obtain with (ipn.V^n).99) can now be rewritten as (frvvA \1"'Vr\ )(cn)=w(cn) (s-iooa) . V^n)\ .^ ) ( ^ m . Then (^. cn(ipm.8.

E)tp = 0. (V>V-</>V>) = const.Vrl>n) .^ ' V = — (•>!>'<t>-<t>'i>). just as there is no degeneracy in the case of the one-dimensional harmonic oscillator (Chapter 6). If one wants to perform a perturbation calculation in such a case.1: Do discrete spectra permit degeneracy? Show that in the case of a one-dimensioanl Schrodinger equation with discrete point spectrum there is no degeneracy. However.ci2.Vl/>m) J \ C22 J m \ C22 J V ' These matrix equations have nontrivial solutions if and only if (lf>n.154 and V (^m. Let ip and < be two eigenfunctions belonging to the same eigenvalue E. cannot be degenerate because there is only one quantum number corresponding to one definite energy and hence cannot be degenerate. At x = ±00 the constant is zero.Similarly Eqs.^Vm) 0. . 9} ip. (8. the secular determinant. An example which emphasizes the significance of the spatial dimensionality in this context is treated in Example 8. Then / > i>" 2— = V ip Hence 0 = V ' V . A one-dimensional system. dx i. ip = c<p. In this particular case therefore non-degenerate perturbation theory can be used (cf.C22.3). It follows that ip and <j> are linearly dependent.e. the degenerate perturbation theory must be used. as we saw. Example 11.5). i. . Example 8.1.6. Time-Independent Perturbation Theory (l/>m. It is then possible that a special perturbation. \nip = ln(j> + c o n s t .101) This secular equation determines the first order correction En • The equations (8.E™ (Vn. if)' (/>' — = —.e. i> 4> Hence there is no degeneracy.. there is degeneracy in the case of the hydrogen atom with Coulomb potential (Chapter 11) which is based on three-dimensional spherical polar coordinates r. Solution: We consider the Schrodinger equation with potential V in the simplified form -i>" + (V . 11.Vlpn) CHAPTER 8.100a) determine the coefficients Cn. d>" -E=^-. Nonetheless we shall see that the Schrodinger equation of the hydrogen-like problem can also be separated in some other orthogonal coordinates. splits the problem into two independent one-dimensional systems.100b) determine C2i.e. 4. i. like that providing the Stark effect (proportional to rcosO).This completes the derivation of the first order perturbation corrections if an unperturbed eigenvalue is doubly degenerate. (8. in particular in parabolic coordinates (see Sec.

and the solutions then have to be matched in their regions of overlap. the method has to be applied in each domain separately. J.106) (8.V0)<t> = (v.V)(p = 0.104) "This method was developed in R. W. also possible in areas not of immediate relevance here. In the following we describe such a procedure as first applied to the strong coupling case of the cosine potential or Mathieu equation.7 Dingle-Muller Perturbation Method 155 8.102) where D is a second order differential operator and E a parameter. (8. Muller [210]. We assume that V can be expressed as the sum of two terms. Muller [73]. H.8. B. and such that one can even obtain a recurrence relation for the coefficients of the expansion. B. (8. anharmonic oscillator potentials and screened Coulomb potentials.102) can be written D<f> + (EQ . with some restrictions on the function V. The subdivisions (8. J. Dingle [72] and H.e)<t>. Muller [216].e. a potential. J. in particular in applications to periodic potentials. T For instance spheroidal functions can be studied with this method. the considerations given below would be valid even if D contained only the derivative of the first order. .104) are chosen such that the equation Dcf) + (E0 . one quickly arrives at clumsy expressions which do not reveal much about the general structure of a higher order perturbation term. It is immaterial here whether D does or does not contain a first order derivative — in fact. W. Eq. cf. Applications of the method are.7 Dingle-Muller Perturbation Method We have seen above that although the procedure of calculating higher order contributions of perturbation expansions is in principle straightforward. i. V = V0 + v.t Since in general a solution is valid only in a limited region of the variable. W.* In various versions this method is used throughout this book. (8.105) (8.103) such that if E is written correspondingly E = E0 + e. of course. See also R. Dingle and H. We consider an equation of the form Dcj) + (E . Thus the method is a systematic method of matched asymptotic expansions. (8.103). It is therefore natural to inquire about a procedure which permits one to generate successive orders of a perturbation expansion in a systematic way.V0)<f> = 0 (8.

(8. Considering (frm+s we have.108) with coefficients (m. m + s)0m+s. since D<f>m + (Em .Em+s)(/)m+s. We write the coefficients "(m.VQ) ^ s c s4>m+s = 22 s C s(Em ~ Em+s)<fim+s = ( u .V0)<f>m+a = 0.110) If we write the next contribution <f>W to <p(°> in the form m+st (8. The next step is to use the recursion formulae of the functions 4>m in order to re-express (v — e)4>m as a linear combination of functions <fim+s. (8.111) s then (see below) 0(1| = E (m. m + s) which are determined by the recursion formulae.Vo)(j)m+s = (Em .V0)<f>^ .V0)4>m = 0.m + s)4>m+s s (8.E0\ (the latter in a restricted domain). Consider (D + Em= (D + E m V0)(4>W + ^(1)) = (D + Em . Time-Independent Perturbation Theory is exactly solvable and has the eigenvalues Em and eigenfunctions (f>m.109) D(j)m+s + (Em .107) represents a first approximation to the solution (f> of Eq. s an integer: (v-e)(pm = 'Y^{fn. the equation Dcf>m+S + (Em+S .m + . the function 4>(0) = <t>m (8.+s ' (8.e ) ^ = 2 ( m . where m is an additional integral or near-integral parameter (depending on boundary conditions).) This result is obtained as follows.102) with E ~ Em. Then if ]e| < \E\ and \v . and so v (8.156 CHAPTER 8.m + s)" in order to relate them to "steps" from m to m + s.113) . v = {D+Em-Vo)<t>m.e\ < \V0 .

m) E J^ ( m ~ Em+S) ' • fa.113) must vanish. and the second expansion is an equation from which e and hence the eigenvalue E is determined (i.118).n\ ^-119) + . The factors (Em — Em+S) in the denominators result from the right hand side of Eq.m) + > ^ .m)4>m in (8. We then find altogether ^ | l-^m - Em+S) + y^y^ SJ.116) m ^ v (m. (8. . (8.Em+S) = 0 (second approximation).m + r) m ^ (8 118) r^O ' ~~ Em+s)\Em ~ Em+r) together with n _ / x y^(m.TP \(T? . (8. (8. (8.. . (8.T? S The first expansion determines the solution 4> (apart from an overall normalization constant).m + s)(m + s.^ (Em .Q (m.„N This procedure can be repeated indefinitely.119) is effectively the secular equation). .e.m + s) 157 (B-\-IA\ s^O: 1 cs = J m _ ^m+s .102) to order (1) of the perturbation (v — e).m) — —. /01_.m + s)(m + s.110) and are therefore related to the inverse of a .8. (8.115) Repeating this procedure with cf)^-1' instead of 4>^ we obtain the next contribution 0(2) =YY together with (m. V ^ V ^ {rn. We observe that the coeffcients of expansions (8.m + s)(m + s.114) To insure that 0 = ^(°) + f^ ) H is a solution of Eq. Eq. £ ^ T^o (^m ~ Em+s){Em — Em+r) ( m ' m + s ) ( m + g ' m + r ) $m+r (8. the coefficient of the term (m.m) — 0 (first approximation). Thus (m.m) 1^1^ IW .m + s)(m + s. .119) follow a definite pattern and can therefore be constructed in a systematic way.m + r)(rn + r. .7 Dingle-Miiller Perturbation Method and so (m.117) „ .

This means the solution is not normalized. as already mentioned after Eq. . The systematics of the coefficients suggests that one can construct the coefficients of an arbitrary order of this perturbation theory. This can in fact be done and will be dealt with later. . Other applications can be found in the literature. as will be seen in later chapters. J. 2 .1)( 2 + 64J3] — . Miiller [211].[q(llq2 + 1) + 2(33g 2 .7 1 9 2 + 32g + 2976) 3•215g2 +32Z 2 (252g 2 .38).423) + Z(2720 9 3 . (f>m in the above. i. has been treated by R. An explicit application of the method (not including matching) seems suitable at this point. and is typical of a large number of cases. .6q + 1)1 + 24(5? . for instance.e. . (8.[ 4 ( 8 5 g 4 + 2q2 . Imposing normalization later one obtains the normalization constant as an asymptotic expansion. One should note that in this perturbation theory every order in the perturbation parameter also contains contributions of higher orders.2). (8.2 therefore illustrates the use of this perturbation method in the derivation of the eigenvalues of the Gauss potential. S.2: Eigenenergies of the Gauss Potential^ Use the above perturbation theory for the calculation of the eigenvalues E = fc2 of the radial Schrodinger equation dr2 + k2 1(1 + 1) V(r) ip(r) = 0 with Gauss potential V(r) = -g2e~a r for large values of g2. The following Example 8. does not occur in any of the later contributions.9) + 4096J4] + 0 ( l / g 3 ) . for the investigation of the behaviour of the late terms of the perturbation series.e. as will be seen later.I2q + 64) + 256Z 3 (41 g . 1 . i. 2 k2 + g2 = ga(2l + q)-^[3(q2 „3 + l) + 4(3q~l)l + 8l2] 3-2ag r .120) n4 . W. Kaushal [146]. The result is the following expansion. n = 0 . An additional aspect of the method is the full exploitation of the symmetries of the differential equation.* Example 8. The eigenvalue expansion obtained is an asymptotic expansion.158 CHAPTER 8. § H. A further conspicuous difference between this method and the usual Rayleigh-Schrodinger perturbation method is that the first approximation of the expansion. since it will be used extensively in later chapters. ^The potential A r 2 / ( 1 + gr2). and when the solution is multiplied by this factor one obtains the contributions involving </>m in the higher order terms (for an explicit demonstration of normalization see Example 17. Time-Independent Perturbation Theory Green's function. this is — so to speak — the price paid for obtaining a clear systematics which is essential. in which q = in + 3. for instance.

Then the equation is Dqi> = -Ah+-y 2 4 ^ 1 1 » fc- i\ . Therefore in the general case we may set k2 + g2 = ga(2l + q) .7 Dingle-Muller Perturbation Method Solution: We expand the exponential and rewrite the Schrodinger equation in the form rfV (ir 2 .2) where 1/ .0(z) = zl+1e-z2^fa.b. Then a?ip dz2 fk2+g2 2ga l(l + l) 1 z*)i> ±(T'(-^*and S=-z2.121). a ._ „!2 ^ ( " « 2 r r 2 > . i/> with Dq = d2 . 3\ 2 V ' 2.2a2 A.123) The first approximation (note the last expression as convenient notation) ^ .• 0 ( ° ) = ipq(z) = zl+1e-z2/4<S>(a. (8.V ! 4 I -z- S *! V2 ipq(z). b.b. Sees. . 1 1(1 + 1) 1 . 2 . Setting q = in + 3 implies k2 + g2 ga(2l + q). (8.123) the contribution fl<°> -Ah + . 11. We then set V>o(2)=2i+1e-z2/4X0(2) Now the function XQ{Z) satisfies the confluent hypergeometric equation (cf. . 2 The solution XQ(S) = $ ( a . \2 . <«»> In the limit |g| —• oo we can neglect the right hand side and write the solution ijj —* I/IQ. (fc2 + <?2) 4ga and b = I -\—.^z2) = i>(a) leaves unaccounted for on the right hand side of Eq.122) where A is of order 1/g.l)ip(a .a + l)if>(a + 1) + (a.6 and 16. The solution i. a)ip(a) + (a. . .-z2 is a normalizable function if a = —n for n = 0 . „2 J„2JL. . 1 . (.. S) is a confluent hypergeometric function. multiply the equation by one-half and set h = —a/g.2 '(l + l) . (8.1). The recurrence relation for the functions ip(a) follows from that for confluent hypergeometric functions: -z2f(a) = (a. . 159 •g'-g'a'V U E Here we change the independent variable to z = y/2gar. We now insert this equation into Eq.8.

a]1=0. a + j)i/>(a + m).l . ( 5 m ( a .3).S m (a. when j = 0. The coefficients S m ( a . a + 111 r [a.2a = I + -q. The above remainder i j ' 0 ' can now be rewritten as oo i 1 R(0)=J2h + i+2 i+2 ]T -(i+2) [a.a + 2]i P . of course.a]i It is clear that the various contributions can be obtained from a consideration of "allowed steps". a + r — l)(a + r — l .a]i = . (a. As a check on the usefulness of the perturbation method employed here. a + i-)(a + r.a — 211 .120). a + r + l)(a + r + l .1) = a . a) = 1. a ] x + L _ 2 J 1 [ « .a—lli. a + r) + S m _ i ( a .b = (g + 3) . If we now evaluate the various coefficients in the last expansion we obtain the result (8. . a + 1) = a = In general CHAPTER 8.a + j}i+1^(a + j). This equation shows that a term fiip(a + j) on the right hand side of Eq. In fact. a + r) = 0 for |r| > m. a . Now we make the following important observation that Dqip(a + j) = jip{a + j).Z. a + r) with the boundary conditions Sn(a. a + r) = i 2 \ m ?n -z J -0(a) = J^ j= — m S m ( a . a + i) = 0 for i ^ 0. fa. a + r) satisfy the following recurrence relation: Sm-i(a. (a. (8.e. Now proceeding as explained in the text one obtains finally the expansion ip = ^(°) + v ( 1 ) +y>(2) H — along with the expansion from which A is determined.160 where (a. a ] i + 0(/i3) [a. the latter determining to that order the eigenvalue.j^0 ' + jU+1 J ^(a + j) with h[a.120).a]2+L ^ J1[a + 2 . Time-Independent Perturbation Theory (g .i ^(i)=£V+i i=0 J2 j = -(i+2). So (a. fa. one can write down recursion relations for the coeffcients of powers of hl as will be shown later in a simpler context. „ .123) and so in R(°> can be taken care of by adding to the previous approximation the contribution fitp{a + j)/j except. . a) = b . 0 = h[a. where [a. Hence the next order contribution to tp^ becomes oo i+2 r [a a . one could now use the Rayleigh-Schrodinger method and rederive the result (8. a]i + h fa. _ J [a . o + r) + S m _ i ( a .^ A + J7^S2(a'a)' K a +J]i+l = 4 / i + 2-)!' S ''+ 2 ( a ' a + j) for i and j not zero simultaneously. i. a ] i + L ' ^ J 1 [ a + l .2.

We know from classical electrodynamics that a planar light wave with propagation or wave vector k and frequency u> is represented by a vector *So far our considerations were restricted to one-particle systems. the generalization to many particles. except for the canonical algebra which has to be formulated with Dirac brackets replacing Poisson brackets. as we shall see in Chapter 27. light. In continuation of our earlier search for an approach to quantum mechanics as close as possible to classical mechanics by looking for the generalizability of classical mechanics — as already attempted in Chapter 2 — it is reasonable to consider also classical systems with a wave-like nature. 161 . although not exactly straight-forward. and this means electrodynamics.Chapter 9 The Density Matrix and Polarization Phenomena 9. and we shall then see that these lead to analogous results as in Chapter 2. proceeds along similar lines. Regrettably this has to be left out here in view of lack of space.1 Introductory Remarks We saw in the foregoing chapters that in considering a particle system which is classically described by the solution of Newton's equation* is quantum mechanically described by states which in the position or configuration space representation are given by the solutions of the Schrodinger wave equation. i.e. 9.2 Reconsideration of Electrodynamics In this chapter we consider some aspects of classical electrodynamics with a view to their generalization in the direction of quantum mechanics. We therefore begin with the appropriate classical considerations with a view to generalization.

(9.3) (9.oei(k-r-wt> which satisfies the transversality or Lorentz condition k • A = 0.i). t) = A 0J R cos(cji) + A 0 / sin(u. 9. When Ao/? is perpendicular to Ao/ and |Ao#| = |Ao/| the ellipse becomes a circle and the field is said to be circularly polarized. t) given by A(r. (9. i. The Density Matrix and Polarization Phenomena potential A(r.162 CHAPTER 9. In general the constant Ao is complex.1.1) Fig. It follows therefore that A(r. t) = A(XR cos(k • r — cut) — Ao/ sin(k • r — cot) and at r = 0: A(0.4) (9. This is the equation of an ellipse as indicated in Fig. 9.2) (9.6) . In the latter case one has the possibility of {A0R x A 0 /) • k > 0.1 The vector potential ellipse. When AQR is parallel to Ao/ the ellipse becomes a straight line and the wave is said to be linearly polarized. (9.t) = 3ftA.e.5) which is described as right polarization as compared with left polarization for which (A0R x A 0 /) • k < 0. A 0 = A0R + iA0I.

the vector potential A. / = 4 0 ) * 4 0 ) + 4 0 ) * 4 0 ) = (A(°\ A<°>) = |A(°)|2. a calcite crystal or a film of nitrocellulose. (AW.A(°)).A« we can construct only the following scalar products with this invariance: (A(°\A(°)). .11) 'This means we consider the polarization. i.2) the vector potential A is completely determined by its two (say (x. determined with the help of a polarizer. Rae [234]. I=<l2x2>:=E40)*l*40)i. we recall. the connection must be linear.A 2 .g. (9. See for instance A. i. (A(°). not.9) The intensity is characterized by the fact that it can be looked at as the "expectation value" (1) of the unit matrix 12x2. but the polarization (state) and the intensity can change (the latter by absorption) from R(0)} - R ( 1 ) }- In view of the linearity of the Maxwell equations. (9.AW). the intensity of a wave. which leads to an outgoing wave which we give the label "1". 17.7) We consider now measurements with polarized light. as e.t in which k and ui remain unchanged.A2 is the sum of the moduli. (AW.9. i.2 *i*4 0) (9-8) with coefficients F^. for instance.k (9-10) We give the initial wave the label "0" and subject this to an experiment. the fields E and B are observables. the direction of polarization.e. A2. Here we are interested in observable properties of light waves — like.8). with k . From the vectors A(°). In classical electrodynamics. represented by a filter "F". (9. Then their connection is given by Eq. p. We set therefore A 0 = Ai + A 2 . i-e. Observable quantities can only be those which are invariant under translations and rotations.e.e. (9.2 Reconsideration of Electrodynamics 163 As a consequence of the Lorentz condition (9. however.y)) components A i .A i = 0 = k .A«). The intensity 7 of a light wave with amplitude coefficients A\' . AW = £ fc=l.

12) i. We write the observable quantity as (F):=Y.13) the quantity F must be an hermitian matrix and thus representative of an observable.17) . (9.k l f c 4 ° \ (9. (9. so that (F) can be looked at as the expectation value of an observable F. (9. i. if (F) is an observable quantity. pki = AkA*. To this end we observe that as a consequence of Eq. it must be real and hence (F) = <F>*.16) i. (F)=Tr(Fp). (9.13).k i (9. The Density Matrix and Polarization Phenomena Only the second last of these combinations describes the experiment. the connection between the initial polarization and the final polarization.13): j2^FikAk = Y^AiF*kAi = Y. (9.164 CHAPTER 9.15) in order to rewrite Eq.k (9.e. Thus. Then according to Eq. But we can convince ourselves that we achieve exactly the same as with our earlier consideration of observables. We use definition (9. (9.13) This consideration of translation and rotation invariant quantities is somewhat outside the framework of our earlier arguments.14) We now introduce a matrix p — called density matrix — which is defined by the relation pik := AiA%. i. Thus (F) = J2FikPki = ^(Fp)a.A*FikAk.e.e.A( 1 )) = ^ ^ F i.15) This matrix is hermitian since (Pife)t = (AiAtf = (A*Akf = (AkA*f = {pkif = (Pik). (A(°). i.A*kF£A* = E and hence F = Fl A p i tkAk.

we can describe the wave function of the state polarized in the direction of x or y respectively by \<P)I =f0 J and \fh = ( i ) A special state is represented by j)=-(J)+<!) with |a| 2 + |6| 2 = l.19). that of a pure state.15) with A\ = a.21b) This state is still a pure state. like that of a bulb. (9. we have (9 19) ' In this case there is no preferential direction.e. For a 45° polarized wave we have 1 . i. mutually orthogonal polarization directions x and y described by the matrices (see also below) * .e. . In the present case of a polarized beam of light we have two possible. i.* For a light-wave travelling in the direction of z. But light.2 Reconsideration of Electrodynamics In particular we have / = Tr(p).d "»=(o ! ) " In the case of an admixture with equal portions of 50%. <»^> (9. This admixture is described by introducing matrices p. The density matrix for the pure state (9.22) For a — 1.9. i. is in general unpolarized. it is a statistical admixture of light rays whose polarization vectors are uniformly distributed over all directions. is given by a single polarized wave. b = 0 and o = 0. A2 = b as </*) . 1 The classical polarization vector or wave vector A corresponds in the quantum mechanical case to the wave function ip. b = 1 we obtain the expressions (9. 165 (9.(AAt) = ( £ £ ) . (9. "incoherently.( J o ) .21a) follows from Eq. The so-called pure case.18) The vector A describes a particular ray of light.e.

b = —=.™ V 0 n kmAm / ^-^jPim-^mk' . 1 = . (9.g.Z^ 3.3 Schrodinger and Heisenberg Pictures As in Chapter 2 we now consider on the purely classical basis the (analogues of the) Schrodinger and Heisenberg picture descriptions.8) describes the relation between the initial and final polarization vectors A^°) and A^1) respectively in a measurement of the observable F.8) A(1)=E^40)> k R = R(<*). We now inquire about the the way p changes in the process of the transmission of the light-wave through an apparatus. Then according to Eq. We shall see that the matrices appearing in the description of polarization properties of waves possess the same generalizability as the particle considerations of classical mechanics.23) 1 . (9.24) The following two 50% admixtures yield the same effect: 1 1 (\ 0\ P = -2P*+2Py=[Q 1 P = 2^ 4 5 ° + 2P135° i J. We set in analogy to Eq. l l\ " i °\ ) ' ^9-25^ 9. The Density Matrix and Polarization Phenomena P45° = I 1 For a 135° polarized wave we have a = and l M . (9-26) where a is a parameter whose variation describes the continuous variation of the polarization from its initial state labeled with "0" and hence parameter value OQ (analogous to t or j3 in our earlier considerations of the density matrix) to the final state with label "1". Pi35° = \ \ 2 ? )• (9. as a consequence of absorption).15) Pik — A i A k . in which the intensity I = Tr(p) is not conserved (e. Equation (9. (9.166 and CHAPTER 9.

(9.27) The measurement of an observable F in the new state with superscript " 1 " then implies according to Eq.9. The corresponding condition here is that no absorption of the wave takes place.pW).28) (9. In the case of Eq. that the initial intensity Jo is equal to the final state intensity I\.e.30) Ti{FRp{®I$) = Tr(F. The Liouville equation describes the motion of an equal number of systems in phase space elements of the same size. 9. (9.4 The Liouville Equation It is natural to go one step further and to derive the equation analogous to the Liouville equation. i.33) i. (9.e. the observable F remaining unchanged.31).17) (F) (1 ) = Ti{FpM) = = Tr(tiiFRpW) F' = RtFR. The two cases are completely equivalent and can therefore be described as Schrodinger picture and Heisenberg picture representations. and p(0' remains unchanged.4 The Liouville Equation 167 pW = RpMRl (9. in the state functional) in analogy to the time dependence of W(p.t) in Chapter 2 in what we described as the Schrodinger picture there.30) the dependence on a is contained in pW(a) = R{a)p^{a0)RJ'(a) (9. We can interpret the result either as (F) ( 1 ) = Tr(F{Rp(0)R}}) or as (F){1)=Tr({RtFR}p<-0)).q.32) (9. on the other hand. i.e.29) where (i.e. In the case of Eq. (9. in other words without creation or annihilation of systems. (9. the dependence on a is contained in F' = R^(a)FR{a). here this contains the dependence on a.34) .31) (9. (9. the observable is transformed. that Io = Tr( / 0 (°))=Tr(pW) = / 1 .

41) We see therefore that the 2 x 2 matrices entering the description of polarization properties of waves possess the same generalizability as the particle considerations of classical mechanics in Chapter 2.2) has to be taken into account.39) and (9. (9.F}. i. For an infinitesimal variation of the state of polarization we have R{a) = 1 + Mda.40) Equations (9. (9. (9.32) in the Schrodinger picture that p^iao + da) = = = or ( w i t h p(°> —> p) (9. The Density Matrix and Polarization Phenomena Tr(p(°)) = Tr(p^) so that = Tr(Rp^R^) RlR = 1.M .21) and (2. the matrix M is anti-Hermitian: Mf = . (9.39) If the dependence on a is not contained in p^1' but in F' (cf.40) can be considered as quasi-equations of motion.38) H = Hl R(a)p(°\a0)R\a) (l-iHda)p(0)(a0)(± + iHda) pM(ao)-i[H.37) (9. Eq.B}:=i[A. In view of Eq. j£ = -i[H. . (9. (9. (9.p]. In Chapter 27 we shall see that the Poisson brackets here are actually Dirac brackets because the gauge fixing condition (9. CHAPTER 9.35) we have 1 = R*R = (1 + M W ) ( 1 + Mda) ~ 1 + (M* + M)da. (2. For p ( 1 ) ( a ) = p ( ° ) ( a 0 + da) follows from Eq.e.35) Thus the matrix R has to be unitary.e.36) (9.B].33)) we obtain correspondingly in the Heisenberg picture the equation ^ = +i[H.168 i. We set therefore M^:=iH.46) suggests to define Poisson brackets of matrices by the correspondence {A. Comparison of these equations with Eqs. = Tr(R^Rp^).p(°\ao)]da.

(10. 10.Chapter 10 Quantum Theory: The General Formalism 10. Heisenberg and interaction pictures.1) This expectation value remains unchanged if \u) is replaced by e%a\u) with a real. we establish their Heisenberg equations of motion. and consider timedependent perturbation theory. To insure that the 169 . Thus in particular we establish the uncertainty relation for observables in general. the Schrodinger.2 States and Observables Every state of a physical system is represented by a certain ket-vector \u) which is an element of the corresponding Hilbert space "K. Thus the expectation value with respect to the vector describing the state in the Hilbert space does not depend on this phase factor.1 Introductory Remarks In this chapter we are concerned with the formulation of quantum mechanics in general. In keeping with our earlier notation the expectation values of dynamical variables are defined as follows: Postulate: The expectation value or mean value of an arbitrary function F(A) of an observable A in the case of a system in the state \u) £ "K is defined as the expression (F(A)) = (u\F(A)\u).

For these the following rules of canonical quantization are postulated: Postulate: The operators q~i. The phase space coordinates qi. [qi. z and for definiteness here — which will not be maintained throughout — the "hat" denotes that the quantity is an operator. Quantum Theory: The General Formalism expectation value of the unit vector or identity operator is 1.2.Pi are postulated to obey the following fundamental commutation relations (again leaving out "hats"): [Qi.Pj] = ihSlj. Naturally one can also write (10. such as for instance of the angular momentum operators Li. E. 10. we must have (u\u) = 1.2.y. classically qiPi = PiQi. q). we assume this now.1 U n c e r t a i n t y relation for observables A. y. which clearly does not hold in the case of operators.* a conspicuous exception being for instance a spin system. B We first establish the following theorem: If A and B are observables and hence hermitian operators obeying the commutation relation [A. (10. where i = 1. In order to avoid multivaluedness one always starts from Cartesian coordinates in position or configuration space.3 represent the three Cartesian coordinate directions of x. Such a correspondence with classical mechanics is not always possible.Pi are called fundamental observables. (10.Pj] = 0. The first step in the investigation of a quantum mechanical system is to specify its dynamical variables A.B] = ih. \pi. Here. .q).3) These fundamental commutators determine the commutators of arbitrary observables A = A(p.2) (F(A)) = {u\u) <"lf. Pi • Pi.170 CHAPTER 10.4) *Even in those cases where this is possible the operator correspondence may not be unique. however.g. and assume that A = A(p.Qj] = 0. We can describe a system as one with a classical analogue if the Hamilton operator of the quantum system is obtained from the Hamilton function of classical mechanics by the correspondence •^i • Qi.

we have A2 = A2 .2A(A)} - {A)2)1'2 (10. Ai = = = ((A2) .. Since A = A— (A).B\u) G IK as vectors) we obtain (for an application see Example 10. Thus AA = AA = ( i 2 ) V 2 ..7b) = = [A.0)1/2 = AA.9a). Hence (AA)2{AB)2 = (A2)(B2) = (u\A2\u)(u\B2\u).2 States and Observables then their uncertainties AA. C = A.. (10.7a) (B) = 0.12) where \u) 6 "K is a ket-vector representing the state of the system. = \\A\u)\\2\\B\u)\\2 (10..2A(A) + (A)2. (A-{A))(B-{B))-{B-{B))(A-(A)) (10. B}= ih. (10. (10. Applying to the expression (10. so that (A) = (A) .(A)2 = (AA)2.9) ((A2) + (A)2 .B] = AB-BA > ^h.(A) = 0. 111 (10. Eqs.2(A)2 .6) B:=B-(B). (10. (4.10.10) .7b). AB defined by the mean square deviation AC=((C2)-(C)2)1/2.12) the Schwarz inequality (cf.{A)2)1'2 {{A2) .5)) (for A\u). (3.5) are subject to the following inequality called uncertainty relation: AAAB In proving the relation we first set A:=A-(A).8) and. (10.B.13) AB = AB = (B2)1/2.11) (A2) = (A2) .1) (AA)2{AB)2 = > (u\A2\u)(u\B2\u} \{u\AB\u)\2. using Eq. (10. (10.(A)2)1'2 (1 °= a) ((A2 + (A)2 . Then [A.

.e. and hence for condition (b) 0 = = tSee Sec. (AB + BA) This means (AB + BA) is real. Zi (10.3. (10.16) I = -h. Then AB = \(AB + BA) + \(ABZ Z BA) (1 = 8) \{AB + BA) + \ih.19) (u\AB + BA\u) = (u\AB\u) + (u\BA\u) c*(u\BB\u) + c(u\BB\u) = (c* + c)(u\B2\u).4). A\u)=cB\u). From this we obtain |(u|iJB|u)| 2 = i22 + / 2 .15) as (u\AB\u) = R + iI.18) >R2 + I2. Zi Zi (10. 4. (10. R= -(AB + BA).17) as had to be shown. (10.172 CHAPTER 10. Zi + BA)\ + ^ih.14) Hence (u\AB\u) = /^(AB where. for a complex constant c. (10. Quantum Theory: The General Formalism Next we separate the product AB into its hermitian and antihermitian components^ and use Eq.13) (AA)2{AB)2 i. The equal-to-sign applies in the case when (a) it applies in the Schwarz inequality.15) = = (u\AB + BA\u) = (u\(AB + BA)^\u)* (u\(AB + BA)\u)*. We can therefore rewrite Eq. The condition (a) is satisfied when. (10. AAAB>I=-h. and with Eq. (10. (10. and (b) when (AB + BA) = 0. since A and B are hermitian.

13) to operators Li — Li. (10. Separating the product into symmetric and antisymmetric parts.e. . the following uncertainty h2{Lk)2. = 5R2 + G 2 > 9 2 = hi2(Lk)2.Lj] = -(LiLj + LjLi) + -ih£ijkLk.10. For these we have in the one-dimensional case the relation [p. we have LiLj = -(LiLj and ^ = y(u\LiLj so that \(u\LiLj\u)\2 Hence {ALx)2{ALy)2 > l + LjLi) + -[Li.i relation holds: (ALi) 2 (AL. Example 10. we obtain {ALi)2{ALi)2 > \{u\LiLj\u)\2. = 0. ALX 10. (ALZ)2(ALX)2 > \h2 = (Ly)2.3 One-Dimensional Systems 173 Thus c = — c*.e. (10.1: Angular momentum and uncertainties Show that in the case of angular momentum operators Li. i. (Lx±iLy) ^ 0.-) 2 > = x.z. for which one component of L is "sharp"? Solution: Applying Eq. (ALy)2{ALz)2 > ~h2(Lx)2. For states \u). + LjLi\u). ALZ = 0 (these are states \lm) with Lz\lm) it follows that (Lx)2 = 0. (A-{A))\u)=i(Zc)(B-(B))\u). i. What does the relation imply for states |u) = \lm). -h2(Lz)2.20) or. 9ftc = 0.3 One-Dimensional Systems We now consider one-dimensional systems with a classical analogue as described earlier. (Ly)2=0. i. The observables like angular momentum are functions of p and q.7a). implying that the vector |it) has to satisfy the following equation: A\u) = i($Sc)B\u) (10.e. q) = -ih. i. {u\LiLj\u) = R + iQ. such that Lz is "sharp". in view of Eq. 9 = -h{u\eiikLk\u).e. (Lz)2^0. Ly are not "sharp". Thus in this case Lx.y. (Ly)=0. m\lm)). implying {Lx)=0. ALy = (u\L±\u) ^ 0.

(10.22) d ih—(p2).21b) (10.174 CHAPTER 10. Next we have a closer look at the operator q. i. Similarly we obtain \p. i. Eq.BC] = [A.34d)) [A. This is why the operator p requires a representation in the space of eigenvectors of q.e. We have (cf. We demonstrate that (a) its eigenvectors have infinite norm (in the sense of a delta function). A — A(p.21b) now follow with the assumption that A(q.qn] = -ih^(qn). and (c) the spectrum is not degenerate. for q\x) = x\x) we have p\X)=\x)(-th^y Now let A be an observable.pn]=ih-^(pn).C\. (b) its spectrum is necessarily continuous.p)]=ih^^and \p.A(q.p}p + p[q.23) = -ih-g-(q2) (10.p] = ^ihp ~ Proceeding in this way we obtain [q.i h ^ ^ . as remarked earlier. (10.q). Q2] = [p. . (3.p] = ih we obtain therefore [q.21a) Equations (10.p)] = . they do not have a common system of eigenvectors. Quantum Theory: The General Formalism Since this says that p and q do not commute (as operators).21a) and (10. Then (see below) we can derive for A the following commutator equations [q. q]q + # > v] = ~2ihq and \p.p) can be expanded as a power series in q and p.B]C + B[A. Using [q.A(q. We obtain these equations as follows.p2] = [q.e.

24) lim SK(x — x') — S(x — x').28) i. .e. (10. Since p = p\ follows t h a t U\a) so t h a t U(a)U\a) = 1. 10.1 The delta function for K — oo. U] = ih^= aU.3 One-Dimensional Systems 6K(x) 175 Fig. i. (10. = eipa/h = Uia)'1 = U{-a). We thus see explicitly t h a t the vectors \x) have infinite norm (in the sense of delta function normalization). Replacing in Eq. an operator.27) (10. we obtain [q. > We begin with (a). 10.f^ ^ J—oo dke ikx —ikx' (10.25) It is shown in Fig.1 how the delta function arises in the limit K —> oo.e. For the cases (b) and (c) we consider the operator U(a — p-ipa/h (10. and a a c-number. U is unitary. 6K(x-x') = — fK dkeik^-x'^ 2TT J-K sin K(X -K{X — x') — x') (10. We have (x\x') = = where f^ J—oo dk(x\k)(k\x'} (4 = 9 ) ±.21a) A by U.26) it Here p is an observable.10.

the spectrum is not degenerate.00).e.31) . CHAPTER 10.q)U^U\ilj). Thus all these values belong to the spectrum of the operator q. or qU\x) = U{q + a)\x) = (x + a)U\x). (10. i.29). q any other observable F(p. In an analogous way by defining the unitary operator U(b) = eiqblh (q operator. cf. or qU(a)\x') = (x' + a)U(a)\x'). Eq.00). we obtain qU\x) = U(q + a)\x) = {x + a)U\x). (10. Evidently every eigenvalue has only one eigenvector. WU = 1. ^Observe that {iji\A{p.29) This means: U\x) is eigenvector of q with eigenvalue x + a.e. we can see that also p possesses only a continuous spectrum (from —00 to 00). Since this holds for any arbitrary value of a in (—00.3. that with U(a) = e~ipa/h = U. this means: With a unitary transformation (which leaves the matrix elements unaffected * and hence the physically observable quantities) one can pass to any arbitrary eigenvalue in the domain (—00.q)\i>) = (i>\U^UA(p. (10. q) can be constructed representing a dynamical variable. The eigenvectors have the same norm as \x): (x\U*U\x') = (x\x') = S(x-x'). Quantum Theory: The General Formalism qU = Uq + aU = U(q + a). and that the eigenvectors do not have a finite norm and are normalized to a delta function according to (jPx\Px) = S (Px-p'x)- With the help of the observables p.176 i. 10.30) (10. b a c-number). which means that the spectrum of this operator is a continuum.1 T h e translation operator U(a) We saw above.

x')5(x" = (x" + a\c*c\x' + a) -x'). (10. We choose the phase such that U(a)\0) = \a).a). (10.x" .5(x> . the operator U(a) acts to shift the value x' by the amount a. i. (10. In the expression of Eq.e) ~ 5{x' .36) i.33) (10. (10. (x'\p\x"). Comparing Eqs.31) and (10.37) With the help of the latter expression we can calculate.10.hm or „'i„i™" s> i ™' ~/ (x'\p\x")\ = -8'{x' -x").e.x") U{x'\p\x" % -pz. It follows that U(a)\x') = = U(a)U(x')\0) = U(a + x')\0) \x' + a).x') = = i. i.x")c(a.e) (x \p\x ) = ./ (10. It also follows that (x'\U(a)\x") = (x'\x" + a) = S(x' . U(a)\x') = c\x' + a).32) we see that \x' + a)ocU(a)\x'). it follows that 5{x" .e.3 One-Dimensional Systems 177 Moreover.32) Since U is unitary.38) h 6(x' . offdiagonal elelemts of the {x}-representation of the momentum operator. for instance. we have q\x' + a) = (x' + a)\x' + a).34) {x"\U\a)U{a)\x') c*(a. \c{a.x" .e. i. (10. The operator is therefore called translation operator.35) (10. (10. \x') = U(x')\0). since q\x') — x'\x').39) .x") .e.e.37) we set (with e infinitesimal) U(e) ~ 1 Then (x'\U(e)\x") = 6(x' . (10.x" .x')\ = l.

isolated from any measuring apparatus which would disturb the system and hence its behaviour in the course of time) and let \tp(t)) be its state vector at a later time t > to (with no interference in between).4. The operator U(t.§ It follows that (x"\p\x') = --d'(x' This expresses that p is hermitian. 6'{x) = -!.45) (10.f dpipeipx.4 Equations of Motion Let \tp(to)} be the state vector at time to of a completely isolated system (i.to)m0)). a conservative system.44) (10.27) and (5. In this case the classical Hamilton function does not depend explicitly on t.40) 10. This is in conformity with our earlier postulate in Sec.42) ^Observe that 6'(-x) = £ f ^ dpipe-** = ± / .35). Differentiating U(t. to) with respect to t we obtain ihjtU(t.° ° d(-p) (-ip) e"« = -S'(x). We introduce another postulate: Postulate: The time dependence of the vector \uE(t)) is given by the relation \uE{t)) = e-iH^-t0^h\uB{tQ)).tQ) = e-iH(t-t<Mh. i. .178 CHAPTER 10. Let \UE) be an eigenvector of the Hamilton operator H with eigenvalue E. Quantum Theory: The General Formalism We mention in passing that.t0). since 5(x) = . H\uE(to))=E\uE(to)).L fdpeipx. (10.4i) We assume first of all.t0) = HU(t. i. it follows that 5'(x) is an odd function of x.e. .to) is called the time development operator. (10. See Eqs. (5.to) exists which is such that^ m)) = u(t. 5.e.x") = (x'\p\x")*. so that U{t. For the way the state of the system develops in the course of time we make the following postulate: Postulate: A linear operator U(t.43) (10. (io.e.

(10.*l)^(*l.47) e.49) . (10.t0) = U(t + we have -U(t t) = lim ^ + Mo)-^(Mo) = n (10-46) = U(t. But obviously (cf.41) we obtain |^(t 2 )> = = J7(t2.10.-z / dt'H(t')U(t'. Hence U(t. Eq.45) has very general validity.tiM*i. (10.e.4 Equations of Motion 179 We now demonstrate that Eq.to) can be expressed as an integral as the solution of Eq.t)-l]U(t.t) = l-^eH{t).44)) we have U(t.t) and therefore [U(t. Then with U(t + e.t0) [U(t + e.*l)|^(*l)> = I7(*2. t0) = E/(t 2 .48) with U(to.t 0 )|^(*0)> U(t2.48): [/(Mo) = 1 . t0) = H(t)U{t. (10.to) (differentiating we obtain immediately Eq. Ufa. (10.t0)\^(t0)).t)U(t.t) = 1.to) = 1. t0) (10.t0)U(t0.t0)]-1 Let us set for small values of e: U(t + e.t). In the immediate considerations it is not necessary to restrict ourselves to systems with classical analogues. From Eq.t0) or ih—U{t. (10. i. The operator H is d efined by this relation. = U(t0. (10.t) = l.to).48)). Evidently U(t.

Comparison with the equation postulated earlier.5o) ihjtm)) = Hm)).30).51) Since |</. i.41).(£ + dt)) = U(t + dt. the Schrodinger equation ju(t. In the following the subscripts S and H indicate Schrodinger and Heisenberg picture quantities. >(*)> = i. Now let \tp(t)) be the state of the sytem at time t. Then \Mt)) = u(t. (io. Quantum Theory: The General Formalism We obtain the differential equation for the development of states of the system in the course of time by differentiation of Eq.52) is the operator of an infinitesimal unitary transformation (recall that U = 1 + ieF satisfies the unitarity condition UU^ = 1 provided F — F^). requires the identification H(t) = H= Hamilton operator. it follows that U(t + dt. the predictions of different descriptions are identical. One describes as Schrodinger picture the present description of a system in which the state of the system is represented in terms of a vector \ip(t)) which changes with time t. and these remain unchanged under unitary transformations. to) can therefore be interpreted as a product of a sequence of infinitesimal unitary transformations. (10.e. Then the probability to find the system in a state \x) is \(xm2 = (xm(ip\x)With unitary transformations we can pass over to equivalent descriptions.180 CHAPTER 10.t) = l % ^Hdt) \il>(t)) -Hdt (10. Eq.e. The eigenvectors of these observables are also constant in time. the instant at which a measurement is performed on the system.t0)\ii>s{to)) (10.53) . The operator U(t. (5. whereas the physical quantities are represented by observables in "K which do not depend explicitly on t. Since the only measurable quantities are moduli of scalar products.t0) m0)). t)\t/>(t)) =U~ and since H is hermitian. Another such description is the Heisenberg picture. (10.

HH]. (10. As(q.With explicit time dependence. we have to take into account the contribution dAs/dt.58) HHAH (U*ASU)(U^HU) [AH. Here H is the Hamilton operator in the Schrodinger picture. we obtain ih^-AH at = = -U^HAsU + ihU^^U at + U^AsHU (10.57) and rf[As.p) in the Schrodinger picture dAs/dt = 0.48). (10.HH].55) Thus AH is explicitly time-dependent.10. It does not always have to be the case that As does not contain an explicit time dependence.54) Observables transform correspondingly with a similarity transformation.H]U = = = U]ASHU AHHH — -U^HASU . even if As is not time-dependent.e. However the associated observables are time-dependent.59) This equation is called Heisenberg equation of motion. .4 Equations of Motion and \^H) = U\t. in order to obtain the Heisenberg picture (description).e. t0). for instance in cases of non-equilibrium or if part of a system disappears through absorption. For As(q. (10. subject to a continuous change.HH]+ihU^U.H]U + iW]^-U. but here.(tf HU)(rfASU) (10. = so that ihjtAH or ihjtAH = = [AH.t0)\ips(t)) = \ips(to)) = constant in time. if ^ = 0. i. t0)AsU(t. In the Heisenberg picture it is HH = U]HU. {AH. i. Differentiating Eq. a scalar product) remains unchanged: AH(t) = U\t. 181 (10. as a rule.e.56) U^[As. Thus. so that a matrix element (i.p. (10. we consider only cases without explicit time dependence of As. the vector space of the Schrodinger picture (description) is transformed in such a way that the state of the system is described by a vector \I/JH) which is constant in time. (10.55) and using Eq.i).

An observable CH is called a conserved quantity. as in the transition from the configuration or position space representation to the momentum space representation with the help of Fourier transforms.61) Thus conserved quantities commute with the Hamilton operator. In particular. Solution: We have for an observable A (A) = (i>H\AH\TpH). In the treatment of the pictures we used unitary transformations of operators (and vectors). We end with a word of caution. The "pictures" just explained — with regard to the time development of a quantum system — are not to be confused with the representations of the theory.Pi): and where the expressions on the right follow from Eqs. . In many cases the Schrodinger picture is more amenable for explicit calculations.21b). [CH. We also observe that the momentum pi is conserved when \pi. if ffCH = 0. we used in essence unitary transformations of matrices. since in the transition from one to the other the commutator remains unchanged.13) to derive the uncertainty relation for energy and time.Pi we have in the Heisenberg picture (replacing in the above AH by qi. Example 10. In our treatment of representations. We see that formally one obtains Hamilton's equations of classical mechanics.2: Energy and time uncertainty relation" Use the Heisenberg equation of motion and the general relation (10. in the case of the fundamental dynamical variables qi. Quantum Theory: The General Formalism Wave mechanics is now seen to be the formulation of quantum mechanics in the Schrodinger picture. i. (10.HH}=0. In the Heisenberg picture essential properties of a system can frequently be recognized more easily in relation to their counterparts in classical mechanics — in both cases the time development of a system is given by the time dependence of the dynamical variables.H] = 0. For an extensive discussion see B.e. This holds also in the case of the Schrodinger picture.21a) and (10.182 CHAPTER 10. provided the Poisson brackets in the latter are replaced by commutators. Finally we add a comment concerning conserved quantities. A. Orfanopoulos [224]. (10.

HH]) Hence we obtain by ih-(A).7a). 2 d{A)/dt " and. since \IPH) is time-independent.2 > -h. (10. We construct the following vectors using Eq.10. of A\<t>) = a|<£)) are then called "good quantum numbers". dt in.62) and dAs/dt > \WAH\i. i.)\2 ( = \{i. this is practically the same as at time t for times t with \t — t'\ < T.e.63) so that with Eqs. > (10.e.p) we then have 0=^M) = U[A. i. we can replace ([AH. This means that for a physical variable A(q. and TA O C (10.15): (AAf(AH)2 Using Eq.H}\f)\2. (10.e. 183 dV ' \ dt >"•"»» + ( £ ) • (10.Such a characteristic time interval can be defined for any dynamical variable. The probability density P ( r ) = | ^ ( r ) | 2 = (^|r>(i#> is independent of t. i. at ^±-AH>h. We defined earlier stationary states as states with a definite energy E and wave function * = V( r ) e x P ( — i E t / h ) .64) = 0.4 Equations of Motion Then.67) 1 d(A)/dt is oo and AE = 0. in a different form: The eigenvalues of an observable are called "good" if [A. a position or momentum observation is independent of t. AW = {A(A»|V>.62) The Hamiltonian H does not depend on time.\HAM\2) -\(nA. H\tl>) = {H(H))\f).13) and (10. then if a measurement is made at time t'. (10.e. (10.H]) = 0. (10. (10. If r is the smallest such characteristic interval of time. Let \ip) = \"4>H) represent the state of the system.68) To put it The eigenvalues a of A (i. H] = 0. if we set AH = AE we obtain TAAE (10. . is displaced by A A in the interval A T = TA. (A).65) and rA AA d{A)/dt' .66) This means. the "centre of mass" of the statistical distribution of A.

71) From Eq.t) — ipE(x. as is observed.e.Eo(x)e-iEot/he'^2h9(t). (10. This means that in the case of decaying particles. Recall for instance radioactive decay.184 CHAPTER 10.(x)fEo(E')e«E-E'Wh 2irh f dE'^E.H ) ( £ . t) = ^(x) exp(—iEt/h) describes states called bound states of unlimited lifetime. i ) = J dE'^El^)e-iEltlhfEo{E').70) we obtain: oo / -oo TPEO (x.69) and (10. The wave function of such a state with energy EQ and lifetime r = h/'y > 0 could. for instance.tyEt/hdt = = = / J—oo dt J dEJrl>E. Quantum Theory: The General Formalism 10.E1) 2irhfEo{E)4>E(x).69) (Observe that if ip consists of only one state with energy E in the domain of integration of E'.. the wave function ipEo(x. (x)/ E o (E')5(E .70) by From Eqs. (10.69): oo poo p / -oo *PEo(x. But the number of radioactive (thus excited) nuclei decreases exponentially. First we obtain from Eq. have around EQ the form ipEofct) = i. i. Energy and decay length of the radiated a-particles are characteristic properties of naturally radioactive nuclei. States with finite lifetime (At = r ^ oo) are those whose probability density falls off after a certain length of time called "lifetime". the wave function ^ ( x . then fE{E') = S(E-E') and integration with respect to E' yields again t()E(x. (10. (10.5 States of Finite Lifetime When E is real and P(x) is independent of t and J dxP(x) = 1.£ ^ + n/2)^ o ( x ) e{^1/2+i(E-E )}t/h 0 o =0 . t)e iEt h ' dt = V>£0 (x) / JO <fte{-7/2+i(i5-£b)}t/ft (10.)exp(-iEt/h)). states with sharp energy (AE = 0).70) we can determine the function fEo{E) integration with respect to t.72) = V'EoW -7/2h + 0i(E-E )/h -| oo .t) of such a state must be the superposition of states with different energies about EQ and a corresponding weight function / so that ^ ( x . (10. On the basis of the uncertainty relation AEAt > h these must have an uncertainty AE in their energy spectrum. This means. the particle density diminishes in the course of time (t > 0). /-oo (10.

to) be an approximate but unitary solution of this equation. The state with lifetime r — h/'y is not a discrete state. (10. i. 7 > 0. (C/(0) unitary). i. Eqs.74) Thus the solution of this equation is the main problem. (10. ih-U(t.toM{to)).41). This description contains effectively parts of the other two pictures and is particularly useful when Eq. .6 The Interaction Picture One more motion picture of quantum mechanical systems is in use.e.10.76) **For specific applications see e.e. the formula says that JEQ{E) possesses a simple pole at E = -Eo — 27/2.75) (10. (10. that** EQ.e. Considered as a function of E. Let U^°\t.45).45) can be solved only approximately (in actual fact. (10. They are called "resonance states". (10. 10. V-E(X) ^ ^ (£ > = -dbfi-ft+T/*- (la73) This result is known as the Breit-Wigner formula.e.105) and (20.g. We obtained the result (10. (14. where U is the solution of Eq. We saw previously that if the state of a system is known at time to.73) precisely by assuming with fEo{E)^8{E-EQ) a "smearing" of states around the energy EQ (with uncertainty AE). We see therefore that the states with lifetime r < 00 and TAE > h belong to the continuum of the spectrum. A state is discrete if its immediate neighbourhood (in energy) does not contain some other state. £/ = £/W(Mo)^'(Mo). of course. W)) = U(t. which means that a nucleous with discrete energy EQ does not possesses some other admissable level close to EQ.t0). 185 i.11). i. and is called "interaction picture" or "Dime picture".to) = HU(t. |^(*o))> then \xp(t)) for t > t0 follows from Eq.e. i. whose real part EQ specifies the energy of the state and and whose imaginary part 7/2 specifies the lifetime r = h/j. every unitary transformation of states and operators defines a possible "picture").6 The Interaction Picture Prom the last two equations we deduce for E close to ipE0(x).

h^U> = at ( io. (10. (10.t0) 0 = l. H(°Ht) = ih^uW.77) C/(o)t itiuWfLu^HUMu'-ih^-U' at or i.81) But [/(°) is unitary.e. i. £/(°)£/(°)t = 1. be defined by the relation H(°)rj(0) _ i h ^ l = 0 (exact)> n (0) = e-iH(o)t/aj (1080) i.78) (10. with Eq.e. varies only slowly with t).82) (10.e.78) at with the initial condition c/(0)t f ^ ( 0 ) _ ^ ^ 1 V \ at J U\t0. Quantum Theory: The General Formalism ih^(uWu') i.e.186 Then CHAPTER 10. dt so that #(0)t ( t ) It follows that H{0\t) = dt = H<®{t).e. i.79) ^'~°i.e. [/' has only a weak ^-dependence (i. For a "good" approximation U « [ / ' ' we have HU^-ihjtU^^0. (10. i. . Now let H^(t).e. = HU<®U'. In addition we have (10. with Hamiltonian H = H^ + H'.83) ih^-U^ ishermitian.

U^H'U^U' (10.86) at = U^HU^U'. (10. (10. lfe(*)}. and obtain from Eq.85) = where we used Eq.92) .6 The Interaction Picture 187 We use now the subdivision of H into unperturbed part and interaction.78). We multiply this relation from the left by and from the right by U^ and obtain umHU(o) = c/wt (^WtjW) + umH>Tj(°) i n u ^ ^ + U^H'U^.87) = We define correspondingly as interaction picture state |iM*)> = t/ (0). (10.80).90) = U^H'\^s(t)).91) Since the Schrodinger equation is given by ihjt\^s{t)) = H\1>s{t)) = ( # ( 0 ) + H'Ms(t)). (10. We also set H^U^H'UW.e.89) where |^s(i)) and As are state and observable in the Schrodinger picture. (10. i.85) multiplied by U') on the right of Eq. H = H{®{t) + H'.84) U^ where H' is also hermitian.U^HU^U' H'fi'. (10. (10.88) and as interaction picture version of an observable A the quantity Ar(t) = tf(°>t AgU^. Then ihjt\Mt)) = ih~u^\Mt)) = u(0)ihi^W and H'^it)) = U^H'U^U^\^s(t)) + ih(^) \Mt)) (10. (10. U<®\Mt)) = \Mt)).78) at (10 85) (10.10. (10. (10. We now multiply this equation from the right by U' and insert the first expression on the right (of Eq.

(10. i. cf.94) _Hf)Al 1 + lhdAl+AlHf). Since H'j is assumed to be small (i. dAs/dt=0) .60a) or (10. the vector varies with time (different from the Heisenberg picture). (10. (10. near zero and \tpi(t)) is almost constant in time. ^ (0) l^(*)>.e.55)) at + imm<Ms_u(o) at dt _umH(. we have i. (10.e. so that with Eq.93) Thus the vector ipi(t)) satisfies an equation like the Schrodinger equation. i.89).90) and replacing the left hand side of this equation by the right hand side of Eq.e.89)^0. a perturbation so that H ~ H(°)). ot i.e. Quantum Theory: The General Formalism we obtain by starting from the right hand side of Eq.H\0)}. we differentiate with respect to time the operator defined by Eq. (10.94) Then we have (normally with dAs/dt dt (io=8o) — 0. (10. comments after Eq.95) Thus in the interaction picture the physical quantities A are represented by time-dependent observables which satisfy a type of Heisenberg equation (10.188 CHAPTER 10.80) (and multiplying by t/(°)t) ih~\Mt)) = u<MH'uM\Mt)).92) ^°)^|V-/(t)) + ^ ( ^ ) | V ' / W > = ^ (0) ^ (0) l^(*)> + ^ . (note the difference between H' and H'j-) ihjt\Mt)) = H'AMt)). H'j = UWH'UW. (10.e. A^t) = U^AsUi0).60b) with iff0) instead of H.o)AU(o) at +uWAsHWuW -UWHWU^UWASUW + iww^uw dt +U^ASU^U^H^U^ (10. In order to obtain the equation of motion of an interaction picture observable Ai.e. (with ihjtAI(t) = [AI. i. (10.

99) (10-98) Instead of the original Schrodinger equation we now solve Eq.100) (10. {ipn\<Pm) = Snm.e. (10. (10. Since .80) as E/(°)(t.t0)H'uW(t. We obtain the operator U^ from Eq. i. (10. H' = H'(t). i. (10.96) in which the perturbation part H'(t) depends explicitly on time and HQ replaces H^0' in the previous discussion.94) Hi = uW(t. We have the Hamiltonian H = H0 + H'. ^ \<Pn){Vn\ = 1n (10. the system is in the Schrodinger eigenstate \ips ) of iJ 0 with energy Em.i0) = e-<Ho(t-to)/ft.87) and (10.10.98). ih±\Mt))=Hi{t)\Mt)). H0\ipn) = En\(fn).101) Iteration of this inhomogeneous integral equation yields the so-called Neumann series.7 Time-Dependent Perturbation Theory 189 10. We assume again that the spectrum and the eigenvectors of Ho are known. (10.e.t0). \Mt)) = +' \Mto))-^fdt'Hi(t')\Mto)) ft J dt' f dt"Hi{t')Hi{t")\^I{tQ)) + (10.97) The equation to solve is Eq.93).102) We assume that before the perturbation H' is switched on at time to.7 Time-Dependent Perturbation Theory We consider time-dependent perturbation theory as an application of the interaction picture. Integration with respect to t yields \Mt)) = \Mto)) ~ \ \ dt'Hi{t')\^j{t')). (10.e. i. where according to (10.

n We wish to know the probability for the system to be in a state \ips ) n at time t after the switching-on of the perturbation H'. The corresponding amplitude or appropriate matrix element is n ( 4 ° V s ) = 5> m (i) n (v4 0) |V4 0) >™ = m fl "W- ( 10 . This probability is the modulus squared of the projection of the Schrodinger state \ips) o n t o \ips )„.190 CHAPTER 10.^ V ™ ) . we can write l 4 0 ) ) m = e-lHot/hWm) =e .105) rt n Jto h ' Jt00 t rt - x - l h ' Jto t0 (10. (10. Hence we set |^> = £°«(*)l4 0 ) >n. (10.103) The actual state \ipg) of the system is solution of ih-\^s) = H\^s). (10. (10. (10.104 ) Here we insert Eq. We express this state \tps) as a superposition of the states \ips ) n of #o with coefficients which as a result of the time-dependent perturbation depend on time. assuming that this supplies a sufficiently good approximation: IMQ) = \fm) -l-z\ n dt'HW)]^).108) . (10. i.106) We insert this expression into Eq. the time dependence of \ips } can be separated as for stationary states.103) for m — n and use Eq. Quantum Theory: The General Formalism HQ is independent of time.107) Jto With this we obtain from Eq.88) and obtain > n ^ V s ) = {<Pn\eiHot/h\ll>s) = {<Pn\eiHot/hUM\Mt)) = <¥>*(*)>• (10. (10.105) In the interaction picture the iJo-state m at time t is \Mt)) = eiHot/hWP)m = \<pm).e.102) and truncate the series after the first perturbation contribution.

Thus we set H'(t) = H'0(t).111) Different from Eq.109) the transition probability Wmn(t) = jp Jo . A further example is provided by a-decay.En (10.110) With this we obtain from Eq. (10. where the momenta of the a-particles form a continuum.10. This is the case. for instance. (10. or if by emission of a photon with continuously variable momentum an atom passes from one state into another.Jr(En—Em)t _ ^ (ipn\H'\(pm) {En — E„ sm{±(En . if in the scattering of a particle off some target its momentum p changes to p'.8 Transitions into the Continuum 191 Hence the probability for the transition of the system from state m into the state n ^ m is K(t)\2 = I fdt'e^.109) 10. (10.25) we here have the positive quantity (note t in the denominator) s i n 2 { ^ ( £ n . t o = 0. < -KO?t .Em)t' (<Pn\H'(t')\pm) = Wnm(t). We assume that the perturbation is switched on at some time to.Em)t} \{tfn\H'\ipv h 2h(En . » — < (10.8 Transitions into the Continuum In many practical applications one is interested in transitions into a continuum.Em)t] _ sin2 at aH {±{En-Em)}H Consider the function St{a) :-1 sin2 at ir a2t t 7T for a —• 0. One reason why we begin with these simplifying assumptions is also to obtain reasonably simple expressions. but that otherwise the perturbation is time-independent. (10.112) for a^0.

if the energies belonged to the discrete spectrum. (10. 10.114) It follows t h a t for large but finite times t Wmn(t) 2vr ~ — t5(En Em)\(vn\H'\tpm)\2. > We see t h a t this quantity has the same behaviour (in particular for a — 0) > as the function 5K(x) we considered previously.2.Em)t} r i 7 TTT. and no other state would be nearby. . Hence lim St(a) = 5(a).113) {±JEn-Em)¥ or n m irtSt En — ^ 2h 1 s i n 2 { ^ ( £ n .2 The delta function for £ — oo.ii5) This expression has a well defined value unequal to zero only if the energies En belong to the continuum. though sufficiently small. t—>oo Hence also (shifting t to the other side) sin2{^(ffn-. so t h a t Wnm would be zero.Em).192 CHAPTER 10.io. In the case of transitions into the continuum we have to consider the transition probability into the interval dEn at En. We assume t h a t the matrix elements for transitions into this infinitesimal element can be taken . Quantum Theory: The General Formalism T h e behaviour of this function or distribution is illustrated in Fig. 10.— = o (En 2h En 2h5{En . Otherwise. Fig. we would have En — Em ^ 0.E7m)t} = (10.

(10. sin 2 (tx) ax i—-( * According to F. Then. Fermi [92]. so that initial and final states belong to the continuum. Solution: For test functions <p{x) € S(R) the delta distribution is defined by the functional I &(x)4>(x)dx = <j>{0). 142. F Jo .3: Application of test functions to formula (10. In the case of the functional (10.4). r = J E ^ W =p(£m)y|(</>n|#W|2n (10-117) The formulas (10. Fermi gave this formula the name "golden rule" .10. Example 10. _ 2irh AE > > Se. Pauli [227].r Z x 1 ^ ) = 4>{Q).W .117) were derived by Pauli in 1928.e.113) Use the method of test functions to verify in the sense of distribution theory the result (10. p. if p(En) varies only weakly with En.t The result makes sense as long as the uncertainty AE in the energy satisfies for finite but large times t the relation . (10.115) and (10.118) where 8e is the separation of states around En which in the case of the continuum goes to zero.116) The transition rate V is the transition probability per unit time. pp. for instance from a potential. '"Recall also the integral J ^ ° dec sin 2 x/x2 = TT. 266. 75. If we are concerned with pure scattering. i. in the article of W. .* In view of the usefulness of Eq. (10. (10.117) in many applications.117) provides the outgoing particle current (see Example 10.113). the transition probability into this set of states is J2Wmn(t) = fdEnP(En)Wmn(t) n •* = p(Em)2lTl{iPnlfliPm)l2t. We evaluate the integral 0(0) = 1. f See E./ t->oc TZ J_00 dx . then Eq. t We choose a test function which falls off at infinity faster than any inverse power of |x|: <Kx) = e .113) this is* hm .8 Transitions into the Continuum 193 as equal. Schwabl [246]. Let p(En)dEn be the number of states in the interval dEn.

V +(a-b)2 a _x 2p6 7 m —^ '.117) for H' = V(~x) the differential cross section da _ / n V dH ~~ \2Trh) f dxV(x)e where K = k m — k n . . . and hence . with n . with v the velocity of the particles. .tan p 2 + a 2 .947. Ryzhik. We wish to know the probability of transitions per unit time into the infinitesimal angular region dO = d(— cos6)d<p around the direction of the angles (9. .^ V v 4 ) = ^ T 3 e * k ° *^ko)e*° * = — v . S. = 0 . Solution: We set ¥>n(x) : so that {<Pn\H'\<pm) = -^3 / d x e i K ' x V ( x ) .a 2 ' i that for a = b = t . The ingoing particle flux is therefore v / L 3 .( 1 / 4 ) ln(l + 4t 2 ) + t tan _ 1 (2<). 77^ + .<p). = ( — ) dk= 2ir (— | 2irJ k2dkdfl. 491. which is the number of particles scattered per unit time into the solid angle element dQ with only one particle sent into the volume L 3 (since the incoming wave is normalized to 1).„I.. 2 T / ^ M.? 2TT / L \3k2dkdUn 1 /V(x). and is obtained as ih.194 CHAPTER 10.. K = k 0 . . Jin § m = See I. x . p.. * ^ = . § We have r Jo -\p\x dx sin ax sin bx - V.k. Hint: Use for ip„(x) a plane wave ansatz with box normalization (volume V = L 3 ) and obtain da from the ratio of outgoing and incoming particle fluxes.( ^ V V • < ih 2 _ i k n . . Hence the result is 0(0) as expected. 1 . . . Gradshteyn and I. . p = 1 we obtain I = . Then ^ . Quantum Theory: The General Formalism with the help of Tables of Integrals.4: Differential cross section from the "golden rule" Obtain from the "golden rule" (10. L3/2 ¥>m(x) : L 3 /2 The periodic boundary conditions ipn(xi) = fn(xi ± L/2) imply kiL/2 = ±nj7r. M. Then p(Ek)dEk Since Ek — h2k2/2[i. E x a m p l e 10. .. ^o n . 2 . formula 3.b2 4 p 2 + (a + b)2 2 b _j 2pa V 1 fc Urn . E = h2k2/2fi.e " 1 * 1 = 2 lim — . dEk = h2kdk//j. L \3k2dkdUn It follows that T./ cfc a n p 2 +fc 2 .ln(l + 4t 2 ) + — t a n _ 1 ( 2 t ) = — tan 4t7T tTV T J-< (oo) = 1. . iknx .

t) = ^On(t)^(x.^°>(E.t).9 General Time-Dependent Method For a better understanding of the "golden rule" derived above.69)./continuum = <San(£)</4°)(x. 1 rfx^°)*(x)^)(x) = 8mn.9 General Time-Dependent Perturbation The differential cross section is defined as da •• Theory 195 outgoing particle current x r2dfl ingoing particle current da / /i V (hko)/(lJ-L3) J dxV(x)eil where for purely elastic scattering |k| = |ko|. pp.119) only the time dependence of HQ can be separated from the state \ip) in the form of the exponential factor of a stationary state. (21. (10. 10. Here S stands for summation over the discrete part of the spectrum and integration over the continuum with the orthogonality conditions J and fdx.x) = S(E-E'). The problem is to solve the time-dependent Schrodinger equation ih^\i(>) = H\r/>). specifically in the case of the Coulomb potential. we present another derivation using the usual method of time-dependent perturbation theory. (10.122b) S e e also L.120) In the case of Eq. see the discussion after Eq. t) = (xl^) therefore in terms of the eigenfunctions of Ho with time-dependent coefficients.122a) (10. 148.^ We start from the Hamiltonian H = HQ + H'. Thus we assume that the eigenfunctions (fn = (x|</?n) of Ho are known.t)^°\E^t)dE (10. We expand the wave function V( x . Thus we set ^(x.t)+ / n a{E.x)^(E'. 193.119) where H' = H'(t) is a time-dependent perturbation term and Ho\ipn) = En\ipn). by which we mean without leaving the Schrodinger picture. For a possible application of this result.10. (10. .121) . Schiff [243]. (10.

. We multiply this equation from the left by </?£(x) and integrate over all space.196 CHAPTER 10.„„-.129) .x). (10.126) + H')<pn^)e-iE^h}. t) = San(t)<pn(x)e-iE^h. / *. i. ^0)(E. (10. since f rJ>(-°>(E.t) = e-iEtlhy(E.E'). f 6kn in case of discrete tp's.„.E') = 5{E . Observe that these conditions do not contradict Eqs.127) i n c a g e Qf c o n t i n u o u g £. H'kn = J d x ^ ( x ) # V n ( x ) = (<pk\H'\<pn).126) ihak{t)e-iEktlh where = San{t)e-iEntlhH'kw (10.e.x.t)rJ><®(E'.120).124) and obtain ihSdn(t)ipn^)e-iEnt/h + San{t)En<pn(x)e-iEnt'h = San(t)(HQ + H')vn{*)e-iEntlh = San(t){En + H')vn(x)e-iEnt/h. . Then J d^*k(X)[ihSdn(t)cpn(X)e-iE-t/h = J d^*k{x)[San{t){En + San(t)En<pn{*)e-iEntlh] (10. (10.123) We now insert these expressions into Eq..i) = e. . = e^E-E'^h5(E Thus with Eq.x. (10.. .t)dx.122b).x. The second term on the left of this equation and the first term on the right cancel out. dnptMrnW = { ^ _ En) (10. (10.iJ5 »*/Vn(x). into dip (10.127) we obtain from Eq.128) .. Next we use the orthogonality relations . (10..125) where an(t) — dan(t)/dt. (10.. V>(x. . Quantum Theory: The General Formalism The discrete and continuous eigenfunctions of HQ are Vi 0) (x.122a) and (10.

e. we obtain the equations: iha[°\t) iha£\t) iha[s+1\t) = = = ••• .^ [a^(t) E h + ---} + \a£\t) + A2a£>(t) + • • • ] .10. let's say am / 0. to perturbation theory in the lowest order. Thus first we have ak0) = const. (10. and we set ak°}=5km or ak0) = 8(Ek-Em). (10. (10. 0.e. (10.132) These equations can be integrated successively. i.130) we develop a perturbation theory.134) .131) and insert this into Eq. i.130) H'kn by \H'kn and at the end we allow A to tend to 1.130) This result should be compared with the Schrodinger equation (10. Thus A serves as a perturbation parameter which permits us to equate coefficients of the same power of A on both sides of an equation. determine the state of the system at time t — 0. In the first place we replace in Eq. We obtain ih{ak°\t) + \ak1\t) + \2ak2\t) = SXH'kn(t)e^. Comparing coefficients of the same powers of A on both sides. We observe that the amplitude ak of a definite eigenfunction tpk has effectively replaced the wave function ip.9 General Time-Dependent Perturbation We can rewrite the equation as Theory 197 iMk{t) = SH'knan(t)e^Ek-E^h. In order to solve Eq.130) along with the parameter A in front of H'kn. (10. they are fixed by the initial conditions. Here we assume (our assumption above): At time t = 0 all coefficients ak are zero except one and only one. (10.133) The numbers ak = const. SH'kn(t)e^~E^ha^(t). Now we set ak(t) = ak0){t) + \a£\t) + X2a[2\t) + ••• (10. We restrict ourselves here to the two lowest order equations.120). since it is equivalent to the latter. SH'kn(ty^-E^l^\t) (10.

Later we shall discuss scattering off a Coulomb potential in terms of the so-called Born approximation of the scattering amplitude which we there write f(9. (10.135) assumes a particularly simple form if — as we considered previously — the perturbation H' is taken to be time-independent except that it is switched on at a particular time t = 0 and is again switched off at some later time t > 0. ip) in the context of time-independent perturbation theory.135)? The quantity ak\t) is the amplitude which determines the probability that as a result of the perturbation ^H'km the system makes a transition from the initial state m (i. i.e. (10. an expression of the following type AS?) = " 2 ^ 2 / e j ( k .Emy in agreement with Eq.136) Comparing Eqs. H'^y^"-^^/hd£.132). we obtain it*™ = f J—oo SH'kn{t')e^-E^'lh5mndt'. Equation (10. We shall see that in this approximation the scattering amplitude is effectively the Fourier transform of the potential.136) we see that both "amplitudes" are Fourier transforms and look similar. Quantum Theory: The General Formalism depending on whether Em belongs to the discrete part of the spectrum of Ho or to its continuum.135) We put the additional constant of integration equal to zero.198 CHAPTER 10. (10. In this case we obtain from Eq. What is the physical significance of Eq. ) ' X V(x')dx'.ml2sin2{^frn (En . Integrating the second of the equations (10.e.137) . (10. ipm) into the state k.k .111).135) and (10. whereas ak is a Fourier transform in time (a distinction which is easy to remember!). (10. so that ihakl\t = -oo) = 0. (10.e. i.135) / Jo L H-frfc ~ Em)/h Hence the probability to find the system at time t in the state n ^ mis (with A-l) 22 i rjv I1 r i r IW |a(l)m|2 K ( j l = i^kl\t) = H'km e^-E^dt> = H'km 6 \nnm\ ei(En-Em)t/h -i(En-Em)t/h _ j [En-Emy 4|g. The amplitude / B o r n is a Fourier transform in spatial coordinates. ifm{kl) = f J —oo (10.

2. Angular Momentum In considering the Coulomb potential in a realistic context we have to consider three space dimensions.y. 199 (1L2) .Chapter 11 The Coulomb Interaction 11.z) be the vector separating two particles which are otherwise characterized by indices 1 and 2 with electric charges Z\e and Z^e. The Coulomb interaction on the other hand is of specific significance for its relevance in atomic and nuclear physics.m. The Hamilton operator is then given by w= jt + ji_^M. Let r : = (x.i be respectively the momentum.2 Separation of Variables. Z<i = Z) y(r) = _ M ^ . the position coordinate and the mass of particle i with i — 1. ( 1L1 ) We let pi.Ti.1 Introductory Remarks The quasi-particle quantization of the harmonic oscillator is of fundamental significance in view of its role as the prototype for the quantization of fields. Both cases therefore also serve in many applications as the basic unperturbed problem of a perturbation theory. The electrostatic interaction of the particles is the Coulomb potential (later we take Z\ = 1. 11. Both of these important potentials are singled out from a large number of cases by the fact that in their case the Schrodinger equation can be solved explicitly and completely in closed form. such as the electromagnetic field (the quantization in these contexts is often called "second quantization". meaning the inclusion of the creation and annihilation of field quanta). r= | r |.

_d_ _ _d_ dX dx2 ~ dXl d dx2 nifi^i [ ' etc. Y. P = pi + p 2 . Similarly d_ _dx^_d_ dx dx dx\ dx2__d_ _m^_d_ dx dx2 mi dx\ mo_9_ m 2 dx2 etc.4). (11.200 CHAPTER 11.3) for the individual position coordinates we obtain ri=R+ •^ —r. mi + m 2 iTi'2 r2 = R ^ -—r. (11. But now P2 2mo = "ip/pi 2 \mi m p2\2 m2J | = m0/pi2 2 \m\ p22 m2. so that as expected we obtain the expression for p in Eq.2/ mi+m2 The mass mo is usually described as the reduced mass. 2Pi-p2 m\m2 pi-p2 mi + m 2 . mi + m 2 va\ (11. z) one verifies that j9_ = dxi_ _d_ 8X ~~ OX dXl dx2. Z) and r = (x. p are such that as operators in the position space representation or {R. (11.5) The momenta P . Solving Eqs. The Coulomb Interaction We introduce centre of mass and relative coordinates R and r by setting R = m i r i + m2r2 • .U7&.3) and centre of mass momentum P and the relative velocity p/mo by setting r» .4) P = p i + p 2 . so that as expected according to Eq. (H-4) \mi m. r = ri-r2 mi + m 2 (11. . /'Pi p = m0 P2 A v. r} representation they have the differential operator forms Setting R = (X. Pi 2 2mi(mi + m2) and P2 2(mi + m 2 ) Pl"»l + pjmi 2m 2 (mi + m 2 ) (P1+P2) 2 2(mi + m 2 ) P2 f f l 2 Pi P2 ^ 1 1 7 ^ 2mi(mi + m 2 ) 2m2(mi+m2) mi + m 2 ' . where m 0 = mim 2 . y.

[Ri.e.Ri.Ri.+ ^(r) 2(mi + m 2 ) 2m 0 (H. 2(mi + m<i) 2mo 2m\ 2m. (11. P on the one hand and those in r.r2) to (R. (11-14) m0 We observe that the equations of motion in R.2 Separation of Variables. Canonical quantization of the classical theory implies that we pass from the Cartesian variables rj. Pj over to operators ?i. The equations of the relative motion have the form of the equations of a single particle of mass mo moving in a potential V(r). (11. etc.j = x. (11. The canonical quantization must always be performed in Cartesian coordinates.pj.il) ""Op' i. (i. (11.12) (11.2 It follows that p2 2 P2 n2 n2 n2 (H.8) and so W =777 v + 7 T . ' d and 'AW OH'ATJ P = - ^ dR <9# <9r ' (ii. r) is a canonical transformation. p = -W.10) One can convince oneself that the transformation from (ri.+ ^2-. r = —. Angular so that Momentum 201 ^ ^ .10) we set H = Hcm + H.pj. we mean n = rx s x. p on the other are completely separated.z).11.= -^. pj.Pj} = ih6ij.15) In accordance with Eq.16) Although we write ri.Pj). (11. Hence Hamilton's equations apply and we obtain. for which we postulate the commutator relations* [ri.9) ft = ftcm + -^ = centre of mass energy + energy of relative motion. The motion of the centre of mass is that of a particle of mass M = m\ + 1712 in uniform motion along a straight line. Ri.+ -P.pj]=iMij. with M := mi + vn^R = ^ P. Pj (which in the following we write again ri. .y.13) * and = M ' 1V1 P = 0.

= rcos9. i. 9.18) In the position space representation the time independent Schrodinger equation is therefore 2M' A fl ) + h2 2mo A r + V(r) * ( R .e.r). (11. 0 < r < oo.19) For stationary states with energy E the equation possesses a complete system of eigensolutions * which can be written tt(R. —n < tp < ir. to that of the Schrodinger wave equation h2 Ar + V(r) tp(r) = Eip(r). where Hcm$(R) = and Hip(r) = with ^total — ER (11. (11.202 CHAPTER 11.21b) + E. The Coulomb Interaction nc H = 2m.24) . V(r) = V(r). (11. In these coordinates we have /\rip 1 d r2 dr \ tdil> dr + r2 sin 9 09 \ S m 1 d_ . 2mo (11.23) y — r sin 9 sin tp. df\ 39 J + 1 d2i/j r2 sin2 9 dip2" (11. as in the case of the Coulomb potential.21a) L*-+v{r)] ip(r) = Ei/j(r) (11. Thus the two-particle problem is practically reduced to an effective oneparticle problem. 0 < 9 < ir. it is reasonable to go to spherical coordinates r. r ) = J E t o t a i*(R.r) = *(R)V(r).22) If.20) r2MAR_ $(R) = ER*(R).Q 2M' + V(r). (11. ip: x z = r sin 9 cos ip.17) (11.

We have I2 = (r x p) • (r x p). n ( .3. (11. k an anticlockwise permutation of 1.29) .32) (11. = (p2 + ^ V (r + 0).3.2.j. 2 1 9 / 1 9 .3.1 it is shown that pr is hermitian provided lim n/>(r) = 0. (11. k = 1. i—>0 (1L25) (11.i ' .26) However.2.11.2 Separation of Variables. (11.j.31) . (l + . is the operator of the relative angular momentum. i.)In Example 11. = —1.24)) 2 + M\ d dr2 J = -tfL^(r^A\ r2 dr\ dr J d2ip (11. that the above expression follows from l:=rxp=-ift(rxV). _ h2 \ . . i( i. = 0. the operator —ihd/dr is not hermitian! The operator pr obviously satisfies the relation [r.2. (11.27) We can now write (see verification below) p V = -h2Ar^ where 2.5r = .pr]=ih. with tijk — + 1 .33) i. (11. if i.28) .12 — I2. otherwise. Angular Momentum 203 We can rewrite this in a different form by defining the operator * : =-«.k a clockwise permutation of 1.e. where (r x p)j = ^eijkrjpk._0di})\ x We now demonstrate that 1.A fc2! d ( d^ ^ = _tfU2dA r\dr and (compare with Eq. j .

ihr • p ) .25)) r • p = —ihr • V = rpr + ih and hence 8 = (r • p)(r • p .44) .ih)pr and x 2 2 9 9 2 2 Q O (11.ihr • p r2p2-(r-p)2 < + ih(r-p).204 It follows t h a t 2_^£ijktij'k' i CHAPTER 11.37) .39) + ih). This means that l2 = so that 9 9 1 (11.36) It follows t h a t (note the part defined as —5) l2 = = ( r 2 p 2 . (11.40) (11.ih5jk)pk (11. * ' -S Since r =xz we have (cf. Eq.41) o = r pr. (11. r— = r • — or or (11.35) j.ih) — rpr(rpr With Eq.43) r2(p2-p2). (11.k ~ rjpkrkPj)- (11.(r • p ) 2 + 3ih(r • p) . (11.15) we have fjPkTjPk = fj(rjPk and TjPkTkPj = = rjPkipjrk rjPj(rkPk + ihSjk) = TjPkPjTk + ~ ihSkk) + ihrjPj.42) (11.27) we obtain r{prr)pr = r(rpr . ih6jkrjpk (11.k Using Eq.j. (11.38) + y* + z .34) W i t h this we obtain — keeping in mind the ordering of r and p — I 2 = 5 Z eijkrjPkeij'k>rj>Pk' = ^{rjPkrjPk i. The Coulomb Interaction = &jj'&kk' — Sjk'Skj'. (11.

! /').Q ' 2mor2 with the condition (11. (pr4>. (11. 8.46) (11.- h f l d —(r<l>*(r. <p)*Pri>(r. 6. ¥ p)r 2 drdn s = (p r 0.6l.l2} = 0. pr = . 0.Pril>) J / r= dQs rcj>(r. i.. 9. (11. 6.2 Separation of Variables.47) Example 11. prip) = = I <j>{r.e.ip) = (<£.28) and (11.6.45) oe Thus in Eq.30) I2 is to be identified with the square of the angular momentum operator. r—>0 il>(r.2.e. ip 6 X>Pr C 7i.[ 4>(r.e lim rip(r) = 0. 9 € [0.<p)drdna " r r=0 J \ll§r~(-r4'(r. <p))r2drdns % J r or r<t>(r. <p)* .30) yields l2 = I2 = sin sine— BinO-^r 09 + d^2 (11. S o l u t i o n : Set d£2s = sin.<P) = Etl>(r.d_ r dr the operator pr is hermitian. Pri/Oi V0.7r].[ i J With partial integration with respect to r this becomes (<l>. H 1 2mo pi + K) +V(r). The Schrodinger equation of the relative motion can therefore be written V1 V 2m.<p) (11.<p))ril>(r. i. Angular Momentum 205 Comparison of Eqs.1 Separation of variables We deduce from the expressions of H and l 2 that [H.— (riP(r. 11.<p))drdns.e.6d0dip. (11. or J . <p) d °° .1: Proof that pr is hermitian Show that if Vv lipe H = C2(R3).. reR3.e'iP^} . Then {<j>. <p)r2dfls = .e. <p)*rip(r} 0.<p G [0. /'(r-.d.e.2TT].48) .26).11.(r. r := |r| : lim rip{r) = o i .<p)*-?-(ri.

(11.lz]=ihlx. H and l 2 possess a common system of basis eigenvectors. ?3 = z).2: Verify that [h. [lyA = 0.206 CHAPTER 11. 11.ly) are simultaneously "sharp" determinable variables and hence have a common system of basis functions. |/z> *xJ = l <i\}ylx + tx'j/Jj [h. [ly.l2] = 0. y. ly\ — —tfi(lylx + Ixlyji [hi h\ = 0. and this implies in the terminology we used earlier that their determination cannot be "sharp" simultaneously. (11. (11. In order to determine the eigenvectors of the operators l 2 and lz one defines l± = lx±ily.ypx) = ihlz. We therefore first search for a complete system of eigenfunctions of l 2 . zpx] + [zpy.e.e. We can now proceed as in the case of the harmonic oscillator and determine their eigenfunctions. they are incompatible variables. z we have altogether [lx. xpz] = y\pz. g i = x. Since 1 = r x p .52) This means that l . Example 11. 2 2 2 = 0.qj]=iheijkqk S o l u t i o n : This can be verified in analogy to the preceding equations (i = l . we have [h.lx] = ihly.e. i. These operators here play a role analogous to that of the quasi-particle operators . 3 .ly]=ihlz. The Coulomb Interaction i.q2 = V.49) By cyclic permutation of x.51) (11.pz]x = ih(xpy . zpx .53) Either of these operators is the hermitian conjugate of the other. i.3 Representation of Rotation Group Our objective now is to develop a representation of the rotation group in analogy to the energy representation of the simple harmonic oscillator. [lz.50) Thus the components of 1 do not commute pairwise. However.xpz] = [ypz.l2} and correspondingly [lx. for any two of the components there is no common complete system of eigenfunctions.lz (or l . z]px + py[z. 2 .lx or l . (11. ly] = [yPz ~ zpy. so that [lz.

i + ] = [l2.e.60) Similarly for l+ with the help of Eq.54) l 2 = \{l+l~ + 1-1+) + ll = l+l. (11. [ig.62) 'At this stage it is not yet decided whether IQ is an integer or not! But we know that IQ is real. (11. A^ in the case of the linear harmonic oscillator. since lz is hermitian. (11. With the help of Eq. (n. (11. l-\l) is eigenvector of lz with eigenvalue (IQ — l)h. (11.Z_] = [ l 2 .* Thus we have (1 being a bounded operator) lz\l) = l0h\l) Then from Eq.54): lzl-\l) = {l-lz . (11.59) = 9) h(l0 -r)\l-r).h~lz(11-56) (11. in particular one defines \l) € CKj as eigenvector of lz with the largest eigenvalue lo. We write or define: |/ .55) One now defines certain ket-vectors as eigenvectors of lz which span the space "K\ of the appropriate states. [i+M = 2%iz.52): 0 = [ l 2 . .11.3 Representation of Rotation Group 207 A.i+] = m+. Therefore (lo + l)h cannot be an eigenvalue of lz. (11. In the present context the operators are called shift operators for reasons which will be seen below.l-h)\l) = {l0 . and with Eq.r) := (Z_)r|Z) so that lz\l . (11.58) {l0 : max.l)hl-\l).49) we obtain the following relations: [iz.51).57) i.54): izi+\i) = (i+iz + i+h)\i) = (lo + i)ta+\i).61) But lo is by definition the largest eigenvalue of lz.i-] = -m-. (11. Similarly we have lg{l-)2\l) and more generally lz(l.Y\l) = (lQ-r)h(l-Y\l).r) (11 = {l0-2)h{l-)2\l) (11.54) On the other hand we obtain from Eqs.+ ll. assuming a finite dimensional representation space. Hence we must have Z+I0=0. y . (11. (11.).

i. IQ(IQ 1 0 (11. [12. (11.Z_] = 0.n) (1 = 60) {l-)n+1\l) = 0. l-\l .68) ..67) i (11 60) = But according to Eqs.n).63) + l ) ^ 2 .+ i2z (1-1+ + 2izh) + 1 \ \i) = (izh + ti)\i) i0(i0 + i)h2\i). l-\l) is also eigenvector of l 2 with eigenvalue Zo^o+l)^ 2 ..e.y\i) l)h2\l-r).n)}h2\l-n).e.56) CHAPTER 11... (cf. (11.64) i.From Eq. Eq. (11.65) We assumed that "Ki is finite dimensional so that the sequence of eigenvectors of lz.h)\l-n) {(lQ-nf-(lQ-.54) (n = 57) (11. (11. i.e. (11.56) that l2|/-n) = (11 6) (l+l-+l2-lzh)\l-n) (ll-l.e. (11.l2\l) = l0(l0 + l)h2l-\l).\l-r) terminates at some number (let us say) r = n (n a positive integer). (11.66) It then follows from the second of relations (11.60) = Vli\l) = l0(lQ + l)h2li\l) i0(io + l0(l0 + i)h2(i.65) for r = n: l2\l -n) = l2(l-)n\l) = lo(lo + l)h2\l .60) and (11.+ 1-1+) + l2z\l) = -i+i. (11. i.64) we obtain on multiplication by Z_: l-Vl-\l) and more generally l2(f-)P|J> (11. The Coulomb Interaction (l+l.\l-l). Since l 2 commutes Hence |Z) is eigenvector of l 2 with eigenvalue with l_.55)) we have 12Z_|Z> = l.208 Thus for l 2 we obtain: V\l) ' = 11. \l).

m) l-\l.m). The phases are chosen such that§ l+\l. .m±l\l±\l.1).. The basis vectors \l)..-l0): l2|Z.---. —loh.n){l0 .m) with — lo<m<lo2 (11.m y/lo(lo + 1) — (m — l)m\l.m + 1)(Z0 + m)\l. (11. lQ = -.m . (IQ — l)h.m)\l. M.m) = IQ(IO + l)h2\l.l)h \/(lo + m + l)(l0 .. We therefore write the 2/o + 1 orthonormal vectors: \l. where lz\l. p. Vol. II.{lo .l-.. The nonvanishing matrix elements of 1 = (l+.65) these are eigenvectors of l with eigenvalues Zo(^o + l)h2. A. lz\l.m\lz\l.n . (11. \l — n) are eigenvectors of lz with eigenvalues loh. we skip this here. 17. p.lz transform the vectors \l).n) 2 . lz) are then: (l.n)} = (l0 ..g.63) and (11..m).e..m) = mh\l. m + l)h. The normalization factor in Eq.. \l — n) into one other. In the following we consider primarily the case of integral values of IQ.m) = = = = y/lo(lo + 1) — m(m +l)\l. y/(l0±m + l)(l0^m)h. (11. Messiah [195]. We write these* for integral values of IQ: \l. *D. According to Eqs. Brink and G. y (lo .l)h.71) can be established in analogy to the method used in the case of the harmonic oscillator. As in the case of the harmonic oscillator we introduce normalized basis vectors. (H-71) These expressions should be compared with the corresponding ones in the case of the harmonic oscillator.67) we obtain (observe IQ(IQ 209 + 1) = —lo(—lo — 1)): *o(*o + 1) = Wo .. 24. The dimension of the representation space is therefore n + 1 = 21Q + 1. -lo = l0-n. R. (11. § See e.m / + l)h ..m) : |Mo).11. Satchler [38].l_.m) = = mh.3 Representation of Rotation Group so that with Eq.m) — mh\l. (11.m) (l.-Z 0 ). | U o ..m).70) they satisfy the eigenvalue equations (with m = l0Jo l.. i.69) Prom this it follows that IQ must be either half integral or integral....1). The operators l+.|2..70) According to Eq.

11.. as this is sometimes described. —lo + 1 . for which the eigenfunctions \ip) are given by l2 l0(l0 + i)h2\ip).75) the variables contained in the expression (11.74) can be separated: 1 d sin 9 36 2 1 sm6— 89 j — sin 0(9) = h(l0 + 1)0(9). d2 sin2 9 dp2 _' (11.yp. and m = —IQ.73) 1 d d sm0 06 \Sm9dd) + (11.1. .77) $((p) = m2<b(y). .e. ip) = e(9)$((p).. . -ih d_ dip' lx ± ily = ±he^ d I ^ ± zcotfl-^ .4 Angular Momentum:Angular Representation In the case of the harmonic oscillator we arrived at the position or configuration space dependent wave functions by considering the position space representation of states. In these coordinates we obtain: h l± so that -h2 = xpv . dp Without prior knowledge about the integral nature of IQ and m we can show that indeed these have to assume the integral values derived above.<p) = i)n2Y^{e^). $ oc eim*. for the simple "rotor". l_\l.72) In view of the spherical symmetry it is reasonable to use spherical polar coordinates r. /o.210 Then. (11. as remarked earlier. 9. where IQ = 0. as above. with i2Y£{e.With the ansatz Y(9. mhY£(9.76) d2 (11.74) We choose the eigenfunctions of l 2 and lz as basis vectors of the ("irreducible") representation. (11.<p) = i0{io + lzY£(0. i. (11. CHAPTER 11. ..-l0)=0. ip. We now perform the analogous procedure for the angular momentum or..2.<p). We . The Coulomb Interaction Z+|Mo> = 0.

(11.78) and the completeness relation oo I ' YrVMYTVrf) = °^~' ± J2 Y7n*(f}.<p) Jo = 6mm'5w.ypx = -tfr-K-(11. (11.(p) are called spherical harmonics. so that im(ip±2ir) The functions Ylm(8. Uniqueness implies immediately that m has to be integral. These functions are defined in such a way that they are normalized to unity on the unit sphere.V>)Yir'(8. orthonormalized system of eigenfunctions.82) .81) V l07T V47T y? = Y? = One should note that the functions Y™ are square integrable only for the given integral values of m and I.=o m=-l sin 6 The connection of these hyperspherical functions with the associated Legendre functions P™. m > 0.77) and o these are determinable by the requirement of uniqueness of the original wave function and its square integrability.n>). and that their phases are such that Yt0(0.(5cos 0-3cos0).\ = W .79) . We mention in passing that in spherical coordinates d lz = xpy .t»'\Y7n((l'. / 3 / 5 J — casO. is given by (2Z + l ) ( Z . The functions Y™ satisfy the orthonormality condition dfilimT' = / Jo < W sm8d9Yr(9.80) y? Furthermore Y°l = V^fcosO). y2° = \ (3cos 2 0--1) 2 V V 4vr V 16?r 3 W—.^ = *(n .11. (11. (11.* ? ^ .r o ) ! YT{e.0) is real and positive.i».76) and (11. This is the case if one demands that l 2 and lz possess a common complete.<p) = (-iy 47r(Z + m)! In particular for m = 0: 1/2 PI11 (cos 6)eimi?. (11.4 Angular Representation of Angular Momentum 211 observe that Z and m define the eigenvalues of Eqs.

For the solution we use the ansatz of a power series. f. Setting y = (lx2)m>2Pr(x) m . If we did not know yet that I = IQ = an integer > 0.m = 0.86) y = 0. i. (11.e. i.. Instead of the values I = 0.1(1 + l)]xK+i = 0.2xP[ + 1(1 + l)Pi = 0.88) we obtain for every value of x ^2 ai(K + 0( K + * . respectively.2(m + l ) x i f + [1(1 + 1) .p. is also in use. (11.. The Coulomb Interaction The number I is referred to as the "orbital quantum number" and m as "magnetic quantum numbed. (11.. (11.e.. oo (11.x2)P[' . Consider the equation for Pi(x). (1 . It is instructive to pursue the appropriate arguments. in the replacement ip —> if + 2-K.212 CHAPTER 11.2xy' + 1(1 + \)y = 0..e. exp(imip). i.2.83) Their connection with the Legendre functions Pi (x) is given by the relation pr{x) with the differential equation =-£^ Pi{x) (iL84) (1 . we would now have to search for those values of I for which the solutions are square integrable. 22 (11.. of Ytm.88) y = ^aiXi+K. The associated Legendre functions or spherical functions PJn(x) satisfy the differential equation (1 .87) we obtain the equation (1 . the designation s.89) Inserting this expansion into Eq. i. (11.x2)y" .Y^O»[(K + i)(/e + i ~ 1) + 2(re + i) .d.85) (11.l)xK+i~2 i .x2)Pf .m(m + l ) ] i f = 0. i=0 (re lowest power > 0).h.e.90) .x2)y" .2xy' + 1(1 + 1) 1-x The number m must be an integer already for reasons of uniqueness of the wave function.g.1.

However. they have to be polynomials (recall the orthogonalization procedure of E.2 + 2mo dr 2mor V{T)-E Viir) = 0 (11.1 we obtain by comparing coefficients QOK{K 213 — 1) = 0. when the series terminates after a finite number of terms. .Qr2 with (cf. . Comparing the coefficients of xK+3 we obtain aj+2(K + j + 2){K + j + 1) = OJ[(K + J)(K + j + 1) .92) This is a recurrence relation.5 The Radial Schrodinger Equation for Hydrogen-like Atoms For i = 0.94) ^For further details see H.93) where Xl(r) is the solution of the radial equation (cf. Eq. (11.1. p. We have therefore (multiplying Eq.78) to be square integrable. Margenau and G. We see that this happens precisely when I is a positive or negative integer or zero. (11. 93. first line of Eq. Schmidt). (11.29)) 2 fe2l 92 Pr = -& r -7Tororz We set (11. (11.11.91) from which we deduce that K = 0. M.*" 11.| > 1.1(1 + 1)]. these are of little interest to us here.94) from the left by r) h2 d2 h2 2 + H! + \)-IL.95) Vl = rxiThe hermiticity condition for pr then requires that (cf. l(l + l)h2 _2mo + 2m.26)) yi(r = 0) = 0.E Xl(r) =0 (11. Eq.5 R a d i a l E q u a t i o n for Hydrogen-like A t o m s The common set of eigenfunctions of the operators H. For the solution in accordance with (11. ai(re + !)« = 0.96) + V(r) . (11. This is precisely the case. Murphy [190]. l 2 and lz are the solutions of the Schrodinger equation of the form (11. For \x\ < 1 we can obtain from this the coefficients of a convergent series. but also those for a different series for |a. (11.46)) p.

r m0 = meMp me + Mv (11.e.99) e ip(r) = Eip(r). s = 1 + 1.1)). for instance.e. Obviously the irregular solution has to be rejected. In the first place we investigate the behaviour of yi in the neighbourhood of the origin.96).101) 2 r ^ We set | 6 = ( _ e ) 1 / a = ( = (11. In order to verify this behaviour we set <VflV>r> yi(r) = rs(l + air + a2r2 -\ ). Z\ = Z2 = 1 in Eq. and substitute this into Eq. (11.1)+1(1+ 1) = 0. L i + + . This applies to the Coulomb potential but also to Coulomblike potentials like screened Coulomb potentials or the Yukawa potential.-1. an analogous consideration applies to hydrogen-like atoms like.214 with CHAPTER 11. (11. oo) of a one-dimensional space. (11. we obtain the so-called "indicial equation!'' -s(s . We assume that V(r) has at r = 0 at most a singularity of the form of 1/r. which behaves like 1/r1 in approaching r = 0. and others. The wave function of the relative motion is the solution ^(r) of the equation -A 2mo Here •4>{r)=Ylm{eM and Vl(r) (11. We now consider the hydrogen atom with V(r) — —e2/r (i.98) 2mo e2 _ 1(1 + 1) yi = o e+ 2 y'l + h r r2 2m0E e = (11.97) / r2dr\Xi(r)\2 = / \yi(r)\2dr. Equating the coefficient of 1/r2 to zero. since it satisfies neither the condition yi(0) = 0.102) . The equation also has an irregular solution. The Coulomb Interaction (11. He + .96) can be shown to possess a regular solution there which behaves like rl+1{\ + 0(r)) in approaching r = 0. nor the condition of normalizability for l ^ 0. Jo Jo We thus have a problem similar to that of a particle of mass m in the domain (0. i.100) with (11. In these cases the behaviour of yi near the origin is dominated by the centrifugal term and thus Eq.

i.e. Oberhettinger [181]. (11.106) W i t h this ansatz we separate from the solution the behaviour around the origin as well as t h a t at infinity.529 x 1 0 " 1 0 meters) 1 V/2 e2f hc\ m0c^1/2 2E 2moE J (11.+-if>-x/2 vi(x).6 The Discrete Spectrum of the Coulomb Potential 215 For E > 0 the solutions oscillate in the region r — oo.105) we obtain the equation for V[(x). (11.107) equation. for E < 0 they > decrease exponentially.105) t h a t for x —• oo there is a solution behaving like exp(—x/2). see e.11. (11. .e.6 11.105) 11.1 Discrete Spectrum of the Coulomb Potential T h e eigenvalues Here we are interested in the solution which is regular at the origin (i. (11. (11. Inserting the ansatz into Eq. d2 x—j 1 + (2l + 2-x) ax d j dx (/ + 1 .g. (11. hypergeometric (11. We can see from Eq. (11. It remains to determine the function vi(x) for a complete determination of the solution. W.108) This equation has the form of a confluent x$" + (6 . We set x — We set as Bohr radius* a = and 1 v = rea rriQe ft V (47Fe 2KT.6. p.e.103) °y moe2 ( = 0.v) vi(x) = 0.100) as^ d2 dx2 1(1 + 1) x v___\ 4 2/1 = 0.a * = 0.104) W i t h these substitutions we can rewrite Eq. We set therefore Vi X J.x ) $ ' . Magnus and F. T This equation is known as Whittaker's equation. i. *The factor 47reo appears in SI units.88. behaves there like xl+1) extended to an exponentially decreasing branch at infinity.

called dupZication formula.(0_i)!(6 + n -l)!n!' (11. ^ ( 2 ^ ) ! = 2 2z z!(z . since the exponential function (generated by the infinite series) would destroy the asymptotic behaviour of the wave function which we separated off with Eq.J. of its argument z with simple poles at z — —n and residues there given by (—l)"/n! as may be deduced from the property (2) above. more precisely.1/2)!.109) In our case we can write the series The expression* r(x + l) = x\ is the gamma function.106). the factorial for an arbitrary argument. (11. meromorphic function. (11. are very strict in reserving the factorial exclusively for positive integers and the gamma function for all other cases. The function T(z) is defined by the integral /-co T(z) := / e-Hz-ldt.112) The gamma function is a nonvanishing analytic. with many of these around in some calculation. (4) r ( i ) = V5F. (5)r(n + l ) = n ! . (11.e. and hence the *Some writers. 02z-l / i\ (3) r(2z) = — ^ r ( z ) r ( z + . . i. (2) r ( z ) r ( l . it is convenient to use only the factorial notation.111) Important properties of the function T(z) are: (1) zF(z) = T(z + l). In general the function $ is an infinite series and behaves at x — oo like » x-{l+l+u)ex_ Obviously such a series is useless in our case.z) = ^f^. However. and for both cases.216 CHAPTER 11. The Coulomb Interaction The regular solution of this equation is the confluent hypergeometric series (also called Kummer series) 1 . (--z)!(z . + ' ' ~ 6 1 ! + 6 ( 6 + l ) 2! + Z. particularly pure mathematicians.! ) ! = imlFI' c a l led inversion or reflection formula.z(z-l)\ = z\.

1 .110) as follows: T(l + 1 + p . 2 .11. if the series terminates for certain values of u. The degeneracy of the levels En is therefore of degree* n—1 1=0 n—1 V(2Z + l) = 2 V Z + n = 2^(n .v) T(l + l-v) _ ~ -K sin{7rQ/ .. i. The quantum number n' is called radial quantum number. This is achieved. n' = 0 . and so to the number of different values of m.V))T(v . . in our case: a = 0. . i. must be a polynomial. every n. the orbital or azimuthal quantum number I can assume the values 0.e.2. This number is equal to the number of finite zeros of the radial part of the wave function (excepting the origin). (11. • • • • <»•»*> It should be noted that the magnetic quantum number m does not appear in this expression.I) r(v-l -p)sm{n(i/ -I . . for z.) 'Observe that with property (2) of the gamma function we can re-express the ratio of gamma functions in Eq. ..1) + n = n(n .. Eqs.1) + n = n2.e.1. One defines as the principal quantum number the number n. 1=0 (11. (11. (11.101) to (11. Thus the infinite series must break off somewhere.p ) } n (-l)Pr(^-Q n'\ K T{v-l-p) ' (n'-p)!' Note that 11 + 1 is the number of possible orientations of the angular momentum vector 1 with respect to a preferred direction. = n = Z + l + n'.104)) 1 n =v = K. For every value En. With (cf.» • » • » .a " \ 2moE we obtain the energy spectrum of the discrete states with angular momentum I ie t m0e2 / 1 N1/2 * = - ( s ) ' ^ . d = 1.e.6 The Discrete Spectrum of the Coulomb Potential 217 latter would not be square integrable.113) This is a quantization condition. .. which are also described as its nodes.114c) (Recall that for an arithmetic series a + (a + d) + • • • + (a + (n — l)d) = na + n{n — l)<i/2. which is thus seen to be a consequence of demanding the square integrability of the wave function. i.

The above treatment of the hydrogen atom has obvious shortcomings: (a) The hydrogen atom was treated nonrelativistically. The spectrum of the hydrogen atom thus has the form shown schematically in Fig.. 11. from an initial state n > 2. (c) The nucleus was treated as pointlike and not as something with structure. ne.1.218 CHAPTER 11... The Coulomb Interaction E=0 degeneracy lEl 3 31s 21s 33p 23p 3 5 d 9-fold IE I 4-fold IE I 1 1 1s none Fig.p.l = 2 to n = 2.4. from an initial state n > 2. hence p). respectively. .'.1 Hydrogen atom energy levels. . 11.1.§ One also writes n2l+1l. (b) The spin of the electron was not taken into account. . The following spectroscopic description of states is customary: nl —> ns.. where the superscript 21 + 1 at the top left of I gives the degree of degeneracy..1 = 1 (called diffuse series. I = 0.1 = 0 (called sharp series..'. ' O n e describes as Lyman series the final state with n = 1. / = 1.. from n. and from n. 2 . . and hence the fine structure of the energy levels is excluded. hence d). nf . stand for I = where n denotes the principal quantum number and s. One describes as Balmer series the final state n = 2.. 0. hence s). .\ I = 0 or 1. nd.1 = 0 (called principal series. I = 0 to n = 2. relativistic corrections are of the order of v2/c2 ~ 0(En/moc2) as we illustrate in Example 11. np.1 = 1 to n = 2.d.

Vol.2. (11. . "1'f„+11" P] „. the polynomial contained in y\ of Eq. Unfortunately.115) we can re-express this function as Lr{z) - r\k\ p=0 \* p\ (r-p)!(fe+p)!- (U -118b) ^This definition of *LJ? is used by A. S.z) = (r + k)\ Lkr(z). Appendix B.{z) is the associated Laguerre polynomial as normally defined and mostly written Lr (z). We distinguish between different definitions in use. k + l. With the help of Eq.= t .115) is effectively the associated Laguerre polynomial normally written Lk(z). Thus Messiah [195] defines the original or stem Laguerre polynomial as : >Lr(z) = *L°r(z) = e'-^(e-'zr). 2 . We mention here three different definitions used in the literature.x). (11.114b) yields the binding energy of the states. it is given by^ *Lkr(z) = ^ ^ f 1 *i-r. Unfortunately the definition mostly used in mathematical literature differs. see. a + 2:*) = t v . so that one always has to check the definition. Messiah [195].r f .6.6 The Discrete Spectrum of the Coulomb Potential The associated radial wave function is yi(x) 219 = xl+1e-x/2<f>{-n'.118a) where Lk. for instance. (11. Each can be recognized by reference to the generating function.11. l .414(3). The mass of the hydrogen atom is given by massif = Mp + m e + E < Mp + me. formula 7. M. and I. p. .110) is usually re-expressed in terms of orthogonal polynomials known as associated Laguerre polynomials. For clarity and later use we denote the function used there by its normal form L^. • (11. in fact. 11. 844. Ryzhik [122]. W.2 (11. . • ( . Apart from factors. I.2l { + 2. Oberhettinger [181]. several different definitions are in use.p)\(2l + 1 + p)\ The spectrum (11. 84.117) The function *Lk(z) is a polynomial of degree r. . 3 .116) Laguerre p o l y n o m i a l s : Various definitions in use! The polynomial obtained by terminating the hypergeometric series (11. however. (u.1. p.r = 0 . Magnus and F. • fc. Gradshteyn and I.115) ' =0 (ri . .

I. and Ll(z) = l. p. L1(z) = l . pp. is defined by^ dk (—'\\k(r^\2 (11. .3 we show in a typical calculation how this normalization condition can be obtained with the help of the generating function.119) dz dz In particular one has (we cite these for later comparison) LQ(Z) = 1. L\(z) = 2-z.120) ' r=0 It will be seen that the orthonormality condition is" z e--zJc*Tk.d z—^z + {k + l~z)—+r Lkr{z) = 0.164.3): e-2t/(l-t) (1 . r r Z r=0 v tr k k (11.(z)*L TJz)dz 3 = KP + kV-}Opqit Jo (11. which we write ^Lk. 162 . The generating function of the associated Laguerre polynomials — obtained by searching for an integral representation of the solution of this differential equation which is of the type of a Laplace transform — is for |i| < 1 and expanded in ascending powers of t (cf.** The associated Laguerre polynomial here.2. Example 11. The reader is there warned that '•'•care must be taken in reading the literature to ensure that the particular convention being followed is understood'. It is said that this is the definition preferred in applied mathematics.122) "See A.220 CHAPTER 11. Ll(z) = 3-3z + ±z2.121) In Example 11.. 163. Messiah [195]. ^ F o r this definition and the following equations we refer to I. For our purposes here it is convenient to refer to yet another definition of the associated Laguerre polynomials. **I. L2(z) = l-2z + -z2.. The Coulomb Interaction The differential equation of the associated Laguerre polynomials is (same for *Lk{z)) d2 . . There at some points n + 1 is misprinted n + 1.z . (11.t) + k l r)! E E (-zY (i +(fc +h)\(r-i)\ i\ r=0 L OO p OO i=0 E (k + r)\ *L (z) = J2L ( y.„\*Kk.„M„ z"*L.1. N. . N. Sneddon [255]. Appendix B. Vol. Sneddon [255].

with y'{i) = — / ps(p)e~ptdp ty'(t) = -t f ps(p)e-ptdp-p2s(p)e-pt s(p)e -Pt J ^(ps(p))e-ptdp.4 + 2z. (11.107) implies k = 11 + 1 and r = n + I. Jfpd>2s{p))e-ptdp. (11. Solution: We consider the following equation of the form of Eq.4z + . i L1(z) = l .119) derive the generating function of associated Laguerre polynomials (11. Then.120)) i-t)k& zt/{l-t) ~*P p=k fc (11. ^L\(z) = .119): ty"(t) + (a + 1 .124a) (i-*) pi The orthonormality condition is (cf.1 . and obtain with this the normalization condition (11. and partial integration.118a) and (11.3: Generating function of Laguerre polynomials With a Laplace transform ansatz for the solution of a second order differential equation like Eq.. The limits will be determined later.124c) Example 11.11. ^L2(z) = 2 .% ' ( * ) + by{t) = 0 with ansatz y{t) = I s(p)e~ptdp.6 The Discrete Spectrum of the Coulomb Potential The differential equation of these associated Laguerre polynomials is** z 221 ~r^z + (k + l-z)-— + dz dz d2 n .120). *L\(z) = .. (11.8): (11..1 8 + 18z 3z2. (11..121). (11. (11.124b) Comparing Eqs. . The generating function of the associated Laguerre polynomials so defined is (observe (—t)k as compared with (11.123) For the purpose of clarity and comparison with the polynomials of the other definition it may help to see that here in particular % ( « ) = !. N d r-k 0.z .122) we see that tLkr+k(z) = (-l)k*Lkr(z) = (-l)k(k + r)\Lkr(z). Example 11. '"''Comparison with Eq. and *L\(z) = .

Thus in the case of Eq." b + 1 e~ptdp = coefficient of pb in g(p) := (p + \)a+be~pt. 2TT_ Jq=0qb+l(l-q)a+l 1 . agreeing with Eq. t > 0.118b)) Ll{t) = = coefficient of pb in ^ (p + l ) a + t > n=o £ ( ~Pf" "" coefficient of pb~n in (p + ! ) « + ' ^ = /g <-*>" (a + 6)! n! —(.p ) « + i q c + i ( i . which removes [.p ) o + i 7 t = 0 . since for n > b : (b — n)\ —• oo.p + 1 = 1/(1 . We demand (to be considered later for integration contours) the following condition. with requirement of a finite solution./ ( l — <?) —— = G(t).119) with a and b as there. this is the rejected irregular solution. Hence one defines now as associated Laguerre polynomial L£ and as the generating function g(p) of these (Cauchy's residue theorem requires the coefficient of 1/p): L b(t) •= — <£ 27T. p and one has (as polynomial of degree 6.222 CHAPTER 11. (3) around p = — 1 if a + b is a negative integer. the only possibility is the last. Then the preceding two equations (inserted into the original differential equation) leave us with I e-ptdp d {p(p + l)s(p)} .a = 21 + 2. e ^ t '?/( 1 -'?) = coefficient of qb in /"OO / = = / Jo Ll{t)Lac{t)tae~tdt tae-t{P/(l-p)+q/(l-q)+l)dL l t I * * [" (2TT. The Coulomb Interaction since d(p2s)/dp = ps + pd(ps)/dp.] from the above equation: C(p) := p(p + l)s(p)e-pt = 0.124d) In order to obtain the orthogonality and normalization of the associated Laguerre polynomials.q) and P~ t ( . j p = (P 0 "V?. n! (6 — n)!(a + n)! An alternative integral representation is obtained by setting p = q/(l — q) or q = p/(p + 1). (11. .) 2 J J p 6 + i ( l . and (4) around p = 0 if b > 0 and integral. (2) from —1 to oo.. p° Possible paths satisfying this condition are: (1) If b < 0 from 0 to oo. Case (1) with p —• oo implies y(t) ~ / dppa~1e~pt ~ t~a. (11. Then dp = dq/(l-q)2.. It follows that 1 d i I _i. s bK> (l-q)°+i (11.( a + l ) p s ( p ) + 6s(p) dp 0. we insert this contour integral representation of the polynomial in the following integral and obtain __?(_) = P da <* . Case (2) implies the exponentially increasing solution ~ e + t . Hence the expression in square brackets vanishes.^ .li I u -{p(p+l)s(p)} : p(p + l)s(p) dp a + 1 b p+ 1 p(p + 1) a + b+l p+ 1 b p Integrating the equation we obtain s(p) and hence y(t): a+b+l^-b ln[p(p + l)s(p)] = In ip+ir+o+'p (p+l)a+b f (p+l)a+b The condition C(p) = 0 now implies for possible contours of integration in the plane of complex p: C(p) = p(p + l)s(p)e'pt („ _i_ -na+b+i = ^ ' e~pt = 0. case (3) similarly.

125a) and the factor a~ 3 ' 2 with a = Bohr radius.111). The wave function of the state (n.6 The Discrete Spectrum of the Coulomb Potential 223 We evaluate the integral on the right with the help of the integral representation of the gamma function. (11. (11. /•OO / Lg(t)£?(t)tae-fctt: (a + c ) ! .3 T h e eigenfunctions According to the previous discussion we obtain n2 orthogonal eigenfunctions in association with the eigenvalue En.?). so that I = (c + a)! 1 f . where Fnl(x) = -yi(x) (11.125b) The normalization condition {J \tp\2dr = 1) determines Nni (which we leave as an exercise!): w "< = Ul^TW F- (1L126) It is useful to know the following integral for certain cases: InmiP) ••= l^ e-xx^k-^Lkn{x)^Lkm{x)dx.c (-a-l)\ d(-c-o-l)! c(c + a)\ c!a! with use of the inversion formula (11. Eq. 11.112). is introduced for dimensional reasons. I.6. Jo (11. .127) .(x).-—: <f h_nJ_-[ dp = (a + c)\ T^ if c = o.—i. f a -p)(i -1) r + 1 l (1-P9) J The coefficient here required is obtained from the expansion as .11. and obtain r tae-t[P/(i-P)+q/(i-g)+i]dt Jt=0 With this the integral I becomes = . and 0 if cj=b. m) is 3 2 &nlm = a~ / NnlFnl (—^Y^O. Clearly only the second form is sensible. Then Nni is a dimensionless normalization constant and (note that we use the associated Laguerre polynomials as in the book of Sneddon [255]) Fnl(x) = xle-x^LZ.

from which we can deduce information on the behaviour of the electron: rF"\nrg)r2dr ^ _ 2 £> n iF&ztfdx *%(*)*** 1 Jo°° Fix (-)r2dr 2 ^i + /„.131) . = 0) we have -) r /u =-..n+l na l2 ° ^ 2 na n+L+lV) (2n + 2/ .129) n2a' Analogously we have roo j = p2 I r n \ 2J ° J0 'V^yr nl _ na J^° xF^(x)x2dx _ na f^SJn+K 3 ) \na) na {6(n + Q2 . 2 (11.(2Z + 1) + 1} = ^[Zn2-1(1 + 1)]. 11. + .128) These expressions permit us to obtain the mean or expectation values (1/r) and (r). (11. thus on the average the electron is the farther away from the proton or nucleus the larger n is.(l) n+l.130) We see that the mean value of r is the larger the larger n is. This behaviour is indicated in Fig. (11.3(2Z + 1) + 2} 2 {2(n + 0 .3fc + 2).6(n + Z)[(2Z + ! ) . a (r)is = ~a.! ] + (21 + l ) 2 .8):* OD - <n!>3 (n-k)\ (n-fc) (n\)3 i£ n (3) = _ . For the ground state (n = 1.1 + 1) (11. and for the explicit derivation of the second case below Example 11.7.2. The Coulomb Interaction One finds in particular (see also Example 11.2Z .224 CHAPTER 11.v (6n2 - 6nk + A:2 + 6n .

2 1 A r = V (^2) . \f{r ) ~ n a.^ We obtain therefore for the quadratic deviation (r) . v ^ V2n + 1 (11.132) (11. 173/174. This corresponds to E ~ — e2/2n2a like classically with circular radius n 2 o. i. Problem 5. / ! J — . with n +1 replaced by n and 21 + 1 by k. pp. = l a nl n N'nll rFnl.6 The Discrete Spectrum of the Coulomb Potential r-.l ) n ] = | [ 2 n 2 + n] an[ n + (11. we have and tp becomes particularly simple. In the Kepler problem of a particle of mass mo in Classical Mechanics with the Newton .20. In this case (r) = = For (r 2 ) one obtains < r 2 ) = n 2 ( n + ^ ) ( n + l)a 2 .. or (r ) ~ n a 2 2 ^ [ 3 n 2 .11.(r) = x W 2 n + 1 =. In the case in which I assumes its maximal value n — 1. S. Sneddon [255].j .133) for large r.2 Hydrogen atom wave functions.J n 1s 2s 3s -*r Fig.( n .134) *These are taken from I.2 I -3/2 Kl r 225 \2 1 R„.e. 11.

pJ = —— and j> Pidqi = nth. The above treatment of the hydrogen atom has obvious shortcomings: (a) The hydrogen atom was treated nonrelativistically. (c) The nucleus was treated as pointlike and not as something with structure. In order to obtain an impression of relativistic corrections in a simple context. However.4 the relativistic equation in cylindrical coordinates.226 CHAPTER 11. „ _ -'Kepler — the Kepler period is given by o Z7ra 3/2 Kepler . in which periods are obtained from the Bohr-Sommerfeld-Wilson quantization condition. but rather like classically in a plane (i.6. \/v vT 7 v^ See also Example 14.3. for these values of I it is not distributed with uniform probability over the spherical surface.135) This expression becomes very small for large values of n. This means. Thus we have to evaluate the following gravitational potential written V(r) = —mofi/r. . Solution: We have H = yjm%c4 + pIn cylindrical coordinates.r. Thus. Example 11. i. r. the electron remains practically localized within a spherical surface of radius (r). Since 6 is a cyclic coordinate. (b) The spin of the electron was not taken into account and hence the spin fine structure of the energy levels is excluded. identifying yj (r2) with aKeplen w e see that T K ep l e r = ^(r2)^ = ^(n2af/2 = *^a3'*n*. we have pg = const. i = 0. The Coulomb Interaction or -TT = ^ = (11. determine the energy E by evaluating the Bohr-Sommerfeld-Wilson integral of "old quantum theory". relativistic corrections are of the order of v2/c2 ~ 0(En/moc2). = ngh. the states Ytl are predominantly concentrated in the xy-plane). p2 = p2 + — pg • eV. 6.— > where ampler is the length of the semi-major axis of the elliptic orbit.e.4: The relativistic eigenenergy using cylindrical coordinates Using the formula for the relativistic total energy E of a particle of mass mo and charge e moving in the Coulomb potential V = ~Ze/r (with E numerically equal to the corresponding Hamiltonian H for a conservative system).e. and separating this in cylindrical coordinates r. we consider in Example 11.

<p = <p. Schiff [243].3. 11. Thus E2 2EeV(r) e2V2(r) nrh = a> prdqr = ilc2 . where ip = 0.11.* See L. y — 0: V ^ c o s tp. ( v c + i ) .V . where the constants are identified by comparison. (11. z=-^-rj).(p . The deeper reason for this are symmetry properties of the problem. the "fine structure obtain the ordinary nonrelativistic energy) Z2ei c2h2[nr + ne^jl- -1/2 E — mac + 1 Z2ei/c2h2n2] -1/2 VTIQC 1+ •+ ng^l-a Z /r. r\ = r ( l — cos#) = r — z.4 Hydrogen-like a t o m s in parabolic c o o r d i n a t e s The Schrodinger equation of a hydrogen-like atom can also be separated in parabolic coordinates £. ip are the spherical coordinates. as also indicated in Fig. (and subtract moc2 to We then obtain with a = e2/hc ~ 1/137.d r K-mlc2 + T U^\-±-AnlK2-^Ur 2 2 \ c / r * + * ! + $*—». In the case of degeneracy with respect to the magnetic quantum number m (as in the present case) one can find linear combinations of hyperspherical functions Yj^(9. which are related to Cartesian coordinates in the following way. Whenever there is degeneracy. but depends very slightly on ng alone which gives rise to the "fine structure" of hydrogen lines. where r = ^/x1 + y2 + z2.136a) (11.136b) £ = r ( l + cos#) = r + z.T).6 The Discrete Spectrum of the Coulomb Potential 227 integral which we achieve with the "method of poles at infinity" explained in Example 16. 11. p. 2 2 Therefore the energy is no longer a function of n = nr + ng..6.2. . 86. It follows that (with h = 2nK) Z2e c h?ni 2 2ixnrfr = — 27r nehtjl - Ze2E >-^m2c2-E2/c constant'. y=y/%qsanp.9. it is possible to separate the Schrodinger equation in other coordinates. which correspond to a new choice of polar axis. (p).

137) One now sets the total wave function ip = £(£)!N"(r7)$(</?) and divides the equation by ip. Multiplying the remaining equation by (£ + r /)/4 and setting the £-part equal to —A.3 — h2 ( 4 r d (rd^\ d_f df\~ + 1 d2xl>\ £?7 dip2 2Z1Z2e< Z+V •ip = Eip. In terms of the coordinates (11. there is really only one basic equation to solve here. and r\ = const.228 CHAPTER 11. one obtains the following two equations: £ ) + TJIQE A m2n | « ' m = o.22) becomes — as we show in Example 11. so t h a t the variables can be separated. 1 . a separation constant.136b) the Schrodinger equation of relative motion (11. . 2 . 11. . d_ dirt) + TUQE fm0Z±Z2e2 2h2 V ^2 A V m (ri'N) Arj1 2n Since these equations are alike. (11. for uniqueness of the wave function. = 0. . We rewrite the first of these equations with the help of the relation M -(w ^ - m + e) . Then 1 d2$ izimip $ oc e $cV = —m where the separation constant m2 must be such t h a t m = 0 .3 Parabolas £ = const. with z-axis as axis of rotation. The Coulomb Interaction Fig.

To obtain a closer analogy with the separation in the spherical polar case. (11. m0Z1Z2e2 ^ 2 = n (say). " .105)) Hence we write I = (m — l ) / 2 .e. 1 . i. In the spherical polar case we had (cf. . ds2 = dx2 + dy2 + dz2.' / 2 p m / 2 L 2 Z + i = m ( / 9 ) ) Analogously we have N oc e-CT/2<7m/2L™ (a).i n +n = n i + n 2 + m + l = we obtain the energy eigenvalues given by -E = n i = n'- \(m + 1) = 0 . . n 2 = n" . . E x a m p l e 11.11.125b) (dividing by p 1 / 2 ) £ oc e .± ( m + 1) = 0 . We leave the calculation of the degree of degeneracy to Example 11. Adding now / .o. ph h2p2 _ h2 m2Z2Z2eA hn 4 2 _ ~~ m0Z2Z2e4 2ft2 n2 ~ 2m 0 ~ 2m 0 This agrees with our earlier formulas like (11. T h e n by comparison with the spherical polar separation the answer is as in Eqs. . 2 .6. .106) and (11. p = /?£ and cr = /^r?. (11. we set (since E is negative for the discrete spectrum we consider) ITIQE =-*"• H J?P ^( <9p: 1 / 2 i m0ZiZ2e2 A = n". . from Cartesian to parabolic coordinates. . 2 . . . 1 .6 The Discrete Spectrum Then of the Coulomb Potential 229 A 22 (£ 1 / 2 £) + moE 2h2 A m2 — 1 +£ - 4£ 2 (cT1/2£) = 0. as well as the Laplacian A = V and the Schrodinger equation. . Eq. T h e n the first of the two equations becomes 1 £) n' m2 — 1 4 p + 4p" ( P 1 / 2 £ ) . .5: The Schrodinger equation in parabolic coordinates Perform the transformation of the metric.114b).

E E -n l „ . i. n the double sum ]T]2 is TI — m. for m / . Solution: For the wave functions ip = £(£)J\f(r])&(ip).l ) ( n . n 2 (1-UJ)2 The coefficient of a .e.1 in this expansion is seen to be n. and attach a factor to to each term introduced in the wave function.230 CHAPTER 11. and one obtains 2 = _ 1 4{ £ ) * (t + 7 )\ j.?e 9£ J 4 t d \ g^gv 9 \ dri \ gv dr] J t d \ g^g-r.2d? dr)}. The Coulomb Interaction Solution: We begin with z of Eq.l)a + (n . z = (£ — rf)/2. 2 + ^dV . .* 71 — 1 n—1 n— 1 n n+ 2 E ( n ~ m ) = n +2 E n _ 2 E m= n + 2n(n .114c).tx2 . 0 there are two. Here we have (see preceding text) 711+77. Similarly we proceed with dx and dy. 2 _ = (9id02 + (gvdvy + /„ ..e. We then sum all possibilities and select the coefficient of ojn~1 in the result. </. Beware! For m = 0 there is only one m-state. i. the total number of wave functions we have is the number n with m = 0 plus twice the number with m ^ 0 starting from m = 1.136a). Thus.[df + dr]2 .g<p 9 \ .e.2)d} = (n. > . Thus dz2 = . see Eq.2 = n — 1.136a).2)d/2.6: Calculation of degree of degeneracy Show that the degree of degeneracy of the wave functions in terms of parabolic coordinates agrees with that obtained earlier for spherical polar coordinates.g. 2 2V? V^ One now performs the square of this and obtains similarly dy2. 712 and m which give n. all cross terms cancel in the sum of 7 the squares. and so dz = (d£ — drj)/2. are elements of length. Magnus and F. {g^Y. Eq. We use the method of selector variables. We consider first the case m = 0. i. 'Recall that the sum of an arithmetic progression is given by a + (a + d) + (a + 2d) + • • • + {a + (n . W. -\— < / — cos ipdr] — \/£r] sin ipdtp. as t h e two signs of t h e exponential indicate. N2 where g^dS. / — cos ipdl. The general formula for the Laplacian is§ ^ follows that 1 [9£\ A = ' 9 \ gr. we have to find the number of values of the quantum numbers n\. Oberhettinger [181]. 9 \ 9<p\ gv 9f\ _ 1 a2 a / a\ a / ay £77 dip2 E x a m p l e 11. (n _ !) n = n2- §See e. In the case m ^ O w e have n\ + 77.. .2 = n — m — 1 and therefore the number of the coefficient of u}n-'m-1 . Since the lines of constant £ are orthogonal to lines of constant 7 (they cut perpendicularly). = e±im^e-("+^'2{par'2L^{p)L^ (a).2) and summing over m. Thus we want to obtain the coefficient of wn~1 in £2 .1) ~ . Observing that the highest possible value of m is n — 1 (i. dx = d{ y £77 cos ip) = . " .e. . (11. for instance. when m = 0 = 77. etc.-2 . (11. / J„^2 . + l 4 {^^)drl f i + v \ . last chapter. (11.

) = ^eF(p . these expressions agree with those of Eq. with Rayleigh-Schrodinger perturbation theory). I. L ni\ ny. (11. we would have to use degenerate perturbation theory and obtain the same result.drid<fi oc (p + o)dpda. s 0/ .9) one finds that _ = %2 jm+2 {n2+m)\ (ni + rre)! +1 : > in. With the help of the wave functions we derived above.. .127) and setting n = r + k.•* n (l) = / Jo It follows that .11. Ikr+k. Apart from the power of the leading factorial (which results from the definition of the associated Laguerre polynomial in the generating function). if we expressed the perturbation in spherical polar coordinates.124c). with the connection » (11.r+k(l) = / Jo e~*xk[(k + r)\Lk(x)f = [(/= + r)\flk. i.eF / nh2 \ {v)=3(n1-n2)-^—lr2j Thus in this case nondegenerate perturbation theory could be used as discussed in Example 8.(p)L™ (<7)]2(p + a)dpda rev where JJ = / Jo dpe~ "p"[I.a). i.128) if one makes there the substitutions n —> n\ (or 712) + m. 8. fc — m.7: Stark effect in hydrogen-like atoms 231 Consider a perturbation v = eFz. and determine the average value of v for a given state.e.—r~n = 3(m-n 2 ). we can relate the expressions Iq to those of Eq.r.128). { 6 ni2 + 6 n i ( m + l) + (m + l ) ( m + 2 ) } .™(p)] 2 By contour integration (see Examples 11.g. = j (2ni+m+l). In fact. .{ n i {2ni + TO + 1} + {ni j=± n 2 } 6(ni-n2)(ni+n2+TO+1) ^±n2} ^7—. . with z = r cos 9. Sec. Thus e. (11. we now have (p-a) J dpdaip* (p — a){p + a)ip J dpdaip* (p + a)ip Jo°° Jo°° e-(e+<r)(p*r[L™ rm+2 i rm 1 n1 n2 _ Jrm+2 1rm "2 n1 (p)L™ (<r)]2(p2 - a*)dpda /o°° Jo°° e-(o+->(p<T)-[L.8 and 11. We obtain the volume element dV in parabolic coordinates from the above results as dV = g£gvgvd£dr)dip = — (£ + rj)d£. with Z\ —> Ze. with Eq. 2(ni +712 + TO + 1) Therefore the change in energy of the state to first order is given by.6). However.e. Solution: Using formulas and results of above we have to determine (v) = eF{z) = l-eF(e. . where F is the electric field.6 The Discrete Spectrum of the Coulomb Potential Example 11. ( ! ^ ± ^ ) l { 6 n 2 + 6 n i ( m + 1 ) + ( m + 1 ) ( ? n + 2)} ny. (11. .e. eF{z). {p — a) — = dxe~*xkvLkn (z)] 2 . Z2 —> e. which will then give the change in energy of that state to the first order (i.1 (cf.

9: Laguerre expectation value integrals Evaluate the following integral with the help of the generating function of Laguerre polynomials: /•oo rSc(i):= Jo dte[Ll{t)Lac{t)tae-\ i.1 ) ! (n .9 we consider integrals with an arbitrary power of t.128).1 ( n + m)!"[ _ (n + m + 1)! (n-iy J n! (ra + m)! _ (n + m)! (n .9) with the help of the generating function G(t) of associated Laguerre polynomials obtained in Example 11.p ) r + 2 [(l-p)(l-g)]"l+1(l-pg). the integral arising here becomes with Eq. The Coulomb Interaction E x a m p l e 11. It follows therefore that the coefficient of (qn — qn~1) in (1 — pq)~~(m~'~2' is p n ( n + m + l)! n!(m + l ) ! Finally / is the coefficient of p n p n . s = at.g )(l.p ) ( l . the integral / is the coefficient of pnqn in the expression _ / .9 )« Solution: Using the generating function for both associated Laguerre polynomials.l ) ! ( m + l)! in the expansion of p " ( n + m + 1)! p " " 1 ^ + m)! 1 _t .7 and (remembering the definition of the associated Laguerre polynomila there!) the second of relations (11.9 ) ] m +h 1 yJo 0 ' 1 —pP 1 -q 11-9 l-pq _ (l-9)(l-p) = a With (see Eq..3).n+2 in the expansion of i n c o e f f i d e n t o W (m + 1 ) 1 ( 1 . tt„:_ (1 .8: Laguerre linear expectation value integral Evaluate the following integral (which is different from the normalization integral of Example 11.x ( n + m)! ( n .111) /*oo /*oo „ — sam+l / e~attm+1dt= (m + 1)! ds = am+2 I ' Jo For use in the following recall that Jo «m+2 {i + X)-n = ~ ^2(-ir .3: /"OO I := Jo e-t[L^{t)]2tm+1dt. L™(t) = coefficient of qn in G(t) e -tq/(l-q) (1 . In Example 11. E x a m p l e 11.1)! n\ This verifies the expression I™t in Example 11.l)!r! -X We now evaluate the integral / as the coefficient of pnqn _ ( m + l)! [ ( l.^~T^—-—^ ) = coefficient of (pn .232 CHAPTER 11.p n _ 1 ) in (n — 1)! J p"(n + m + l)! n! p n . (11. and is a special case of Example 11.3 by one power of t.b. (n + r .124d) of the alternative integral representation obtained in Example 11.a. _ „ . (11. .0 ° e -*[l+P/(l-P)+9/(l-<j)] t m+l di) t | P | 9 l [ ( l . ) (l_pg)(m+2) Now the coefficient of qn in (1 — g)(l — p<j) . c positive integers. dt = ds/a.pK — — —L .p ) ( l .( m + 2 ) is equal to the coefficient of (qn — qn_1) in (1 — pq)~(m+2\ The coefficient of qn in the expansion of (1 — pg)~ M is pn(n + fi — l)!/rt!(/i — 1)!.

a _ 1 = [1 + a ( l . ° ^ + a ( l . = c (q + fe)! W ° (^)!(^)!(^)! ~" (2) In the case of one power of a. i.w(r) = and / a ) = r\(P ~ r . (a) j / = 1.r ) ! a ' . c + b odd).q(p + a - The coefficient of q in this expression is 1 [l + a ( l -p)]a+l where f 9 / p + a — ap\c (c + a)! _ ac(c + a)\g(p) \ l + a-apj c\a\ c!a!(l + a)a+c+1' [l + ^ P ] " [1-T^P]Q+C+1 y>^ d(a + c + C .p ) 0 + l 9 c + l ( l _ g ) a + l ^ a ) ' where j(a) contains the integration with respect to t which may be evaluated with the help of the integral representation (11.L. .124d) of an associated Laguerre polynomial. we have 2r = 6 + c = even.e.(i) = i ! ( . so that = (q+fc±£)| . In case (a) we have u \ 7 V^ ir fb + c\ 7(a) = Jo / Tli u.p ) ] .™( r > (1 + a)P r £^0y^w Jc+6_2r = . Using the contour integral representation (11.p) _l-pq It follows that (using Cauchy's residue theorem in integrating around q = 0) 1{a) = (2^ / / ^ ^ [l-pq c [ 1 .111) of the gamma or factorial function. c + 6 — 2r = 0 (i.Pq + Q(1 " 9)(1 " P)1_a_1 = ti f p^ X COeffident f ap)]"0-1.?(t)t a e.f ]. (l-p)(l-q) + a(l -q)(l . i.a f + c)\(v-r)\(l + ay-rP „ • ^0^r(c-r)\r\(a •• Ha) a\ f dp ac(c + a)\ -^ =%*??* c!a!(l + a ) ^ c + l 9 ( p ) = g J c+*-"-«a>- where with /3 = a + c + l + b (observe that for a = 0 a term with c + b — 2r = 0 remains) „_ = 2 ^ 2 ^ K «.(*) from the coefficient of a1: />00 233 1(a) = / Jo dte. 7£.q)(l .^ ( l .e. i(a) = / Jo dttae-t{p/(l-p)+q/(l-q) + l+a} = Q .l ) * x coefficient of a* in the expansion of 7 ( a ) . c + b even) and/or j / = 0.e. we can re-express 1(a) in the form: /(a) = (2^)2 J J p t + l ( l . c + b — 2r = l (i.6 The Discrete Spectrum of the Coulomb Potential Solution: One considers the following integral and obtains /£.11.e. — (c-r)!r!(6-r)! Kv.1 + y w)\(-a)y (r — w)\w\(j3 — r — l)!(j/ — w)\ Special cases: (1) In the case a = 0.p) .a t [Lg(i). . we have to have y + (c + b — 2r) = 1. where (a+**£)! c j n Jo = .

the standard principles of nonrelativistic scattering theory. . . is the work of J.7. R. Taylor [267]. The Coulomb Interaction b + c\ 2 J (b-¥)\(P-b-¥-w)\{-a) ($±£ -w)\w\(J3 -^±£-1)1(1-11. ^A very readable source to consult. Since such a rigorous treatment is beyond the scope of this text.~* Hence (recall that 7£ c (i) = i!(—l) l x coefficient of a% in / ( a ) ) (a+^)!(-/3a) r r in (£f^)!(^)!(^±£)! t J = ( a + b)\(-(3a) > 6! _ (a + ft)! = [a + lb + l ) a . One should actually consider a screened Coulomb potential* and consider quantities like the partial wave expansion (to be defined in a later section) as distributions.-^)(-.234 and (only w = 0 and 1 arise in the sum) CHAPTER 11. 2mo Here we set (this defines the velocity VQ) j? E ^ 1 (11. Zie given by h2 _ A _ ZiZ 2 e 2 V>(r) = Ei/>(r).1%^ 11. like the asymptotic condition and definitions of the scattering amplitude and the phase shift. In view of the slow fall-off of the potential at infinity. we consider the traditional treatment here and refer the interested reader to literature in which it is demonstrated that the derivation given below finds its rigorous justification on a basis of distribution theory. t We have the Schrodinger equation of relative motion of particles with charges Z\e. cannot really — and this means in a rigorous sense — be applied in this case. also with regard to earlier literature.138) " 2 ^ =2 ° ' 2 m Uo 7 = ZtZ2e2 -^T- (1L139) * Screened Coulomb potentials are considered in Chapter 16. however parallel to the Coulomb case.)!' f^)-(. b\ V cited in Example 11. + (!±-<)«-.7 Continuous Spectrum of Coulomb Potential In this and the following sections we consider the scattering of a charged particle in the presence of a Coulomb potential. These results are seen to agree with the expressions 1^.

z)).b.143) and v = ikrj.138) becomes 235 ^A + k2 .140) The simple Coulomb potential permits particularly simple solutions. that we encountered previously.144) *(a. with the asymptotic expansions (11.108) and (11. (11.z) = Wi(a. . i. a(a +1) z2 ] (11.b.142) This equation is of the form of a hypergeometric equation. .109) we can therefore write the solution ipr = Aeikz$(-i>y. pp. z) for \z\ —> oo? We can find the appropriate formula in books on Special Functions:* $(a. (11. 86 .140) and using (11.146) See. Oberhettinger [181]. (11.b. z = rcos<9.11. v = ik(r . for r — oo) • ipr = Aeikzf(r . (11.z).7 The Continuous Spectrum of the Coulomb Potential Then Eq. for instance. i. . W.. A = const. b.145) U ' ' j " T(6-a)1 J ^ 71=0 r(a)r(a-6+l) n! ' (-7T < axg(-z) < vr).z) + W2(a.e. (11.136b)..141) Inserting this expression into Eq..M) = l + . „ az 1.^ W ) = 0. Magnus and F. (11. (11. Our question is now: How does ipr behave for r — oo.. writing Vv = Aeik^-v)/2f(ri) we obtain the equation d2 x d f(v) = 0.87 under Kummer functions. what is the > behaviour of 3>(a. where ^.z).e. According to Eqs. (11.+ ^ - + . In particular the equation possesses a regular solution of the form (also ip ~ exp(ikr).

e. The Coulomb Interaction W2(a.153) The asymptotic solution represents t h e stationary scattering state (E = const.z) + l-a)T(n + b. (11. if tbr were „ikr ibr ex el"z + f{6) r (11.149) (11. = r cos 0.152) determines the quantity f(9) as /(*) = 7 r(i + i7) exp k(l-cos6)r(l-ij) 2sin2(f) ?7 In sin 2 7 2 exp 2 i a r g r ( l + ij) — ij In sin ( ^ 2&sin2(f) .i 7 ) (11.147) (—7r < arg(z) < 7r) It follows t h a t for r — oo. J\") ci(kr-~/\n2kr) (11.l. a comparison of Eq. i.236 n\Zza-b CHAPTER ^T(n '£ fo 11. Without t h e logarithmic phase factors.a) (z)- na)"~ r(l-a) T(b-a) n\ ' (11.e.148) . A(-j)11 A where \ " r(i + \i) 1 j(kz+"flnk('r-z)) 7 f ( l + Z7) _ e i(fcr.z)T(l .152) Since z = r c o s # .ik(r-z))} =^ + ^d.l.) of a particle with incident momentum hk in t h e direction of z for large distances away from t h e scattering centre. (11.154) . We now define a quantity f(9) by t h e asymptotic relation lb O < ^(.b. > A = Aeikz[W1(-i1.7 ln*(r-z)) k(r .kz+l^ktr-z)) C .(11.^l*7 l + O (11.151) exp[—ryln fc(r — z)] — exp[—ijIn = kr(I — cos 6)) exp[—ryln 2kr — ry In sin 2 (0/2)]. with 2. (11.151) with Eq.ik(r-z)) where ^ ~ Aeikz——^—f't-^ e ife(r-z) + W2(-n.150) ^(r-z)]-1"^ 1 + 0 T(-«7) i.

The problem of the logarithmic phase.n .7 The Continuous Spectrum of the Coulomb Potential 237 the wave could be looked at as the sum of an incoming plane wave exp(ikz) and an outgoing scattered wave. z — — oo. 11.. In other words. 2im. The quantity |/(#)| 2 would then be a measure for the scattering in the direction 9.7. We define as current density the quantity h J := 2imo [^(v^)-v(v^n. For the incoming wave ipi = exp(ikz).155) Since the gradient is proportional to the configuration space representation of the momentum. also called phase anomaly.156) . the current density is * h . do dQ K 12 K Fig. (11.1 The Rutherford formula The above considerations permit us to calculate the differential cross section which is a measurable quantity. As a consequence of the masslessness of the photon. we see that this expression is the exact analogue of a current density in classical considerations.4 Variation of the observed differential cross section with 0. This cross section is defined by the ratio of the outgoing particle current (in direction 9) to the current of the ingoing particles.Q hk mo v0. the Coulomb potential has an influence on the incoming wave even as far as the asymptotic domain. (11. the effective range of the Coulomb potential is infinite.11. that we encounter here is a characteristic of the Coulomb potential. 11.

in the current this would in any case contribute only something of order 1/r. We obtain the total scattering cross section by integration: „2 — /£<" = £/777*V sin 4 « 1 »«°) We observe that this expression diverges in the forward direction.157) The differential scattering cross section da into the solid angle element dVL is defined by the ratio da ir which in the present case is with \f{6)\ obtained from Eq. We see that a depends only on the modulus of the potential (i. In the context of quantum electrodynamics one refers to "vacuum polarization" and relates this to the idea that a charged particle like the electron polarizes the otherwise neutral vacuum by attracting virtual charges of opposite polarity and repelling virtual charges of the same polarity.e. (11. i.159) as the Rutherford formula which can also be derived purely classically. is called scattering amplitude. In reality this divergence does not occur.153): 4fc2 s i r - The expression /(#).238 CHAPTER 11. We make a few more important observations. (11. which yields the scattering cross section.154) we chose the prefactor of the incoming wave in (11. The Coulomb Interaction In the case of the Coulomb potential we now ignore the logarithmic part of the phase (see discussion later). since in actual fact the Coulomb potential occurs only in a screened form.e. This implies effectively that the Coulomb potential becomes (virtually) a screened potential or Yukawa-type potential of the form V . This implies that scattering takes place for the attractive as well as the repulsive potential.152) as 1. for 0 —> 0. In Eq. Correspondingly the other part of (11. We recognize the result (11.152) yields as density of the outgoing current Jr = ^\f(0)\2vo(11. 'screened ZlZ<2e2 c-r/r0 ° . the square 7 2 ). That this formula retains its validity here in quantum mechanics is something of a coincidence. thereby screening itself off from the surroundings.

e.2 fr2 . A wave packet with momentum maximum at k = ko (with momentum p = hk) is therefore given by the following expression in which xo can be considered to be an impact parameter.*'^ e1 p-iEk{t-to)/h (x x i)= (2^3^ ( k ) roT ^ k o l * . t 0 ) .e. (11. be given by ^ f dk AL.164) Ek~Eko + {k-k0)-v0h. t 0 )^k(x. whose maximum moves with the particle velocity.8 Scattering of a Wave Packet We saw earlier that in quantum mechanics a particle is described by a wave packet. i.165) ^ ( x ) = £ fe Vk(x). i.e. t 0 ).t). / dk 7^340(k)4(x-xo. of continuum wave functions. (11. _ x o _ v o ( t _ toh io)_ e-m0-^-VoKt-t0)^o)ix (1L167) 0 l/> () ) fx-Xn t)~ gik-(x-xo) .x° ' r j . 11. and if H is time-independent one has the stationary wave function ^(x. which in the case of free motion (zero potential) are simply plane waves. 11. i.4. m0 m0 and V.k(x) is a solution of the continuum with energy h2k2 E = Ek = — >0. of the equation r fi.i i f ('-'°>/Vk(x. and which provides the uncertainties arising in quantum mechanics. The particle velocity v is v = — . t) = U(t. t0) = e. 2mo V + F(x) k (11. the wave packet in the absence of a scattering potential. The wave packet is a superposition of waves ip(x.yJ ^ j A k o W 727)3/2 e 0.166) Let the incoming free wave packet at time t > t$. v0 = . 2 (11-162) where Vk(x) = ^ ( x .11.t).^ ^ % ( x ) .8 Scattering of a Wave Packet 239 This screening of the Coulomb potential insures that the scattering cross section in forward direction is actually finite as observed experimentally and indicated in Fig. (k).161) where Ak0(k) ~ A(k — ko) and ^ k (x.t) = e . (11-163) (11.

169) The wave packet in the presence of the potential is obtained from the free form (11.k0) ~ k0 + and the expansion of E^ in Eq. (11.167) we have ^k0(x-x0.154)) that for |x| — oo.ip) is the scattering amplitude with respect to the incoming plane wave.k 0 ) 2 ~ Jk20 + 2k0 • (k .xo.i) dk (2TT (11.240 CHAPTER 11. i.161) and (11. The Coulomb Interaction In Example 11.165) ko • (k . ^k(x-xo) ^k(x) ~ = e.169) and (11.ko) k0 (11.ik . the stationary scattering wave.e. (11.<p) + Ol Akr l (11.t o ) / V k ( x .171) The phase of fk(0. k = kez.x °Vk(x). ip) could be carried along but we ignore it. (11.e.x 0 ).t) = J ^ A k o ( k ) e . With Eqs. i.x + _ / k ( e j ¥ ) ) -ik-xo (2TT)3/2 -zk-xo = <>(* r y (2vr)3(27r)3/2e M ^ ) e -*o)/^] e -ik-x 0 (11.^ ( * . Eq. From Eqs. (11.iB *(*-*°)/ R °'tj + 1 pikr e * . Then with k = ^(ko + k .167) by replacing the plane wave exp(zk-x) by this scattered wave.173) .166).163) we obtain </>ko(x . can be written A (X) = ( 2 ^ + -W J ^ ^ ^ ( ^ ( x ' ) - eikx 2mn f P»fe|x-x'| (11-168) We saw earlier (cf.172) h h k0 {r-v0(t-t0)} (11.170) A ko (k) e . (2TT)3/2 eik-* + —fv(P. this solution > has the following asymptotic behaviour in which f(6. (11.10 we show that the solution V'k(x) of Eq.

to). r ' ) = -47r<5(r . '.167).(«)/ ipko(x .r ' ) .154). 11.r')U(r')ip{r')dr' /4T7 (as we saw earlier) . r 241 (0 LO) f ^ Afc o(k)r-ik-xnri[fcor-£fcoa-to)/ftl i(k-ko)-x' = /k0^. t) .10: Schrodinger equation as integral equation Demonstrate the conversion of the Schrodinger equation into an integral equation. with the energy of the maximum momentum of t h e wave packet. (A + fc2)t/.r')V(rW).x 0 .to) has t h e form of t h e wave packet in radial direction as indicated in Fig.fco(t-to)/ft]^0o)(x/-xo. E x a m p l e 11.5 Scattering of a wave packet with scattering centre O.r')l/(r>fc(r').*0ko (x .174) r -/k0(^. and determine the Green's function G(r. 11. *. t) -f.11. Adding a suitably normalized solution of the free equation. i. (11.^e (2^3(27r)3/2e y (27r)3(27r)3/2e (11.r').x 0 .(r) 2rrao V(r)V(r) 7/>(r) (2TT)3/2 ^|dr'G(r.^y /k0^. the solution can be written iMr) : (2TT)3/2 ^|dr'G(r.y)e-iko'Xoe'[fcor-£. A particular solution of the Schrodinger equation is — J G(r.z Fig. U(T) = ^ V ( r ) . (11. . however. (11. Solution: We define the Green's function G ( r .5. T h e scattering amplitude is t h e same as t h a t given by Eq. Here ^ (x' — xo.171).e. r ' ) by the equation (A + fc2)G(r.8 Scattering of a Wave Packet so that with Eqs.

6.fc ) oo dfc' 1 d nik r 7rr dr / . G+(r) 1 d dk' lim e->0 TTT dr Jc+ fc'2 — (fc2 ~ik'r 1 d -2-ni nr dr \k' + k+ ie/2kj ik r J__d_ 7rr dr 2ni N ^ residues k. 9{k') = 1 1 S(T) = j G(r) 1 ^ A j dk'e i k ' r .e \ r dr\ k ikr 1 ikr 0 With the choice of an analogous contour C— in the lower half plane.242 CHAPTER 11.fc2 ' 2 ^ i fc72^"fc2 Integrating out the angles we obtain 1 G roo />7r 2n fc' rZn j. e > 0 small. These are the so-called outgoing and ingoing Green's functions. we relate it to a properly defined contour integral with the two simple poles at fc' = ±Vfc 2 + ie = ±(fc + ie/2k). 11. dk' ik'*r 2TT2 fc'2 . Thus we consider the contour integral taken around the contour C+. r ' ) = G ( r — r ' ) with the Fourier transforms G(r) = J dk' 3 (k')e ik ' r . one obtains another Green's function given by G_(r)= r -e-ikr. . The Coulomb Interaction We find a set of Green's functions G(r. « = 5-2 / J _ Z100 7rr 7o / A fc'dfc' 2 / 2 • fc2 (fc' *(fc' . Hence we write G(r)-G+(r).Rek' -k-fc/2K Fig. which is such that the contribution along the infinite semi-circle vanishes.6 The contour C+ in the upper half of the k'-p\ane.=k+ie/2k 1 d f. 11. Imk . displaced slightly away from the real axis as indicated in Fig.O O ™ fc2 (e*fc'r _ e- ifc / r 1 fc ) = - / fc'dfc' i(fc' 2 -fc 2 ) Since this integral does not exist./2 dk'd(—cos 6)dip iki.

9 Scattering Phase and Partial Waves 243 11.0 + W2(l + 1 + i-y.11. 21 + 2. i. z) = W^a. b. where Mo is a parameter in Chapter 16. Then £^<fc + (2l + 2 £ = ~2ikr. b. -2ikr) s For comparison with calculations in Chapter 16 note that for h = c = 1 together with 2mo = 1 and hence E = fc2 the quantity 7 here contains a factor k. e. (11. -2ikr)] [Wi{l + 1+1-1. z) + W2(a.176) " 0—A ~ (I + 1 + i-y)<t>i = 0.21 + 2. (11.g. b. (r) (11-177) Similar to above we obtain a regular solution y) ' with 0j = F(Z + l + i7.2Z + 2.145)) F(a. under the stated conditions 2/07 = Z\Zie2 = —Mo.£).e. .178) (11. Thus we proceed as above. the logarithmic phase as a consequence of the slow decrease at infinity and a special symmetry which permits separation in other coordinate systems.e. i.175) yl' + k Here we set (as in the case of the hydrogen atom) yi = eikr{kr)l+lUi). Before we introduce the general form of the so-called partial wave expansion.140) (previously we solved this only for the bound state problem) § . (11. Then yW (11.9 Scattering Phase and Partial Waves We observed above that the Coulomb potential is associated with some special effects which one does not expect in general. If we proceed from the radial Schrodinger equation. z). (11.z2 27fc 1(1 + 1)' yi = 0. we define first the scattering phase for the case of the Coulomb problem proceeding from our previous considerations. Eq. Eq.2l + 2. whose continuation to infinity is again given by (cf. (11.179) = = elkr{kr)l+1F(l e (kr) ikr l+1 + l + ij.100) — and we saw that in the particular case of the Coulomb potential this is not essential — we have to solve the following equation for the scattering case obtained from Eq.

180) ^ ' (11.183) r(Z + 1 + 17) := Reie.i f c r + i 7 In 2/cr+i7r(i+l)/2 + We set T(l + 1 + 17) = r(z + i + i 7 ) ' r(z + I .i 7 ) „ikr—ij\n2kr T(2l + 2) e -e* jTv(-l-l-i-y)/2 2l+i —ikr r(z +1 .ry) = e"^r(Z + 1 + ry). Then (r) r—»oo Vi 7T7/2 T(2Z + 2)e*7'*e i8i 2'+1I\Z + l + i 7 ) i5.146) and (11. (2kryr{-i" V 7 .147) this becomes asymptotically (r) Vi lkr eikr(u„\l+l (kr T(2Z + 2) '—(2ikr) \T{1 + 1 . Then e2i5< = Reie Re~ J2i9 9 = Si. The Coulomb Interaction Using Eqs.«7) „ .l -e*"" + T(l + 1 + H ) T(l + 1 . .181) so that eldlT{l + 1 . (11.' .n) • e „—ifer+i7ln2fcr _e-Mr(i7-J-l)/2 r(z + I +17) T(2Z + 2) /2 • ikr-i-y In 2fcr-i7r(J+l)/2 r(z + 1 .i 7 ) I T 2l -l—X—i^j ( + 2) -2ikr( ?.244 CHAPTER 11.j syy-l-l i-y il( _ .ii)' (11. +ikr—iry In 2fcr—i7r(Z+l)/2 _i_ — iSi— ikr+i^ln2kr+iTr(l+l)/2 ! We set I^TiT^rsin(J:r"7ta*"^+4)'<1L182) 6 = arg T(Z + 1 + ij) (11. T(2l + 2) (2fcr)-*T« (A-l-l-i-y ikr .

n ' .. .e.u\ respectively with the following asymptotic behaviour: u(±) r^o e±i(fcr_7ln2fcr. kr) the regular solution with the following asymptotic behaviour: r 0 (11.187) contains almost the entire physical information with regard to the Coulomb potential.e. However. with Eq. considering I as a function of E extended into the complex plane.114a) reveals). i. as we saw. considering I —• an(E). That the energy E appears in Eq. i. .139) n' = 0.1._W2)_ ( 1 L l g 6 ) The factor e2lSl between the Jost solutions (in the regular solution continued to infinity).e. (11.188) in the form y/E. 245 (11. Squaring this expression we obtain the wellknown formula for the binding energy of bound states (as a comparison with Eq. a positive integer. i. In the present quantum mechanical considerations I is. i. g = e«* = rSi + 1+ *V*ri (11. Analogously one defines as nonregular spherical Coulomb function or Jost solution the outgoing or incoming wave u\ .9 Scattering Phase and Partial Waves Hence 5j = argr(Z + l + i7)..2. (11..185) J F} ^° sm(AT-7ln2£T + ^ ) (also called regular spherical Coulomb function).e.11. where the continuum of the spectrum starts. One defines as regular Coulomb wave Fi(j.184) The phase 6i = 8i — lir/2 is called Coulomb scattering phase. (11. The expression e 2i5j _ 2i5i—iirl is called S-matrix element or scattering matrix element. indicates that — looked at analytically — the scattering amplitude f(6) possesses at E — 0 a branch point of the square root type. For instance the poles of S are given by Z + l + z7 = .

In these cases. Singh [252]. We shall return to these in Chapter 16 in the consideration of screened Coulomb potentials or Yukawa potentials and indicate other important aspects of these.190) and the scattering solution „ikr Vwt = / ( 0 ) — + 0 V. The large r asymptotic behaviour of the solution which is regular at the origin in the case of the scattering problem can be written ^ r e g = V'in + V'scatt." These Regge trajectories can be shown to determine the asymptotic high energy behaviour of scattering amplitudes and hence cross sections.7.246 CHAPTER 11. For the Coulomb potential a typical such trajectory is indicated in Fig. we obtain the partial wave expansion of the scattering amplitude as follows. and thus in general. 11. we obtain trajectories which are known as Regge trajectories. (11.189) where as before the incoming wave in the direction of z is ikz ipin = e (11. (11.191) . We mentioned above that the logarithmic phase does not arise in the case of short-range potentials.7 A Regge trajectory of the Coulomb potential. 11. lman(E) jump from E<0 toE>0 o A LU -infinity - E<0 Rea n (E) Fig. The Coulomb Interaction and plotting these for different values of n.

pp.^2l + l^S^k) lik (11.eikr\P^cos9).9 Scattering Phase and Partial Waves 247 For a spherically symmetric potential the amplitude / depends only on 9.193) where r\x e2l5t determines the change in amplitude and phase of the outgoing spherical wave.i * r 1 v-^ f ne2ibl — 1 \ = k^2l W +1 \^r^)Pl{cosd)—. The incoming plane wave can be re-expressed as a superposition of incoming and outgoing spherical waves with appropriate components of / and with a definite relative phase. .i f c r .195) with the scattering matrix element S(l. 21 .g. It follows therefore that pikr ^scatt ^ f{0) = Aeg-eikz = ~ ^(21 + l)Lilne-ikr - rHe2iS<eikr\pl(cos9) eikr\pi(coa9) eikr " 2 ^ E ( 2 Z + l){e i ' .192) Thus the incoming and outgoing spherical waves have a definite relative phase. r e. W. Magnus and F.~ J > Z + l)Lil*e~ikr .^(cosfl) (1L194) This therefore yields the following partial wave expansion of the scattering amplitude = ^lY. The potential disturbs this phase relationship (by delay or absorption or redirection).22. Oberhettinger [181].196) "See e. Legendre functions) and its asymptotic expression for r — oo and z — r cos 9 is given by" > eikz -^rJ2(21 + 1 ){e i i ' r e.11. There the expansion is given in terms of Bessel functions Jv which have to be re-expressed in terms of the asymptotic behaviour (1 2) of Hankel functions H„ ' . When we solve the Schrodinger equation and calculate the asymptotic behaviour of the regular solution we obtain correspondingly V'reg .r ^ V ^ Wostf). (11. The mathematical expression can be found in Tables of Special Functions (there cf. (11. (11.k)=me2iS^k\ in which r\i^\ indicates absorption.

This page is intentionally left blank .

however. which in one form or another. In particular we consider cases which illustrate the occurrence of tunneling and that of resonances. It is very instructive. one of the main and now wellknown methods to perform such calculations. can be found in any traditional text on quantum mechanics. which a particle in classical mechanics would never be able to intrude. that quantum mechanically a particle may have a small probability of being in a spatial domain.Chapter 12 Q u a n t u m Mechanical Tunneling 12. to explore first much simpler models in order to acquire an impression of the type of results to be expected. Such a vital quantum mechanical phenomenon is socalled tunneling.g. i. Thus in this chapter we consider traditional "square well" potentials in one-dimensional contexts. For various general aspects and applications of tunneling we refer to other literature. It is not surprising. i. was developed only in the last two to three decades. M. therefore. These newer methods will be presented in detail in later chapters. in texts on quantum mechanics is rare.1 Introductory Remarks One of the main objectives of this text is the presentation of methods of calculation of typically quantum mechanical effects which are generally not so easily derivable in nontrivial contexts but are important as basic phenomena in standard examples. the computation of finite lifetimes of a quantum mechanical state. states of a finite lifetime. 249 . that the explicit derivation of such quantities in nontrivial contexts. namely the instanton (or more generally pseudoparticle) method.e. Razavy [236]. In fact. Similarly difficult is.e. as — for instance — for trigonometric or double well potentials. in general.* *See e.

1) as the probability described by ijj(x.250 CHAPTER 12. -ih-r(x.5) L j • dF IFo is the probability that per unit time a particle passes through the surface FQ. We assumed this already in the preceding since we interpreted the normalization integral / Jail space dx\ip(x. t) was developed by Born in 1926.3) we can obtain the continuity equation which describes the conservation of probability in analogy to the case of the conservation of charge in electrodynamics.t) (12. r (12.t)\* (i2 2) - With the help of the Schrodinger equation ih-iP(x.2 Continuity Equation and Conditions The statistical interpretation of the Schrodinger wave function ^(x. (12. i. According to Eq. Quantum Mechanical Tunneling 12. Correspondingly one interprets as probability current density the expression j x ( < ' >=iH v ^7 v <4 = [W{x. M^ Then dt + V-j 1=0. so that p = |T/>|2 is the probability density for a spatial measurement at time t.e.t).t)\2 =1 (12. .4) (12.t) that the one particle concerned is somewhere in space at time t.2) we have = o + ~v-[(vv>*)V'-V>*(W)] = -v-j.t) = HiP(x..

finite and single-valued at every point x in order to insure that V( x ) i s a unique representative of the state of a system. However. 2m E 2 V0 > 0.2 . the equation can be integrated at every point. Tarrach [119]. even for discontinuous potentials. which is the case we consider here. (12.12.3 The Short-Range Delta Potential It is very instructive to consider in detail a potential V(x) = VQ6(X) with the "shape" of a one-dimensional delta function. Nonetheless.33. pp. the problem is analogous to that of the scattering process from a potential in a one-dimensional case. Such a region is for instance the domain where V is larger than the total energy E. From this follows that both p and j have to be finite and continuous everywhere. Thus it is possible that the probability for the particle to be in such a domain is nonzero.3 The Short-Range Delta Potential 251 The time-independent Schrodinger equation is a second order differential equation in x.6) . with regularization their study is very instructive for illustrating basic concepts of modern quantum field theory. we can do this also here. 29 . 'Maybe the reader dislikes being confronted again with the one-dimensional case. t For the singular delta function potential a modification of these demands is necessary.2a6(x)]il> = 0. It is therefore quantum mechanically possible for a "particle" to pass through a classically forbidden region. An important consequence of the wave-like nature of the wave function ip(x) is that it can differ from zero also in domains which are classically not accessible to the particle. 12. Gosdzinsky and R. Schiff [243]. as we mentioned above. If there is no absorption. If ^>(x) and V^>(x) are known at every point x. It is therefore sensible to demand that V( x ) a r i d V'0(x) be continuous. as shown by P. This effect is known as "tunnelincp. The wave phenomenon which permits transmission and reflection is familiar from electrodynamics where the laws of reflection and refraction of optics are derived from continuity conditions at the interface between dielectric media applied to the fields E and B of Maxwell's equations. This case plays a special role. where „ mpVo V(x) =+V0S(x). although in general small. In much the same way that we define there reflection and transmission coefficients. The Schrodinger equation for this case is* ip" + [A. Delta potentials in more than one dimension do not permit bound states and scattering. since this potential implies a discontinuity of the derivative at its singularity (as will be seen below). ~hr> k = n^- (12 7) - . higher dimensional delta potentials do not have properties like that in one dimension. 1 0 ~x o a= f See L.

i.ikC = i.e. [<//]0+ . e Q (12.11b) -2aC. (12. In the region x < 0 we then have both types of waves: the incoming or incident wave to the right as well as the wave reflected from the potential. i. Quantum Mechanical Tunneling Integrating the equation over a small interval of x around 0 we see. a 2 — a 2 a — ik a2 + k2 ' . Consider the case of a purely outgoing wave in the region x > 0 (no reflection back from infinity).13) ~A R= for x > °(12.12b) a=—^— C = -a + ik and hence B = C-1 Thus for i/> we obtain = ?—. 2 (12.e. that the derivative of i/> is discontinuous there: W\U + k2 J i>dx .(ik .ikB) = 2aC or 2a(l + B).9) States with k > 0 correspond to those of free particles that move with constant velocity v = hk/mo. The continuity conditions at x = 0 are therefore for V : and according to Eq. (12. (12. i.14) One defines respectively as reflection and transmission coefficients R and T the squares of the respective amplitudes.11a) ikC .e. _ fQr x < Q^ eikx _| ge-ikx ^=\ ~ Blkx *x .12a) (12.[^']o.2 a J 5{x)4>(x)dx = 0.252 CHAPTER 12. .e 1 + B = C.1) . (12. except in domains of the potential V.= 2a^(0).ik(C . a — ik £Qr < { eikx a e-ikx •t" .9) for ij.: (i2-iQ) Ce for x > 0.' : It follows that ik .8) (12.

19) Thus the delta function potential supports exactly one bound state. (12.a l s l.e. possesses simple poles at a ± ik = 0. as coefficient of the outgoing wave.2 = . 2m 0 (12. = . Thus ^ = Ce.12. poo nc s~t2 / -oo dxe dxC2e~2aW = -2a\x\ _ u 2C2 / JO a Hence C = y/a and therefore ^ = Vae-QK (12. or h2 E = --—a2. K = a. ( 1 2 . f Ce~KX c. .g.16) The corresponding wave function proportional to etkx for x < 0 associated with the pole for k = — ia is^ rl){x) = V^e~aW. z<0.2 o ^ ( 0 ) .3 The Short-Range Delta Potential and T= -^-rr a — ik 2 253 =1-R.(K)C = . i. at k2 = —a2. 'Note that the physical pole has negative imaginary part. . / -oo In the bound state problem the Schrodinger equation is ip" + [k2 + 2a6(x)]ip = 0. Newton [219]. See e.K 2 < 0.ip'(0-) = .) and V>'(0+) . the quantity T.2 a C . i. Omnes and M.e. when normalized to 1. and C follows from the normalization: OO A.K C . R.i 8 ) .15) Furthermore we see that similar to the S-matrix which we encountered in the case of the Coulomb potential. Proissart [223] or other books on scattering theory like R. We write the solution ^ so that ^(0+) = ^ ( 0 .e -«|x| = J \ CeKX for for x > 0.17) da#(s)| 2 = 1. oo (12.

The Schrodinger equation is iP"(x) + 2 ^[E-V(x)}iP(x)=0. Similar treatments of square well potentials or barriers can be found in most books on quantum mechanics. . or in the lectures of E.22) + De Feikx + Ge~ikx *We follow here largely F. Fermi [92].1 The square well potential. \x\ < a. p.1. The latter contains also specific applications in nuclear physics. 58. p. x€[-a. V(x) V=E -a v=-y Fig.254 CHAPTER 12. oo]. (12.21) The solution of the Schrodinger equation then consists of the following parts: Aeikx + Be tp{x) = { Ce~ iKX lkx iKX with with with xe[—oo. Schwabl [246]. 12.4 Scattering from a Potential Well We can immediately transfer the above considerations in a similar way to scattering from a potential wen* as indicated in Fig. Quantum Mechanical Tunneling 12.20) for for \x\ > a.a]. x G [a. —a]. (12. We set 2m0(E + VQ) h2 so that ip" +fc2V>= 0 ip" + K il) = 0 2 (12. We want to consider the case of stationary states for E > 0. 55. 12.

M(-a) .12.25) ^\=M{a)lCD where 1 / f l — « \ irea+ifca /i .f )e" t-\ K\0—tKa—ika . (12.-iKa ? ikAe~ika and .23) Ce~iKa + DeiKa -iKCe~iKa + DiKeiKa = = Fe ifca + Ge"*fca. (12.24) The first set of equations can be rewritten as / ( p—ika / p—ika pika pika U pika A \ \ / \ / 1 pina Kpina p—ina Kp—ina \ \f ° ){D ) \ B ) ° or A N \ Bt i. ( V C n vD ). 2 l ^i i K\ iKa-ika O \ (1 + %)e ina—ika (1 .. (12.ikGe~ika.26) K^g-ifoa—i/ca fc. .27) v ' Similarly Eq. fc Af(a) = .24) yields F ^ G. ) A l ( pika p—ika pina K ina kC p—ina \ _p—ika J \ Kp—ina J ( c\ U/ (12. ). I (12. Fikeika .4 Scattering from a Potential Well 255 The connection relations obtained by equating ip(x) and its derivative from left and right at x = — a and a yield the equations Ae-ika + Beika = CeiKa + Df.e. (12. (12.28) Prom these matrix equations we obtain ( B ) = MWM(-fl)_1 ( G ) ' where M(-a)-1 = v ^12'29) -i / [i + n i K\eiKa+ika *)e (i t-\ K\„iKa—ika fcje \ ] J fl2 30) • J ' 2 \ ("I — K'Jg-ita+Jfca Q _i_ K\p~iKa-ika \ .ikBeika = -inCeiKa + DiKe~iKa.

.(1 + |e 2 ) sin2 2na 1 .34) and as transmission coefficient (note e is complex!) \T(E)\2 = 1 .32) [cos2«.a . the quantity T{E) := £ = A -lika cos 2/ta — je sin 2/«x (12.e sin 2na IJlika e B = -r)F sm2na. we obtain A = F( cos 2na — . (12. = < Ce~lKX + DeiKX .33) One defines as transmission amplitude the ratio of outgoing to incoming amplitudes.34) into Eq. (12. i. (12.^esin2«a]e 2lfca \r\ sin 2/ta Thus if we now set G = 0 in order to have only an outgoing wave proportional to exp(ikx) to the right of the well.22) we obtain ' Aeikx + AT(E) I sin 2Kae~ikx T/>(X) (12. Quantum Mechanical Tunneling Hence with the complex quantities IK (12.4T(£)e ifci. \R\2 = T(E)^sm2Ka (12. x € [-a.—2ika [cos 2/ta + \e sin 2«a]e~ (12.256 CHAPTER 12.37) .e.36) One defines as reflection coefficient \R\2 the modulus squared of the reflected amplitude divided by the amplitude of the incoming wave. i.a].31) we have A AT(a) = )-MFo B -^r)sin2na .a ] .i^ 2 sin2 2KO 1 + |ry?7* sin2 2/ta Inserting Eq.35) with with with x G [-oo.oo].e. (12. xe[a.

However. an equation of the type f~1-f = g-1-9. for -VQ < E < 0.38) as expected. they have solutions. (12.12.e.e.39) 2 i.12. in the domain of A imaginary and K real.\ sin2 2KQ •n 4 sin2KaT(E) — \K\ > - IRI2 (12. (12. . (12. and k real. nor when both parameters are purely imaginary. 2\k K.40) we obtain cot(na) — tan(/«j) — i K IK -. 83. i. (12. Li ( — H— I sin 2na = 0. where cos 2/ca With the relation§ cot(2«a) = -[cot(«a) — tan(/ta)]. (12.41) i.44) .e. We see in particular from Eq. which is satisfied if either / = g or / = — g'1. cot(fca) = i— or tan(fta) = K K . p.34) that the amplitude F = AT{E) of the outgoing wave has a simple pole where cos 2KCL e sin 2KCL = 0.35) and using Eq.4 Scattering from a Potential Well On the other hand according to Eq. B. 2m0\E\ h? ' (12. formula 406. i. (12. Dwight [81].43) Thus we set k — —i ' H .42) These equations have no solution for K.31): 1\T(E)\2 257 1 .e. (12. (12.

(12. i. We now return to the scattering states.42) which then leads to bound states with associated even and odd wave functions.e.20).g. (12. Eq. from Eq. (12. n = 0 . (12. The infinity of the amplitude of the outgoing wave corresponds to a zero of the amplitude of the incoming wave. and is itself infinite there. ± 2 . This can be understood.e. (12. 2Ka = nn. (12. The amplitude of the transmitted wave assumes its maximum value where (cf. The vanishing of A implies that because k is imaginary the factor exp(ikx) does not diverge exponentially for x — .34). t&n(2Ka)ER = 0. around the values given by Eq. . i. Schwabl [246].e.l ) n . t2 2 (12. (12.c o . = 2/rn -(E-ER). . (12. A = 0. The remaining wave part becomes exponentially > decreasing.45) These are the equations that one obtains from the connection relations for the case of Eq. "Note that dtan(2K<z)/'d(2na) evaluated at ER is 1. F. „ .l .50) See e. Then according to Eq.49) Taylor expansion of the denominator in powers of (E — ER) yields .' We observe that T(E) has poles at those values of E which supply binding energies. ± l . Also. T(E)e2ika = — r .47) These states in the continuum are called resonances.46). Quantum Mechanical Tunneling .48) We expand T{E) around these energy values. .46) This condition implies the eigenvalues E^ER = n2^-^ -V0>0.65. k\ d(2Ka) K) dE \ E R 1v/2^a 2 h 2ER + V0 ^ER~(ER + V0) r 1 (12. (12.+ -\tan(2Ka) where T is given by" 2 1 (K 2\k 1 (K k\ /n .258 and obtain the equations CHAPTER 12. . 2m0\E\ and Kcot(Ka) = y—-^—L. 64 . At their positions cos(2/«z)|£.fl = ( . pp.35)) for E > 0 sin2Ka = 0. i. 2m0\E\ Ktan(«a) = — y——^— .

a 3 a. -2a<x<2a. . = 0. 12.e. We restrict ourselves to the period — la < x < 2a. Since we have reflection as well as transmission there are states with repeated reflection and finally transmission. V(x) x=-2a x= =2a . . T(E) has poles at E= ER-i: (12. -s \ V0 /• ~\ 'X . V(x) = 9(x + a)VQ6(a-x).5 Degenerate Potentials and Tunneling In the following we consider a potential which is very similar to a periodic potential and can serve to illustrate a number of aspects of the process of tunneling in quantum mechanics.51) i. A potential of this type is known as a Penney-Kronig potential.52) Here ER is the energy of the resonance and 2/F the lifetime of the state.2 The Penney-Kronig potential.' ''' -5a -3a \ -a ~~-~.2.12.^ l / . These are precisely the resonance states. 12. The Schrodinger equation to be solved is ^"{x) + ^(E-V(x))i.5 Degenerate Potentials and Tunneling We obtain therefore T(E)e2ika = = (-^ 259 iT/2 E-ER + iT/2' (12. 12. We consider a potential consisting of a chain of rectangular barriers as depicted in Fig.' ' 5a \V- / Fig. This result is physically very plausible.

e There is the additional boundary condition (for Vo = oo) ^ ( . Hip(-x) so that with Hi/)(x) — Eip(x): H[^{x) ± t/>(-x)] = E[^{x) ± xl>(-x)\.1.. the differential operator is even in x.55) The Hamilton operator of the problem. The particle of mass TUQ whose quantum mechanics is described by the Schrodinger equation is free to move in the domains under consideration to the left and to the right of the central barrier. i..a ) = 0 = ^(a).. . x + 3a <> ™ since the particle cannot penetrate into the infinitely high wall.56) Thus the problem has a two-fold degeneracy: Associated with every eigenvalue E there is an even and an odd eigenfunction and h2 n 2 7T 2 E — En > 2m0 4a 2 (b) Next we consider the case VQ > E.2 = W h 2 _„2 = 2mo(E . with V as parity operator.e.260 CHAPTER 12.2 < a .Vo) n.. This boundary condition implies the quantization of the energy with ^ ( 2 a ) = n7r. .2. E^En = ^ ^ t n = 0. We set .e. and hence for = Eip(-x). In view of the periodicity we have ^(x) = ^(x + 4a).53) *w«w^( « j)'*^{i i ^ . VHV-1 Hip(x) = Eil){x): VHV~lV^){x) = EVtl>{x). = EVip(x). and we choose (12. (12. i. (12. Quantum Mechanical Tunneling (a) We consider first the case of VQ = oo. HV^(x) = H..

60) The boundary conditions that ip± have to satisfy. we shall not exploit here.Be Ka (12. that these are such that for VQ — oo they approach those of the previous case. — a]. We can write the solution as consisting of the following pieces: { Asink(x + 3a) BeKX + Ce~KX with with i £ [ . BeKX±Be~KX with xe[-a.59) Asink(x + 3a) with x & [~2a.62) = nBe~Ka T nBeKa = ^n[BeKa q= B e " M ] . (12. (12.2a].12. which. ^kAcos{k2a). At x ~ — a the connecting relations are Asin(k2a) kAcos{k2a) and at x = a: BeKa±Be~Ka K. This wave function is no longer restricted to the interior of a box. We demand these for even and odd wave functions ip±(x) which we can clearly construct from ip(x) (as examples see the dotted lines in Fig. are ^_(0) = 0. ±Asink(3a — x) with x€[a.61) = Be~Ka ± BeKa = ±[BeKa ± Be'Ka]. VV(°) = °. V+(0) = const.58) D sin k(3a — x) with xE[a. The central region of the solution is the classically forbidden domain.a].2a]. We therefore have different boundary conditions. We expect. We > obtain even and odd functions ip± by setting ^ = ±1. At x = ± a we now have to connect ip and ip' of neighbouring domains. (12. V'-(O) = const..5 Degenerate Potentials and Tunneling 261 We restrict ourselves again to the period — 2a < x < 2a.2a. however. = Ka ±Asin(k2a).. xe[-a. of course.a]. 12. (12.2).il>-]?0.63) = KBe~ F = . — a]. so that (W meaning Wronskian) W[tl>+. (12. This means that i>±(x) = < | = ±1.

However.+ • • • 1! cos^(n7r) 2ak . (12. It is plausible therefore to use appropriate expansions. i. » we expect the eigenvalues to approach asymptotically those of the case Vo = oo. In general one now rewrites the equations in matrix form.e. (with re-insertion of nir/2a for k) 2ak ~ nir .nir + 0[{2ak . we obtain k v ' eKa^e-Ka | tanh(«a) v ' These are transcendental equations for the determination of the eigenvalues of the even and odd eigenfunctions. = AA. Taking the ratio of the equations in both cases. and is — as we see — less than that belonging to the lower sign case.Thus we observe a splitting of the asymptotically degenerate eigenvalues: kesympt.3 Level degeneracy removed by tunneling. i.nir) + OU2ak .— (1 ± 2e~2A + O (—). F ^odd (12.262 CHAPTER 12.68) . say. i.64) we obtain ~[(2ak k .even.4 x ) and cothx = 1 + 2e~2x + 0 ( e " 4 x ) .2 a . + 0 ( e .e. we set tanhx = l .nir)2}} = 1 ± 2e~2Ka. the odd solution with.. the limiting value for K = oo: tan(2a/c) = = tan(n7r) -\ (2ak — nir) 1 — ^-—. say A.e. Quantum Mechanical Tunneling Fig. (12. in the present case it is easier to continue with the above equations. 12. is the one belonging to the even solution.mr)2}.65) and we expand tan(2afc) about nir.2 e . where Akoce'2™.66) Inserting these expansions into Eq. For large values of Vb. (12. k = fc0dd.67) The value of k belonging to the upper sign. for VQ — oo. (12. i.e.

5 Degenerate Potentials and Tunneling 263 Thus the effect of tunneling — here an exponentially small contribution — removes the degeneracy of the asymptotically degenerate states. p. Solution: The wave functions i/>o>Vv in regions V = 0. exp r rb rb / •J a 2mo (V(x) . b > a.^ 0 = °' Hence in dominant order •00 — exp . 12. .1: A car's quantum mechanical probability to pass a bump A very slowly moving car of mass mo (kinetic energy almost zero) encounters on the road between x = a and x = b. 57. Fermi [92]. .12.4 A car meeting a bump in the road.3. .** " W ~^7~ Fig. exp .4. 12.69) where the potential V(x) is (with g the acceleration due to gravity) V(x) y'(xo) Setting w(x) = 7r**E. (12. i.E)da amplitude divided Thus with E ~ 0 the probability P is given by the square of the transmission by the square of the incident amplitude. Compute this probability. 2m0E „ 2mnE .2 dxy/V(x) (12. '. 1pV 2ro 0 4>o '.e. 0.v = 0. w(x0) •• = = mogy(x). and is a reformulated version of a problem set in lectures of Fermi. ^v + -^-[E..70) xo = -(a + b) 2' . a sinusoidal bump in the road with a height of one meter at its peak as indicated in Fig. V ^ 0 are given by the equations ^o + " J .. as indicated in Fig. but quantum mechanically there is a finite probability for this to succeed. 12. . y(x) = sin and y(xo) = 1. 2mn. Classically the car is unable to overcome the bump. The following example is an attempt to transcribe the quantum mechanical effect into a macroscopic situation. Example 12.V{x)\i.

4 x 10 3 9 ].2 . which give /•TT/2 dx sin'M . 369 and 8. n I . p. Emde [143]. dx sin ' x •• Looking up the value of the factorial in Tables. and assuming a length of b — a = 100 m of the base of the bump. 950. S. Gradshteyn and I. We have 8m 0 : exp ^(b-a)/4 x 2_1 -^^W(s x °' 9191 (12. .72) Assuming a mass of 1 ton of the car (i. M. formulas 3.17)).l . h = 1. ^ ) = 2 ^ . we obtain ( . 14. Quantum Mechanical Tunneling = y/mog 2(1) — a) /""'' / dw sin1/2 n Jo w.1.1.y) is the beta function (see Eq.71) The integral can be looked up in Tables of Integrals. p.384. : 2 ^ B ( ^ .e. (15. We find™ (with ( . Ryzhik [122]. (12.**.1 / 2 ) ! = v^r) TT/2 /.1 / 4 ) ! = (4/3)(3/4)! = (4/3) x 0.807 m s . 1000 kg) and using the following values of the natural constants: g = 9. Jahnke and F.9191 ~ 1.621. We can now evaluate the probability.2. one evaluates an approximate probability of exp[-6. p. **For instance E.264 we have ro / dxyfV(x) Ja CHAPTER 12.055 X 1 0 _ 3 4 J s .2 [fr/2 ~ I)'] (M-l)! 2 ' where B(x.

we consider the corresponding case of a particle above the flat surface of the Earth. in the book of G. in Chapter 15. 13. we consider the linear potential in three space dimensions that forbids the particle to escape. Siissmann [266]. This is the quantum mechanical version of the problem of Galilei in one space dimension. p. 144. Then considering the probability distribution determined by the squared modulus of the wave function. 265 . This is the problem of Galilei in quantum mechanics. Our treatment in the present chapter aims predominantly at an exploration of the quantum mechanics of the freely falling particle in a domain close to its classical behaviour.3. An additional reason to consider the linear potential at this stage is its appearance in the following chapter in connection with the matching of WKB solutions across a turning point. Thus.* In Chapter 14. the last case allows only a discrete spectrum. Finally. In the present chapter we consider in some detail the first of these which is the potential of a freely falling particle in one space dimension. * A highly abridged treatment of this problem is given.Chapter 13 Linear Potentials 13. each of which propagates with the classical momentum and energy of a Galileian particle.2 The Freely Falling Particle: Quantization Under quantization of the freely falling particle we understand here the solution of the Schrodinger equation for the linear potential. for instance. whereas the first case permits only a continuous spectrum.1 Introductory Remarks We distinguish here between three different types of linear potentials. Example 14. We shall see that in this particular case the wave function can be written as a superposition of de Broglie waves.

e.t) = . (13. V(x) = —mogx. (13.3) In general the momentum representation is of little importance. x > 0.= l V 2-7T J_oo 1 dkeikxijj{k.2.266 CHAPTER 13.1).= l V 2-7T J-oo dxe-lkxip(x. (13. Linear Potentials we shall see that at large times t its behaviour is determined entirely by the parameters describing the classical motion of the Galileian particle. The present case is an exception and a good example of its applicability. In the present case of quantum mechanics we have h2 d2 2mo dx2 -m0gx ip(x. 13.4b) . i.1) where g is the acceleration due to gravity and we can put Vo = 0. c> 0.t) = c\x\ + V0. from d_ -m x2 0 dx + V(x) 0.t). A different > "linear potential' is the so-called "confinement potentiaF V(x. (13.t). (13. We see " that the case g < 0 can be related to the reflection x — —x.t).t) = -mogx + V0. g > 0.4a) f°° ${k.2) which is the case with a discrete spectrum.1 S u p e r p o s i t i o n of de Broglie waves We consider a particle of mass mo falling freely in the time-independent homogeneous field of the gravitational force F(x. and yields rriQX = mog.t) = . Therefore we use the Fourier transforms il){x. The equation of motion follows for instance from the derivative of the constant energy. We consider here the case of Eq.t) = m0g with potential V(x. Recall the classical treatment of the freely falling particle. (13.

6) the argument k is then replaced by this. d \ ~f1 dkr[ k + m0gt ^r't (13.4a) d ~ f I dk ikx dk 2vr (im0g) iP(k.t) m0gx /2TT m0gxip(x.9) . We rewrite this equation in the following form th (13.-d m +imo9 .t).e.t). multiplication by a test function and subsequent integration).t) dketkxiP(k.7) In the wave function on the left hand side of Eq.e.< (13. To this end we consider the following substitution in Eq. (13.5) m+tm9Wcr{ktt)=2^^ktt)t (13. (13.t) h2k2 d ~J(k.6) and wish to solve it. we obtain the differential operator on the left of Eq. The vanishing of the boundary contributions is to be understood in the sense of distribution theory (i.8) With this result and the substitution (13.6).t) the expression (13.3) as a partial derivative d/dk: d lKx dkeAkx{imQg)—4){k.2 The Freely Falling Particle: Quantization 267 and consider the following partial integration in order to re-express the last term in Eq. • f-d ~(' m0gt %h . Applying the total derivative d/dt to this function.3) yields us now by inserting for ip(x. (13.6) (and to avoid confusion we do not introduce new functions and variables): k -> k + mogt (13.13. (13.6) the latter equation becomes d ~( m0gt ^(^M<=+=?. This can be achieved by first converting this partial differential equation with two partial derivatives into an ordinary differential equation with the one total derivative d/dt. (13.t)-im0g-^k.7) applied to every quantity in Eq. Equation (13.t) dk nikx k=oo '2-K (im0g)il)(k. i.

t ) = ipQ(k)exp Jo fdt' k+ m0gt' H (13.12) .i) .11) *[k-?f.t) = i>0[k 1— ) exp 2m0i J0 k+ ^(t'-t) (13.268 CHAPTER 13. Linear Potentials This is now an ordinary differential equation of the first order and can immediately be integrated to give an exponential. i. We define the wave number kt as kt-.10) k by m0gt back : k —> k h The solution of Eq. (13.10) Note the amplitude function ipo(k) in front.1: Verification of momentum representation solution Verify by direct differentiation that the solution (13.gh [ d t ' j f c + ^ ( t ' . E x a m p l e 13.e. Thus addition of both indeed leaves the expression on the right hand side of Eq. the third term from the integrand) > / -mog \ o exp dk h J h dt y ' 2mo k2i>(k.6). t ) .6).6). Thus we have first (the second term arising from the upper integration limit. We observe t h a t t h e solution is actually a function of kt — k — mogt/h. (13.6) is then $(k. Our next step is to re-express the solution (13. To obtain the solution of the original equation (13.= k- rmt. the following result h 2moi tp[k-\ ^—.t) ok lm og-^— exp dk •ghf dt'|fc+^(t'-t)W(fc.t). We have thus obtained the solution of the transformed equation.11).6).11) in a neater form. Solution: We apply the two operator expressions on the left of Eq.6) to the solution (13.11) solves the partial differential equation (13. On the other hand the second operator expression implies dtj>o imog^—ip(k. we have to reverse the substitutions and replace in (13.t ) ' U ( f c .t In Example 13.1 we verify t h a t this solution indeed solves Eq. (13. (13. (13.

.t) into this integral and recalling the property (13.4a). 27T ^ .— [It(k) .t) = .gfci + 2 g2m0 2/i 2 ' (13. .18) h =k mogt p h S' h .17) = J_/ C X ) h2 [It(k) .14) (13.o ut = 2mo -—kf 2mo A.+^f (13. Inserting ip(k.13.} -fA ' _ _tkt 2 k+ t) 2 k . m0g (13. (13.I0(k)].= 1 / f°° dkelktx^{kut) dkipol k V —-t 1 exp iktx K ) (13./„(*)] 2m^ig Inserting the explicit expression for the argument of the exponential.2 The Freely Falling Particle: Quantization 269 Then the integral in the exponential of the solution (13.t) defined by Eq.11). (13. m0g ~7T' k Z k _ r ^ h t Y + i n Smog (13.2„2+2 mfaH 3^3+3 .o o 2 -— | exp iktx h mZgH' . _ k t | h2 3 Inserting here the explicit expression for fct.t) = 1 / dkipo ( k mogt 2 h ( 1m\ig \ With .16) Next we evaluate the Fourier transform of ip(k. we have ip(x. . we obtain rp(x.13) rn0_t+2 + mlg2t3 . we can write the integral / = l h l h (hr + 7 7 (h 3m0g 3 m0g \ 1 H 3m0g With the definition « * ) • ! ( * = * .11) can be written -j'A ^.19a) .15) we obtain / = .

fm0gt\ \ 2h ) (13.2.o o Jo This result can be interpreted as follows.t)\2 for t — oo (distant future or past).18) into a • different form in order to be able to perform the integration with respect to k. Linear Potentials f dt'. (13.2 Probability distribution at large times Next we inquire about the behaviour of the probability density \ip(x.23) .t) as 1 I-00 ip(x.t) as a decomposition into de Broglie waves.19b) h3 3 we can rewrite ip(x. These wave vectors vary in accordance with Eq. 7 u>t'dt' (13. (13.gk— + 2h 3 2?7lo 2 2 h mQgk ^ _ mjfo2 2 -i + ft2 -fcr 2mf)g 03mg £3 .22) We can rearrange these terms in the following way with the ^-dependent contributions contained in a quadratic form: ht 2moi ht 2moi k mpx ht mogt 2h mogt\ 2h J + IX . Such a decomposition is possible only for homogeneous force fields. ipo(k) is the probability amplitude with wave vector k at time t = 0 and varies in the course of time with the time-dependent wave vector kt = k — mogt/h. To this end we convert Eq.19a) exactly as for a falling particle.270 and CHAPTER 13. 13. With the help of the de Broglie relation this wave vector corresponds to the particle momentum Pt = P — rnogt or to its velocity vt = v — gt.v = kh/mo.21) In other words.The associated kinetic energy of the particle is -mv+ = huJf.20) expresses the wave function tp(x.t) =-y= dkip0(kt)) exp i< fctx V27T J . The fc-dependent terms in the argument of the exponential are T:=ikX--^{-^fk2+r^k 2rriQig [ n nr (13.20) (13. the relation (13.ut> = Jo t2 g2m0 i 3 k t . (13.

integrate and then take the limit fi —> 0). rnogt m0g2t3 tX^r~.t) = 1 1 oo 271 27T /. It follows that ip(x.28) (for a verification replace a by a + ifi. (13.26) with the formula 1/2 (13. being now independent of £.18) thus becomes ip(x.QX ht V ht 2ft For |i| — oo the definition of £ defines a distance xc\ given by > k i.24) We now set fit / 2m 0 V" 0 V m0a.13./m 0 a. (13. Then ip(x.26) d£e ie £ kt^ m0' 0. m0xcl ht m0gt 2h -gt O v: °° 2hti (13.e. ftf fti m0gt\ 2h J' " J' m0x ht mo5* 2ft + 2mo ~ftT f. (13.t) 1 exp r .2 The Freely Falling Particle: Quantization The integral (13. TOO#A ft and.QX 2 ^ 0 7 (m.25) The amplitude ipoi^t) then becomes for |i| — oo: > 4>o(h) \t\—»oo 4>o[ k \ . fi > 0.t) exp i\ mogxt mox2 — h 2h 2ht mo<?2£3\] r (vn§x 24ft mogt\ (13.29) ./-oo dkip0(kt) exp l x ixI exp oo . 2h h 24ift ' mogt^ r m. We evaluate the Fresnel integral in Eq. may be put in front of the integral. /m0gt\ \ 2ft J ° ) I m09H3 ft J 6ih m0g2t3 _ m0x2" 2Aih 2hti m gt IT / dk^oik t) exp -oo ht k2mo« ~fti 2ft~] (13.27) xd = vt- Thus xc\ is the distance travelled by the particle as determined in classical mechanics.

3 Stationary States In this section we show that the Schrodinger equation with the potential (13. For stationary states we set xl){x. xc\). These are incoming and outgoing waves.e.272 CHAPTER 13. 13.3) the time-independent Schrodinger equation i^ + ^ [ E + mogx]<l>(x)=0. (13. this expression can be brought into the following form: if *\ 0 ip{x.t) = e-iEtlhct>{x) (13.32) For x — oo the function cf){x) behaves like > 4>{x) ex exp exp dx { 2mo (E + mQgx) \ H2 -.1 A turning point at x = —XQ. kt and ut- 13. 1/2- - ±^(2mlg/h^x^ / 2i (13. For x exponentially increasing and decreasing solutions are possible.33) -oo for E <C \mogx\.30) We see therefore: The large time asymptotic behaviour of the wave function is determined entirely by the classical values of x (i.1) has only a continuous spectrum. Linear Potentials Replacing here x by the large time value x c i.dA 4>o(h (13. V(x)=-m 0 gx V=E Fig.t) ~ W —-exp i V irlt m ktxd J cot.31) and obtain from Eq. . (13.

e. and hence is continuous and extends from —00 to +00.13. we .3 The Time-Independent Schrodinger Equation 273 Classically the particle coming in from the right with energy E cannot penetrate into the potential barrier since (where V > E) with conservation of energy E — V = p2/2m. so that ih dtp{k. t) = — = / V27T J . at which the classical particle bounces back. The Fourier transform is given by the relations (13.36) m0g \ 6m0 with c = const. t) = Eil)[x.e.35) It follows that /W tm0g \ 2m0 i. Fig.e. is reflected.e. its momentum p would be imaginary there. i. which is not possible for real particles. at which E — V = 0.e.1. Akx (13. However quantum mechanically this is possible with a small. E)dk. probability. In view of the simple form of the potential in the present case the equation can be integrated in the momentum or k representation. i. J \ „ (13. Thus the spectrum does not contain states which are normalizable over the entire domain x e [—00.t).e. The point — XQ.4a) and (13. t). (13.34b) It therefore follows from Eq. and we can establish for the resulting continuum solutions the orthonormality and completeness relations. i. but also d 00 dt Eijj(k. 13. exponentially decreasing. 1 f01 dketlcx^{k. H2k3 4>{k) — cexp Ek (13. We determine the constant from the condition that the continuum solutions are to be orthonormalized to a delta function.4b). i.t) = dt (13.34a) Then d &k 1 f°° ih—il>(x.6) that also for 4>{k. i. cf. is therefore called a turning point Quantum mechanically this is a point at which periodic solutions of the Schrodinger equation go over into exponential solutions or vice versa.0 0 dkelkxE^(k.00].t).t) — (p(k) > E +^og~Mk) = 1 (h2k2 ^m.

41) J—oo Next we insert (13. apart from a constant phase which we choose to be zero. / dE(j)E(x)(t>E(x') = ^ s .L dxe-ikx(f>E(x).39) we obtain dkcc*exp m0g It follows t h a t cc = -&(# .38). the relations oo /*oo dE4>E(k)4>E(k') = 6(k-k').. </>E(x) = .(k) = / J ~oo dx<t>E{x)<\>EI(x) = S(E - E') Inserting (13. Linear Potentials OO dx<ffE(x)<i>E>(x) = -oo 8(E-Ef). h? h2 .L [ dkeikxMk). / <>~--ikx VZ7T J (13.e. In a similar way we can verify the validity of the completeness relation. i. 13.M *K J-oo Setting dt cos ( -ts n Jo -st).37) (eikxdk. (13.43) y=2^x+2^E.e.36) (with c replaced by (13.36) into (13. V 2lT J Mk) = .a : ' ) .274 have / Using d(x) = — one verifies t h a t with CHAPTER 13..£') = <*(#-£')• 1 27rm0g' = /o .38) we have for <j)E{k): OO /"OO / -oo (13.42) The Airy This integral can be rewritten in terms of the Airy function function Ai{s) can be defined by the following integral: 1 f°° Ai(s) = — / eft exp i ( st . (13-40) i. so t h a t -oo / <fiE(x) J dke ikx s/2nmog exp m0g \ 6m0 Ai(s).39) dk4>*E{k)4>E. (13. (13.40)) into (13.

. and in Chapter 15 in the computation of energy eigenvalues for the three-dimensional linear potential. (13.44) is then expressed in terms of the Airy function: . / s M . the function 4>E{X) i and setting s y = — = 1 .J * . whereas the other is of exponential type.^ . . (13.^x + ^ E (13.13.43) / • '**«(» + * ) * • < * + *) ds dtexp i{s(y —TIT (27 + x)- -s3 ^Y J ds_ f dt_ f J exp — i< t(x + z) -r 2^rJ 2W dxe^-Qe^-^e-U*3-*3) _^Pi(Sy~tz)e-i(s3-t3) S(y = j£j**w -I ^LJt{y-z) 2TT _ We will encounter the Airy function Ai(s) again in Chapter 14 in the matching of exponential W K B solutions to periodic W K B solutions. so t h a t the integral becomes 4>E{X) = 2vr ^/mQg / n oo dtexp -oo h? * 2m£g y \ 7 5 * " 3 M' Choosing /2mfo\1/3 A= ^ . For instance dxAi(y / • + x)Ai*(x + z) = 5(y — z)./ 2m 0 E (13. For a further solution Bi of Airy functions we refer in Chapter 14 to Tables.46) This follows since with Eq.45) One can now derive normalization and completeness integrals for the Airy functions. In the next section we derive the important asymptotic expansions of Ai(s) for both positive and negative values of s. and we shall see t h a t one branch has periodic behaviour.3 The Time-Independent we can write <J>E{%)'4>E{X) = Schrodinger Equation 275 2ir^/m0g J dkexp h2 yk — -k '2m20gV~ 3 W i t h the substitution k = [it we change the variable to t.

(13. o (13.2 The contour C for s < 0 with ends in the angular domains (a) and (b).f dzJ*W 2TT JC with z = reie. i.e. or -n <6 <--TT o for r -> oo.51a) and since with sin 30 < 0 also sin(30 + 2ir) < 0: -TT < 36 + 27r < 0.. 13.51b) . (13.43) contains the phase (with x — z): > *(*) := sz . Thus we must have sin30<O.\zz. (13.49) (13.47) Consider the following integral along some as yet unspecified contour C in the plane of complex z: / := i . We want the integral to exist and so desire that lim e**^ = 0.276 CHAPTER 13. (13.50) Im z domain (b) / ' ^ ^ ~C domain (a) 0 ^ ^" ~ ^ \ x Zs=-i^S Fig. (13. Linear Potentials 13.r 3 sin 30 ).-TT < 6 < 0 for r -> oo..r 3 cos 30 J + i ( sr sin 9 . .e. i. .4 The Saddle Point or Stationary Phase Method The integrand of Eq.TT < 30 < 0.48) We have $(z) = sre1 1 A S„3i6 -r e sr cos 6 .

and the main contribution to the integral comes from the a region around the saddle point.45) exists for the contour C. It is reasonable to choose the contour in such a way that only a short piece of it contributes substantially to the integral.54) where a is the new variable with —oo < a < oo.51b).2.48).55) s < 0.e. (13. (13. (13. where the derivative of the phase function vanishes. we choose zs = -iy/^s.58) . so that exp(±2i9s) — 1 and — s — rl. and we set with a real: z := zs + a (-S)V4' (13. (13. (13.-z3s and V^s + .56) 2n(-s)^Jc But I daefr-"?*"^. 13.57) $"(z s ) = 2% s. —i ${zs) = szs .( \ / ^ ) 3 -2z. provided its ends at infinity satisfy these conditions. Hence _ 2 / -s)3/2 /•oo 2TT(- s)!/4 doe J—oo -CT2- i*3 3(. Such a piece can be found in the neighbourhood of the so-called saddle point zs.54) i ¥~s)3/2' (13. i. Inserting (13.51a) and (13. at which the phase becomes stationary. oo] as indicated in Fig. Then 6S = ±7r/2.55) into (13.e.S )V2 (13.4 The Method of Stationary Phase 277 The integral (13.13. i. Then the ends of the contour lie in the domains specified by Eqs. We now choose the path of integration C at fixed O z = —iy/^s with a e f [—oo. 0 = &(z8) = s~zt s = za „2„2i6>.52) = exp(±i-7r) Let s be real and s < 0. we obtain da 2KJC (-*)V4 c 1 e **(Zs) exp i{$>{Zs)+l-(z-Zsf<$>"(Zs) + (13. Then Ze TeC : V^se±i7r/2 = ±i\fzrs.53) Since we shall have the contour C in the lower half plane.

Linear Potentials where the contribution oc er3 is obtained from &"'(zs) whose evaluation we do not reproduce here. We observe that the Airy function has a periodic behaviour in one direction but an exponential behaviour in the opposite direction. 2 s 3/2 .. (13. but would increase exponentially for imaginary values of a — thus describing a surface around that point very analogous to that of a saddle. S.62) This is the Airy differential equation with one solution 4>{z) = Ai(z). With the help of the substitutions (13.60) The reason for describing the point zs as a saddle point is that the integral in Eq._ 1 . cos .59) In Example 13.2/3-2u = 0 (13. as one can see from the following equation^ ri' + F-fz •2^2. This is an important aspect in the WKB method to be considered in Chapter 14.491.278 CHAPTER 13. Gradshteyn and I.56) decreases exponentially around a — 0 for real values of a.s ) V 4 exp \(s) 3/2 (13.63) The Airy function can be re-expressed in terms of cylinder or Bessel functions Zv(z). (13. M. > so that Ai(s) 2 v ^ ( .44) we can convert Eq.64) tSee I. 8. (13.61) 2m0E = f 2 \1/3r 2 2 2 fi h \h mog2 Equation (13. V 2h2 (13.42) into an equation with a more appealing form. (13. For (—s) — oo the integral converges towards T/TC.32) therefore becomes dz2 + zd) = 0. V^s / 4 V3 (13. (13.2 we show that for s — oo: > Ai(s) 1 .32) with the solution (13. According to Eqs. Sec.44) and (13. .45) 4>E(X) ~ Ai(z). where — with e = (2/h2mog2)1'3 — we have /2^m3g3y/3 x\ = e[E + mogx}. Ryzhik [122].

2: Periodic asymptotic behaviour of the Airy function Obtain with the saddle point method the asymptotic large-s behaviour of the Airy function Solution: We have (cf. the conditions (13. .13.43)) Ai(s) = — [ dzeiit(-z\ *(z): 1 Ai(s)._ . cos | . Eq.51a) and (13. 27T Ic Jc 3 This time s is real and positive. Thus there are two symmetrically placed real saddle points on the real axis. changing to the variable p. with *"(z) = -2z.51b) are satisfied. (13. Hence the saddle points are given by 0 = * ' ( z s ) = s — z2. * " ( z ± ) = =F2Vi. Then.s 3 / 2 - .4 The Method of Stationary Phase with solution u = z1/2Zi/2/j(7A 279 (13-65) With expansions (13. we obtain f° J-oo dp=--—[a+ i\a\]da. Choosing the contour C to lie along the real axis. Example 13.^ e ^ / V2 4 [°° Jo dae'2^.60) we have thus obtained (with the help of the saddle point method) the asymptotic expansions of the Airy function Ai(z) which play an important role in Chapter 14..z ± ) 2 * " ( z ± ) .59) and (13. kl dpe*' = ~e^'4 V2 f° J-oo dae-W2 and f°° dpe'^ JO = . We have therefore r /*0 poo zs = ±y/s = z± / Jc dze^M = / J-oo dze^^ + JO dze^z\ In these integrals we set respectively P with *(z) = #(z±) + i ( z . P Introducing a new variable a by setting p = a-i\a\. It follows that for s —> oo: Ai(s) ~ .

3: Derivation of the Stirling approximation of a factorial Obtain with the saddle point method from the integral defining the gamma or factorial function the Stirling approximation.*. ~ ef<-z ex 3f(*)dz / -x2f"(z0)/2 dx = i 2TT P/(*o) -c The gamma function Tin + 1) or factorial function n! is defined as in Eq. i.e. z o = n .3. 13. (11. Then.3 The saddle point at z = n + ix. and hence integrating as described above through the saddle point indicated in Fig.111). . Thus here / ( z ) = — z + n l n z . the method of steepest descent evaluates the integral -J^dz by expanding f(z) about its extremum at ZQ with z — ZQ = ix. one obtains n! ~ where we used f^° dxexp(—w2x2/2) = V2^nn+1/2e-n.dz = idx (observe that this implies an integration parallel to the axis of imaginary z through the saddle point). n!= / e~zzndz= / e~z+nlnzdz. since f'(zo) = 0 and (z — ZQ)2 = —x2. 13. / " ( z o ) = —n/z2 = —1/n. Solution: Briefly. Jo Jo \ . ^/2-KJW1.280 CHAPTER 13.Be 2 Fig. Linear Potentials E x a m p l e 13.

For the history see pp. The central issue of WKB solutions is their continuation in the sense of matched asymptotic expansions across a turning point (i.317. Jeffreys [144]. will there be that of a coupling parameter.Chapter 14 Classical Limit and W K B Method 14. *R. 281 . 316 .* Descriptions of the method can be found in most books on quantum mechanics and in some monographs.e. i. and the role played by h in our considerations here. The WKB result is therefore frequently described as the semiclassical approximation. Kramers and Brillouin. of misunderstandings and other aspects may be found in the monograph of Dingle. § See e.g. Froman [99]. Brillouin [37]. H. where *G. ' T h e most prominent earlier expounder is H.295.1 Introductory Remarks One of the most successful methods of solving the Schrodinger equation is the WKB method. A. Kramers [153] and L. for misunderstandings pp. B. been developed previously by others in different contexts.* The method had. O. The WKB method we consider in this chapter is a precursor to our uses of the path integral method in later chapters in the sense that the expansions are around the classical limit. A critical overview of the history of the method.§ Our main interest here focusses on the use of the method as an alternative to perturbation theory (supplemented by boundary conditions) and to the path integral method so that the results can be compared. named after its first promulgators in quantum mechanics: Wentzel. N. Wentzel [282]. It goes without saying that such comparisons are very instructive. 292 . ^ The central problem of the method is the topic of "connection formulas".e. however. Froman and P. Dingle [70]. the linkage of solutions above a turning point to those below.

for which the observation of position and momentum is independent of the time at which the observation takes place. • {VA + %-AVS [ h B iS(r)/ft +^{vS-VA+^A(VS)2 *See e.t) = e-iEt/h7P(r) and \V(r. the derivation in Example 18.V(r)]^(r) = 0. A(r) and S(r) being real functions of r. It is for these states that one obtains the time-independent Schrodinger equation A^(r) + ^ [ E . (14.2) dip(r = I VA{v) + ^ ( r ) V 5 J eiS^lh. In this equation we set. For stationary states ^/. We shall see that in this limit the > Schrodinger equation describes a steady stream of noninteracting Newtonian particles. It is then possible to derive in a general form the relation known as Bohr-Sommerfeld-Wilson quantization condition. We start from the Schrodinger equation ih <9* =HV at with H=-£— p2 2mo + V(T). so that _ Aip AA+^V• h AA+-CVAn (AVS) + IVS h VS + AAS) iS(r)/h (14.7.* We illustrate the use of this condition by application to several examples. the energy has a definite value E with V(r.2 Classical Limit and Hydrodynamics Analogy In this section we consider the limit h —• 0. the function S being called phase function.g.282 CHAPTER 14. 14.t)\2 = \iP(r)\2. Classical Limit and WKB Method E — V changes sign). V>(r) = A(r)eiSWn.1) (14.3) .

12) VA A (14. These equations are equivalent to the Schrodinger equation. We can identify Eq. we obtain the following two equations: E .* -VA-VS+-AAS + V(r)A. (14. (14.+ VA-VS + -AAS = 0.7) h2 AA 2mo A 1 (VS)2 2mo = 0 (14.5) Apart from these we also wish to consider the equations that follow in a similar way from the time-dependent Schrodinger equation. (14. We set in this equation.2 Classical Limit and Analogy with Hydrodynamics 283 Inserting this into Eq.t)}2.4) (14.t) = and obtain dA i_dS_A + dt h~dt 2 h r A(VS)2 AA h2 2mo ih A(r.10) with a continuity equation by defining as probability density p:=*** = [A(r.11) We define as probability current density the vector quantity J:=K where V* = ijf* -V* rnio i_ „ *. h h (14.13) . (14.t)eiS{r't)/h.Q A 2i 8A 1 (14. (14.6) 2mo dt now with A also a function of t. h2 ih— = H$ A + V(r) * .V(Y) + and 2VA • VS + AAS = 0.1) and separating real and imaginary parts.14. (14. V(r. h h Taking real and imaginary parts of these equations we obtain the equations: + dt and 2rriQ 2m.10) m 0 ^ .

16) BA 1 moA—.17) With Eqs.284 CHAPTER 14. Next we investigate the significance of Eq.14) A2VS.01.20) With this we can rewrite Eq.21) . (14.. m0 But now (14.9) for the phase function S. Classical Limit and WKB Method This means # * — V * = — AVA + — and so J = — A2VS.15) and (14. (14. we have mo—(A2) + V(A2)-VS + A2AS = 0. Since we obtained this equation from Eq. (14. (14. (14. (14. the implications of this equation are also those of the equation of continuity.10).19) m0 m0 it is suggestive to define the following vector quantities ± = at v := I p = J-VS. . i-e. TTIQ (14. ot 2 Multiplying this equation by 2 and recalling that the function A was introduced as being real. (14.+ AVA • VS + -A2AS = 0. Since J = — A2VS = -P-VS.9) as 9S 1 -mo^ 2 + T. + h2 AA V = — ^ .n. %(1"1' I<-2) and V J = — V • (A2VS) mo Prom Eq. (14.10) we obtain = — [V(A2) • VS + A2AS}.J = 0.16) this becomes m 0 1 rr + m o V . mo <™> (14. (14.18) ot ot This is the equation of continuity that we encountered already earlier. ^ + V • J = 0.

2 Classical Limit and Analogy with In the limit h2 — 0 this becomes > Hydrodynamics 285 ^ + W 2 + F = 0. One should note that in no way do we simply obtain the classical equation of motion from the Schrodinger equation in the limit h2 — 0.V ) v + V 7 = 0. Such series are typical for expansions in the neighbourhood of an essential singularity. This is the analogy of the Schrodinger equation for h2 — 0 with hydrodynamics. r o t v = ( l / m o ) r o t g r a d S = 0. and hence''' ^-(m 0 v) + m 0 (v. > In the case of stationary states we have dp/dt = 0 (which is analogous to the condition of stationary currents. . (14. An expansion in ascending powers of h therefore starts with contributions of the order of 1/h2.e. However.23) (14. In the first • place we require the ansatz (14. where. with the help of Eq. v dt dt so that the equation can bedt written dr dt ~(m0v) + W = 0. We deduced this equation solely from the phase S in the limit h — 0. > The motion of the particle or rather its probability is altogether given by the above equation of continuity.24) d 1 dr d •— = d Lv _ .22) (14. This result is very general. i. (14.e. This means.14. This expression with h in the denominator shows.20). 'Recall the formula V ( u • v) = (v • V ) u + (u • V ) v + v X rotu + u X rotv. so that V ( v • v) = 2(v • V ) v + 2v x rotv. We see therefore that in the limit h — 0 the particle behaves > like one moving according to Newton's law in the force field of the potential V. The equation of continuity then describes the stationary flow of a fluid. we obtain ^ V 5 + | v ( v v ) + V V = 0.2) for the probability amplitude. The particles have no interaction with each other. which for very small values of h yield the dominant contributions. the wave function \I/ describes effectively a fluid of particles of mass mo. d — = (14.1/h. a singularity different from that of a pole). dt 2 Constructing the gradient of this equation. they are asymptotic expansions. V • J = 0. that the quantum theory implies an essential singularity at H = 0 (i. or that of equilibrium in electrodynamics) and dS/dt = —E (from ^ oc exp(—iEt/K)).25) dt We recognize this equation as the classical equation of motion of the particle of mass rriQ.

Sec. As remarked at the beginning. i. §We follow here partially A.3. Kramers and Brillouin.1 The W K B Method T h e a p p r o x i m a t e W K B solutions Without exaggeration one can say that (apart from numerical methods) the method we now describe is the only one that permits us to solve a differential equation of the second order with coefficients which are nontrivial functions of the independent variable over a considerable interval of the variable.2. The expansion in powers of h2. 318. (14. p. z (14. 6. (14. similar to our earlier procedure with A and S real functions. the method is widely familar under the abbreviations of the names of Wentzel.27) (14. h2 l 2m0 {E-Vy corresponds to an expansion in decreasing powers of (E — V).26) We observe already here that an expansion in rising powers of k.e.In A(x). Dingle [70].§ y = A(x)eiS^'h where W(x) = S(x) + .30) ax *For details see R.5)) and 2 ^ f dx ax + All=0.$ The basic idea of the WKB method is to use the series expansion in ascending powers of h and then to neglect contributions of higher order. Messiah [195].[E-V(x)}y = 0. such as "Liouville-Green method' and "phase integral method'. B. We consider the one-dimensional time-independent Schrodinger equation y" + ^.28) The calculations we performed at the beginning imply the following equations (compare with Eqs. (14.4) and (14.3 14. There are also other more specific descriptions. Classical Limit and WKB Method 14. i = eiW(-x^h. We set. is an asymptotic expansion. or h2/2mo(E — V).286 CHAPTER 14. This means that the WKB method can make sense not only where one is interested in the classical limit but also for E — V(x) and hence E large. .

A = const. . (14.V) and (14.35) .36) Equation (14. (14. + (14.10)). (14.\ In S'. 3 / 5 .3 The WKB Method 287 Equation (14. (14. Li This expression can be substituted into Eq.35) can be integrated.V) = h2 In the WKB method one now puts S := S' = S0(x) + h2S1(x) + (h2)2S2(x) S'0 + h2S[ + ••• . Eq.31) \a.34) Equating coefficients of the same power of h we obtain (S'0)2 = 2m0{E .33) 3 (S")2 . Since the expression (14.4 (S>)2 1 S'" 2 S> (14.29) becomes (note that this is still exact!) (S')2 .28).e. r//\ 4V^ 2 Up to and including contributions of the order of h2 Eq.2mQ(E .14.32) s" = sz + h2s'( + •••. But first we note that: A dA dx2 2 = c{S')-ll\ = A 15^ 2 S' + ^ = A ax 3 (S") 4 (S')2 2 2 S' With this Eq.32) is therefore (S'0)2 + 2H2S'QS[ .30) corresponds to the equation of continuity (i. Integrating this second equation we obtain ll[dX so that = -\l^dX' A = c(S')" 1 / 2 . (14.2rn0(E -V) = h2 lS'A 2S'0 (14.

i. The solution is seen to be periodic provided k(x) is real.I In S' + In c h 2 2 2 exp ^{S0 + h Sx) .27) we obtain now with Eq.40) rx dx v "7 W)\.and So = ±h / — — A J k(x (14. y = c'{k)1/2 exp or y(x) = a \ A ( z ) cos I / dx W) dx + 13).19). p = h/X) k{x) :-Then S'n +.e. and instead is asymptotic which we shall not establish in detail here. we expect that in any case So » ft2Si. (14.' For the expansion (14.33) to make sense as an asymptotic series. Chapter XIII.± ln(S'0 + h S[) + In c (s'0y/i and hence exp :5 0 + 0(h2). B.39) and (14. Eq.288 CHAPTER 14.41) We now ask: What are the conditions of validity of Eqs. Classical Limit and WKB Method has the form of a momentum. (14.39) where a and /3 are constants.E)' Then y(x) = V^(a 7 exp dx + 6 exp T(x) (14. y/2mo(V(x) .41)? We note first that the expansion in powers of h2 does not converge. Dingle [70]. it is reasonable to set (cf. (1. E > V(x).38) h y/2mo(E-V)' (14. 1 S e e R.31): y = exp = ~ ~iW(x) h exp L nH™ exp ^S . (14.37) For Eq. For V(x) > E (this is the classically forbidden domain) we set l(x) = -^==^====. (14. . (14.

36). we obtain -1/2 -1/2 2S[ and hence ±2hS[ = [kL/2]"kL'2 = lk~1/2X = Al/2 [lA-l/2A//_lA-3/2(A/)2lAl/2 so that It follows that (14.3 The WKB Method We have 289 S0 = ±h and Si follows from (14. Since k(xy 1/52\S'0 (14. Setting R := |A'| = \moKV'{x)\ \2m0{E-V{x)\W .45a) .V{x)) ^ This condition is satisfied provided E or E — V{x) is sufficiently large or (14. E > V(x).42) [(^).3/2 s&T + 3 {So Il« we have (cf.1 / 2 ]" [-§0SS).36)) 2s'0s[ = [(s&r 1/2 ra) 1/2 With S£ = ±ft/A.14.43) The condition So 3> ^2<Si therefore implies that /f>^-i/-^)k = H/y/2m0(E is sufficiently small. (14. Eq.

45b) so that the condition So » h2S\ is also satisfied by demanding that i ? C l .V(x) = V(a) . Roughly speaking one therefore requires E to be large in both cases.V{x) ~ -Or . How do the solutions of Eq. Classical Limit and WKB Method we can rewrite the inequality (14.a)V'{a).290 CHAPTER 14. In a similar way we have for E < V(x): l'(x) <C 1. We can derive the equation whose exact solutions are the WKB approximations of the Schrodinger equation. (14. around the classical turning point.e. In terms of X the original Schrodinger equation is y" + -= y = 0. V(x)) (14. 14. i.2 Turning p o i n t s a n d m a t c h i n g of W K B solutions Let x = a be the value of x at which E = V(a). (14. (14. the WKB approximation is in general invalid (see below). at which E — V{x) changes sign? We note first that in this case near x = a: E . *2 A(z) = V / 2m 0 (£ .44) as Idxsj2mQ(E-V{x))l 1 + ^R2j » ^Rh. and hence (with UIQ = 1/2. In the neighbourhood of E = V(x).48) The (dominant) WKB approximations are exact solutions of this equation.48) behave in the neighbourhood of such a point called "turning point'.46) The desired equation has the solutions y(x) = aXll2 exp ±i dx / (14.3.47) Differentiating this twice — we omit the details — we obtain the equation y" + l A2 (A 1 / 2 )" *i/2 y = o. h = 1) .

The transition is provided by matching relations. in the domain E > V the WKB solutions are oscillatory. (14.48) possesses a singularity of the form of l/(x — a)2 at every turning point x — a.14. 14. (14. in the domain E < V the WKB solutions are exponentially increasing or decreasing. On one side. We shall not derive these conditions here in detail but rather make them plausible in Sec. In the immediate neighbourhood of such a point the Schrodinger equation can therefore not be replaced by Eq.1.1.1 The regions around the turning point at x = a. i.1 there are WKB solutions on either side of the turning point and some distance away from it. As indicated in Fig. We assume we have a turning point as indicated in Fig.48). since this requires different approximations there.3 The WKB Method with the derivative relations 1 1 -a)5/4' 4(x 1 5 16 (x-.3. 14.3 after stating them first here. V=E Fig. 14.e.a ) 9 / 4 ' 291 (v^r 1 (x — a>2- These relations imply that Eq. 14. and on the other side. 14. It is therefore necessary to find other solutions in the neighbourhood of x = a and then to match these to the WKB solutions in adjoining domains as indicated in Fig. which connect one domain with the other. We define solutions ?/i and ?/2 with branches to the left and to the right as follows (for .

49) x 3> a -Vxsin and 2 exp V2 VACOS (L X dx A ra dx X for for i « a . (i x dx 7T ~~k~ 4 / One should note t h a t the solution ^4?/i + By2 for A ^= 0. Thus. This means the asymptotic part suffices only in t h e exponentially decreasing case.50) x S> a. however with the limits of integration x. P u t differently: We could add a lot of contributions to Ay± without affecting the asymptotic form of Ayi.51) It can be shown t h a t the second W K B matching condition in the opposite direction (and for the potential rising from right to left as in Fig. has the same asymptotic behaviour in the domain i « a a s Ay\.1) is given by the following relation with the + sign in the argument of the exponential on the right and no factor of 2: VACOS I / —. Note also the extra factor 2 in yi.292 CHAPTER 14. one could simply replace everywhere /•a / pa • • • b y / Jx J x .H— J =^ v/exp + m (14. Classical Limit and WKB Method definiteness note the + signs): Vtexp yi ~ S + Jx I X for for i « a .1) is then: VI exp {~[T) Vxcos fx dx TV Ja ^ _ 4 (14.52) For a potential rising from left to right. 14. 14. and E ^ V for a ^ x the same formulas apply with the same direction of the arrow. (14. as in Fig. (14. The first W K B matching condition in the direction indicated by the arrow (and for the potential decreasing from left to right. in order to make the formulas independent of whether the potential is rising or falling. a interchanged.

(14. If higher order contributions are to be taken into account. the equation becomes approximately y" = (xa)xiy. the equation V approximately linear around turning point V=E Airy solution WKB . the matching relations have to be altered accordingly.48). as we saw.Airy overlap* Fig.e.2 The overlap regions around the turning point at x = o. y" + Toy = o. (14. One should remember that the matching relations above are those for the dominant terms in the WKB expansion.46).3.14. is singular at a turning point x = a. 1 ~ (x . Thus around that point we can not use the WKB solutions and hence have to find others of the original equation (14. 2 1 X (14. i.a). 14.3 Linear approximation and matching How do we arrive at the matching relations given above? We observed above that Eq.53) Since at x ~ a. 14.3 The WKB Method 293 In that case Eq.54) .51) — always with the decreasing exponential on the left — is valid in both directions. (14. Therefore we consider now this original equation in the domain around x = a. whose exact solutions are the dominant WKB approximations.

or to the literature:H 3/2 2^{-zy/* Ai(-z) ~ < ~~FL~T77 y/TTZ1/* 1 1 s e -K-*) n for ~. as > may be seen by referring back to Eqs. R.59). Stegun [1]. (14. Dingle [70]. we have r x * f x ' /2(x -a)i/2d*=^(x . In the following we require the asymptotic behaviour of the solutions Ai(—z). > for a.56) We have discussed some aspects of solutions of this equation in Chapter 13. (14. p.55) the equation becomes the Airy differential equation d2y dz2 zy.60).1 / 4 by \^. Abramowitz and I.294 CHAPTER 14.Bi(—z) for \z\ — oo. (13. In particular we encountered a solution written Ai. B. — a ^> 0.59-60. .291.a)X\/3. or M. (14. 448. Classical Limit and WKB Method where xi is a proportionality factor which we choose to be constant.^ 4 I 7T x — a <C 0. This is given by the following expressions.4. (14. A.57b) UV2 3 11 by J dx . J (X (i4 58) - Replacing in Eqs. formulas 10. With the substitution z = (x . (13. i I 77-2 \3 + > for i-a>0.57a) COS 2^2/3 _ SZ 1 1 ?(-z) /2 3 es^ Bi(-z)~< 1 1 for x — a -C 0. A second solution is written Bi.57a) and (14.a)3/2 ^ /2 =^ 3/2 o Q. Cf.57b) To the same degree of approximation as the equation y" = (x — a)xiy. p. (14.

Example 14.1 to the quartic oscillator. by Ai(z). however.1 the Airy solution becomes proportional to the exponential WKB solution and to the right to the trigonometric WKB solution.l. (14. In the neighbourhood of a turning point the Schrodinger equation becomes an Airy equation (in the leading approximation).l v/2m0(E-V(x))dx 7r = (2n + l)-.1: Quartic oscillator and quantization condition With the W K B matching relations derive for the case of the quartic oscillator with potential V{x) = x 2 + x4 the Bohr-Sommerfeld-Wilson quantization condition given by rb dx 1 fb i / — = .59a) . We summarize what we have achieved. The latter domain is a small region where the solutions from either direction overlap. are only valid in a small interval around the turning point at x = a as indicated in Fig.52). At and around the turning point the solution of the Schrodinger equation is therefore given by Airy functions. These solutions.3 The WKB Method and 295 •.2.2. As an illustration of the use of the WKB solutions we apply these in Example 14.14. 14. n = 0. 14. In the direction to the left as indicated in Fig.(-zf2 by J we obtain the matching relationsX (14.51) and (14. e. In this way different asymptotic branches of one and the same solution are matched across a turning point. and hence must be proportional (in view of the uniqueness of the solution there). 14.. F T' dx (-Z)-1^ by ^ harmonic oscillator V=x 2 +x 4 V=E linear approximation Fig. In this sense we have now verified these relations.g. we enter the domain of validity of a WKB solution.3 The anharmonic potential well. As we approach a limit of this interval in going away from the turning point.

61) Evidently these functions have to continue themselves into each other (since the wave function has to be unique). whereas the approximation of the eigenvalues by comparison with the harmonic oscillator in the domain of the minimum. . regions 1 and 3 in Fig. In the neighbourhood of the minimum at x = 0 we could approximate the potential by the harmonic oscillator. .44) this implies large values of n. . extends only over a length of very few wave-lengths and is therefore valid for n small.l. b are the two turning points.3.296 CHAPTER 14. In order to see this we consider r dx TT\ _ Jx T " 4 J ~ COS / r" dx_ _ Via ^""1 ^~4J" COS Ua I + 4 ~ i ( .62) The condition for this to be exactly satisfied is — as we show now — that the Bohr—Sommerfeld— Wilson quantization condition holds and c' = ( — l ) n c . ^ A = (2n + l ) . I " ' (-K rx dx iz\ _ ( (b dx_ 7T _ fx dx _ • f / fb dx\ f* dx\ ( fb dx\ .60) Using the matching relation (14. In these domains the dominant WKB approximations are 1 /j/i(x) = -cVlexp / fa dx\ I —/ — \ r for i « a. Let E = V(x) at x = a and x = b.^ . as in the case of the harmonic oscillator itself. yz{x) = -c'vlexp ( [x dx\ I — / — ) for x 3> b.b). j/i.y o dx TT T provided 6 _ 4 /. According to Eq.. (14. One can also express this by saying that the linear approximation of the potential around the turning points must extend over a length of several wave-lengths and is therefore valid for large values of n.2. Then around these points the potential would be well approximated by a linear potential (as we saw above). i. Consequently only the lowest eigenvalues would be reasonably well approximated by those of the harmonic oscillator. for which the WKB approximation surprisingly yields already the exact energy eigenvalue as one can verify by evaluating in its case the Bohr-Sommerfeld-Wilson rule. Expressed as an integral over a complete cycle from one turning point back to it the relation is: <b dx^2m0{E-V(x)) = (n+-jh.3. 14. The potential now has to be inserted into this condition and the discrete eigenvalues En of the problem are obtained. (14. We argued earlier that the WKB method is suitable in the case of large values of E. Sometimes this distinction becomes imprecise. (14.(x) = c ' v Acos I / — I for a < x < b. The corresponding eigenfunction would be approximations of the proper eigenfunctions around the origin.59b) Solution: We consider an anharmonic potential well as depicted in Fig. 2 n = 0. Classical Limit and WKB Method where x = a. 2 . 14. /n = s u ^ TJcos^--ya TJ+cos^a T J . (14. The condition for this is ya{x)=yb{x) for xe(a.e. n = 0 . 1 . . the continuations of these functions into the central region 2 are: ya(x) = c v A c o s I / J. (14.. Here we are interested in the discrete states (the only ones here).51). Hence we search for solutions which are exponentially decreasing in the far regions. .

14. n = 0.4 Bohr-Sommerfeld-Wilson Condition 297 14.e. .4 Three cases with different pairs of zeros.63b) *The author learned this in lectures (1956) of R. A. (c) V Fig. . pdq = n + 1 — . before Heisenberg's discovery of the canonical algebra and the formulation of the Schrodinger equation) this condition was always given as §pdq = (n + l)h. 14.1.4 Bohr—Sommerfeld—Wilson Quantization A quantization condition which is very useful in practice and in a wide spectrum of applications is — and remains in spite of its old fashioned reputation — the quantization condition established by N. (14. B. with n — 0.63a) or. . . Dingle. ri2 I Jqi :pdq n + 1 — . . This relation is sometimes wrong.64) (14.x number of turning points TTH.1. The corrected form* is. Sommerfeld and W. 2 . Consider the Schrodinger equation in the abbreviated form d2iP(q) dq2 f(q)t/>{q) = 0.x number of turning points h. Bohr. In this so-called "old quantum theortf (i. equivalently. and knows of no published form or script. (14. Wilson by supplementing classical mechanics by Planck's discretization. 2.

this is an exponential zero.4(a). i. In this > case the wave function for g > go is .e. this means dqyf—f = (n + l)7r.e.4(b). —/ = K J h It follows that / pdq = (n + l)Trh= -(n + l)h.66) J<1\ Thus in this case the so-called old quantum theory is correct.51) the linkage between the trigonometrical and the exponential solutions across the position go in space at which /(g) = 0. -0(g) ex . We have the case of trigonometric solutions.65) We use the knowledge of these solutions to find the eigenvalues of the equation. as illustrated in Fig. Since sine and cosine have zeros spaced at intervals of TT. / • n an integer > 0. i. In quantum mechanics we have dq2 + Ki/j(q) = 0. Case (a): The case of two trigonometrical zeros. i. the wave function is to vanish at points q = q\ and qi with the function /(g) remaining negative in between as illustrated in Fig. Jqi ^ fpdq J =2 pdq = (n + l)h. (14. (14.67) Case (b): Next we consider. (14.e the relation (with a shift of TT/2) 1 2/V4 exp Jqo f9^W)dq 1 1 1 4 (-Z) / sm Jqo (14.e. i. if /(g) is negative. this is a trigonometrical zero. the case of tp(q) — 0 at q = gi. The WKB Method One pair of solutions is (as we know from the earlier sections of this chapter) tpT(q) oc 7T7Iexp 1 sin (_J)l/4 Cos 1 = JdqVJiq) F if f(l) is positive. where n —-. 14. Before we continue we recall from Eq. where /(g) is negative.298 CHAPTER 14. where /(g) is positive. 14. together with q — — oo. There are three cases.

E x a m p l e 14.3. (14.e. Since both cases refer to the same region qo < q < q'0.63b). Solution: As in the case of the quartic potential.2 and 14. and • • for the wave function to be exponentially decreasing to the right of q'0. 14.67): • the wave function for q > qo must be proportional t o (-Z)1/4 sin [ ^W)dq+->x Jqa . Hence we obtain the condition I'o V-f(l)dq 90 (n + 1) IT.-77T (where we reversed the order of integration to obtain a positive integral and multiplied through by minus 1).4 Bohr-Sommerfeld-Wilson This will vanish at q = q\ if rii / V^f{q)dq+j^ Jan It follows t h a t i 4 Condition 299 = {n+l)TT.2: The harmonic oscillator Use the relation (14. Therefore their arguments differ only by (n + l)ir. i.68) (14.70) We can therefore summarize the results in the form of Eq. i. ri\ / J do y/-f(q)dq (n + 1) 1" (14. As illustrations we consider Examples 14. the sines must be proportional. Case (c): Finally we consider the case of i/j(q) = 0 at q = ± o o exponentially as illustrated in Fig.63a) or (14.63b) to obtain the eigenenergies of the quantized harmonic oscillator. ^~f(q)dq Jqo LJqo + -7T J'% On fq / i 1 V~f(<i)dq-j-K (n + l)7r. Hence we have pdq= i n + - .e. (14. the wave function for q < q'0 must be proportional to ri'o (-Z) / 1 4 sin J V W)dq+^ Z i • oc (-7) 1 / 4 sin / y/~f{q)dq . or Pdq (n + l ) " 2 h.14.69) pdq = (n + l ) " 4 h.4(c). this is the case of two turning points. For the wave function to be exponentially decreasing to the left of qo and using Eq. (14. Thus in this case the so-called old quantum theory is wrong.

3: A particle in the gravitational field A particle of mass mo is to be considered at a height q above the flat surface of the Earth.300 With potential V(q) = 2ix2mov2q2. 4 / x CHAPTER 14. Solution: This is the case of one trigonometrical zero of the wave function ip at q = 0 and an exponential zero at q —> oo. The WKB Method this requires evaluation of the integral with qi = ^' E/2-K2mQu2 / 2 m 0 ( B . ^Thus a trigonometrical zero is the condition of absolutely no penetrability beyond it.5. 14. t Hence we have the case of the integral J> pdq (n + 1 ) - Inserting the potential this becomes fqo=E/™0g 2 / dq^2mo(E Jq: Jq=0 Evaluation of the integral yields the energy 1/3 2/3 — mo m) • (n + 1) En = ( — m0g2h2 n+ • (14. whereas a turning point does. 14.63 x 10~ 3 4 J s .5 A particle above a flat Earth. and h = 6.28 x 1 0 . .71) For m 0 = l x 1 0 _ 3 k g . where g is the acceleration due to gravity. Thus there is one turning point at E = mogq. in principle. permit this although with rapidly diminishing probability.807 m s .2 .2 3 J x (n + 3 / 4 ) 2 / 3 . so that E = hu{n + 1/2). one obtains E ~ 2. g = 9. Example 14.V(q))dq The result is E/u. Calculate its quantized energy. :: V=m0gq Fig. as illustrated in Fig.

73) Jx! rp2a dx I^-<E -2 v) = ^ n — = .2.l. In the case of screened Coulomb potentials (cf.( 1 — = — = it — = h\ n H — 2TT OJ 2-K J p dE V 2 In the case of the harmonic oscillator we have hi n E (14. Chapter 16) one obtains the quantization relation (for Mo = const.) M0 N = l + n+1 : + • 2-/B Hence dE K 2 d dE[ M0 2VB.AE(q0 2 = 2n + 1) = — exp 7r dzy/-E -zn + V(z)/h where q=zo are the left and right barrier turning points.7. and u> is the oscillator frequency in either well. (14.4: W K B level splitting formula Derive from the WKB solutions of the Schrodinger equation with symmetric double well potential the WKB level splitting formula A f B £ = .f V ft h JXl h 2TT (14. 9 .= — <*— = h(n+-\. 2TT W 2-K J p dE V where p = ^/2mo{E Solution: We have — V). d dE V)={n+-)n. with p = modx/dt.5 Further Examples 301 14. (14.74) p hjtl *2 —t\ _ mo f 2-rr 2-K J" X2 dx p~ ~ d dE For a period T from x\ to X2 and back to x\ this implies T 1 mo / d a . E x a m p l e 14. 1 .14. ra 2/ = 0. Solution: The proof is contained in Example 18.75) which verifies the formula immediately.5 Further Examples E x a m p l e 14.5: Period of oscillation between two turning points Use the Bohr-Sommerfeld-Wilson quantization rule to obtain for the period T of oscillation of a particle of mass mo between two turning points the relation T 1 m0 f dx d ( — = . h M0 _h 1^3/a ~ 4 0 / 2 J V \ 3 _ 2N3 \M0) ~ ~Ml ' .72) \\2dxp=Qdx^{Eh Hence.

q = y+ -a . Example 14. (11.<?2i<?3 for E > 0 given by E — V = 0. The q -term in this equation can be removed by transforming the equation to a cubic in y.y2)(y .77) Determine (a) the turning points.6: WKB method applied to the cubic potential Consider the cubic potential V(a)=1-b\\a*~q). Fig. where . where 1 2 • 1 2 y = Q. (14.~ \ (14. 14. Solution: The potential has a finite minimum at q = 0 and a finite maximum at q = 2a2/3 as indicated in Fig.-a .e.y3) = 0.6. Eq. The WKB Method ^ 2 JV 3 2TT ~ ~Mjf (14.302 and T CHAPTER 14.76) in agreement with the result obtained from the classical Kepler period for the Coulomb potential (cf.6 The cubic potential. 14. and (b) evaluate the WKB exponential exp[—2/b arr j er ] and the WKB prefactor 2/ w e u.78) (a) We determine first the three turning points at qi = <?i.133)). (14. i. which determine the imaginary part of the energy. The equation E — V = 0 then becomes after a few steps of algebra y3 ~ Py + Q = (y P = ^ a 4 and Q = 2 ^ yi)(y . It follows that the energy between the minimum and this maximum lies in the range 0 < E < £ m a x = ^aSb2.79) .

212.9 3 — = = 2 " 3 " [ . ui=sn 1 \l—- — = l. (b) W e see t h a t gi a n d 93 m e r g e t o 0 w i t h e — 0.a 2 [ l . q3 ~ | .2/3 i n t o q = y + a 2 / 3 .k 2 2tan# \/3 + tan! . ' P . M i l n e . F r i e d m a n [40]. .q)(q . tan60°=%/3. p .e. 78: sin 120° = v ^ A c o s 120° = .93).9 l ) f f / du s n 2 u en2 d n 2 u .g i ) ( g 2 . (14. f o r m u l a s 236.\tan0^ v/3. 92.5 Further Examples 303 T h e r o o t s of t h i s e q u a t i o n a r e k n o w n * a n d for 4 P 3 > 27Q2 a r e r e a l a n d e x p r e s s e d in t e r m s of a n a n g l e 6 given b y (observe t h a t w i t h t h e choice of t h e m i n u s sign. B . a n d e x p a n d t h e r o o t s in p o w e r s of e.3/2. F .{ V 3 e } > 0.c o s ( 0 + 12O°) + c o s ( 0 . D . _ \ / 3 — tan(9 2 V92 . § T o h e l p we cite H. O n e o b t a i n s t h e o r d e r i n g in F i g . h = 1) ^barrier := / d( }J-^2~(E~v) = ^~= dq^/(q . (14.85) T h e angle 9 = 60° c o r r e s p o n d s t o E = 0.08. (l_fc' 2 a + fe'4)l/4. T h e r e f o r e E = 0 c o r r e s p o n d s t o fc2 = 1.93 2a2 [ . 0<6O°.2 c o s ( 0 .' : ± 6 0 ° w h e n E = 0.6. H e r e cos(<9 ± 120°) = . 80 a n d 361.78)) c o s 30 •• 4P3 ( ' \ aeb2} .80) I n s e r t i n g t h e r o o t s yi.9 2 [ . 92 = -a2[l . p . 91 < 93 < 9 < 92(14. V3 + t a n 0 2 .1 2 0 ° ) w i t h 0 < 6 < 60° for 0 < E < £ .2 c o s 0 ] .g i < 1. q3 = -a2[\ . 14. T h u s we c a n n o w w r i t e t h e W K B i n t e g r a l from > o n e b a r r i e r t u r n i n g p o i n t t o t h e o t h e r as (recalling t h e factor y/2/b in front of E in E q . gi ~ y {-V3e} < 0. M . we h a v e qx = .81) m a 91 .1 / 2 .< 1. W i t h t h e o t h e r i n t e g r a l from T a b l e s we o b t a i n (K(k) b e i n g t h e c o m p l e t e *L. g 2 ~y{3}>0.cos 0 + cos(6> + 120°)] = a2 92 .84) 92 .9 3 a n d t h e elliptic m o d u l u s k a n d t h e p a r a m e t e r g a r e given b y k2 k'z = 92 — 93 _ 92 . 93.gi = 1.ui=K(k).82) In o r d e r t o d e t e r m i n e t h e r e l a t i v e p o s i t i o n s of t h e r o o t s qi. 37. (14.§ ( c o s 0 ± \ / 3 s i n 6 0 a n d § sin 6 sin# (14. W e o b t a i n ^ -fbarrier : = 6(92 ~ 9 3 ) 2 ( 9 2 . 6 = 0 i m p l i e s E m a x of E q . e > 0. p .120°)] = a 20? 2a2 3 2a 2 • cos 0 - 91 . D w i g h t [81].2 c o s ( 0 + 1 2 0 ° ) ] . i. p . V cose' + sin 8 (14.T h o m s o n [196]. (14.14.120°)] = .0^-6O°. a n d t h e a n g l e 0 = 0° t o E = £ m a x .= s i n t 3 v 3 (14.04.83) T h e i n t e g r a l m a y n o w b e e v a l u a t e d w i t h t h e h e l p of T a b l e s of I n t e g r a l s .c o s O + cos(6» .79) a n d t h e n s e t t i n g m o = 1. B y r d a n d M . we set 0 = 60° — e.

(14.~ y ^ ) a< [ COS 5 + .87) Inserting G(fc) together with the expressions for the prefactors into (14.k). 1+72 = and J o s 0 + ^ = (1 . The WKB Method elliptic integral of the first kind. E(u) the incomplete elliptic integral of the second kind.fc'2 + fe'4)"1/4. as in (14. p.90) hi JlnynE-v) V(q-qi)(q2 Vz b Jq -q) -I" bJqi where P fe = 93-91 92-91 = . Byrd and M.89) This expression will again be obtained later with the help of configurations called "bounces" (cf.sm t %/3 £\.^ + ^ ) (l + fc'2)2 .91) a2[cos0-^] y/3-tanfl ? L .32)).92) ''P.91 F(TT/2.2)*w+2(fe4+k'2)E{k)] v ' G(k) For one complete round from one turning point back to it. formula 233./ dq~ ^2 b JQl 1 \/(q-qi){q2-q)(q. we obtain in the limit of E = 0: j d ( ^ ( E „ y ) = 2/barrier = ^ ! .?)(?3 . 72.2(fc2 . F. G(fc) =86aS (ITS£ li[fe.= —^ = k a2[cos0+^] v ^ + tanfl V=".S . E(u = K(k)) = E(k). Thus Iwen=l-gF(ip. V 93 .84) we obtain Carrier = 2a 2 sin6> 2 b( . Whereas the integrand of the above barrier integral is effectively a momentum.304 CHAPTER 14. g= 2 . E(u = 0) = 0. Friedman [40]. . 0 (14. (14. Eq.qz) (14.^ V3 / V V3 y a / c o s e J. the integrand of the integral across the well is effectively the inverse of this momentum.00. 91 < q < q3 < 92- bJc. V?2 . 2 2 . A:') = # ( * ' ) • (14. sinV = q3 qi ~ = 1. (24. (14.) . with K(k = 1) = oo. D. V3 cose + ^ = V3 2 1 + 72 v/l + 3 7 4 C (cose) -2 = 1 + 3 7 4 = 4 d . Next we evaluate the required integral across the well from 91 to 53 at energy E.2)K(k) + 2(fc4 + k'2)E(k)) k -^i — .91 . Thus Jwell : = [13 1 f13 rf 1 / dq J^^E-V) 1 y/2 1 V 2 f[13 = — —. E(k = 1) = 1) (•u\ = K{k) -i Q G{k) = / Jo Next we set s dusn2ucn2dn2u = j[k'2(k2 15fc4 . .86) 15 tan.85). We can evaluate this integral again with the use of Tables of Integrals.

To obtain the latter we have to use the quantum mechanical expression approximated by E = hw(n + 1/2).113. .63b).5 Further Examples Hence (using K(0) = rr/2) 305 ab ab (14. we obtain 15 . With q = 1/z the integral (14. Gradshteyn and I.94) C °S3 2 (l-fc2 + fe4)3/2 - l+ g f c ( l + fc) + Comparing this equation with Eq.99) **For the potential approximated around the origin as V ~ a2b2q2/2 the eigenvalues are E = fat)(n + 1/2) with it) = ab. M.85) (14. Thus one obtains from (14. 906. 32 (14. pp. and hence at E = 0. 905. we obtain E -a6b2k'4(l + k'2).! With these expansions one obtains ^barrier = 15 -ahb 2 + 8A .96) Expanding the coefficients of the elliptic integrals in Eq. (14.89) can be expanded to exhibit a simple pole at z = 0 allowing evaluation with Cauchy's residue theorem: 2 dz 2Ez2 a2b2 %2dz Ez2 2nib\ -o?E o?b2 2-KE ab tt I. S. a°b n = 0. so that even in the case of the ground state the zero point energy will contribute to the prefactor. in fact through logarithmic contributions contained in the argument of the exponential for k ^ 1. (14. (14.** However.114.3. we obtain > . With algebra — which it is impossible to reproduce here in detail — one can derive expansions in ascending powers of fc' (which is small) of all relevant quantities. formulas 8. The same expression for E is obtained by applying the "method of poles at infinity" to the Bohr-Sommerfeld-Wilson condition (14.3 and 8.14.l. (14.2.4 2n + 1 fc' = 4 — ^ — .95) Comparing this with the harmonic oscillator approximation E — En = ab{n + i ) . (14. to represent a physical decay rate.88) in rising powers of fc' . 2a 5 b K(k) + 2 E(k).2 IE 4 -fc' ln 8 (" «V iU^fc' 4 . Ryzhik [122]. we cannot expect the ratio exp(—2Jt>arrier)/2J'well oc w exp(—27barrier) evaluated at k = 1.80).97) 4 32 4 32 Here we insert the corresponding expansions of the complete elliptic integrals'^ (note the argument of K and E is fc): K(k) E(k) fc'2 = = < $ ) fc'2 1+ f l n\k'l f H + -1 4 5 3fc' 1T Hfc^J ~2_ 1^ \k'J 12 (14.93) where w is the harmonic oscillator frequency in the well.. .

ba / 25a5b\n+2 ±l_l e-T5 a f >. oa ba Thus with the W K B method we obtain exp[-2/barrier] — barnerj = 8 s. It follows that (for one complete orbit back to the original turning point) . _ exp[-2/barrier] >.> tf = = 2/wfiii = ±tv8a°b—-—P. We observe that with \j2je as 1.e x p [ . (0) i hw En = E\> . so that we obtain 4 .„.11)) dq 2TT y/2-§?(E-v) It would be interesting to derive the same quantity with the perturbation method and to compare the results.( n + l ) e _ JL a * 6 and we obtain for the ground state (ro = 0) . 7 ~ —. n ~ V27T .2 7 b a r r i e r ] Z Z7T where (cf. this has not yet been done. (14. i. The WKB Method For k' —• 0 the first and the third terms dominate.— (14. _ T T • (14.73). /25a5b\™+5 e exp[-2/barrier] ^ ( r ) Correspondingly we obtain for the full period 2/ w e ii: 2/well = ? K ( f c ' ) ( l .103) 27 w e l l 27rVn+y ' fn With Stirling's formula in the form of what we later (with Eq. i. (18. +3 the result becomes exP[-2Jbarrier] = ± i _&g_ (25a5br+J e . 15 a .fc'2 + fc'4)1/4 * ± t ^ . ...Wigner formula (10. V (14.96) for the imaginary part of the energy as in the Breit.178)) call the Furry factor set equal to one.e./^~^l^ab -^aub .j . 2n + 1 — / 26a5b \ . the result agrees with the ground state path integral result (24. ..100) 'barrier = ~0°b In [ ^ — j .101) (14.7 .102) . 8 Bfc .105) As argued in Chapters 24 and 26 one expects this W K B result to agree with the one loop path integral result using bounces..e. also Eq.J + 0(VoJb). (20.306 CHAPTER 14.

C. L. See particularly E. [231]. after the realization that quarks as their constituents might not exist as free particles and. led to a spectrum which contains only scattering states. the Coulomb potential. in fact. Yan [82]. K. Rosner [230]. C. There are numerous. Eichten. led to a classification of quark bound states which is in surprisingly good agreement with a large amount of experimental data particularly in the case of heavy quarks. The study of this potential. Lane and T. along with inclusion of angular momentum and spin effects and their interactions. In the present chapter we consider briefly the three-dimensional potential V(r) oc |r| = r. Quigg and J. indications from the nonabelian generalization of quantized Maxwell theory.* In the present chapter we consider various aspects of this potential and extend this consideration to ' T h e s e investigations became very popular after the discovery of the heavy charmonium bound state ^ and were naturally extended to sufficient complexity to permit comparison with experimental measurements. Quigg and J. Thacker. This potential became widely popular in the spectroscopy of elementary particles. 307 . at least indirect. which permits only bound states. the classification of the various states of nucleons and mesons and other particles. Rosner [269]. Gottfried. K.g. now known as quantum chromodynamics. This would mean that the force binding the quarks together would not decrease with increasing separation as in the case of e.1 Introductory Remarks The particular linear potential we considered in Chapter 13. B. [232]. D. i.-M.e. that it may not even be possible to extract individual quarks experimentally with any finite amount of energy. that this is indeed the case. [233] and H. describing free fall under gravity.Chapter 15 Power Potentials 15. L. T. Kinoshita.

</>)R(r).V 7 = -const.m2. and the normalization [ dr\y(r)\2 = 1.1) i. 15. (15.3) with the boundary conditions''' u(0) = 0. This force maintains this value irrespective of how far apart the particles are separated. For a central potential V(r) we write (15. A > 0. ~ (I + l)rl (15.4>)-ru{r) = Ylm(9. f dr{u(r)}'2 1.£ ] * ( r ) = 0. Power Potentials general power potentials.r acting between two particles is constant and directed towards the origin (of the relative coordinate). These particles therefore cannot be separated by any finite amount of energy. we return to the Schrodinger equation in three dimensions. i. h2 2n V 2 * ( r ) + [V(r) . and obtain 2/x 1(1 + l)h2 u"{r) + E-V(r) u{r) = 0 2 2/ir (15. We leave consideration of the logarithmic potential to some remarks and Example 15.u ~ r i + 1 (/ + l)R(r). mi + m2 and r is the relative coordinate.2 The force The Power Potential F = .1.2) tt(r) = Ylm{9. and so u'(r) . v > 1. we consider immediately the more general power potential V(r) = \r\ 'We have u(r) = rR(r). In order to have a clear starting point. (15.e. and u'(0) -> [u(r)/r]0. R(r) = ^ . m\m<i M= : . (15.308 CHAPTER 15.6) = (I + l)u(r)/r = R ~ rl.5) u'(0) u[r) r=0 = R(0).4) Instead of considering the linear potential.e. where \x is the reduced mass of the two particles of masses mi.

(15. (15. (15. N = 0."-MA. x > 0.15.V(r)) = (2n-l) +1 N.. (14. 3 .1.9) Three-dimensional case: We transcribe the above considerations of the onedimensional case into that of the three-dimensional case by restricting ourselves in the latter case to S waves (no centrifugal potential!).7) k(x) = h/y/2m0(E-V(x)).10) Here rc is the turning point and n — 1. and there identify the S-wave states with those states of the one-dimensional case whose wave functions vanish at the origin — these are the nonsymmetric ones with N odd — as required by Eq.8b) This condition implies that we have only one pair of turning points. odd integer ^=(n-ll)vh.2 The Power Potential 309 In Chapter 14 we obtained the Bohr-Sommerfeld-Wilson quantization condition (14. (15. which has been reduced to an effective one-dimensional problem in the polar coordinate r. and the turning points are at a — — b := — xc.4) in the three-dimensional case. We therefore proceed as follows..9) into the three-dimensional case in the following form: / Jo C dry/2n(E . Then k(x) = k(-x). We now want to find a quantization condition for the three-dimensional problem.. (15.2.63b): 2 and V(0) = 0 (15.63a). d5. . so that with Eq.8a) ^ JO = (2Ar + l ) £ . This means that we transcribe Eq. We also assume that V'(x) > 0.8a) we can write the quantization condition (14.2. One-dimensional case: We consider first briefly a particle of mass TUQ in a symmetric one-dimensional potential with the symmetry V(x) = V(-x) without loss of generality.63b) for the one-dimensional Schrodinger equation. (15. is to be interpreted as the principal quantum number in three dimensions. . In order to to obtain the . . or f l H " ^ ) ! .

Power Potentials eigenvalues E —• En for the radial potential (15. dr = .6) we have to evaluate the > integral / : = [ ° dr^2n{E-\r»)= Here the turning point rc is given by E .y) defined by (x-l)\(y-l)\ y -^ (x + y .12) and (15.13) we obtain _ X_ „ _ so that A£_ (15. c (n-j\irh.14) Prom Eqs. _ —vr" *dr.\rvc = 0 . s.15) I ' = ^ 1 (( 1) B{x. Now. (15.y) T(x)T(y) _ T(x + y) ("-!)/" /•! / d"(1"")/. (15-16) (15. 1W or Xv \E I It follows that (I) / ^""^(l-*)1/2.310 CHAPTER 15.11) rc = (15. (15.17) The integral on the right is the integral representation of the beta function B(x.'(1-*)1/2= J1F-\l-t)y-1dt. E A fv. (15.13) dt . .12) i/z Jo We set dr 1 l A v ~ E \ (Et\l/v so that (15.

$T(z) = {z.55). One can convince oneself now.18) \v \E or 3 E2- r(§)r(i)V2M (n ' (15.20) These are precisely the eigenvalues which one obtains from the vanishing of the (periodic) eigenfunction at the wall of the square well (of course n — 1/4 has t o b e replaced by n since the W K B approximation is only accidentally correct for small values of n ) .19) or En = _ 1)^(3 + 1)^/^2^+2) r(|)ir(i)^7^ • v 0 1 Fig.1)!. ^ '"Compare with Eq. T(z + 1) = z\. Weinstein [281].1 Approach to square well with v —> oo. 15. t h a t in the limit v —• oo the potential approaches the shape of an infinitely high square well. and n — — ]7r/i. The case v = 4 is that of the pure anharmonic oscillator also discussed by M.1.2 The Power Potential 311 so t h a t by comparison y = 3/2 and x = 1 + (1 — u)/i/. (12.* In this limit the eigenvalues become:^ E„ (n-h^ni) Lr(f)r(i)^7^ rv"ir •n2 2/i 2^2 (15. 15. (15. for instance simply graphically as indicated in Fig. i r ( i ) = r ( l + i ) = ( i ) ! . .15.

the case of the linear potential. = (n->r(§)A 2/3 3TT \h n 2/3 L r(§)v^7^ J (15. For the following we set in Eq. (15.2) implies A s fj./-/L This result agrees with that for the one-dimensional harmonic oscillator r2 in the form" cf>"(r) + ^(E-\r2)cf>(r) =0 (15.23) The eigenvalues for the linear potential can also be obtained from the zeros of the Airy function. Power Potentials In the case of v = 2.2. The potential V(x) = X\x\ (15.oj2/2. > and with Eq.1 + \)hx> = {An • l)£fiw. (13.oj = ^/2X/JX. Physically it does not make sense for the probability amplitude to have a discontinuity there.41) for the .25) "Comparison with Eq. But for our present purposes this applies only in as far as the solutions (j)(x) which are exponentially decreasing at infinity vanish at x = 0.61) h2 = I. It is therefore necessary to demand its continuity there in the sense of equality of the first derivatives from either direction (apart from the equality of the values of the functions there from either direction). In the one-dimensional consideration these are precisely the odd wave functions as illustrated by an example in Fig. we obtain En = (n . In the case v = 1.24) is discontinuous at x = 0 (this means the derivative there jumps from positive to negative). we obtain E„. 15.mo = \. eigenvalues (JV + h)tkj — (2n .> r ( 2 ) A V 2 in \)*\W \it^T& (15. x > 0.21) (4n-l). the harmonic oscillator. since these are the ones we are interested in (cf.22) 0 which selects from the usual and with the boundary condition 0(0) eigenfunctions the odd ones.g — —2. discussion at the beginning of this section). (6. (6. so that the appropriate Schrodinger equation becomes d2(j) + (Edx2 x)<t>(x) = 0 .312 CHAPTER 15.

15. we must have as a result of the boundary condition <p(x) = 0: 4>(x)\x=0 oc Ai(E . Eq. i.2 Behaviour of an odd wave function at the origin.27) i. i.57a) one solution of this equation is the Airy function <l>(z) on Ai(z) = Ai(E .26) For z = s := E — x —> —oo.60) applies. i.59) applies. Ai(s) 1 2v^(-s)V4 exp x —> +oo.29) .28) In particular at x — 0. dz2 According to Eqs. (15. \(s?» (15.e. i. (14.x)\x=0 = 0. E > V or s > 0.e.e.e. Ai(s) "v^ G ' cos s 3/2 _ t (15. W i t h z :— E — x this is d24>{z) + z<f>{z) = 0. In the domain —a < x < + a .x). s = E. (13.2 The Power Potential 313 V(x) V(x)=|x| turning points exponential fall-ofl Fig. this wave function has the required exponentially decreasing behaviour at infinity.56) and (14.e. (15. (13. and for s —> oo the trigonometric behaviour of Eq. 15.

3.02137 10.1 demonstrates the quality of the WKB approximation by comparison with the exact values.30) ! ^ / 2 .29) implies that cos ( | ^ .51716 6. Rosner.e.08181 5.78671 7.e.7 T Tl * 2 V 4 Actually this expression is only valid for E and hence n large. 201. [231]. so that n — 1/4 ~ n. n = l..^ = !(2n-l).33811 4.32025 4. Quantum Mechanics with Applications to Quarkonium. *This Table is reprinted from C. p.. It is interesting to note that in the dominant approximation both methods agree.08795 5.e. (15.. the eigenvalues En are determined by the zeros of the Airy function. .04017 11.00852 11. i.82878 1 2 3 4 5 6 7 8 9 10 2. of course. but for large > values of E Eq. Table 2.28) is really only valid for s large and s — oo.52056 6.2. (15. Table 15.94249 9.314 CHAPTER 15.31) (an odd function has an odd number of zeros!). It is possible.93602 12..00767 11.03914 11. L. with permission from Elsevier.En from Ai(£„) = 0 2.94413 9. Quigg and J. i.93528 12. copyright of North-Holland Publ. The expression (15.= ) = «. Power Potentials Table 15. (15. = im — (7r/4).1: S-Wave Energy Eigenvalues for V(r) — r (with h = 2/x = 1) from f 3 7 r ^n 2 ( i>j2/3 . Co.02265 10. (2/3)E3/2 or 3 / 1 2/3 E — En — . to obtain the zeros of the Airy function numerically.78445 7.82814 i.

have only a finite lifetime and decay into other particles such as electron-positron pairs. can be shown to be proportional to the modulus-squared of the particle wave function at the origin.g. The decay widths T. In general these quark-antiquark pairs.15. Jackson [140]. the charmanticharm meson ^ . van Royen and V. See also J. r <53) 1 3 - dr cos 2 (---)- (15.50)) that in the domain V < E the leading WKB approximation of the wave function u{r) is given by the periodic function *>-"^-(JC'£)-I)' ^vm=rr (1532) where iV is the normalization constant. We saw previously (see for instance Eq.e.e. We indicate briefly here — without entering into further details — how these quantities can be calculated. of course. . e.34a) **The most frequently quoted reference for this result is R.3 The Three-Dimensional Wave Function 315 15. D.• •) by a mean value.* We can determine the constant N in the leading approximation by demanding that = 2 / Jo i. P. cos2(---)^— / 1 fr° 2 > w . ee. *Note that this makes N a W K B normalization constant. in Chapters 24 and 26.3 The Three-Dimensional Wave Function The linear potential has in particular been used in the investigation of the spectrum of heavy quark-antiquark pairs.e. dr[u{r)f " f** «" (i £>-i)4 Here we replace the oscillatory part cos (. i. The resulting masses agree in most cases very well with those extracted from experimental observations. i. e. We shall encounter WKB normalization constants at numerous points in later chapters. which are inversely proportional to the lifetimes. had to be supplemented by inclusion of relativistic corrections as well as other contributions arising for instance from spin and angular momentum interactions. F. (14.g. Weisskopf [239]. In order to be comparable with experimental data these investigations.** r ( # ^ e e ) o c |*(0)| 2 .

2) and (15. & (15..38) N : sin V ^ Jr dr' 7T (15.40) VWY . Power Potentials this averaging implies a numerical factor like 1/2..1 )n .39) in / .l N n . (15. A = (15.36) \i dEn irh2 dn According to Eqs.35) N2 .\ \ * n = l. i. (15.37) (15.316 In accordance with t h e relation CHAPTER 15. so t h a t approximately -l N2 / Jo k(r)dr .2.. (15. (15.34b) VME-vy Variation or differentiation of Eq. From Eq.35) u T (•T=2TT/OJ dt cos 2 cot 2' (15.Jo dr' n M?) ~ 4 in sin sin JTCl-drW2v{E-V)-j l \ 7T N 7W)sin = ( .32) we obtain to leading order u'{r) and therefore «'(0) = N N (1JU0) N (15.rcu w X{r)dr = -Kh .e. dE a—— / dn J0 so t h a t with Eq.4) we have for I = m = 0: |^(O)| 2 = |n'(O)| 2 |y O o(^0)| 2 = ^ K ( O ) | 2 .10) with respect to n implies (which is permissible for small separations of neighbouring levels) 2/i ldEn rrc dr ^/2ii{En V) 2 dn J0 = irk. (15.

g. The case of arbitrary positive power potentials as above has been considered by R. J. Kaushal and H. Hence Eq. Bose. . we have ^'(0) = ( . An enormous literature exists on the subject in view of its relevance to the spectroscopy of mesons and baryons made up of quark constituents and the necessity of checks with results derived from experiments.15. Itt^pocn2*"-1^"-^.23) for the energy levels of the linear potential. S. J. S. 'For one such investigation which includes the Coulomb potential in addition to a confining potential see e.* Using the result (15. the expression (15. G. for various potentials: (a) For a confinement potential of power v. W.43) or the results of corresponding calculations one finds the following dependence of (^(O)!2 on the quantum numbers n = 1. Bose and H. W. (IM2) l*(o)r = M dEn dn irh 2 ~^WEn ~*T (15 43) ' Inserting here. Hite and H. W.l ) n . (b) for the logarithmic potential^ l*(0)| 2 ex ^ v (15. K.44) (15. § S.1 ^ ( 2 ^ „ ) 1 / 4 .41) I*(0)|2=_L^)^.37) we obtain: 1 (2M^)1/2 47r ft. J. for instance. . 3 . E.38) implies 317 (15. and hence with Eq.3 The Three-Dimensional Wave Function Since we are considering potentials with V(0) = 0. MiillerKirsten [147]. Miiller-Kirsten [29]. (15. K. Miiller-Kirsten [28].2. we can investigate the dependence of the decay rates on n (note that these would be decays into particles different from those bound by the linear potential — a consideration we cannot enter into more detail here).45) 'See the references of Quigg and Rosner cited at the beginning of this chapter. . Thus only theoretical investigations performed immediately after the discovery of the particles \I/ and T are basically of an analytical nature* and therefore have not been performed largely with numerical methods and the fitting of as few parameters as possible. . (15.

q = 2n + l. ip = exp[(z — c)/2]. 7 With the substitutions r = exp(z — c). Example 15.2.L 2 + U{z)]<f>. and there Eq. (15. i. K.e. W. {S = ^ > 0.^ +•••.133). Landau and E. = 1(1 + 1).46) We can now link the value of the wave function at the origin to the oscillation period T using the arguments of Example 14.318 (c) and for the Coulomb potential^ CHAPTER 15. Eq = i q .^ ( 3 ^ + l)-55p^«(3^-l) + . in agreement with Eq.. (15. We observed there in the case of the Coulomb potential that the period is proportional to n 3 . D.1: Regge trajectories of the logarithmic potential Consider the radial Schrodinger equation for the potential V(r) = g\n(r/ro). Miiller-Kirsten [29]. J.1. . (11. Bose and H. U(z) = | dz -zf3e2a/?e2*.75). Lifshitz [157]. E' = E + gln(r0).j . M. Solution: Details can be found in the literature. rewrite the equation as ^ z + [ ... (14. . as well as Expanding U(z) about its extremum at ZQ = —1/2 and using the perturbation method of Dingle and Muller derive the expansions N)2 = ^-^ 3 .46) and the classical Kepler period of Eq.g ^£+(<*-Pl»r-iy = 0. = (I + 1/2) 2 . Power Potentials |*(0)| 2 oc .— oo < z < oo.5. n = 0. ." 1 11 L .L2 h4 = 4/3exp[(2a — /3)/f3]. *=^f. S.I l n f ^ p ..c = —a//3.

Chapter 16 Screened Coulomb Potentials 16. It is clear that for ro — oo the > potential becomes of Coulomb type. in microscopic and hence quantum physics with the possibility of real and virtual creation of particles and hence a vacuum state. A very similar potential is the Yukawa potential V{r) = -g2 — . r in which /i has the dimension of an inverse length or equivalently that of a mass in natural units.Phenomenologically the screened attractive Coulomb potential may be written V(r) = -g2p-r/ro . so that like charges are repelled and unlike charges are attracted by each other. This implies.1 Introductory Remarks We observed previously that the infinite range of the Coulomb potential — the concept of range being defined more precisely below — leads to a scattering phase with a logarithm-r contribution. The result is a screening of the Coulomb potential which thus attributes it a finite range ro. This effect sounds unphysical. it will always notice the latter's presence. it is. Historically this potential arose with the realization 319 p-Vr . and in fact. In reality. that no matter how far the particle is scattered away from the source of the Coulomb potential. r where r is the distance from the source charge and the charges and other constants are collected in the coupling g2. The exponential insures a rapid fall-off of the potential at large distances and for this reason the parameter ro can be looked at as a measure of the range of the potential. a charge as source of a Coulomb force leads to an effective polarization of the vacuum like that of a dielectric.

and that.1) and is written in spacetime-dimensional Minkowskian notation l/{jpvpu + /i 2 ). A Green's function is effectively the inverse of a quantity called propagator which is intimately related to the expression in Eq. in fact the parameter [i represents the mass of such a spinless meson.320 CHAPTER 16. and in view of their relation to the exchange of virtual elementary particles have therefore been objects of intense study in elementary particle theory. Thus in the following we consider a generalized Yukawa potential which can be expanded as a power series in r and is written oo V{r)= }Ml+1(-r)\ (16. . Yukawa potentials and superpositions of such potentials play an important role in nuclear physics.3) where for the real potentials we consider here. the mathematical expression with an exponential is not so easy to handle analytically. who later — as is wellknown — deviated from this idea and invested his efforts into the study of nonlinear spinor field theory.e. (16. Heisenberg [133].1) We encountered Green's functions earlier.2 ) or (p = hk) 1 /" . one frequently resorts in calculations to the expansion of the potential in rising powers of r. however. from the relation -Pi + P 2 + M2c2 = 0. Regge trajectories — which we encounter *The S-matrix theory of strong interactions was actually initiated by W. all coefficients Mj are real and independent of the energy E = k2.ii k .* The relativistic equation of motion of such a meson when free is the Klein-Gordon equation which results from the quantization of the classical relativistic energy momentum relation of a particle of mass /x. In the literature reference is sometimes made to the so-called static limit of a relativistic propagator. i. Since. Screened Coulomb Potentials that the strong nuclear force is mediated by the exchange of mesons. The realization that mesons and baryons are made up of quarks does not really change that picture at lower energies. (16. What is meant is that the four-dimensional Minkowskian Fourier transform of the propagator is given by the relation dk (2TT) J 4 e x ^r—2 = 7Z^u o)^n k + fi2 4TT v y |x 2 ikx ^ e-^W ( 16 .x j 36 x 4 (2TT) k + p? 4 2 (2TT) J d x- e-^xl 4TT|X| ' We observe that the Fourier transform of the propagator is effectively the Yukawa potential.

mo being the reduced mass of the system. Froissart [223]. . 2 . and hence that the confluent hypergeometric series there obtained has to break off after a finite number of terms in order not to destroy this behaviour. Squires [258].r)=0. (16. Basically this argument amounts to an argument similar to that used in the case of the Coulomb potential where the integer n arose from the requirement that the wave function be normalizable. physical) values of angular momentum as functions of energy E. S S. . . however. Regge trajectories were realized to play an important role in the high energy behaviour of hadronic scattering amplitudes.§ Our intention here is to consider these potentials as a generalization of the Coulomb potential and hence to proceed along similar lines in the derivation of Regge trajectories.Q. only for the cases of n = 0. Frautschi [97] and E. Cheng and T. bound state eigenenergies and the 5-matrix.^ Regge trajectories arise as poles of the Smatrix in the plane of complex angular momentum. to be of the order of 1/k.1 Introductory Remarks 321 in this context — are functions which interpolate integral (i.3) have been considered by various authors. for the Regge trajectories or Regge or /-plane poles of the 5-matrixll / = ln(K) 'Standard references are the monographs of R. "These were first considered by C. C. £+*-*P-VV t/.e. l+n+l = ^P-.{l. S. Masson [180].r).16. Lovelace and D. (16.5) where K — ik and A n is an expansion in descending powers of K. Kinoshita [21].3) in the radial Schrodinger equation for the partial wave ip(l.4) where E = k2 and h = c = 1 = 2m. i.e.k. where n is an integer and we referred to these already in Chapter 11 in the simple case of the Coulomb potential. Bethe and T. f H. (16. i. ''The approximate behaviour of Regge trajectories for the Yukawa potential has also been calculated by H. Potentials expandable as in Eq. Thus in the present case it turns out that it is easiest to calculate first the expansion for the expression l + n+1. Wu [47]. Mandelstam [185]. J. Omnes and M. |if|->oo. i.1. . They are usually written I = an(E). A.1 in the above. As discussed in detail by Bethe and Kinoshita^ one can start by arguing that a countably infinite number of Regge poles may be defined in the region of large negative energies E = k2 of the radial Schrodinger equation by requiring I + n + 1 for n = 0.k. T. (16.e. Naturally it is easiest to familiarize oneself with these by studying solvable potential models.e. See also the other references below. $ In the following we consider screened Coulomb or Yukawa potentials of the type of Eq.

we naturally assume conditions on the potential V(r) which are such that the ^-matrix is meromorphic in the entire plane of complex angular mommentum. Miiller and K.k.k. (16. 16. Under these conditions the S'-matrix is meromorphic (i. Bottino. in the present case at fc2 = 0 or k2 — Mi = 0. Longoni and T. J.6) 2K{M°~ An(K))x+ 2K^(^K) MiX ' (16J) **H. J. (16. has only simple poles) in the plane of complex angular momentum and in the complex &-plane (E = k2) cut along the imaginary axis.)/j.e.2 Regge Trajectories Since our treatment here aims at obtaining an S'-matrix as a generalization of that of the Coulomb potential. Section 11. W. ttThus it will be seen that with perturbation expansions the problem of the screened Coulomb potential can be solved practically as completely as the Coulomb problem. < oo for all |0| < vr/2.** The energy is later obtained by reversion of the resulting series.4) to z = —IKv and set TP(l. J. Regge [31] and E. for all n. W. H. *See in particular A. o rV(r) I dpp\V{peie)\ regular at r = 0. . < const. In the following we follow mainly the last two of these references. W.z) Then x is a solution of the equation VaX= (16.19). Muller [201] and [202]. starting with the power r _ 1 of the Coulomb potential. 'In the case of the Coulomb potential (cf.15) and (16.9) the cut starts at E = k2 = 0. J. Schilcher [203]. Screened Coulomb Potentials as a function of the energy.z) = e-z2/2zl+1X(l. A. M.ndp. as may be seen from Eqs.322 CHAPTER 16. Squires [258].t Considering such superpositions of Yukawa potentials with /•oo / a(/j. Proceeding now as in the case of the Coulomb potential we change the variable of Eq. Muller [204]. one can assume an expansion of the potential V(r) in ascending powers of r. The conditions for this have been investigated in the literature* and may be summarized as follows: /•oo V(r) = / d/xa(//)e-^7r. tf H .

a) —b — 2a. (16.z) = Ha). (a. which means that the hypergeometric series breaks off after a finite number of terms. (16.a+ l)$(a + 1) + (a.awhere (a.j) = (a + j -l.5) is equivalent to a = —n.5)) a = + 323 . . Miiller-Kirsten [249].2 Regge Trajectories where d2 Va = z-^ and (cf. By a repeated application of the recurrence relation (16. Sharma and H.j +(a + j + l.z) is known to satisfy a recurrence relation which we write here for convenience in the form z$(a) = {a. (16. a + 1) = a — b+ 1.b. where $(o.11b) ^ The associated boundary conditions are: So(a. as in the case of the Coulomb problem.10) zm$(a)= ^ j=-m Sm(a.j -l) + (a + j. anharmonic and cosine potentials have been derived in L. b\ z) is seen to be a confluent hypergeometric function which for reasons of convenience we abreviate in the following as 3>(a).1).7) is seen to be of order 1/K.L ^d {b-z)--a l+ l + ^ ^ l 2K = - n &ndb = 2l + 2 = -2n-^4^-- K (16-8) The right hand side of Eq.a + j)Sm-i(a.a + + l).0) = 1.j)$(a + j). (a. j)Sm-i(a.a)${a) + {a. {0) X = Ha.a + j)Sm-i(a. so that to leading order we have VaX{0) = 0.* Recurrence relations for coefficients of perturbation expansions for Yukawa.j) for \j\ > m are zero. W.10): Sm(a. all other So(a.9) (16. and all Sm(a. (16.11a) The coefficients Sm(a. We also observe that the ansatz (16.b. J.9) we obtain m l)$(a .j) may be computed from a recurrence relation which follows from the coefficients (16.16. K.i 0) = 0. The function <&(a.j) (16. Eq.a — l) — a — l.

aj 2 (2K)A _ [a. Any term /j.12a) where [a. the latter can be written -. 1 . Evaluating the first few terms of the expansion (16. (16. This equation is seen to be 0 = 7n7L a ' a Jl + P a + n $ ( a + n) = 0.aj 2 -\ [a. [ a . Mn The coefficient of the sum of all the remaining terms in 3? (a) is then set equal to zero and determines to that order of approximation the quantity An(K). [a.n. Screened Coulomb Potentials Substituting the first approximation x^0-* = 3>(a) into the right hand side of Eq. #(a + ra) fjL$(a + n) V.324 CHAPTER 16.a + l ] 2 [ .7). and hence Va$(a + n) = ra$(a + n).1] 2 . oo _ . + (16. (16.' — .5) the Regge trajectories I = ln(K).<&(a + n) on the right hand side of Eq.a\4.l. of course.13) + ••• • One can now construct coefficients with their recurrence relations for the individual terms of this expansion. a .j). . so that 1 1 2K— ' (2W[a'a]2+(2^F[a'a]3 1 r i . i.12b) The usefulness of this notation can now be seen in the ease with which it permits the calculation of any number of higher-order perturbation terms.o]i = M 0 . 0 < \j\ < i.e. Va+n = Va . [a.a + j]i+l = MiSi(a. for the coefficients M^> of the expansion Details can be found in the references cited above.[a>a + J]i+i*(a + 3).7) and hence in Ra may be cancelled out by adding to x a contribution /x<I>(a + n)/n except. (16. (16.a]l^a) + ^2(2K)i+1 Y. i R »] = 2K[a.An{K). We first cite the final result .[ a + l.13) one obtains the quantity An(K) and hence with Eq. (16. This follows from the fact that D a $(a) = 0. when n = 0.

Ahmadzadeh. is an asymptotic expansion for large values of the energy — is not very useful in the very interesting domain around energy zero. i.1) +6M 2 Min(n + 1) + 2M2M02 + SM^Mo] + g^s [3M 4 M 0 (n 2 + n .2 Regge Trajectories 325 and then demonstrate the calculation in Example 16. it is clear that one expects the trajectories to be finitely closed curves with asymptotes given by those of the Coulomb potential.^ The plots confirm the expected behaviour for strongly attractive potentials but exhibit also a superficially unexpected departure into the lower half of the complex i-plane in the case of the first few trajectories (counting in terms of the quantum number n) for weak coupling. G.l)n(n + l)(n + 2) + 2M 3 M 0 (3n 2 + 3n . With numerical methods one can achieve more. Tate [6]. Such plots of numerically computed Regge trajectories for specific values of the overall coupling constant and energies varying from minus infinity to plus infinity have been given by various authors.16. The result is AnW = M0-'[n(n + l)M2 + MlM0] <2» + ^ ^ 1 + — ^ [ 3 M 4 ( n .^ p [ 3 M 4 ( n .1) +6M 2 Mm(n + 1) + 2M2M02 + SM^Mo] . Burke and C.1) + 3M3M02 + M22n(n + 1) (16.l)n(n + l)(n + 2) + 2M 3 M 0 (3n z + 3n .5). Inserting this expansion into Eq. Obviously the expansion — which.1.14) +4M 2 MiM 0 ] + 0 ( K ^ 7 ) .2.§ This is an important observation which indicates the equivalence of the methods. P.1) + 3M3M02 + Mln{n + 1) +4M 2 MiM 0 ] + 0(K~8). we obtain the expansion for the Regge poles. including an exploration of the domain of small energies E. (16. Thus the expected appearance is that shown in Fig. of course. 16.( 2 1 ^ 6 1 ) [ 3 M 4 M 0 ( n 2 + n .e. Analytical Observe there the quadratic form of the centrifugal potential! "See in particular A.15) The same expansion may be derived by the WKB method from the BohrSommerfeld-Wilson integral for three space dimensions as explained in Example 16.1 by evaluating the first two terms. . (16. . However.

—" + In = — .O) = Other terms vanish since So (a. (16. a] 2 = M i Si (a. From Eq.0) = (a. so the imaginary part ai of I — an represents its width in angular momentum.13). and the lifetime At of the resonance satisfies the relation TAt ~ h. For a resonance with a long lifetime.1: Evaluation of perturbation terms Use the above formulae to verify the first two terms in the expansion of A n for the Yukawa Regge trajectories.0). (16.^ . and the angle A#. one can expect a resonance with the lifetime determined by this imaginary part. [a.326 CHAPTER 16. C. ai is small and A6 is large.1). The conjugate variable to energy is time. Example 16. Similarly the conjugate variable to angular momentum is angle. through which the particle orbits during the course of the resonance. satisfies the relation ajAO ~ h. Solution: We evaluate the first three terms of expansion (16." Re I 1 0 1 2 Fig. just as a decay width T represents the width of the resonance in energy. For a bound state a j = 0 and the orbit becomes permanent. a] 3 = M 2 S 2 (a.12b) we obtain [a. .a) = b . p. Hence Si (o.11b) we obtain Si(a.a)So(a. . Frautschi [97].2 a = -In See S. where the imaginary part of I is very small. this is a particularly interesting domain since at the position soon after this point at integral I (as at I — 2 in Fig. 16. Screened Coulomb Potentials expressions of the behaviour of a Regge trajectory in the immediate neighbourhood of E = 0 are practically unknown. 0) and Prom Eq. 16. (a. ±1) = 0. 0). However.1 Typical Regge trajectory for a strongly attractive Yukawa potential. In fact. 114.

Thus (with terms which are 0): Si(a.4) with mo = 1/2 and around the classical orbit the integral in the following relation I h = <b ^J pdr: dr •K2 Mo {l + \)2h2 1/2 ^classical orbit The integral is most easily evaluated by the method of "poles at infinity". a)Si (a. we have to evaluate (cf.6 + 1 ) .n ( n + l Inserting these expressions into expansion (16.A „ ) H -Mi 2An 2K if -n + (2i^): r M 2 2K Hence 1 .2 a ) .65) and observing that in z there is one pole at the origin and one in approaching infinity in view of the expansions VA + 2Bz + Cz2 _ VA y2 -.0) + 0 + 0 = ( 6 . Solution: Ignoring higher order terms in r.2 Regge Trajectories 327 Analogously we evaluate S2(a.0)+ 0 + 0 = (a-1).-1) Si(a.1 ) + (a.(ra + 1) [n + (2K)3 A n+1+• n K K (M0 .16. One sets z = 1/r so that. 14. 0) + a S i ( a . using the Cauchy formula (6. -(n+l)U+^p-J .6)Si(a.6 + l)a + A.a)5o(a. a)Si (a.An) • [ M i M 0 + n(7i + l)M 2 ] 1 A n = M 0 .1) + (a . 1) (a .0) (a . 1) Hence 52(a. We obtain the quantities S i ( a .b)(a .0) Si(a. 2 : E i g e n v a l u e a p p r o x i m a t i o n b y " p o l e s at infinity" Verify the dominant behaviour of the Regge trajectories of Yukawa potentials by contour integration of the Bohr-Sommerfeld-Wilson integral in which 1(1 + 1) is replaced by (I + 1/2) 2 .0) = (a . 0) + (a + 1. .a-l)S0(a.11b).^ 2 M " + ! ) M 2 + M i Mo] + • E x a m p l e 1 6 . A" An An A" = = = (a.2o)Si(a.1 ) + (6 .2 V~Az and z \ zz dr\ A 2B C dz V'A + 2Bz + Cz2 (6.a + l)S0(a. Sec. i ) again from Eq.65) -2m dz V'A + 2Bz + Cz2 • Z— 0.13) we obtain 0 (Mo .00 B + Cz 2ni VA + 2Bz + Cz2 B 2niI —= + vC with ambiguous signs of square roots I.1.0) + 0 + 0 = ( a . (a. a)Si (a. 1). (16. . (a. .

1)1(1 + 1)(Z + 2)M 4 + 2M3M0(3Z2 + 3/ .3: Calculation of Regge trajectories by the WKB method Use the WKB method to verify the first two terms in the expansion of A „ for the Yukawa Regge trajectories. t?{l + \f M0 ±2K 1/2 -2TT 2J 2V^2 (n + l + l)h = 2TT Mo h.16) = n n=o 1 2 With this inversion we obtain = M0-^^[l(l + ^ [ 3 ( Z . (16. Example 16. Thus we can write down the S'-matrix for the present case by exploiting the limiting case of the Coulomb potential.15). n = 0.(iO mp~ ~ ' ' ' '---+ l)M2 + M0Ml] (16.1) +61(1 + l)M 2 Mi + 3M2M02 + 3Mx2Mo] + 0(K~6). Boukema [34] and [35].e. In the second paper the second order WKB approximation is used and shown to yield complete agreement with the terms given in Eq.. evaluate — with 1(1 + 1) replaced by (I + | ) 2 and V(r) = —Mo/r + v(r) — the following equation dr K2 Mo (l+\? 1/2 .l.14) for An(K) in order to re-express the latter in terms of I so that i+i+ A. ±2HK in agreement with our expressions above. Singh [252]. we can use Eq.t First. This shows that 1(1 + 1) has to appear in the BohrSommerfeld—Wilson integral as (/ + 1/2) 2 . I. Screened Coulomb Potentials Mo . tSee Chapter 11 and V. Solution: For details of the solution we refer to papers of Boukema. i. however. .3 The 5-Matrix It is clear that if we now work through the usual procedure for the derivation of the S'-matrix we pick up a logarithmic phase as in the case of the Coulomb potential. (16. We know that corresponding to every Coulomb Regge trajectory we have a corresponding one in the Yukawa case. The turning points in this case are at r = 0 and r = oo.2.5) together with the expansion (16.* 16.328 Thus here dr K' CHAPTER 16. *J. Here we do not perform this procedure. by expanding the right hand side in rising powers of v and evaluating the individual integrals.

J.19) we obtain the energy. Now reversing the expansion (16.l)n(n + l)(n + 2) + 2M2M02 . such as. we can now write down the S'-matrix in terms of the scattering phase 5i as the expression TV/ I 1 I ^l(J<l) S ( W = e -r(/ + i-^#>)e ' (16 18) ' We observe that the poles of the S'-matrix are given by l + l + ^p- = -n. J. Miiller [211]. the behaviour of the scattering phase in the domain of high energies. W.g.. One can use the explicit expression (16.19) +2M 3 M 0 (3n 2 + 3 n . which are precisely the expressions yielding the Regge trajectories of above. Paralleling the case of the Coulomb potential in Chapter 11. Expanding the square root V-fC2 + Mi for \Mi/K2\ < 1 one regains ln(K) with the expansion (16. for instance. We observe also that the S'-matrix is unitary as a consequence of the result (16.2.4 The Energy Expansion It may have been noticed that in the above considerations we could have combined the constant Mi with the energy into a combination K2 + M\. the following *See e. § See H.l (two additional terms are given in the literature). Miiller [204]..17) and is therefore real for real values of the potential coefficients Mj.16.4 The Energy Expansion We observe that Ai(K) has the property MK) = M-K) 329 (16. This has been done§ and one obtains . M 0 2y/K2 + Mi n(n + l)M 2 4 ( # 2 + Mi) 3 /2 (2ra + 1)M 0 M 2 8(K 2 + Mi) 2 + 16(if2 + Mi) 5 / 2 3M 4 (n . W.17).* 16. . H.0[(KZ + Mx)'6} (16.e.. n = 0.14). i.18) of the S-matrix now to explore further aspects. and could then have carried out the perturbation procedure not in inverse powers of K but in inverse powers of y/K2 + M\. .l.

Screened Coulomb Potentials expansion.5 The Sommerfeld—Watson Transform The basic theoretical tool for the exploration of Regge poles is a representation of the scattering amplitude given by the so-called Sommerfeld-Watson transform.330 CHAPTER 16.l)n(n + l)(n + 2)(n + 3) +2M 5 M 0 3 {5n(n + l)(3n 2 + 3n . in which the leading term is the usual expression of the Balmer formula for the Coulomb potential: K2 = -Mi+ 1 ° l-4n(n + l ) ^ ( / + n+l)2 4(7 + n + l ) 2 Mio +4(2n + l ) ^ ( / + n + l ) 3 4(Z + n + 1)4 ( l) ^ — . W.l)n 2 (n + l) 2 (n + 2) 1 + • • • (16. Mendelsohn [135] and E. N. Analogous expansions for specific energy-dependent Yukawa potentials have been investigated by A.^ 16.2)(n .* We recall from Chapter 11 the definition of the scattering amplitude F(9. B.20) Extensive investigations of the energy eigenvalues for the Yukawa potential can be found in the literature. 'According to S. in the following expression with an "See in particular G. C.11) + 2M22M03(9n2 + 9n . Warburton [279]. the transformation was introduced in the form given below by G. Watson [280] in 1918 and later resurrected by A.10) + 12} + 4M3M05 +2M 4 M 0 4 (6n 2 + 6n . E. Zauderer [286]. J. J. k) as coefficient of the outgoing spherical wave in the asymptotic behaviour of the wave function tp(r).1) + 2M2M. p.3 M 2 V ( n + l) 2 M(o I +2M 3 M 0 2 (3n 2 + 3n .e. Miiiler-Kirsten and N. p. A. Iafrate and L.f 24(2n + l)(Z + n + l ) 5 M4M0(nz M06 -M$n(n ^ M9 + n . Frautschi [97]. 282. Large coupling expansions derived in H. 107. i. .l)n(n + l)(n + 2) .1) -10M 3 M 2 M 0 2 n(n + l)(3n 2 + 3n + 2) + 20M23n3(n + l ) 3 -30M 4 M 2 M 0 (n .< 3 M 4 M 0 ( n . Vahedi-Faridi [212] contain as special case the expansions of the first pair of authors.1) + Af3Mff + 1) ^ |lOM 6 M 0 2 (n . Sommerfeld [256].

187)) (16. \x\ <C l ..1.187).k) = 2zfc L " v "'' v .iJ 11 —[S(l. dn Iml 1 I c ( \ x o x 1 A 2 A 3 x4 X 5 Rel Fig. so that sin -KI = sin7r(ra + x) ~ (—l)n-KX = (—l)lir(l — n). c = 1. Actually 5j l-rr/2 as in Eqs. (cos 9).22) = \F(9.k)~l] = ~ k" -PiSiW sin 5i(k) and S(l. Eq. ~:' sin nl (16.21) The scattering amplitude determines the experimentally measurable cross section o given by da (16. z=o in which (cf.k) = ^T(2l + !)/(*.23) as the contribution of the residue of a pole in the plane of complex I of some suitably constructed contour integral.TOO= 1/2): ip(r] r ~*°° Jkz elkz + F{0.25) Setting Note the minus sign in the argument of the Legendre function.. I = n + x.2 The integration contour C. 16.2 in the complex Z-plane: *<«•*> = s / c c K <u(21 + 1) vf{i.k) — e 2i5[(k) ~" p—iirl (16.23) f(l.24) Si(k) being the phase shift.''' The idea is now to consider each term of the expansion (16. (11.k)\2. .fe)P .2.5 The Sommerfeld-Watson Transform 331 ingoing plane wave in the direction of z (here with h = 1..km-cose). For this purpose we consider the following integral taken along the contour C shown in Fig.184) to (11.26. 16.k)- 0ikr (16. (11. n = 0. The scattering amplitude possesses the partial wave expansion F(0..

27) a(p) © b(q) Fig.28) . u={p-q')\ (16.fc)P„(cos0). be elastic scattering of a off b. In order to see the relevance of Regge trajectories in a reaction of particles. t = (q-q')2.3 Reaction channels and their respective Mandelstam variables. d(q') (16.k) Zi Jc TT{L . Screened Coulomb Potentials Then integrating with the help of Cauchy's residue theorem. For the momenta indicated we set (with metric +.332 CHAPTER 16. we obtain m® = ± I dl^±^-f(l.—) s = (p + q)2. as indicated in Fig.—. 16.26) P((+COS0) = 2 ( 2 n + l)/(n. for instance. Thus the direct reaction may. 16.3. Consider the reaction of a particle a{p) with four-momentum p^ colliding with a particle b(q) having four-momentum q^ and producing a particle c(p') and a particle d(q'). The reaction therefore describes the following processes which are also described as (reaction) "channels" and in which an over line symbol stands for the appropriate antiparticle (like IT meson with positive charge and that with negative charge): s: t : u: c(p') a + b —> c + d. a + d — • c + b. n=0 which is the usual partial wave expansion.3. we have to digress a little and introduce a few simple ideas which played an important role in the development of particle physics. 16.—.n) oo {-l)1 Pt{-cosO) v ' (16. in which case c = a and d = b. a + c —> b + d. Such a direct reaction and its "crossed channel reactions" are shown schematically in Fig.

0s). the dynamics is described by the exchange of quantum numbers. One now makes two hypotheses. t. L. 16. one then has A(s.t. the only kinematics we require in the following is given by the relations (e. Thus. N. Mandelstam [184] and R. which is an analytic function of the variables s.t) = VtFt(t.2 ^ ( 1 + cos6 s ). s = -2qf(l-cos0t)(16. Symbolically we then have the situation shown in Fig.et) (16.9s)\2.The variables s. A(s.4.u channels) are determined by one and the same relativistically invariant scattering amplitude A(s. if s describes the square of the total energy in the s channel.. the quantities carrying these quantum numbers are the Regge trajectories. the variables t and u would describe momentum transfers in the crossed channels. N.g. B.g.29) and with self-explanatory meaning for the cross section of the reactions ~ = \Fs(s.32) t = 4(q? + ml). (16. For a Lorentz-invariant normalization of the initial and final states.g. A(s. In the Yukawa picture of a reaction the dynamics is described in terms of the exchange of mesons IT (called pions). S. (16. e. u = . except for poles and cuts which characterize the reactions in the three channels. u. (a) Mandelstam's hypothesis: All three processes (described by the s.^ (b) Chew's hypothesis: All (composite) particles lie on Regge trajectories. M.(<fe)^(cos 6t).5 The Sommerfeld-Watson Transform 333 and for simplicity we assume here that all particles are spinless and have the same mass mo. Blankenbecler.§ Since the three variables s.u) (if any of the external particles has nonzero spin. In a Regge theory on the other hand.t. t = -2q£(i . Khuri and S.u are not independent.t.31) i l Taking all particles to have the same mass mo. Chew [48]. T See e.16. Goldberger.t). etc. Treiman [25]. § See G.u are known as Mandelstam variables. F.30) with partial wave expansions (observe the non-invariant factor k has been removed) Fs = ^ ( 2 2 + l)F z (g s )PKcos 98).t) = y^Fs(s. Ft = £ ( 2 J + l)F. . there will be several such amplitudes which together describe all three processes).c o s 0 s ) . one can write the amplitude as depending only on two. the momentum k there would correspond to qs) s = ~4(ql + ml). if we had been considering the s reaction originally.t.

It is known from the Mandelstam representation that the amplitude possesses a branch point at cos 9 = 1 + TOQ/2(^. in the domain of s-channel physical values of the kinematical of the i-channel). So what can one do? 1 H. .5 for Yukawa potentials V{r) u: lm cose dr a(/j. the Legendre polynomial Pj(cos 9) does not possess a cut. Lehmann [162]. the partial wave expansion is not valid in the physical region of the (i. But > > wave expansion X)(2Z + l)F.5 The Lehmann ellipse.e. 16.)e~ •fj. For s — oo : cos 9t = 1 + s/2q^ — oo. Thus one is interested in establishing a representation of the amplitude in terms of i-channel Regge trajectories.((fc)/Kcos0t) s-channel i-channel s-channel variables the partial converges only within the so-called Lehmann ellipse^ which is shown in Fig.4 Yukawa versus Regge theory. 16. However. 16.334 Yukawa CHAPTER 16.r 1+m 2 /2q 2 o ^t "" Re cost: Fig. But for integral values of I. Screened Coulomb Potentials Regge oc(t) Fig.

16. i. to use the Sommerfeld-Watson transform.33) . k) have to the right potentials one can show t h a t the integral along the curved porA' of the circle at infinity tends to zero. N F(6. the contour representation (16. Thus consider now the same integral but taken along the closed contour C shown in Fig. and there is only a finite number N + 1 in the domain of 9W = .6. all poles of Ql > 0. Moreover.t).e. „ i.6 The integration contour C". 16.k) »E n=0 [2an(fc) + l]/3n(fc) p sin 7ran(k) an(k){- cos 6) 2i y _ i _ i o o = F(s.e. Again we have / Jc> • • • = 1-ni y^ residues j3n „ For Yukawa tions A and f(l.5 The Sommerfeld-Watson Transform 335 T h e answer is.^ B 1 A Fig. contour C / / D ^ ^ .1 / 2 . B Iml 1 \ i All \ r\ 0 / closed . sin nl [16.26) but for a different choice of the contour. Then / JDD' ••• + j JBB' • • • = 2TTI V ^ residues f5n. 16.

if ao(t) is the Regge pole with largest real part. Miiller-Kirsten [29].35) This result is valid for s — oo and t negative (cf.e. i. 16. Regge trajectories have. for plots and discussion see R. as we discussed briefly in the above. at high s-channel energies. Predazzi and T.e. W. Regge [229]. Omnes and M. The slope of the Regge function an(k2) is an " Simple cases are the harmonic potential and the square-well potential. Screened Coulomb Potentials Since for s — oo : cos Qt = 1 H n ~* °° > 2?* and (from Tables of Special Functions) farM-oo: Pa(z)* ^+l]\(2zY. the amplitude can be represented as a sum over ^-channel Regge poles. ^M)^E i 2 a f: a „t ( t ) ^ m <-^-'F(s. we have P-\/2+IR{Z) ~ 0(l/y/z) — 0 for s — oo (i.t)&B(t)sa°V. J. also been investigated in the case of other potentialsJI The significance of Regge trajectories is evident from the fact that they determine the high energy behaviour of scattering amplitudes. Eq. The > result demonstrates that the high energy behaviour of an s-channel reaction is determined by the leading Regge trajectory in the crossed ^-channel. (IM4) This is the asymptotic Regge expansion of the invariant amplitude for s —>• oo. Apart from more refined details. we have (16. of course. It follows that under these conditions. for high energies in • > the s-channel). The logarithmic potential has been investigated in a way similar to the Yukawa potential above by S. K. In particular. Limic [177] and E. i. this behaviour has been confirmed in high energy hadronic reactions. Bose and H. .6 Concluding Remarks The potentials we considered above have wide application under the name of Yukawa potentials in nuclear physics and as screened Coulomb potentials in atomic physics.e. In subnuclear physics they played an important role before the advent of the quark idea and thus of quantum chromodynamics. (16. 42.32)) and finite. Potentials with a short-range repulsion more singular than 1/r 2 have been considered by N.336 CHAPTER 16. p. Foissart [223].

The small imaginary parts of eigenvalues or lifetimes of resonances have not yet been calculated. Singularities appear also in the form of cuts in the plane of complex angular momentum. W.** Whereas the former are related to absorptive properties of a reaction. Miiller-Kirsten [209]. like a meson or a baryon.16.1. called "Regge cuts". and is composed of quarks. For an overview see e. a composite particle being one with structure. H. The way to do this is similar to calculations in Chapters 18 and 20. . Regge poles are not the only possible singularities of a scattering amplitude in the plane of complex angular momentum.g. and also as so-called "fixed poles" . J. The above treatment of screened Coulomb potentials is incomplete. the latter are related to the distinction between "' elementary and "composite particles". Related aspects are discussed in Sec.6 Concluding Remarks 337 important parameter in string theory. 20.

This page is intentionally left blank .

p. is effectively the Mathieu equation whose solution has for a long time been considered as being very difficult. Thus Jacobian elliptic functions also pro*To be precise: Separation of the wave equation Aip-\-x2. that the Mathieu equation lies outside the scope of equations which can be reduced to hyper geometric type. since many analogous relations hold.Chapter 17 Periodic Potentials 17. K being the complete elliptic integral of the first kind. 339 . The Schrodinger equation with this potential. 2K or AK.) which are functions that interpolate between the n. the ellipsoidal wave equation reduces to the trigonometric form of the spheroidal wave equation. The reason for these difficulties is.lP = 0 in ellipsoidal coordinates leads to the ellipsoidal wave equation with sn 2 z and sn 4 z terms. depending on the parameter k. called elliptic modulus. See F. 19 and p. and are themselves periodic with period e. Taking in the ellipsoidal wave equation the limit k —> 1 (fc elliptic modulus) and putting tanhz = sin 6.or 27r-periodic trigonometric functions on the one hand and the non-periodic hyperbolic functions on the other. Arscott [11]. Apart from the discontinuous Kronig-Penney potential consisting of a periodic repetition of rectangular barriers. separation of Laplace's equation Ai/> = 0 in ellipsoidal coordinates leads to the Lame equation (no sn 4 z term). however.g.0 < k2 < 1. example 14. In essence these periodic Jacobian elliptic functions. motivated mainly by the regularities of the crystal structure of matter. the most immediate candidate is a trigonometric form like that of the the cosine function. M.1 Introductory Remarks Prom the beginning of applications of quantum mechanics periodic potentials of various types were immediately considered. are not much harder to handle than trigonometric functions. 25.* These elliptic equations involve Jacobian elliptic functions like sn(a. The Mathieu equation can be obtained as limiting cases of spheroidal wave equations and the elliptic Lame and ellipsoidal equations which represent a further level of complication.

The most important formulas for our purposes here are collected in Appendix A. and to relate these to the weak coupling bands or regions of stability.7..1. 17.1) where h2 is a parameter and — IT < z < IT. each with its own characteristic eigenvalue. it is this difference which implies in the case of the Mathieu equation the existence of the parameter function v called Floquet exponent^ The Mathieu equation can be obtained from the Lame equation in the limit of n — oo. The first of these conditions is already tMathieu equation: R.340 CHAPTER 17. A main objective of this chapter is the calculation of these level splittings following the original calculations of Dingle and Muller^ with the perturbation method described in Sec. i. Miiller. or 5-wave Schrodinger equation with the cosine potential is conveniently written d2v — | + [A .2) where K2 = n(n + l)k2 and n real and > —1/2 and 0 < z < 2K. Muller [73]. J. which does not possess any polynomial solutions. 8. Both of these domains are important for a host of other considerations. Here the Jacobian elliptic function sn z is a periodic function analogous to a sine function and has real period 2K} For n = 0. B. W. the three Jacobian elliptic functions s n z . (17. ..2 . k2 — 0 which means that in > > this limit the Lame polynomials degenerate into the periodic Mathieu functions.e. W. [205]. the Lame equation possesses 2 n + l polynomial solutions. ''The reader who encounters these Jacobian elliptic functions here for the first time. need not be afraid of them. those of integral order. Periodic Potentials vide periodic potentials like trigonometric functions. For large coupling. The finite heights of the periodic functions of the potentials permit tunneling from one well to another and thereby produce a splitting of the asymptotically degenerate harmonic oscillator levels. In fact. cosine and tangent. such as sn 2 z + cn 2 z = 1 and double and half argument formulas etc. We might mention already here the much less familiar but more general Lame equation. The Mathieu equation. (17. c n z and d n z are handled with analogous formulas. See also the discussion in Sec.4. J. Like the trigonometric functions sine. Dingle and H. This is a significant difference compared with the Mathieu equation. which one looks up in books like that of MilneThomson [196] or Tables when required.1. Lame equation: H. or S-w&ve Schrodinger equation with elliptic potential. which is conveniently written ^ + [\-2K2sn2z]y = 0. these periodic potentials may be approximated by series of degenerate harmonic oscillator potentials. The cosine potential of the Mathieu equation can therefore more precisely be described as Mathieu or trigonometric potential and the potential sn 2 (x) of the Lame equation as Lame or elliptic potential.2h2 cos 2z]y = 0.

we see that for sufficiently large values of h2 the periodicity or boundedness of the solution is destroyed and hence becomes one of instability. i. we see that in a plot of h2 versus A. a bounded function. The instability of an orbit would be evident either from an unbounded spiralling away to infinity or from a collapse into the centre. i.e. Although the mathematics literature on the subject uses the physical concepts of stability and instability. Vachaspati [242]. Salem and T. we compare the For interesting related discussions see M. .2.2 Cosine Potential: Weak Coupling Solutions 341 seen to rule out polynomials.17. these concepts are rarely explained as such there. Here an important aspect is the determination of the domains of stability of the solutions and the boundaries of these domains. Our considerations below are deliberately made simple and detailed because their treatment in purely mathematical texts requires more time to become accustomed to.e.e. In classical mechanics stability is treated for instance in connection with planetary motion. h2 = 0 belongs to the domain of stability.1.e. Looking at the Mathieu equation for h? —> 0. It is this additional parameter function appearing in the solution of the Mathieu equation which attributes the equation its reputation as being particularly hard to handle. the case of h2 small.^ 17. Thus a bounded trigonometric solution is indicative of stability and an unbounded exponential solution of instability. and one can imagine that a nontrivial parity factor exp(in7r) then turns into a complicated phase factor exp(iz/7r). as in Fig. i. Thus the solution of the appropriate Newton equation is essentially a periodic function like cos 8 (in the case of planetary motion this yields the polar equation of an ellipse). The orbits of planets are stable in the sense that deviations from the recurring elliptic orbits are small. Correspondingly quantum mechanics — which requires the electrons of an atom to move around the nucleus — explains the stability of atoms.2 Cosine Potential: Weak Coupling Solutions We consider first the weak coupling case. 17. y" + \y = 0(h2). the semi-axis A > 0. Adding the negative term — 2h2 cos 2z to A. since the solution there is of the form cos vAz.1 The Floquet exponent We consider the given Mathieu equation with argument z and compare this with the same equation but with z replaced by z + ir. i. 17.

y(2vr) 2 a = —7—^ = 1. d2y(z + 7r + [A . ° ^ [1 + 0(h2)}.2h2 cos 2z]y(z) = 0. TT.342 CHAPTER 17.e. (17. (with Tn as translation operator) Tlxy{z) = y(z + TT) = cry{z). i. b: y(z) = A cos vAz + B sin \f\z and y+(z) = a cos yXz.2/i2 cos 2(2 + vr)]y(z + vr) = 0. we obtain y(vr) y(2?r) a = — — a = —^^r. — • .e. 2/(0)' yfr) . a = const. y~(z) = 6sin vXz.B and a. i. Periodic Potentials equations ^ p + [A . y+{z)- .4) with (for convenience in connection with later equations) y'+{z) y'_(z) = avAsin-\/Az = = bv\cos^f\z = —y-(z). Setting z = 0. We set (here and in the following frequently apart from contributions of 0{h2)) with constants A. sin V Az or linear combinations. y(0) Now suppose we write the equation for h2 small y" + Xy = Then solutions are y(z) oc e ± ^ [ l + G(h2)] or C S 0(h2).3) The parameter a is therefore the eigenvalue of the operator TV Our first objective is its determination which amounts to the determination of the parameter called Floquet exponent below. (17. d(z + TT) cos 2z We see that there are solutions which are proportional. A solution of the second order differential equation is determined completely only with specification of boundary conditions which determine the two integration constants.

yV(0) = 0. 7 y\z) = ay'+(z) + (3y'_{z). we obtain V (TT) = ay+ (TT) + (3y _ (TT) = < [aa]. we choose these with the following set of boundary conditions: y+(0) = a.2 Cosine Potential: Weak Coupling Solutions 343 The solution y(z) is an arbitrary solution. The roots a± are therefore obtained as 2bV\y+(ir) ± J4b2\yl(Tr)-4. ay'~(ft) — bVXy+(n) and the Wronskian w + {y+(7r)y/_(7r) . We can write therefore.W or fy+(n)-aa J/_(TT) = W ^ ^ J a ^\ = n For linear independence of a and /3.aa][y'_{-n) . the determinant of the matrix must vanish. i. Using Eq. y{z) = aty+(z) + 0y-(z). 2/'(7r) = a ^ W + ^ .y+(7r)j/_(7r)} = 0. and by'+(?r) = —ay/Xy-{n). Setting in these last equations z = 0 and using the previous pair of equations and the boundary conditions.4).oWX] . This means.a2b2\ a± ~ 2a6\/A .{ay'_{ir) + bV\y+(ir)}a Now from Eq.y-{ir)y'+{K) = abVX in agreement with its value at z = 0 above.e. (17. y'(z + n) = a[ay'+(z) + py'_(z)}. with constants a and /3.17.y+(7r)y_(7r) = 0 or abVXa2 . y'_(Q) = bV\ with Wronskian W[y+. \y+>y-\ = y+{ir)y'-{ir) .3) we obtain y(z + ir) = a[ay+(z) + @y-(z)]. (17. whereas the solutions y+ and y_ have here been chosen specifically as even and odd around z = 0 respectively. This expression is equal to the coefficient of a2 in the quadratic equation for a. 2/_(0) = 0.y_] = y+(0)y'_(0) — y_(0)y+(0) = ab\f\. [y+(v) .

5) is that for h2 = 0 we have v = A/A and hence more generally v2 = \ + 0{h2). 8. 2 _ hA 2(A-1) (13A-25)/i8 32(A-l)3(A-4) 12 ift (45A3 .344 Setting / = 2by/\y+(Tr).0 This series may be reversed t o yield the Floquet exponent.e. T h e result for A is the expansion A = h4 2 . Setting cr+ = e .1.6) We can derive various terms of this expansion perturbatively with the method of Sec. Thus this Floquet exponent is determined by the value at z = -K of t h e solution which is even around z = 0 and is independent of the normalization constant a of t h e even solution. This is the easiest application of t h a t method since only simple trigonometric expressions are involved. This calculation is demonstrated in Example 17. (5i/ 2 + 7)/i 8 vz + — .22)) the boundary conditions of periodic solutions. Eq. . .e. (17. this implies y+(Tr)/a = ± 1 . Periodic Potentials g= 2aby/\. i.5) a The parameter i/ is known as Floquet exponent. ** An important consequence of Eq. Prom this condition we determine later (cf.8) "Observe that for integral values of v (in lowest order of h2). i. <T_|_ + <T_ = 2y + (vr) . we have for the sum of the roots a+ + <7_ = 2 cos TTV or" C O S T T I ^ ^ ^ C O S V A T T .1169)/i 12 64(A-l) (A-4) (A-9) 5 2 + 0{hw) r 16 (17.7. (17. 5 a Prom this we see t h a t (as one can also verify explicitly with some manipulations) one root is t h e inverse of t h e other. <7_|_cr_ = 1. cr+ = 1/<T_. where the constants are usually taken as unity from the beginning. (17. we have <7± — CHAPTER 17. (17. **This point is not immediately clear from mathematics literature. _ / ± v 7 ^ V .= — .455A2 + 1291A ..+ 2 2(i/ -l) 32(i/2-l)3(v2-4) ( 9 ^ + 58.2 + 2 9 ) ^ 2 + 64(1/2 _ 1 ) 5 ( z / 2 _ 4 ) ( l / 2 _ 9) + u ^ ^ ^ .

(17.jl2.17.1) 15A2 . Example 17.1: The eigenvalue A for non-integral v and h small Use the perturbation method of Sec. as will be done later.3). Schiifke [193].2 Cosine Potential: Weak Coupling Solutions and so 345 (8 . 8. = J—^ + o(hr). p.1 below one obtains the following expansion for cos7rz/ which is fairly obvious from Eqs.9a) 4(1-A)>/A 64(A-4)(A-1)3A^A One observes immediately that these expansions cannot hold for integral values of v or A. To lowest order the solution y is j / ° ) = yv = cos vz n where d2 Dv:=—~+u2. It follows that in these cases i? = n2[l + 0(h2)].11) or smvz or e±ivz.12) S e e J.35A + 15A2)/l8 /.10) A = n2. W. (17. rr +h* 3 =7rsin7rvA 64(A-1) (A-4)AVA +Q(h12). (17. which can then be rewritten as Dvy = 2fe 2 (A + cos 2z)y.„„ x n. 124. Meixner and F. cos TT\/A 32A(A-1) 2 . .5) and (17.e. (17. These cases therefore have to be dealt with separately.35A + 8 . — 2/i 2 A and insert this into the Mathieu equation y" + [A — 2h2 cos 2z]y = 0. we must have a2 — 1 and hence e 2iv/A7r = 1} i.9a) and is therefore not derived here in detail^ ri/ = VA + rCOS7Tf = COS7TVA + /l 4 h4 7rsin7rVA -= 4\/A(A . For these integral cases of v we demonstrate in Example 17. Solution: We write the eigenvalue equation A = v2.2 how the expansions for small values of h2 may be obtained perturbatively. With a perturbation theory ansatz for y+ around h2 = 0 as in Example 17. VA = ±n. (17.9b) TT2 Next we observe that if we demand that the solutions y+ (z) and y_ (z) satisfy the condition (17.7 to obtain the eigenvalue A as a perturbation expansion for nonintegral values of v and b? small. (17. n an integer.

Periodic Potentials The complete solutions of these cases are written respectively in self-evident notation cev(z. (17.3 and 19. The right hand side of Eq. v) = 0. Meixner and F.3.e. Proceeding in this manner we obtain the solutions ce. . W.h2). (i/. . so similarly yW leaves uncompensated / # > = ft2 ("• v ~ 2 ) R (o) . and (v.2 — we can write down recurrence relations for the perturbation coefficients pa(2j) together with boundary conditions. as shown.h2).v±2) = l.11) terms amounting to i?i 0 ) = = 2 h 2 ( A + cos2z)y„ = 2h2Ayu + h2(yl/+2 + y1y-2) h2[(v. (17. Schafke [193]..(2) + . in each case 2cos2z2/„ = 3/„ +2 + j / „ _ 2 .346 CHAPTER 17. We observe that in these cases 2cos2z cosvz 2cos2z smvz 2cos2ze±il/z i. (17. = . sev(z. (17.2)j/„_ 2 + {v. where except. se and me as* oo i y = „«>) + yW + i. h2(u.2) — 2 "^ -2(2i/ + 2) " + 2 vanishes. Next we treat terms yv+a in j / 1 ' in a similar way.!/) = 2A We now observe that DvVv = 0. for instance.13) The first approximation y(°) = yv leaves unaccounted on the right hand side of Eq. ("• ^ + 2) (o) 2(2^ .2. (17. With our perturbation formalism — as demonstrated in cases considered in Sees.11) and so in R}?' may be cancelled out by adding to j / ' 0 ' the new contribution -a(2v + a)' = 0. h2). me±v(z. when a = 0 or 2v + a = 0. v)yu + {v.16) directly into the original equation in the form of Eq. of course.2) _ _|_ _ (y.11).2) which applies when v is nonintegral.14) = = = cos(y+ 2)z + cos{y— 2)z. since j / 0 ' = yv leaves unaccounted R\. 17. . (17. see Example 17. The recurrence relation can also be obtained by substituting the right hand side of Eq. (17.e.v— 2) 2 ^ + yu+2 Vv 2 y (D = h _2(2u-2) -2(2i/ + 2 ) ' Then up to 0(h2) the sum j / 0 ' +y^ is the solution provided the remaining term in Ri. by J.„ + £ h2i £ P2i(2j)yv+j (17. i. v .e.15) Hence a term fMy„+a on the right hand side of Eq. so that also Dv+ayv+a and hence Dv+a = Dv + a(2v + a) and Dvyv+a = -a(2u + a)yv+a(17. or A = 0(/i 4 ). e ± i ^ + 2 ' 2 + e±i("-2)z. siniy + 2)z + sin(i/ — 2)z. We assume for the time being that a ^ 0 and 2v + a ^ 0 (the latter case requires separate consideration.v + 2)yu+2]. up to 0(h2) we have A = 0.e.16) *These small-h expansions are convergent. v . '.14) therefore leads to the following next-to leading order contribution (v. i. i.

e. „ . Meixner and F. + y~i) (17. we obtain immediately 0 = 2Ah2+ 2(^T)+-' which verifies the term of order h4 in Eq.2: The eigenvalue A for u=l and h2 small Use the perturbation method of Chapter 8 to obtain the perturbation expansion of A for v = 1 and h2 small. We consider the specific unperturbed solution yv = cosvz which is the dominant contribution j / ° ) of our solution y. fo^ + 2) . For v = 1 we then have yi = cosz = y~\. thus determining the quantity A. . 0 = h2(u..g. This completes the determination of the solutions for nonintegral values of v in the domain of small values of h2. These expansions are actually convergent with a definite radius of convergence as explained in the mathematical literature.1 with A = v2 — 2h2A but with v = 1.17) Inserting 1 for the step-coefficients of Eq...v)-\ . Hence we obtain 0 = /i 2 (2A + l ) f h4 + ••• .. i.14). . is set equal provided the coefficient of the sum of the contributions in 3/1 contained in R\ to zero.R{ . the solutions are not yet normalized. 121. Solution: We start as in Example 17. .17. v) ' ( 2 v + 2) (17."-2). Schafke [193]. (17. Example 17.. J. p.18) Again applying Eq. one has to deal with each integral case separately as in the following example.1 we now have the situation that the contribution y\ to the entire solution leaves uncompensated on the right hand side of the equation the terms amounting to R(°li = = 2 h 2 ( A + cos2z)3/„=i =2h2Ayi+h2(y3 h2[(2A + l)yi+y3]. * Note. W. See e. however.2 ) v(y-2. 2 ( 2 i / .2—r——— (i/ + 2.u) + h4 (". The higher order terms naturally require a little more algebra.7). . Following the first few arguments of Example 17. (17.2 Cosine Potential: Weak Coupling Solutions 347 together with the equation determining A.15) (but with v — 1 and a = 2) we obtain the next to leading contribution to the solution as (i) = h2 =_fe2 V {-2(2v + 2)}„=1y3 8 y3' This contribution leaves uncompensated the terms amounting to R\ = ~ — 2h (A + cos2z)j/ 3 o = -—[Ay3-\ 4 \ 2 Hence the next contribution to the solution is 4 V -8 2(-4)(2 + 4 ) / 4 V -8 -48 Proceeding in this manner we obtain as the complete solution the sum y = y(0)+yW+y(2)+. In the case of integral values of i>. (17.

COS 6Z • 128 h2 COS 2 • COS 3 2 - h / c o s 32 16 cos 5z -48 cos z 32 in agreement with Meixner and Schafke [193]. and v is to be found in ascending powers of h2.348 From this we obtain -2h2A = h2 CHAPTER 17. In the above we considered the case of v an integer and calculated the eigenvalue A.n integers. J. The specific normalization here is taken as I /*7r r-n 1 = — / dzce2(z. W. h2). Eq.19b) This expansion agrees with that given in by Meixner and Schafke [193] (p. Jo dz cos z h2 cos 3z - iC2T It J--* cos z H 64 cos 2 3z - 64 It follows that 128 and the normalized solution is therefore h2 Vc . With normalization (not to 1. (17. 120. 123) (there the solution is called cei(z. Schafke [193]. h4 (17.24c) below) . *" J-ir Thus we have and uses / dz cos mz cos nz = — <5 mn . Periodic Potentials h4 r ••• . . h2)) except for the overall normalization which implies that in our (still unnormalized) case above there are no contributions cos z in the higher order contributions. m. Inserting this into A = v2 — 2h2A 8 for v = 1 we obtain (cf.19a) This expression agrees with the result given in the literature (there the eigenvalue is called ai). i.* The associated solution is ym+ym.e. We introduce a normalization constant c and rename the normalized solution y then yc. the case of A an integer. We consider this case again in an example. The reverse situation is later of importance. Meixner and F. p. see below!) one obtains the additional terms given by Meixner and Schafke [193]. cos z h cos Zz 8 h2 h4 ( cos 3z cos 5z — cos iz A 1 8 4 l -8 -48 h4 (cos 3z cos 5z 1 4 I 16 -48 (17.

l ) 2 (A .. (11A2 .4 ) + . (17.4) 2 7T 2 /l47T2 + • 8 h87T2 8A(A .Q . fe47T2(A-4) 4%/A(A . so that 2 2 cos nu = cos 7r(2 + S) = cos 27r cos TC8 = 1 and comparing with the above. we obtain^ S 3 V 2 and therefore i/ = 2 - iy/E ( h 3 \2J 7i -.3: The Floquet exponent i> for A=4 and h2 small Show that the expansion of the Floquet exponent v around \f\ complex expansion: v = 2- 349 = 2 is given by the following iVE/h 3 \2J 7% h (108^/5 \2J 8 1185H (h 31104^X2 Solution: We proceed as follows which demonstrates explicitly how the singular factors cancel out systematically.2 Cosine Potential: Weak Coupling Solutions Example 17. J . H.9b) v — 2 + <5.1) .17.1)2-/A TT2 +h8 (15A2 .('108\/E\2/ Expansions of u for integral values of vA have recently been given with a larger number of terms:^ l//i\4 3 el 2J 1 fhy 15 1 2 / + \/A = 4 : v § 311 4320 V 2 ' + 1555200 V 2 + 137 (h\° 305843 (h 12 27000 I 2 ) + 680400000 V 2 133 (h 12 + ••• R .9b) and considering the approach A — 4 gives > -(A-4)2 + . o 1-• • • .3 1 x 4 + 8) 274233 (. Substituting these expressions into Eq. W.= l 2vA sin27r + (A — 4)(cos vA-7r)x = 4—= + ••• = j= h •• • . 128A2(A-1)3 + 0(h12) in the limit e — 0 > Hence — observe the cancellation of factors (A — 4) in the term of 0(hs) h TT (ll C O S TTV = 1 H S 2 x 1 6 . (17. S.1) (A . ^S. Liang and Yunbo Zhang [189].3 5 A + 8)TT2(A-4) _ 3 / / 64(A . Manvelyan.• 1 + 5n h 2 s 2932 Setting in cosm.3 1 A + 8) ( A . We thus obtain the expansions cos"\/A7r sinvAvr = = cos27r + (A — 4 ) ( — s i n V / A 7 T ) A = 4 — = + --.4)Av A2v A ~ 32A(A . . Gubser and A. Miiller-Kirsten.9b) about A = 4. on the left hand side of Eq. . J. We set A — 4 ~ 4e and expand the cosine and sine expressions appearing in Eq.. (17. Hashimoto [124].

Periodic Potentials Our next immediate aim is to specify the solutions for which Eq. (17.29. in F.4) into account.1 Boundaries of domains of stability.6.2*0 y~(z ± 7t) = 6 sin VA(z ± TT) = y_(z)cos VA7r ± 6 cos vXzsin v\ir. for instance. (17. "These conditions.w^.2. These are defined by specific boundary conditions.e. Arscott [11].10) applies. We now derive these boundary conditions. bV\ y+(0) 2/40) 07. as also conditions (17. 17.2 Four t y p e s of periodic solutions The Mathieu equation allows four types of different solutions — briefly: Even and odd solutions of periods TT and 2ir. . M. i.350 CHAPTER 17. we have" y+{z ± 7r) = a cos vA(z ± TT) = y+(z) cos vXir = TV-{Z) F sm v<\7r. Fig.21b).21a) and (17. pp.W4W=UtWSta±I(. solutions for integral values of v. 17. We verify these for the large-/i 2 solutions in Example 17. 28 . may also be found.. Taking in particular Eqs. = and s+w 2tW±I. concentrating again on the leading term in the even and odd solutions y±(z) for ease of understanding.

f^\y+(j) /vr\y_(7r) and for the derivatives ^(i)^^'-(D^ . o y+(fy-(f)-ob\/A _ = 1 + 2I&A/X . /7r\j/!_(7r) y L l 7TJ .21a) Since (with the help of the Wronskian) *W4)MM?)-^ we also have y+(v0 _ . Then y+ -y- 351 U . ^\y±M /vr\yL(7r) . 1+ y+(f)^(f)-y_(f)yV(f)' 7T Using the Wronskian (which we actually had above!) W[y+. /TTW^) 2j^T±y+V2 From these equations we obtain y+{ir) in terms of functions at TT/2 by eliminating y'+(Tt) from the first and the third equations. .2 Cosine Potential: Weak Coupling Solutions We select the equations with lower signs and put z — TT/2. which after some rearrangement can be written y+(vr) _ . Then y+ -K 1 _ 2/+00 7T y- V + (f)[i + 2/±Ml ^ ] y'-il) 2y_(f )i/+(f. .4(f) (17. we obtain y+(7r) _ i | 2y_(f).y- V+ \ 2 7"^ V 2 f-la^+U a cos V A | ] 6^A + a&( sin\/A-j ^A = abV\.17.

These may be subdivided into classes depending on whether the integral value n — 0..1. (17. IT. in this order.1 for the first few eigenvalues which are boundaries of domains of stability (as discussed in Sec.. Thus for these solutions the right hand sides of these equations imply the vanishing of the functions y± or their derivatives at z = ir/2. A -*• b2n+2(h2).2. se2 = sin 2z 2 12 13824 ' h2 z —— sin 3z + • • • . (17.2): hA 7h8 29h12 + -Y l28-^04. y -> se2n+i=qo. There the second term on the right is nonzero.. 2-K. (17. was treated in detail in Example 17..— — — • • • . A 2 y -• ce2n=q0-i.23) We can now see how the domains of stability and instability arise.2 we require this relation for nonintegral values of v. . the case of ai with solution cei (of period 2-7r).21b) must be ± 1 .24d) v 12 ' **To avoid confusion we emphasize: In Sec.24c) 8 h2 sin 4z -\ . The ordering of the eigenvalues which then results is a0 < bx < ai < b2 < • • • < bn < an < • • • for h2 > 0. y -> ce2n+i=qo. We see that there are four possibilities and hence four different types of functions. A -> b2n+i(h2). 17. This is shown schematically in Fig.+ .5) that the left hand sides of Eqs. 17. and whether the function is even or odd. Mr7nA (17. (17. where the notation for the corresponding eigenvalue is put alongside on the right: y+(f) = 0 with y+(f) = 0 y'-{\) = 0 with y _ ( § ) = 0 with with A . y -> se 2n +2s 9o +i The functions so defined have respectively period 7r. (17..21a) and (17. (17. ce± = cos 8 64 h4 5h8 b2 = 4 1 . of v is even or odd.24b) 8 h2 z —— cos 3z + • • • . sei = sin 8 64 /i4 h6 a\ = 1 + h2 — .1). 27r.T + . These are defined by the following boundary conditions. Finally we cite here some expansions of the eigenvalues along with those of the associated solutions (an explicit example. . Periodic Potentials and hence** a afc\/A For integral values of v (in lowest order of h2) we know from Eq..24a) ao = hA h6 b\ = 1 — h —— + — — • • • . 19.352 CHAPTER 17.' fh2 V2ceo = l .3.• a2n(h2).22) -* a2n+i(h ). (17.

17. Thus in Fig. 17. 353 . Miiller [73]. W. of course.24e) 17. 17. Dingle and H. Since we have already a definite notation for the levels (boundaries of regions of stability) in the domain of small values of h2 we naturally want to be able to relate these correctly to those in the asymptotic laige-h2 domain. The main objective of the present section is therefore the calculation of the tunneling effect in the form of a splitting of the otherwise asymptotically degenerate harmonic oscillator levels j t Fig. .3. 17.2 suggests that for very large fluctuations the potential can be approximated by a number of independent and infinitely high degenerate harmonic oscillator potentials. . B.> (17.17.17„.3 Cosine Potential: Strong Coupling Solutions „ 5/i4 a2 = 4 H 12 763/i8 .2 The cosine potential.2cos4.2 or by calculation that the potential 2h2 cos 2z has (harmonic oscillator-like) We follow here R. 13824 .3 we show schematically how the low energy solutions for integral values of the Floquet exponent are related to the asymptotically degenerate harmonic oscillator eigenvalues. Consequently one expects the eigenvalues in the large h2 domain to be given approximately by those of the harmonic oscillator. In the case of the cosine potential. The cosine potential cos 2z depicted in Fig.hz — 12 .3 17. 17. the walls are not infinitely high and hence tunneling occurs from one well into another. J.1 that the boundaries of regions of stability merge to lines.1 Cosine Potential: Strong Coupling Solutions Preliminary remarks In the case of large couplings h2 we observe from Fig.z-3 ce2 = cos 2z . We can see from the inverse of Fig. .

. (17.1. Sees. 4 (17. 8.2.25) This is a Weber (or parabolic cylinder) equation which we encountered earlier (cf. Periodic Potentials minima at z — ±ir/2. but only approximately so.. simply the one-dimensional Schrodinger equation for the harmonic oscillator. Ion (17.3) and is. In the case of finite heights of the potential barriers. there is tunneling from one barrier to the next.2 and 8.i r ) ~ 1 — 21 z ± -7r Changing the independent variable now to the equation is approximated by A + 2/t2 dx2 + 4/i 2T y = 0. of course.2 The solutions We insert Eq.26) 17. n 0.. This equation has normalizable solutions for X + 2h2 2h q0 = 2n + 1.. (17. Thus in this case the quantity on the left is not an exact odd integer.26) into the original Mathieu equation and obtain y" + 2h q + rrx-r . where A / 8 is a remainder.3. Thus we rewrite the Mathieu equation as y" + A + 2hz cos 2 \z ± -7T y = 0 and take c o s 2 [ z ± .27) . as in the case of the cosine potential. and we set this equal to q.354 CHAPTER 17.2/i cos z V = 0. We therefore wish to expand it about these points. we therefore set 9 A A = -2h2 +2hq + —. Taking the remaining terms of the cosine expansion into account.

31) We make the important observation here t h a t if one solution of the type (17. another solution.29) the equation can be written in the form D^A where 1 2 6 /i (16A" + 2 A ^ ) . (17. (17.27) we )A = 0.29) is known. z). 16/i/ (17.+2h': 2h level splitting oscillator 2nd excited state associated solutions A^9 h=4 oscillator 1 st excited sate ^1 oscillator ground state A^Oh 2 large h 2 =0 Fig.3 Cosine Potential: Strong Coupling Solutions 355 For h —> oo the solutions of this equation are y ~ exp I ± 2h / Writing y = obtain J 4exp(±2/isin cos zdz I = exp(±2fasin z). the linearly independent one. (17. can be . 17.17. Choosing y(z) = A(z)e2hsinz. and substituting this into Eq.28) A" ±4hcos zA' + 2h[ q =f sin z -\ A.30) D A): COS « = 4z + 1 iq sin 2.3 Schematic picture of Mathieu equation eigenvalues. 2 (17.

of course.35) not yet taken care of is set equal to zero.356 CHAPTER 17. .(°) + A^ is the approximation of A to that order provided the coefficient of the term in (17.Q)Ag + ( ? .34) The leading approximation A^ = Aq therefore leaves uncompensated terms on the right hand side of Eq. ? . (9. 1 + 4 ) ^ + 4 + (^. i.e.9-4) = (9 + l)(9 + 3).g + 4) .\z To lowest order the solution A is A^> = Aq. 2 A A = («.35) But since and so a term iiAq+n on the right hand side of Eq. (9-l)(9-3).32) cos^'+i) [\K .9) = 0. ? .4)A. ^ ) A + ( ? . 2[(92 + l) + A]. (17. (17.30) can be cancelled out by adding to A^> a new contribution — fiAq+4i/2i — except.^ 6T [(«>« + 4 ) A ? + 4 + (Q. (17.9) (9.36) Then ^4.33) (17._ 4 . 2 /i D^Ag = 0.35) can be cancelled out by adding to A^ the next order contribution A& = 1 (g.4 ) ^ .30) amounting to Ri ] q = .9-4) -\ — Aq-4 (17. Aq+4 2Jh . D{q%Aq+4i = 0. The solution Aq(z) of the first order equation Dq Aq(z) — 0 is c o s ^ 1 ) ( \K+\Z Aq{z) = sml(9+i) ( ivr + \ z ) COS 2 9 |( -D I\n+\Z (17. With some algebra one finds that 16^ + where (9.4 ] . (17.9 + 4) = (9. Periodic Potentials obtained by changing the sign of z or alternatively the signs of both q and h throughout. (9. when i = 0 — the terms not involving Aq in (17.

35) and (17. can be taken care of by adding to A^ + A^ the next order contribution A^2\ where A (2) 2h u 2 (q. q.. A(z. Then for A = A(0)+A(D+A(2) + ..39) which is the equation from which the quantity A and hence the eigenvalue A is determined. . h) = A(-z.4. except those involving Aq.4. (17.g) 26/i L_ 2 13 /i 2 ^±A{q + ^q) + —1 ^A{q^q) '17. h). (17.4) (q . . (q.40) .q 8)Aq+8 (o)' ^±V(q 1 4.g + .8)A.37) Here all the terms.g-4) 4.q + 4) (g. . q. q . (17. A.)+ & ! = « ( . to satisfy Eq. the sum of the terms in Aq in Eqs.30) amounting to RW 9 = fag+ — 2?h 1 ^ 2™h? +< 4 ) R(o) \ . -h). q (17.q 4)Aq+4 q + 4 1 q + + + + + <^(.39) is obtained by changing the sign of z or alternatively the signs of both q and h throughout. q ..g + 4)(g + 4. Thus in particular A(z.38) Aq-8 + -1 -2 Clearly we can continue in this way.<J+4 Aq+& + 1 2 1 . h) = A(z. Thus 0 = (g.4) -1 -1 (q._ 8 (17.17.30) and hence y = Ae2hsmz to be a solution of the Mathieu equation.37) and so on — left uncompensated — must vanish. q. q . +4. an associated solution y — A(z)e~ with the same eigenvalue and thus the same expansion (17. -q. In its turn the contribution A^1' leaves uncompensated terms on the right hand side of Eq.4) (q .3 Cosine Potential: Strong Coupling Solutions 357 This equation determines the quantity A in the eigenvalue equation to the same order of approximation.4 . Q . (q. We note here again that if one solution y = A(z)e2hsmz 2hsmz is known. ) ^ 4. (17.q + 4)(q + 4.

(17.43) 9 2eh .41) so that the early successive contributions decrease in magnitude." Hence in dominant order the solution B is JB^0) = Bq[w and it is convenient to set Hei Bq(w) = 1 (9-1) H (17.358 CHAPTER 17. where Hen{w) is a Hermite polynomial when n is an integer.42) (17. The differential equation there is Eq.44) where D(B) 9 = dw2 $_ !)• (17. Before we study the eigenvalue equation in more detail. Note different definitions in use. Periodic Potentials Finally we note t h a t the above trigonometric solutions of t h e Mathieu equation are valid in the domains cos 1 4 1 -Z -7T ± » 1 2 VK' (17.i ) We encountered Hermite polynomials earlier in Chapter 6.46) 2s(«.45) {B) We recognize Dq as the differential operator of the Hermite equation.28) and write the upper equation for a solution y = Bexp(2hsinz) in new notation B" + 4/icos zB' + 2h( q-sin Changing now the independent variable to w{z) = iVhcOS I -7T + -Z I . (17. 2 w2 + H* (17. it is found that D(B) z+ — ]B = 0.67). i. a solution Bq(w) of the equation Dq Bq(w) — 0 is Bq(w) oc He^^^iw). (6. We return to Eqs. this means the solutions are valid in particular around z — 0 since then l / \ / 2 3> l/Vh ~ 0.e. .d2B w dw2 + w(l w dB_ dw d dw 1. We observe that for the large values of h we are considering here. we show that two other pairs of solutions with a similar coefficient structure and the same eigenvalue expansion can be found which are valid in adjoining domains of the variable z.) i ( Q . and discussed there.

.3 Cosine Potential: Strong Coupling Solutions 359 because then as shown in Example 17.sin z + by changing the independent variable to w(—z) = 4v/icos I -7r z (17. forms a rapidly decreasing expansion provided that \w (z)\ « Vh.. i.qi)Bq_4. also the same coefficients in the case of functions C below. Finally we obtain a third pair of solutions.47) where the coefficients are the same as before. Comparison with the case of solution A now shows that the form of that solution can be taken over here except that everywhere Aq{z) has to be replaced by Bq(w).e.48) This shows that this solution is valid in particular around z = ir/2..q + 4 ) £ g + 4 + (q.51) . i.50) C = 0. q)Bq + (q.4 (using the known recurrence relations of Hem{w)) the perturbation remainder can be linearized. y — Cexp(2hsinz) and Cexp(—2/isinz).17.e. The solution with contributions B^°\B^>. as in Eq. (17. (17.e. 16hJ (17.49) Then the dominant order solution is C^°\w) = Cq{w) with and — again choosing the multiplicative factor suitably — Cq(w) = 2i(" +1 ) ^ .7 T H 1 2 Z 4 <1. i. Again a change in the sign of z throughout yields the associated Mathieu function y = i?exp(—2hsinz) with B[w(z)] = B[w(—z)] and domain of validity in particular around z = —ir/2. which is an amazing and unique property of the Mathieu solutions — cf.34). i. the expansion for the quantity A being identical with that of the previous case.e. with the equation C" + 4h cos z C + 2h ( q . (17. 1 COS I .d2Bq + w{l dw2 "» 2 >^-(» 2 + ^ = {q.3 ) \He* ! (9+1) H. A{wz (17..

59d) of Hermite functions Hen(w).4.54) where the proportionality factor is complex. each of them associated with one and the same expansion for the eigenvalue. A closer look at that expansion is our next objective. (17.4: Hermite function linearization of perturbation Using the recurrence relations (6.n2(n- l)Hen-2 . We note that C(z. The solution y = C exp(2/i sin z) is a decreasing asymptotic expansion provided that |w(-z)| ^ v^/i. .52) and the coefficients are again the same as before. can be re-expressed as 72 = -(n + 1 ~-(q + l)Hen+2 l)He.-h). 2 J where n=-(q-l). we have a remainder •R:=w2He„+w(l Using the recurrence relations wHen(w) = Hen+i(w) + nHen-i{w). Again the solutions have the same structure as in the previous two cases and the eigenvalue expansion remains unchanged. q + 4)C 9 + 4 + (q.7 T 1 2 Z 4 <1. show that the perturbation remainder of Eq. Example 17. 2' the remainder 72.4)C g _ 4 . q)Cq + (q.q.(q-l)2(q-3)HeX (9-5)- i(« 2 + i) + iA He-i (9-1) . Periodic Potentials since then (using the known recurrence relations of He^) ~Mw2^ +w(l+w2)^ + (w2 + o dwz dw V 2 (q. (17. (17.-q. He'n(w) = wHen(w) Hen+i(w). These domains of validity are indicated in Fig.h) ocB{z. Solution: Inserting the dominant Hermite function solution into the right hand sid of Eq. Again a change in sign of z throughout yields the associated solution y = Cexp(—2hsinz) with C[u»(z)] = B[w(—z)] and domain of validity in particular around z = ir/2.(w2 + -A)Hen. i-e- 1 COS I . (17.q. 17.53) This solution is valid in particular around z = —7r/2.44). (17.44) can be linearized with the same coefficients as in the case of the trigonometric solutions. We have thus determined three pairs of solutions of the Mathieu equation. (9+3) (2n 2 + 2n + 1) + .w2)He'n .A Hen .360 CHAPTER 17.

-l)[l(q _!)]. with (. We see that (a) every coefficient Mr results from a sequence of r moves from q back to q.l)(q ..3 Cosine Potential: Strong Coupling Solutions It follows t h a t with the definition of Bq (w) as Bq(w) that He K . q . (q. q)Bq(w) + (q. ~l)Aq-4} 2h 1 { P 2 ( < Z 2 + + P2(<? lAq+i ) 8 A ) ' " ' +P2(q.q + 4) (q + 4.i ) ( w ) 2l(. -2)Aq_8} + • (17.17. The coefficients have thus been constructed in a way which now allows the formulation of a recurrence relation.q-4) (17.56) (9-4. (q.j^0 e 2ft sin z A + ~ww 1 7 i + ^7ir{ p i(9' ^ ^ + 4 + Pi(q.3 T h e eigenvalues We saw above that along with each solution the following expansion results: -2A = M 1 + ^ M where Mx = 2(q2 + l).g-4)(q-4.. ^ 1 Hv" '^ ^' ^ ' . q + 4) = (g + 1)(? + 3). 17. In addition (c) each move except the last has to be divided by the final displacement from q divided by 4.q + A){q + A. We can write the solutions obtained previously. ) .g) -1 -1 and so on. q) = 2[(q2 + 1) + A].0 and —4.)Bq-i(w).3. (q. -l)Ag-4 + P2(q. 2 + ^ M 3 + (17.' 361 in 2*f*-1>[i(g-l)]! = (q. the solution y = exp(2/i sin 2)^(2). e. r j=-r.3).1i . (?) 1 1 + (9.g. (b) The allowed moves are +4. no intermediate move to or from q being allowed. -.q- 4.57) .4) = (q . in the following compact form with coefficients Pr(q.j): y = = = e e2hsinzA{z) oo 2ftsinz r=0 V . .55) M2 = M.q + 4)Bq+4(w) + (q.

34) we deduce that Pr(q. W.e.g + 4)(g + 4. l ) .g + 8)(g + 8. Periodic Potentials PlM)=(M+i). and the contribution from a sequence of r moves from q + 4j back to g.362 Clearly CHAPTER 17. 2) 1 2 ' ' ^ ' ^ 1 (g.j). g + 4j)P r _!(g. i. as jP r 2 (g.g + 4)(g + 4.j). ._1) = Mzil. (17.g + 8)(g + 8.P 2 ( g .g + 4)(g + 4. x)_ (g.l ) . j .60) M2r+l = = *R. j) = (g + 4j . for instance.-j) = (-l)rPr(-q.i (g. and in general* M2r = J]j[P2(g. Dingle and H. 0) = 1 and all other P0{q. (17.j)Pr+1(g. Miiller [73].j).j)-^2(9.g + 4) 2 1 (g. p = (<?><? + 4 ) ( g + 4 >g + 8) .59) One can now write down formulas for M^r and M<ir+\ in terms of the coefficients Pr{q.58) with P 0 (g. B. r ^JtPr^jO-Pr+lfejV^rfe-jO^r+l^-j)] r 2^jPr(g.1) + (g + 4j. g + 4 j ) P r . M2 = P 2 ( g . P2(q 2) Pl(g.j). By the above rules we can write down the recurrence relation for the evolution of a coefficient Pr by steps from the coefficients Pr-\: jPr (g.g + 12) 1 2 3 (g. g + 4j)P r _i (g.3) ^3(9.j).4. Thus we have. From the coefficients (17. the factor j removing a duplication of factors in the denominator at the junction. .g + 4) ^3(9.g + 4)(g + 4. J.Each term in M2r can be considered as the product of the contribution from a sequence of r moves starting from q and ending at q + 4j. j) +(g + 4j + 4.g + 4) + 1 1 1 and so on.j ^ 0) = 0. j + 1) (17.-j)] = 2j]jPr2(g.g + 4 ) ( g + 4.

In fact we demonstrate in Example 17. L. . [137]. h — —h.^q(33q4 1 220 hA 1 (63g6 + 1260g4 + 2943g2 + 486) 225/i5 rg(527g6 + 15617g4 + 69001g2 + 41607) —3T76 (9387g8 + 388780g6 + 2845898g4 + 4021884g2 + 506979) —WTfs q ( 175045qS + 9702612 ^ 6 + 107798166g4 (17. This means that the in approaching such a domain the appropriate solutions must become proportional in view of their uniqueness. of course. Miiller [73].62a) f R . Ince [136]. as a result of tunneling between wells.4 The level splitting Above we obtained three pairs of solutions along with their domains of validity as decreasing asymptotic expansions. Up to > > this point. It will be seen in Chapter 18 that in the case of anharmonic oscillators.92q3 + 70g2 .2 * +3) + ^ 2 (9g4 . We observe that there are regions. *E.61) +288161796g2 + 130610637) These are the terms given explicitly in the reference cited above.3. as pointed out earlier.3 Cosine Potential: Strong Coupling Solutions In this way one obtains the eigenvalue expansion A = 363 _2h2 + 2hq-^(q2 1 . q is only known to be approximately an odd integer qo = 2n + 1. a= (8/i)3(9-x) [1(9-1)]! 1 . W. S. J.^ .284g + 57) + (17.5 the following proportionalities there: B[w(z)] = aA(z). 17. t Terms up to and including those of order l//i 5 had been obtained by others and by different methods.( 3 ^ . 17.(5q 4 + l)-^Jiq(q2 + 3) + 410g2 + 405) + 34q2 + 9) . Its precise deviation from qo. Dingle and H.* We observe that the expansion remains unchanged under the replacements q — —q. These domains in the interval —7r/2 < z < 7r/2 are indicated in Fig. where solutions overlap. Goldstein [115].4. B.17. the solutions and eigenvalue expansions again have the same type of symmetries. [116]. remains to be determined next.

We write and expand this function as follows: .32).5: Proportionality of solutions in domains of overlap Show that in their common domain of validity B = aA and C = aA.364 and C[w(z)] =a~A(z). Periodic Potentials + _(3<?+2<? + 3) + 2 15 /i 2 (9g 4 + 92g3 + 70q2 + 284g + 57) + • • • (17. Solution: We begin with the solution containing the function A{z).(17. Example 17.Z (g+1) 1 H 1!4 4 2 1 (g + l)(g + 5) .COS -7T 2!42 \4 2 COS /-l -7T V4 Z H 1 2 Z H 1 2 .62b) *-z domains of overlap Fig.4 Domains of solutions and their domains of overlap. and determine the constants a and a.1 ) ( -7T+ -z 1 — COS 2 1 \-l-^+1> 2 Z C O S S ^ . 17.7T H Aq(z) = = C O S ^ .1 ) ( -7T + ._ ^ l CHAPTER 17. a .1 4 . The first term of the expansion of A(z) is the function Aq(z) given by Eq.

the coefficient of the dominant factor cos' 9 _ 1 ^ / ' 2 (7r/4 + z/2) in the result is seen to be J (8/i)3 ( g .2g + 3) 1 [£(9-1)]! 2?h 215h2 (9g 4 .53) for the expansion. (6. \q-l)/2(w) 2i(9-D[i(.92q 3 + 70q 2 .2. with e. As we saw there.284g + 57) + • The other relation is obtained similarly. we have as even and odd solutions y±ocA(z)eZflsmz±A(z)e -2/isin. the coefficient of the dominant factor c o s ( . we have to impose the appropriate boundary conditions. with (8h)li''-1) (3g2 .36). Hence we have to construct even and odd solutions and impose the boundary conditions at z = 7r/2.17.g + 4) (q + 4.46).q) 2 7 /i 1 -1 4(g + 3) 1 (g + l ) 2 ( 9 + 3) + -29ft ti(9-l)]l i (8/1)4 ( 9 . (17.61) for the connection and Eq. (17.6 at the end of this section we derive for these \axge-h2 solutions the conditions similar to those given in various equations in Sec. (8.-3) 1!(8™2) )[i(g-l)]!L Inserting this into the expansion of B[w(z)] = B^ + BW -f • • • where w{z) — 4 / i 1 / 2 cos I -IT + -z J.3 Cosine Potential: Strong Coupling Solutions 365 Inserting this into A = A(°> + A' 1 ) + • • •. Since the solution with A is obtained from that with A by a change of sign of z.i ) 1 (g.g. i.63) In Example 17. we obtain Bq[w(z)}- H.2 for the case of small-/i2 solutions and thus insure that in both cases the same boundary conditions are obeyed.-!)(. This is essentially the expansion of the parabolic cylinder function with the exponential factor removed — see Eq. 17. In order to link our asymptotic solutions with the Mathieu functions of integral order defined earlier in the consideration of small values of h2. s .i ) [|(9-1)]! It follows therefore that over their common range of validity B = aA with a = b/a. these are conditions imposed on even and odd solutions at the point z = it/2._i)]! w*te-» 2l(91 (. (17. These are the conditions of Eqs.g-4)fa-3) -1 1!4 / I V/i 2 = 1 (g-l)(g-3)2 2 9 /t Similarly substituting the asymptotic expansion of the Hermite function^ into the expression (17.e.22). For further details and explicit expressions of higher order terms we refer to Dingle and Miiller [73]. _ 1 " 2 ( 7 r / 4 + z/2) in the result is seen to be a = 1 2 /i 7 fa. A*-1) given by Eq.

22) are therefore given by the following boundary conditions: 1T y'+ v+fa-o I'=° y- a a\dwjw=0 B[w = 0]c2h a a -1h dw\ ! ' .62a) and (17.64) These are now the solutions around z — ir/2 which permit us to apply the boundary conditions (17. Periodic Potentials We can extend the solutions (17.66) ^E. The second of relations (17.63) to the domain around z = ir/2 by using the proportionalities (17.e.66) then follows from the recurrence relation (6.65) For the evaluation of these conditions we require the following expressions involving Hermite functions which we obtain from Tables of Special Functions:^ ^- ( =0) = 2TT2I//2 Fi(^ijii' He2n(0) = ( Ur w=0 •W+m' (17. Magnus and F. He'v(0) = -Hev+1{0). a ± C[w{z)}c_2hsinz^ a (17. and then replacing 2n by i>.59d). . Oberhettinger [181]. The functions defined as in Eq. Here v is the not necessarily integral index of the Hermite function Hev{w) with Hermite equation ( -j-ir -w—+v)Hev(w) \ aw1 aw J = 0.22) and to correlate the solutions to those for small values of h2. 80. i.62b).» a \dw _ C[w = 0}c_2h a w=0 = Q (17.66). He'u{w) = wHev{w)-He„+1{w).e. i.366 CHAPTER 17. (17. Then y± oc ^ M e ^ i n . W. one obtains the first of relations (17. give the following expressions for polynomials with argument zero: ~19)"(^)!. p. With the help of the duplication and inversion formulas of factorials. Thus y+ = ce and y_ = se.g. He2n+1 (0) = 0.

68) Inserting here expansions (17.67a) and (17.39 2~^h? (17.67a) g -106g -87 2 14 /i 2 (17. we have cot j ^ ( g ~ 1)} = ~\<(l .106g .7. f^(16h)i/2e-4h 2 [Uq-1)]\ 3(g2 + l) + 2^h2 (9g4 .67b) <f ~ 82g^ .40g3 + 18g2 .87 4 2 ¥w (17. B[w = 0] 2TT _ _g_ ¥h+ g 24/* _ JL g4 .3 Cosine Potential: Strong Coupling Solutions 367 With these expressions we obtain by insertion into B and its derivative.17.14) from which one obtains 1 (q + i) TV cos{f(g-l)}[i(g-3)]!' and the duplication formula (2Z)\\/TT = z\(z — l/2)!2 2z . and similarly for C and its derivative. and using the reflection formula (8. the solutions would be given by q = go = 3.67c).-» 2 6 /i the condition reduces to the equation cot{ .65) we have B[w = 0] _ a C[w = 0] ~ ~ ^ -4/i (17.67d) Considering now the second of conditions (17.1) 7T .11.g0)3].39 (17. from which one obtains "1 1) 4<*-3) 2(9-!)/2 2 ^ \b.69) If the right hand side of this equation were exactly equal to zero.\4 q + m { 4« VJ-L ^[±(g-3)]!sin{f(g-l)} [J(ff-l) "(^L-^-»{^-': _ _q_ ¥h + g .136g + 9) + • • • .( g .! ) ] ! [ . .82g2 .67c) 4 2 \dw)w=Q C[w = 0] [\(q . Hence expanding the cotangent about these points. (17.Qo) + 0[(q .

40gg + 18gg . Thus we have 2 (16/i) qo/2 T2 -4. Considering now the first of conditions (17.h q-qo = * [£(*>-!)]! 1 3(902 + 1) 26h +^^{9q^ . .528gg + 3307Q .136g0 + 9) 2 13 /i 2 1 4 2 19 3f(9gg . The remaining condition (17.70) except that now qo = 1. we use \(q)~\(q0) + (q-q0)(^J . = _a a 6 _4h (17. Periodic Potentials ' " Vn [i(go-l)]' 3(g02 + 1) 26/i . (17.65) leads to the same result except for a change in sign.136q + 9) + .70) + ^ 2 (9?o . i..40g3 + 18q2 .368 Hence we obtain q-q0 a [2(16h)i°/2e-*h CHAPTER 17.5.72) This leads again to the result (17..65) we have (dB/dw)w=0 (dC/dz)w=Q Proceeding as above one obtains tan{-(q-l)\ 7T .136g0 + 9) + • Clearly the last of conditions (17.65) yields the same result except for a change in sign.73) We now return to the eigenvalue X(q) and expand this around an odd integer qo.(17. .120g£ + 467g .40gg + 18gg .e.408g0 + 1089) 2 /i +• (17.71) = 2 13 /i 2 fc{l6h)i/2e-4h 1 3(g2 + 1) 2 [4(5-1)]! L (9g4 ..9..

A_(g0) ^ . (17. p. results from the boundary conditions.1. Arscott [11].17. made evident by the splitting of the asymptotically degenerate (harmonic) oscillator levels. W. (17. (17.3 Cosine Potential: Strong Coupling Solutions Differentiating \(q) of Eq.e. and that for the odd Mathieu functions se qo+ i and se9o (lower. quite apart from applications. solution of the Schrodinger equation for the basic and nontrivial periodic potential is important for a thorough understanding of its quantum mechanics. We see here that *R. i...e.t From Eq. 120. The dominant contributions were also found by S. The tunneling effect.3. J.3.61).74b) The above results for the cosine potential are in many respects important. though approximate.88q0 . + ^ 2 (9<?o + 89o " 78g02 . plus sign) give the so-called level splitting AA(go) as a consequence of tunneling indicated schematically in Fig. In the literature one finds frequently the statement that the splitting is a nonperturbative effect. example 3. B. 17. M. n ' r 1 / -rrre 4/l ^odd(go) Seven (<Zo) Wfaqo-m (16/i)T n\V2n e~4h. Miiller [73]. .3 we call this difference 26E„n. (17. and not the separation of these as a result of tunneling. In particular we saw that the perturbation expansion yielded only the degenerate eigenvalue expansion. 20. *In Sec.2.74a) The difference between the value of A (earlier called an and 6 n +i) for the even Mathieu functions ce?0 or ce go _i (upper. In the literature that we follow here* in all cases several more higher order terms have been given.^ ( 3 g 0 2 + 8go + 3) A ( Q 0 ) T (87r)V2[i ( g o -i)]l L2 w . minus sign)..87) = Azpteo). Goldstein [116]. 9o = 2n + 1. The complete. i. Effectively the degenerate eigenvalue expansion results from the anharmonic terms contained in the power expansion of the cosine potential. n = 0.74a) we obtain for the level splitting in dominant order* x r ^ \ / N 2(16/i)>+1 _4h A+(go) . this becomes 369 = (16/i)^+1e-4h l . . See also F. just as any solvable problem is important for the insights it offers in a concrete and transparent form. Dingle and H.

the Schrodinger equation with the cosine potential is a special case of several other more general cases. In fact the perturbation theory developed in this context and tested by application to the cosine potential is of such generality that the recurrence relation of the coefficients of the eigenvalue expansion can be looked at as a difference equation. The exponential factor represents. The splitting can also be obtained with some methods of large order of perturbation theory. .74b). results from the imposition of boundary conditions.2. Example 17. and one can compare the methods. The associated or modified Mathieu equation with cosh z instead of cos z is another important equation which we shall study in detail in Chapter 19 in connection with a singular potential. (17.2: y+(±ir) J4(±TT) y+{z±n) y-(z±7r) = = ± —y+(z)± 2/+(0) . (17.370 CHAPTER 17. As mentioned at the beginning.6: Translation of Solutions For the large-h 2 even and odd solutions (17.32) we obtain Aq(z ± TT) = (-l)±i<-9^'Aq(z). the exponential factor we encountered in the semiclassical method where it has the structure of the factor exp(classical action/ft). Thus we do not reproduce every step. y+(0) 2/_(0) Solution: The derivation requires a cumbersome tracking of minus signs. which one can then try to solve in order to obtain the behaviour of the coefficients of the late terms of the expansion. 17. We begin with the solution containing the function A(z). We shall not consider these in detail here but do consider briefly the elliptic potential in the next section without going into extensive calculations which can be looked up in the literature. The splitting can also be derived by the pathintegral method as will be shown in Chapter 26. T y'_(0) y'A±*) y-{z).63) — and using only the leading terms — derive the following set of equations with shifted arguments similar to those of the small-/i 2 case of Sec. . From Eq. M±TT) . in effect.74b) is also important for various other reasons. We consider for simplicity only the dominant contributions and replace throughout —1 to some power by exp(i7r) to that power and combine such terms into cosines and sines. _3/+(z) + ____2/_(z). Then we obtain first y±(z) = yA(z)±y^(z). yA{z) = Aq(z)e2hsin '. evident through the exponential factor exp(—4/i) in Eq.§ The boundary conditions we imposed here are those of a self-adjoint problem which has real eigenvalues. Naturally one expects the solutions there to be related to those of the periodic case considered here. y^{z) = Aq(z)e~2hsin *. ^ ± n ) = _ ( _ 1 ) T i(<JTl) A 9 ( 2 ) (the extra minus sign in the second equation coming from Aq(z ± 2-7r) = — Aq(z)). The explicit derivation of the level splitting (17. Periodic Potentials this nonperturbative effect.

z ) .c o s | ~(<1 T 1) f y .VA DV 371 1 V±.1 ) | .T ( ? =F 1) p + (0).) cos < -7r(q 1) [ = .q).4ft) sin j ^n(q .4ft) ~ -8(q .g A ? ( z = ±TT).31). Aq(z) The derivatives can be handled with the help of Eq. 17. y+(0)~2V2.y-]z=„ = [y+yL .1 Elliptic and Ellipsoidal Potentials Introduction Our intention here is partly to deal with other periodic potentials.Ah) ~ 32ft.4 17.[A'qM + 2hAq(n)] = iV2(q . where Aq(0) = Aq(0) = V2.7 r ( q = F l ) } j / . — j / + ( 2 ± 7 r ) = ± i s i n | ^ ( g = F l ) U + ( z ) . (17. The Lame equation has not been a widely known equation of mathematical physics so . etc.( 2 ) .7 r ( g ^ l ) U + ( z ) q : i s i n { . Finally for the derivative of the odd solution we obtain y'_{ir) = K ( T T ) . since A q (0) = \/2 = Aq(0): y+(±n) = ±isin|^(9=Fl)J3/+(0). we obtain — with Aq{z + 7r) = (-1)^ ^ 2 A .4.17. the operator having solution Thus ^ ( 0 ) = -(q/2)Aq(0) = -A~'q(0) and A'q(z = ±TT) = ~qAq(z -qAq(z = ±7r).4ft) cos i -TT(<J + 1 ) 1 . but also to enable a brief familiarization with the Lame equation. (-l^ifr^i^O + C-l^ifo^j/aCO cos | . we obtain the conditions stated at the beginning.( ± 7 r ) = c ° s | .4ft.y-V+\z^ = -[j/+(0)] 2 (<? . Similarly we find y-(z±ir) = = and by putting z = 0: V.. From the equations for the sine and cosine we can deduce the value of the Wronskian at z = n in the leading approximation for large ft: W[y+. Replacing sines and cosines by the functional expressions.%/2(g . A' (z = ±ir) = --(.• Then y'+ (0) = 0.( z ) From this we deduce for z = 0. A ?q( Z = ±TT) = .2hAq(n)] . ( . y'_ (0) = \/2(4h . which is easiest by comparison with the foregoing treatment of the Mathieu equation. = ±ir).3 The Ellipsoidal Potential Replacing VA. and y+(z = ±TT) = V2(q .

Periodic Potentials far.0 and .* and the work of both prepared the ground for present-day investigations. Erdelyi [85]. but also cubic potentials. Arscott [11] (p. J. § F .0.372 CHAPTER 17.98. W.0. H. Arscott [11].fi2A. Liang. p. the new stage of the development of the investigation of the Lame equation was really initiated by Ince^ around 1940.0. for general n the solution of Lame's equation is not single-valued owing to the singularities in the finite part of the u-plane. But more recently it has been observed to arise in various contexts. inverted double well and cosine potentials (J.. Ince [138]. soon encounters grave difficulties and there has not been developed up till now any general theory of Lame's equation at all comparable with that of the Mathieu equation . however.. This line of investigation. (17. 194) remarks to parallels with the Mathieu equation:"..75) ' T h i s means double well. .5 The potential sn2(z. Tchrakian [165]). so that Floquet's theorem cannot be immediately applied . and was continued by Erdelyi..K2sn2u . 19. The most conspicuous difference compared with the Mathieu equation is.4sn4u]y = 0. 194) remarks. k) for k = 0. 17.5. as alluded to at the beginning. the non-occurrence of the Floquet exponent. H.75. *A. M. L.1.* As Arscott [11] (p. In particular the Lame equation was recognized to arise as the equation of small fluctuations about instanton solutions for practically all basic potentials... Muller-Kirsten and D.5 < z < 5. If one separates the wave equation § V 2 * + J2* = 0 in ellipsoidal coordinates (which we do not need to consider here).. one arrives at three equations of which one is the ellipsoidal wave equation pt + [A ." Fig. f E .-Q.

75) reduces to Lame's equation and one writes K. where n is real and > —1/2 (and n is an integer in the case of solutions called Lame polynomials) where k.e.\k\ < 1. In order to distinguish the above equation from that with the Lame potential. for very > high barriers (harmonic oscillator approximation around a minimum of the potential).K2sx?u}y = 0. 17.2k.76) which can be looked at as a Schrodinger equation with periodic potential K2sn2u where snu is one of the Jacobian elliptic functions of period 2K. i. K being the complete elliptic integral of the first kind.2 Solutions and eigenvalues In the following we sketch the main points of the method of deriving asymptotic expansions of the eigenvalues and eigenfunctions. in front of the second derivative has to be kept in mind (mo being the mass). and of the derivation of the level splitting'. (17.77) where q —> go = 2iV + 1.3 The Ellipsoidal Potential 373 Here K2 and A are separation constants and U2 = l2(a2 — c2). (17. . N = 0.17.2 = n(n + l)k2.78) We follow in this brief recapitulation the description in H. For barriers of finite height the parameter q is only approximately an odd integer qo in view of tunneling effects. . 17.e. we refer to the potential consisting of the two terms with sn 2 n and sn 4 n as the ellipsoidal potential. is the elliptic modulus of the Jacobian elliptic functions snu. . J. 2 . Although the results have been calculated for the ellipsoidal wave equation.a > b > c being related to the lengths of the three axes of the ellipsoid in a Cartesian coordinate system. The second step is to insert (17. Miiller-Kirsten. The first step is to write the eigenvalue A as A(q.K) = qK+^±. Eq.5 for several values of fc. we consider mainly the Lame equation. W. Jian-zu Zhang and Yunbo Zhang [206]. i. cmx and dnu. (17. the equation ^ | + [A . The range of the independent variable u is 0 < u < 2K. (17. the usual factor —h2/2mo. in the case n2 — oo.76) and to write the solution y = A(«)exp | f Ksnudul = A(u)[f{u)]K.4. In comparisons with the Schrodinger equation. (17. . If we put fl = 0.77) into Eq. The function sn2u is plotted in Fig.1.

ll^ ^A(u)[f(u)]^k±A(u)[f(u)}-^k.K — > A(u)=A(u + 2K). (17. more precisly for dnu = cnu F 1 » -. A into equations in terms of the variables z(u) = ^—{k' .82 . k' = Vl~ k\ \dmi±cnu/ 17.q.Es(ii).81) Since these expansions are not valid at the extrema of the potential (where the boundary conditions are to be imposed). i. A. i. determine their proportionality factors) in domains of overlap (their extreme regions of validity). dnu =F cnu . A second solution is written y = A(u) exp < / nsnudu >. C which are valid for dnu±cnu <1. B and C. . A(u.e. _ Solving the resulting equations iteratively as before. (17.79) The very useful property of these solutions is that for the same value of A (which remains unchanged under the combined replacements q —>• —q..-q. by transforming the equations for A.-K). Periodic Potentials . ^ (17.80) The domain of validity of these solutions is that away from an extremum of the potential. . / -z J . \/8K (dmz = cn« \ ' F .e. A are derived.K)=A(u.374 where CHAPTER 17. the other in terms of those of an imaginary variable. (17. dnu± cnu K Thus one can construct solutions Ec(u). /dnu + kcnu\ \ dnu — kcnu J For large values of K the equation for A(u) can be solved iteratively resulting in an asymptotic expansion for A(u) and concurrently one for the remainder in Eq. one has to derive new sets of solutions there and match these to the former (i. Thus in the third step two more pairs of solutions B. C replacing yl. but solutions B. one pair in terms of Hermite functions of a real variable.77). one obtains again the same expansion for A.e. which are respectively even in u (or snu) or odd.

85) The first three terms of this expansion were first given by Ince [138].3 T h e level splitting In the fourth and final step one applies the appropriate boundary conditions on these solutions. Tricomi [87]. (17.e.83) Then Ec(ti) Es(it) KB[fM^(u)r/2fc±C{zH] a uy n a n S/Ms Each of the solutions thus derived is associated with one and the same expansion of the eigenvalue A. Es(2f0 = Es(0) = 0. 17. Erdelyi. Oberhettinger and F. i. Miiller [205]." A = qK-±(l 1 + r {(l k2)(q2 + l)-^-{(l + ifc2)3(5g4 + 3 V + 9) + k2)2(q2 + 3)-4k2(q2 + 5)} 2WK2 -Ak2{l + k2)(5qA + Uq2 + 9) . Magnus.6(5g4 .4. £2 = 0).63)} + ••• . one sets at it = 0 and u = IK altogether** Ec(2A") = Ec(0) = 0. (17.38g2 . G. W.384n2k4(q2 + 1)} q r {(l + A. p. C = a~A. F. 64. J.^ ^ { ( 1 + k2)5(63q6 + 1260g4 +2943g2 + 486) -8k2{l + k2)3(49q6 + 1010g4 + 1493g2 + 432) +16fc4(l + fc2)(35g6 + 760g4 + 2043g2 + 378) -64fi 2 fc 4 (l + fc2)2(5g4 + 34g2 + 9) + 256f22£. This expansion is found to be in the more general case of Eq. a of _ _ B = aA. *A. who obtained this expansion for the eigenvalues of Lame's equation (i. .e.75).17. the ellipsoidal equation.3 The Ellipsoidal Potential 375 In their regions of overlap one can determine the proportionality factors a. (17. as well as \duJ2K 11 (17.2)4(33g4 + 410g2 + 405) -24/c 2 (l + k2)2(7q4 + 90q2 + 95) + 16fc4(994 + 130g2 + 173) +512f22fc4(l + k2)(q2 + 11)} . W.86) \duj0 o h r 2K > \duJ =[-sr 0 =0\duj (17-87) H.

Muller [205]. from which the difference q . go being an odd integer.40g03 + 18g02 .l n 2^+ i J + 0(«-i) (17.89) one obtains the eigenvalues from which the level splitting can be deduced. .4k2(q2 + 2g0 + 5)} + • • • ].91) tt H. J.376 CHAPTER 17.i Here the upper sign refers to Ec*> l or Ec£°.1 / 2 (2TT) 1 /22« l + (l_fe)^«( i .90) + 3.j 26K2 (1 85) = A(g0) + (q . Periodic Potentials These conditions define respectively functions EcX°. + fc )(3g + 8g0 + 3) k2)2(9q% + 8g^ .:J {3(1 + k2)2 (9q* .go)« qo(l + k2 22K {3(1 + k2)2(q2 + 1) .2 n K 2 {3(1 + 1 +128fc2(2g^ + 9ql + 10g0 + 15)} For the two lowest levels go = 1 and one obtains for their separation AA(1) 2(4^) 3 / 2 (l-fc)*.88g0 .2 /c +256fc2g0(g2 + 5)} (17. One obtains with go an odd integer Q-Qo =F2 2fl + ir\l-kj 3(g02 k\-K/k 8K qo/2 \l-k ) 2 1 [i(g0-l)]!L + l)(l + fc2) 25 K ^ 2 J J .136g0 + 9) n 3.87) (17. 2K and 4K respectively.g0 is obtained by expansion around zeros.78g^ . 2K. Finally expanding A(g) ~ A(g0) + ( g .Es^ 0+1 .g 0 ) ( ^ . and the lower to Es*>+1 or Es*5. One finds^ A(?) A(g0) T :(l 2K 2 2 fl + k 7T V 1 — k 2 -njk 8K -k 2 yo/2 I 1 Hqo-m 25. Evaluating these one obtains (from factors of factorials in q and —q) expressions cot{-7r(g — l)/4} = • • • and tan{7r(g — l)/4} = • • • (in much the same way as in the case of Mathieu functions). (17.Ec£ 0_1 and Es2° of periods 4K. W.

* Apart from the choice of notation. however.86).74a). that calculations with the Schrodinger equation are not only easier. J. Replacing u by x ± IT/2. Ince [138]. . *H.92) reduce the periodic ellipsoidal wave functions and their eigenvalues to corresponding Mathieu functions and their eigenvalues.4 Reduction to Mathieu functions Under certain limiting conditions the ellipsoidal wave equation reduces to the Mathieu equation. it = x ± ^ . Thus in the case of the dominant contribution of Eq. h .2h2 . (17.Ah2 sin2 u}y = 0.17. (17. i. Miiller [206]. but also more easily generalizable.* M G .2h2 = A.4 h ( i e M 9 0 7 2 \.J1 O. (17. .3 The Ellipsoidal Potential 377 This result agrees with a result of Dunne and Rao** who calculated this expression using instanton methods. Thus if k — 0 and n —> oo in such a way that > K2 = n(n + l)k2 ~ finite. L.2h2 cos 2x}y = 0. K = 2/i. this equation becomes y" + {A . Without going into details we mention again (as at the beginning) that there is also a specialization from the ellipsoidal wave equation and its solutions to spheroidal wave equations and their solutions. A = 0. W.4.f ln[(l + fc)/(l-fc)] _ e -4h) ( 8" \ ^^ __ (16h)«. the conditions (17./2i and hence A^ A/ A AU I2 . 17. (17.e.93) (17.90) one has in this limit: /l±±\ ~K/U = e . in which they dubbed the classical configurations Lame instantons. A(g)->A(g0)T4/i\/-e ^ [Hqo ~ !)]•' in agreement with Eq. Dunne and K.87) agree with those of Ince. + E. K2 = h2 (say). then snit — sin u and Lame's equation (f2 = 0) becomes > y" + {A . V. Rao [79]. A . One can see. Hence the conditions fi = 0.t One can verify that under the conditions stated the results of this case of the ellipsoidal wave equation reduce to the corresponding results of the Mathieu equation.

** The associated Mathieu equation appears also in string theory in connection with fluctuations about a L>3-brane (see Chapter 19). H. A generalized associated Lame potential has been considered by A. thus revealing an unexpected significance of this not so wellknown equation of mathematical physics. W. N. Khare and U. Miiller-Kirsten and D. J. Sukhatme [148].e. Gu and S.-W.§ This is therefore a very important equation which in the limit of infinite period becomes a Poschl-Teller equation. for instance. Tchrakian [165]. § . In particular we shall encounter the Lame equation as the equation of small fluctuations around classical configurations associated with cosine. **Y. in the problem of two parallel solenoids the lines of constant electromagnetic vector potential | A| are elliptic with the Hamiltonian separating into a Mathieu equation and an associated Mathieu equation.g. J. W. See e.378 CHAPTER 17. D. double well. all basic potentials. Periodic Potentials 17. Liang. inverted double well and cubic potentials. A. S. M. Quian [123]. J . Kan and K.5 Concluding Remarks The potentials we considered above will re-appear in later chapters. Rana [287]. Ganguly [103]. Miiller-Kirsten and J. A host of related elliptic equations has recently been discovered and studied. de Veigy and S. A.-Y. . H. i.Q .^ The equations considered above also appear in diverse new problems of physics. 1 1 See Z. Thus. H. See Chapter 25. Ouvry [276] and Jian-zu Zhang. Cho. Shiraishi [49]. also because the role played by the Floquet exponent in this problem is not yet well understood Jl Another recent appearance of the Mathieu equation is in the study of the mass spectrum of a scalar field in a world with latticized and circular continuum space. This is an interesting problem.

W. T. In the cases treated most frequently in the literature the anharmonic *We follow here largely P. Miiller-Kirsten and A. T. Priedberg and T. [19].1 Introductory Remarks The anharmonic (quartic) oscillator* has repeatedly been the subject of detailed investigations related to perturbation theory. 379 . in nearly a thousand of physics articles the problem of the anharmonic oscillator was touched in one way or another. M.-Q. Bender and T. J. The recent work of R.* The fact that a main part of their work was concerned with the calculation of the imaginary part of the eigenenergy in the non-selfadjoint case which permits tunneling.". H. Turbiner [273] remarks: "It can not be an exaggeration to say that after the seminal papers by C. In particular the investigations of Bender and Wu. f C . Lee [98] referred to it as a "long standing difficult problem of a quartic potential with symmetric minimd'1. D. Bender and T. ' T h u s A. This lack of popularity of the calculation of complex eigenvalues even in texts on quantum mechanics may be attributed to the necessity of matching of various branches of eigenfunctions in domains of overlap and to the necessary imposition of suitable boundary conditions. Liang and H. Miiller-Kirsten [163].t which related analyticity considerations to perturbation theory and hence to the large order behaviour of the eigenvalue expansion attracted widespread interest. This seemingly simple problem revealed extremely rich internal structure . Wiedemann [4] and a revised version of parts of this reference by J. There is no end to this: An entirely new approach to anharmonic oscillators was recently developed by M.. demonstrates that derivations of such a quantity are much less familiar than calculations of discrete bound state eigenenergies in quantum mechanical problems. Achuthan. Wu [18]. J.. Wu. W.Chapter 18 Anharmonic Oscillator Potentials 18. both of which make the calculation more difficult. Weinstein [281].

The three different cases are: (1) Discrete eigenvalues with no tunneling: In this case V{z) = \\h'\z* + \\<?\z\ (2) Discrete eigenvalues with tunneling: In this case. (3) Complex eigenvalues with tunneling: In this case. which are nonetheless linked as a consequence of their common origin which is for all one and the same basic differential equation. Anharmonic Oscillator Potentials oscillator potential is defined by the sum of an harmonic oscillator potential and a quartic contribution.380 CHAPTER 18.1 The three different types of anharmonic potentials. To avoid confusion we specify first the potential V(z) in the Schrodinger equation dz2 + [E-V(z)]y(z) =0 for the different cases which are possible and illustrate these in Fig. 18. described as the case of the double well potential.1. V(z) = \\h*\z2 I 2i 4 \cr\z . and thus lead to very different physical situations. with the potential described as an inverted double well potential. V(z) = -\\h*\z* + \\<?\z*. These contributions may be given different signs. *~z (1) (2) (3) Fig. 18. Case (1) is obviously the simplest with the anharmonic term implying simply a shift of the discrete harmonic oscillator eigenvalues with similarly .

and this behaviour is that of asymptotic expansions our treatment largely terminates this much-discussed topic. as is sometimes hoped. The imaginary part will be rederived from path integrals in Chapter 26. The eigenvalues of this case are given by Eq. . (18. Case (3) is seen to be very different from the first two cases. In the case of the double well potential our aim is to obtain the separation of harmonic oscillator eigenvalues as a result of tunneling between the two wells.86) below. It is this expansion which led to a large number of investigations culminating (so to speak) in the work of Bender and Wu who established the asymptotic nature of the expansion. (18. except for a change of sign of |c 2 |. The result is an expansion in descending powers of h?. Since these exponentially small contributions are related to the behaviour of the late terms of the eigenvalue expansions (as we shall see in Chapter 20). i. We therefore consider in this chapter and in Chapter 20 in detail some prominent examples and in such a way. however with decay as a consequence of tunneling. The question is therefore: How does one calculate the eigenvalues in these cases from the differential equation? This is the question we address in this chapter. Thus we are mainly concerned with the double well potential and its inverted form. We begin with the latter. We do not dwell on Case (1) since this is effectively included in the first part of Case (3). The eigenvalues of this case are given by Eq. The shift of the eigenvalues is best calculated with straightforward perturbation theory. In this case our aim is to obtain the aforementioned complex eigenvalue. it will not be possible to obtain the exponentially small contributions with convergent expansions. and we present a fairly complete treatment of the case of large values of h2 along lines parallel to those in our treatment of the cosine potential in Chapter 17.e.1 Introductory Remarks 381 normalisable wave functions.175) below. Case (2) is also seen to allow only discrete eigenvalues (the potential rising to infinity on either side). that the general applicability of the method becomes evident. Calculations of complex eigenvalues (imaginary parts of eigenenergies) are rare in texts on quantum mechanics. The boundary conditions are non-selfadjoint and hence the eigenvalues are complex.18. The level splitting will be rederived from path integrals in Chapter 26. however the central hump with troughs on either side permits tunneling and hence (if the hump is sufficiently high) a splitting of the asymptotically degenerate eigenvalues in the wells on either side which vanishes in the limit of an infinitely high central hump. If the barriers are sufficiently high we expect the states in the trough to approximate those of an harmonic oscillator. since the potential decreases without limit on either side of the centre. This type of potential allows tunneling through the barriers and hence a passage out to infinity so that a current can be defined.

3) "1/2 harmonic oscillator h8/25c2 . Anharmonic Oscillator Potentials 18. (18. We take here h = 1 and the mass rriQ of the particle = 1/2. h 4 2 2h1> ~1/2 2c?. we can pass from one case to the other by making the replacements: 4 E1/2 = 2Ei.1 and more specifically in Fig. The potential in this case is given by V{z) = -v(z). If suffixes 1/2.2 The inverted double well potential with (hatched) oscillator potential.2 18.382 CHAPTER 18. 18.1 refer to the two cases. 18. v(z) = -^h4z2 + ^c2z\ (18.2.2. . a point which has to be kept in mind in comparisons.z Fig.2) We adopt the following conventions which it is essential to state in order to assist comparison with other literature. and the Schrodinger equation to be considered is C L^ + [E + v(z)]y = 0.1 The Inverted Double Well Potential Defining the problem We consider the case of the inverted double well potential depicted as Case (3) in Fig.1) for h4 and c 2 real and positive. This implies t h a t results for THQ = 1 (a frequent convention in field theory considerations) differ from those obtained here by factors of 2 1 / / 2 . 18. (18.

i.18.7a) The problem then reduces to that of the pure harmonic oscillator with y(w) a parabolic cylinder function Dn(w). \h?\ — oo and c finite. (18.8) h8 v"{z±) = h and V(z±) = 5 2 2 c ' 2 2 Thus for c > 0 and relatively small.2.1.4) we can rewrite Eq... (18. to match these in domains of overlap. (18.5) (18 6) "^'-"dw* ' * 2' 2 In the domain of w finite. The positions z± of the maxima of V(z) on either side of z = 0 in the case c2 > 0 are obtained from h2 v'(z±) = 0 as z± = ± — with (18. then to specify the necessary boundary conditions and finally to exploit the latter for the derivation of the complex eigenvalue.-1-/7 - Vg(w)y(w) = o(J^j.2) as Vq(u >)y(w) = with T)Jin\ 1 (A + c2w )y(w 0 (18.2. (18.7b) The perturbation expansion in descending powers of h suggested by the above considerations is therefore an expansion around the central minimum of V(z) at z = 0. y{w) oc Di. /i). n = 0. and a variable w defined by setting E= -b* 1 + A A and w = hz. and h large the eigenvalues are essentially perturbatively shifted eigenvalues of the harmonic oscillator as is evident from Fig..d2 . _i\(w) and q = qo = 2n + l. 18. The result will be that derived originally by Bender and Wu. although our method of matched asymptotic expansions here (which parallels that used in the case of the cosine potential) is different. the harmonic part of the > potential dominates over the quartic contribution and Eq. (18. 4 .e. The problem here is to obtain the solutions in various domains of the variable.5) becomes .2 The Inverted Double Well Potential 383 Introducing a parameter q and a quantity A = A(qr..

10) E= -qh2 ^ j.9) z =0 these Thus these c2z4 1/2 + • 2 (18.+ ^ .9) we obtain 1 h4z2 c2z4 A L2 (18.11) V = 0+ dz2 + 2qh + W The solutions in terms of parabolic cylinder functions are valid around and extend up to z ~ 0 ( l / / i 2 ) . as encountered and explained earlier.2. y 4 2/i Here again q is a parameter still to 2 determined from boundary conditions.13) — that one equation (of the two alternatives) follows from the other by changing the sign of z throughout. This construction is simplified by the consideration of symmetry properties of our solutions which arise at this point. Before we return to solutions we derive a new pair which is valid in the adjoining domains. Inserting (18. h) is obtained from the perturbation expansion of the eigenvalue.384 CHAPTER 18.4(2.13) + 2qh u2 A + W A{z) Later we will be interested in the construction of wave functions which are even or odd around z = 0.| 1 d r A!{z)±iA{z)±{ 0.14b) . We observe — before touching the square roots in Eq. as we shall see.11) y(z) = A(z)exp ± z dz h4z2 4 + E hAz2 + cV y(z) = 0 (18. •A~2 hrz" 2^-\ c^z V2 —— + (18. This observation allows us to define the pair of solutions yA(z) = .12) Then A(z) is found to satisfy the following equation r /iV c V V/2 A"(z)±2<|-^.10) into (18.2 A (18. (18. Anharmonic Oscillator Potentials T h r e e pairs of solutions 18. be and A = A(q.) exp Z ifdz[ifdz[- h4z2 4 h"z2 + + cV cV 1/2' (18. In order to arrive at solutions we set in Eq. these solutions are not valid around z = 0. (18.2 We are concerned with the equation d2y(z dz2 where ^ 1 .14a) 1/2- VA( ) = A(z)exp (18.

15) where A{z) is the solution of the upper of Eqs. Since these higher order contributions are of little interest for our present considerations. we do not pursue their calculation. (18. Thus we now have the pair of solutions yA(z) = exp yA(z) = exp dz I^4 + l„2„4 1/2-.2 The Inverted Double Well Potential with A(z) = A{-z) and yA(z) = yA(-z).34) below) — to be derived in detail in Example 18. 1 2 4 2 Aq(z) + 0 (18.e.1 and as a verification again in connection with the solution y# — as the other solutions.20b) . 1 (18.4 . One finds that these solutions are associated with the same asymptotic expansion for A and hence E (given by Eq. (18. (18.Aq(z) approximate the solutions of Eqs.20a) These expansions are valid as decreasing asymptotic expansions in the domain \z\ >0 h.19) We see that one solution follows from the other by replacing z by — z. i. / -c z Aq{z)+0 1/2 .18) M*) z-f(9+l) A_ g (s) = (^2)1/4 exp ?/" dz (z ) / 2 1 2 (18.zh2 =(-)*-2"(18.. 385 (18.13) and we can develop a perturbation theory along the lines of our method as employed in the case of periodic potentials.17) Aq(z) 1(9-1) Z2 (Z )V4 exp 2 We define correspondingly w dz (z2)l/2 (18.13) TzA'(z) T ^A(z) + -qA{z) = O We define Aq{z) as the solution of the equation zA'(z)--(q-l)Aq(z)=0.16) For large h2 we can write the Eqs. We take the square root by setting z2h^1/2 +.18. Clearly Aq(z).13) and A(z) that of the lower of these equations. /-H 4 Z h +~C Z 21. (18. (18.

1 . Anharmonic Oscillator Potentials i. z) ± yA(q. 128 8 16 A = i.18) we obtain A » = H^-r-L^z) 2 z .[yA(q.20a) and (18. (18. procedure.l)A(z) 1 = -tfA"(z) A .e. ft = . i. A = .1) 2 2 2 A?(Z) " ^^^^^ = liq~ 1){q ~ VA"-^- The lowest order solution A^ 0 ' = Aq (z) therefore leaves uncompensated on the right hand side of the equation for A the terms amounting to i ^ / 2r2z2 2z\ y 1 °° (2c 2 l i=3 Clearly one now uses the relation z2iAq{z) = Aq+ii(z). (18.. h2 —> —h2). (18. a2 = 2 1 . -zA'(z) where the expansion coefficients are given by ao = l. „ .e.20a). h2.:=-*£ +i(5-l). X » / A q + 4 i = Mg+4i- .21) Example 18.• Using Eqs. z)).20b). i. As always in the In this way Rq is expressed as a linear combination of functions Aq+4i(z).. A A _ 3 . W i t h proper care in selecting signs of square roots we can use the solutions (18. 04 = 5 .f . 2 . we write > V±(z) = .v ' = o (« .13) in the following form with power expansion of the square root quantities and division by h2: 1 + -(q .1: Calculation of eigenvalues along with solutions of type A Use the solutions of type A. c*i = 1 1 . 03 = . to obtain in leading order the eigenvalues E. one observes that with £>.2 ( * ) 2 and from this or separately A"(Z) : 1 (q . h2.20b) to construct solutions y±{z) which are respectively even and odd under the parity transformation z —> — z (or equivalently q —• —q. .. away from the central minimum.! ) ^ ..e.386 CHAPTER 18. Vq+ii = Vq + 2i. (18. * = — .17) and (18.2 . „f 1 \ S o l u t i o n : We rewrite the upper of Eqs.^A(z) °° /1r2z2\i + J2 { -fir ) \**A'(*) 1 + jA^Wl.| . (18.

. [-1(9 + 1)]! . give an equation from which A is determined. _1J±w) and D_\. of course. (18. Rq '.| c V + It follows that l) + 0 (i The same result is obtained below in connection with the solution of type B. 2ft4 + e)Av -—rAq+i q+8 6 2/i + ---\+- The sum of terms with Aq in Rq are calculating here „2 must then be set equal to zero. In this way we obtain the next order contribution A^-1' to A^°>.—77 [i(9-l)].22) are parabolic cylinder functions D i .2 The Inverted Double Well Potential Thus a term /j. (18. We return to Eq. Hence to the order we 0: {^)^) [{ 1){ 2 ^ ^ ^-3)-^ 0 +1 ^ + 2^ + 3)]+0 {h) 2ft6' A = -24 1 2 ^ + 1>-2^+°G9A= . ^-f( g + i)(±H2^ + 1 > \ — 1.5). +1J±iw) (observe that Eq.'2H9-i) ^fa-Dfr™) and . (18.18.(q + 3)(q + l)Aqv--ri(q '^ ' .. and the coefficient of terms with i = 0..23) .. (18. r ( CJw) = ^ . those in Aq(z).22) is invariant under the combined substitutions q — —q. = -•>— r. Hence in the present case In its turn A^ leaves uncompensated terms amounting to A „ /2c 2 \ 2 9 „ M 2ft4 -2 { 4/i2^ . In the next two pairs of solutions the exponential factor of the above solutions of type A is contained in the parabolic cylinder functions (which are effectively exponentials times Hermite functions).w — > > ±iw) or functions R B w q( ) . The solutions yq(z) of the equation Vq(w)yq(w) = 0. w = hz.e. when i = 0. i.Aq+4i in Rq 387 can be taken care of by adding to A(°> the contribution — _ ^ 4 ' except.

3.0) = = = ±(q±3)(q±7).q + 4j] = c2SA{q. l(q2 + l). we also have Vq+^yq+tj = 0. See Example 18.q}=A + c2S4(q. G.4j)yq+Aj.( < ? .27) Now. (18. and so Vqyq+Aj(w) = -Ajyq+Aj(w).q + 4j}yq+4j. Erdelyi.0). pp.123.29) ' A .23) have been inserted to make this recurrence relation assume this particularly symmetric and appealing form.Aj)(18. S2i(q.l)2/?-4. Magnus. . (18.±4) S4(9. Comparison with our notation is easier if this reference is used. (18. W.28) (A + c w )yq(w) E 2 4 1 - 2 ^ [q. F. since Vqyq = 0. and for j ± 0 : [q.±8) 5 4 (g. (18.25) where in the case i = 2: SA{q. (q±2)(q±S). Tricomi [86]. (18. Oberhettinger and F. 115 . As an alternative to Bq(w) in Eq.388 CHAPTER 18.26) The first approximation y(w) = y(°>(w) = yq(w) = Bq(w) therefore leaves uncompensated terms amounting to 1 4°) = where [q.24) The extra factors in Eq. Vq+4j = Vq + 4j. Actually these factors can also be extracted from W K B solutions for large values of q. j=-i (18.^ For higher even powers of w we write i w2l V<i = Y.3)y g _ 4 . Anharmonic Oscillator Potentials The solutions Bq satisfy the following recurrence relation (obtained from the basic recurrence relation for parabolic cylinder functions given in the literature^) w2yq= -{q + 3)y g+4 + qyq + -(q . (18.23) one can choose the solutions as Bq(w) with B QM = ~ 77[|(9-3)]!24(9-1) These satisfy the recurrence relation w2yq{w) = -(q+l)yq+4 + qyq + .

(18.e.10) we obtain E(q. so that the entire expansion is invariant under the interchanges q .31) Proceeding in this way we obtain the solution y = yW („. .27) can be removed by adding to T/°) the contribution (—^/4j)y<j+4j. (18.) + j/ 2 ) (W) + • • • with the corresponding equation from which A can be obtained. .• -q. 2 . ( 18 .q]=0. p i(9»9 + 4 J'WK7». A = -^(q2 + l)c 2 + o(J^.35 ) . Thus the next order contribution to yq is y H = ie 2^ —z. (18. This type of invariance is a property of a very large class of eigenvalue problems. n — 0.) -t-y^^w) to be a solution to that order we must also have to that order [q.—A(q + 1) . i.^ g ( 4 g + 29) + O(^ 2 1 2 Qr2 2 4 2 / 1 \ J. h ) = -qh .—v*+*j- ( 18 .3 °) For the sum y{w) = y(°)(«.32) Evaluating this expansion and inserting the result for A into Eq.1. and |c 2 | replaced by — |c 2 |. h2 —> -h2.34) is the expansion of the eigenenergies E of Case (1) with q — qo = In + 1. Equation (18. We can now write the solution y(w) in the form y{w)=yq{w) + Y.) + j.(!) («. . In Case (3) the parameter q is only approximately an odd integer in view of tunneling. i.2 The Inverted Double Well Potential 389 Hence a term (iyq+4j on the right hand side of Eq. (18.[-rp ) Y.18. (18. and even powers of q in combination with even powers of 1/h2.e.34) We observe that odd powers of q arise in combination with odd powers of 1/h2.

there is another solution y(—z). ^VO(<L<?) ^i(?. g ± 4 ] T4 ±4 and so on. Miiller-Kirsten and A. q + tt)= J2 i=-2 p i-l(?> Q + ±3 + 4*)[q + 4j + 4t. Achuthan. q) = = 1. W. Anharmonic Oscillator Potentials ±4 ±4 ±8 ±4 P2{q. their recurrence relations and the solutions of the latter we refer to the literature. i. and for j ^ 0 all other P0(q.argz=0 (18. q + 4j) = 0." Since our starting equation (18.)]arg2=0- These solutions are suitable in the sense of decreasing asymptotic expansions in the domains They are linearly independent there as long as q is not an integer. We observed earlier that these are obtained by making the replacements q —> —q.g±4] ±4 ±4 [g.e. q + At] (18. (18.—. We thus have the following pair of solutions VB(Z) VB(Z) = B w .36) with the boundary conditions p o(q. H.390 where for instance v J CHAPTER 18. o. ? T 4 ] [ g T 4 . Our third pair of solutions is obtained from the parabolic cylinder functions of complex argument.q±A) = MM!i±MM+[M±8][g±8.38) [yB(z)]argz=7T = [?/s(-2. Wiedemann [4] . 2 4tPi{q. Again we can write down a recurrence relation for the coefficients Pi{Q. / l \ 1i i( ) + J2 ( ^e ) J2 P ^q + 4j)Bq+lj{w) iu=/iz. 5 + 4j) = 0 for \j\ > 2% or \j\ > 2i + 1. we may infer that given one solution y(z).i 1 + 4j) in complete analogy to other applications of the method.37) For further details concerning these coefficients. J. w — ±iw.11) is invariant under a change of sign of z. > "p.

the solutions of type A being valid away from the minimum.q + 4j) as in ys. Aq(z)) does not appear in any of the higher order terms. so that the eigenvalue expansion remains unaffected by the interchanges q —> —q.h^ih. h2 — — h2 as long as corrections » resulting from boundary conditions are ignored.** First we deal with the exponential factor 1/2- exp exp dz z z h* + -c z J2c z 2 2 .g.3 M a t c h i n g of solutions We saw that the solutions of types B and C are valid around the central minimum at \z\ = 0. In this bordering domain the adjoining branches of the overall solution then differ by a proportionality constant. Integrating and expanding as follows since h? is assumed to be large. (18.z2 .5) are known in some mathematical literature as "stretching variables" and are there discussed in connection with matching principles. Mauss [192]. h2) in which odd powers of q are associated with odd powers of h?. Such contributions arise. 4 . with normalization since the normalization constants are also asymptotic expansions.2 _4 A } 2cVl1/2 h4 8c 2 1 in the solutions of type A which are not valid around z = 0. We add parenthetically that all our solutions here are unnormalized as is clear from the fact that the function in the dominant term (e. (18. . we obtain: exp 8c2 3 < 2cVl3/2 /i4 h2z2 „(zA exp 12c2 (18. Vc(Z) = [V!B(Z)}q^~q.g.2 The Inverted Double Well Potential These solutions are therefore defined by the following substitutions: VC(Z) = [yB(z)]q^-q.The solutions yc.g. In order to be able to extract the proportionality factor between two solutions. like w of Eqs.39) are with the same coefficients Pi(q.4)).h^ih 391 (18. see e. 18. one has to stretch each by appropriate expansion to the limit of its domain of validity. Thus in the transition region some become proportional.40) "Variables like those we use here for expansion about the minimum of a potential (e.18.2.Vc suitable asymptotically decreasing expansions in one of the domains \z\ < O 7T argz We emphasize again that all three pairs of solutions are associated with the same expansion of the eigenvalue E(q. in fact. J.

) ( ^2 „22)\ z £(</-!) -e 4re z 4 4 = hz: [i( 9 -l)]!24(«-D l+O h? (18.z). 4/T 4 Prom (18. 41) The cases of the solutions of types B and C require a careful look at the parabolic cylinder functions since these differ in different regions of the argument of the variable z. (18.44) we see that in their common domain of validity yA(z) = ~yB{z) a (18.41) with the solution ys(z) of Eq.(«.l ) ] ! ( .43) W + 1)]! w^+V A . 2 ^i![i(g-4i-l)]!(-2«..44) In the solution y~g(z). y^(z) we see that in the direction of z = 0 (of course. The function Di. > -n.w yB(z) ~ B.43) (since z — — 2 implies > argz = ±7r).42) H W2 (q-l)e-$W I „ . Anharmonic Oscillator Potentials Considering the pair of solutions ju(.4 i . We do not require this at present. we would have to substitute correspondingly the expression (18. Thus from the literature [86] we obtain Di_{q_1){w) = w*{q-l)e-\w* but D W-i) ^=.45) . (18. Comparing the solution VA{Z) of Eq.2)i (27T) 1 /2 e -f(9-l) 1 eb Uq + m (18.2 « .0-x ! [ £ ( g . (18. with z in ys(z) replaced by — z.-ir_I + M + 1)]! (2w*y U i\[~W • i 5 with -7r > argw.7 r . -^Jw) has a similarly complicated expansion for 1 5 — — > argw > . . e-h* /12c* e\z*h? (18. not around that point) VA(Z) VA{Z) = ehVl2c 2 e-\z*h* z^) + o(± z-fr+i)+o(±\\.392 CHAPTER 18. 2 ) i ' ir! i iHi-m 1 3 j argw| < -7T.42) we obtain for the solution ys(z).

.46) However. ^-^+i)Ah^ze* [_i( g + l)]!2-4(9+D Inserting the ex- l +O h2 (18. we see that in their common domain of validity yA(z) where a = =Vc(z).48) (-h2y 4(«+i) [-i((? + l)]!2-4(^i) l +O /i2 an (18. We proceed similarly with the solutions yc(z).2 — we see that near the origin the potential behaves like that of the harmonic oscillator in fact — our large-h 2 solutions require this for large h2.7. Recalling the solutions y±{z) which we defined with Eq.49) d yc{z) Again there is no such simple relation between yA(z) 18.5. 18.11.50) and y+(0) ^ 0.1 ) e 12c 393 Kg-l)]^"1) l + O h? (18.e. i. and the second qo = 3.50) will be seen to imply qo = 2n + 1 = 1.21) as even and odd about z = 0. (18.2 The Inverted Double Well Potential with a 2 (K ^ ( 9 .41).. we see that at the origin we have to demand the conditions yV(0) = 0 and y_(0) = 0 (18. Looking at the potential we are considering here — as depicted in Fig.47) yA(z) Comparing this behaviour of the solution yc(z) with that of solution of Eq.9. large probability for the particle to be found thereabouts.18. (18.y'_(0) ^ 0. — For instance qo = 1 (or n = 0) implies a ground state wave function with the shape of a Gauss curve above z = 0. the ratio of yA(z)iVB(z) i s n ° t a constant.42) into y~c(z) w e obtain Vc(z) = (-h*. At z = 0 the solutions of type A are invalid..yc{z)pansion (18. Thus the boundary conditions to be imposed there are the same as in the case of the harmonic oscillator for alternately even and odd wave functions. The first of the conditions (18.2. hence we have to .4 B o u n d a r y c o n d i t i o n s at t h e origin (A) Formulation of the boundary conditions The more difficult part of the problem is to recognize the boundary conditions we have to impose. (18.

(«. (18.394 CHAPTER 18.-i)H = e.-l)(°) Thus with the help of the reflection formula cited above: D B.( g + 3)}./o^Tp-if (q+i) .(0) + ^ ( 0 ) l z—>0 2.57) [i( 9 _l)] ! 2 i(9-i) B.s i n { .i ) -..u=o follow with the help of the "circuit relation" of parabolic cylinder functions given in the literature as (q r — ) Di(.. + 1)]!' D ' j .2: Evaluation of yB(0). C. . A.58) diu [ | ( g .SxW) = \ i»fl(0) .5 ( 9 + !!)(-*"')• i ( 1) . M.oi(.55) Solution: Prom the literature. one finds that [ 1 IT (q + 1)1! = -.2.l ) ] ! [ .(0) Cq(w) y' c (0) r^ Show that — with w = hz — the leading terms of the quantities listed are given by B n\\ia-m' 4(«Z + l)]l[i(8-l)]!' i\ -sin{-(q-3)}. y'B(0).52) # 1 = -^ y' c (0) a and 2/B(0) = a (18. Z (18.56) D 4(.1)(-'») + ^ r + l)]! ^. 1 [7(9-3)]! n «(°) = .54) dio q(w) ..! ( .) [i(«-l)]![-i(?+l)]! 2TT V5F[i(g-l)J! V (18.g. — C.F ? . 4W >S [ I ( ( ? _ 3 ) ] ! s i n { f ( g + 1)} and hence =^= 1.' '.(0) (18. with the reflection formula (—z)\{z — 1)! = IT/ simrz.| ( 9 + 3)]! s i n { . Stegun [1]. [ . Anharmonic Oscillator Potentials use the proportionalities just derived in order to match these to the solutions valid around the origin. IT 4 dio In fact.53) yc(0) a' Clearly we now have to evaluate the solutions involved and their derivatives at the origin.(0) = . Example 18. Then imposing the above boundary conditions we obtain 0 = y'40) = limJ-[y'A(z)+y'A(z)] = i»i.i ( g + 3)]l (18. We leave the detailed calculations to Example 18. e. [C.51) and 0 = V-(0) = Jim i|»xM .|fc(0) Thus we obtain the equations (18. y c ( 0 ) .. Abramowitz and I.(0) = and hto-i)(°) ^—'-."*(«-!) ^ ' ~ [ . Expressions for C 9 (0).^{-(9+1)}.(iu)]. = -= V^ >—= = — i i ^ i [1(9-3)]! ^_ sin{_(q + i)} (18. we obtain i V7T24( g .( g + 3)}.

+i)(°) [ _ I ( .i ( « + i ) [ _ i ( .46)._l) From this we derive C. (18.sin{^(g+ 1)} = J . ~ ^ 1 -5(9+1) 2i(''.60) (18-61) [C>)]o = -iV2i [-1(9 + l)]![i(9 " 3)]! (18.sin{^( 9 .(0) = *>-i(„+i)(0> 2 .53) in dominant order and insert the appropriate expressions for a and a from Eqs. using the above reflection formula and the duplication formula ^/TT(2Z)\ = 22zz\(z . 7T 4 V 7T 4 (B) Evaluation of the boundary conditions We now evaluate Eqs.2 The Inverted Double Well Potential From this relation we obtain 395 ^^^i^-w^-vy ^(9-i)(°) (18 59) - Inserting from (18.i .3)}. we obtain 2 i(»+i)[_i(.)[i(g-3)]! and ^-i(. W .+1iH]«=o = . (18. 4 r T (18. + i)]!2-i(«+ ) 1 [|(«-3)]! V5F[i(9-l)]! sin{-(q + l)}. (18.1/2)!.18. isin{f(g + 3 . Starting with the derivative expression we obtain (apart from contributions of order 1/h2) lsin{f(g + 3)} i Sin{|(g .56). _ l)]!2l(9-1)(-/J2)-|(9+D We rewrite the left hand side as 1 sin{f(g + 3)} r„.6 ) } ^ — i t a n < — (q + 3) >. + i)] V* H(?+l)]'[£(9-l)]l Similarly we obtain [°'-i(.62) .49).3)} (h?)fa-V[-\(q [Uq U + l)]\2-*(q+V _£ e 6c^. A __f7r \4 'J We rewrite the right hand side of the derivative equation again with the help of the inversion and duplication formulas and obtain _ (^)g/4(_l)j(g+l) e -/> 8 /6^ _ p (/i4)g/4(_1)|(g+l) _^_ .

.64) the right hand side is an exponentially small quantity.. (18. . this is the case of the purely discrete spectrum (the differential operator being selfadjoint for the appropriate boundary conditions. . we again obtain (18. i.9o) ^ cos | .64) for q = g0 = 3. For c 2 < 0 and the solution y(z) square integrable in —oo < ?Rz < oo.11.^ . 18. (18. (18.1.e.53) becomes 8 m(I(. 7 .9. equivalently.( g 0 + 3) | + • • • ~ (g . 18. 7.65) but now for the odd function with these values of goWe have thus obtained the conditions resulting from the boundary conditions at z = 0.. ^ .64) In each of Eqs. (g-g 0 )^±^ y e-5?.^ >^_>m e J*. by the rotations E -> einE = -E. With a Taylor expansion about go the left hand side of (18.. + 3)} = .. (18.2.53).5 B o u n d a r y conditions at infinity (A) Formulation of the boundary conditions We explore first the conditions we have to impose at \z\ — oo.. Our next task is to extend the solution all the way to the region beyond the shoulders of the inverted double well potential and to impose the necessary boundary conditions there.396 CHAPTER 18. . (18..5. The analytic continuation of one case to the other is accomplished by replacing ±c 2 by =Fc2 or.. the energy E is real. . we obtain cos(-(g + 3)) = . (18. the vanishing of the wave functions at infinity).64) about go — 3. 1 1 .2) with potential (18.1).. z2 -> -z2.i .63) vanishes for q = go = 1. In fact the left hand side of (18.go) ^ (-1) It follows that we obtain for the even function with q — go = 1. Thus we have to determine these conditions first. Anharmonic Oscillator Potentials Then the derivative relation of Eqs. z^ ein/2z.5.63) Proceeding similarly with the second of relations (18.63) and (18.9....63) becomes (? .L ^ i . This is Case (1) of Fig.65) Expanding similarly the left hand side of Eq. Recall > the original Schrodinger equation (18. and the left hand side of (18._ e =*.

—}. replacing sin 30 by " 0 + ^) 1 IT = . i. the resulting wave functions vanish at infinity and thus are square integrable. 6 6 In the case of the inverted double well potential under consideration here (i. i.+00 in arg z € (-!•" * 0 for 5ftz — —ex) in arg z € » N> (18.e. (18. y{z)~exv\-i[-) 7T dV'V.sin 3(9 This expression vanishes for \z\ —• +00 if the angle 0 lies in the range — ir < > 36 < 0. i. r / c 2x 1/2^3 = exp{ ± i — —\. i. or — < 6 < -. in the domain — TT/2 < 36 < > 7r/2.67) Rotating z by vr/2.e. + i s i n 3 6 0 i y ) ~3~^ c2\l/2\z\3 oc exp <M — J — . 2 c2>0. Then c 2 y/vi ^p^My) yf = exp _ r HvcV2!*!3 cos36. It is then necessary to insure that when one rotates to the case of the purely discrete spectrum without tunneling. if -TT/3 < 6 < 0. Now.e. (18. we set z w r = \z\eie. we see that the solution with the exponential factor is exponentially decreasing for \z\ — 00 provided that cos 39 > 0.2 The Inverted Double Well Potential 397 One can therefore retain c2 as it is and perform these rotations.e. we therefore demand that for $lz — +00 and • —7r/3 < arg z < 0 the wave functions have decreasing phase.e. Thus in our case here the behaviour of the solutions at infinity has to be chosen such that this condition is satisfied.68) . Thus exp exp- (-<£ff} [+<£ )%} sin * 0 for $lz —s. the case of complex E).66) z3 = \z\3em.cos 36.18. for fiz -* ±00 we have y(z) ~ exp < ± i / dz In order to decide which solution or combination of solutions is compatible with the square integrability in the rotated (c2 reversed) case. 1/2.

18.21) to + infinity and to demand that they satisfy the condition (18.398 CHAPTER 18.e. We therefore explain our procedure first. i.Z\. for z -> Too. For c 2 < 0 we have correspondingly y(z) ~ exp I ± ( — 1 — y c2 < 0. (B) Evaluation of the boundary conditions The following considerations (usually for real z) require some algebraic steps which could obscure the basic procedure.68) on y±(z) (by demanding that the coefficient of the solutions with other behaviour be zero). Anharmonic Oscillator Potentials V(z) Fig. (18. We have to remember that we have various branches of the solutions y(z) in different domains of z. Eq.21)) were defined in terms of solutions of type A which have a wide domain of validity.3 The inverted double well potential with turning points ZQ. Equating to zero the coefficient of the term with sign opposite to that in the exponential of Eq. and there impose the boundary condition (18. Our procedure now is to continue the even and odd solutions (18.Z\.68) will lead to our second condition which together with the first obtained from boundary conditions at the origin determines the imaginary part of the eigenvalue E. This is not the asymptotic behaviour of a wave function of the simple harmonic oscillator. 18. We do this extension with the help of WKB solutions.68) for c 2 > 0. This is the boundary condition also used by Bender and Wu [18]. (18. Thus we have to match the solutions of type A first to solutions to the left of z\ and then extend these to solutions to the right. Looking at Fig. we match the WKB solutions .3 we see that at a given energy E and to the right of z = 0 (which is the only region we consider for reasons of symmetry) there are two turning points ZQ. Our even and odd solutions y±(z) (cf.

e.2. B. . z\ ~ . 4 2 T ' i.z\) ( \ y^NKQyz) -qh2 -z 4 dz h H—c z 2 x sin ^ . (22) or A.4) for E and ignoring nondominant terms) t o . Messiah. *R.zi)/ \ -1/4 y-wKB\ ) z — -g/i* . ^ 2 4 •K + —(r.4 h + -c z 2 -1/4 < . equations (21). A J-24 1 ? 2 • "^ •-z h -\—c z ~ — q h .V 2 iL44 + -<?z / „2 4 X COS < f / il/2 G?Z -g/l 2 1 2L4 -z 4 M 1 2i.70) where z < z\. 1 2i4 1 2 4 / ~ 2 4 ~ 2 C * (18. p. We begin with the exponential factors occurring in Eqs. Dingle [70]. 6. t To the right of the turning point at z\ these solutions match on to (r.*l) Z 1 2L4 1 -qh? + -zzh* -1/4 _2i4 VWKB\ ) .e. and then use the W K B procedure (called "linear matching" across the turning point) to obtain the dominant W K B solutions beyond z\.^ T rz\ 1 + -yh* . Dingle [70].B. 18.. ^2? 1- 2qc2 /2? .4. From there or the literature* we obtain in the domain V > ^RE to the left of z\ as the dominant terms of t h e W K B solutions -1/4 ^WKB^ \qh2 Z\ + \z2h4 dz - \c2z4 1/2 „2 -czz 4 x exp •r(I. Sec. (18. 291.18.3 is given by (using Eq. In using these expressions it has to be remembered t h a t the moduli of the integrals have to be taken.69) The W K B solutions have been discussed in Chapter 14. 291.=-c z 4 dz 1 L2 9 + 2 4 1/2 x exp . Vol. i. (18.70).^ 4 1/2 + icV 7T Jz + (18. The distant turning point at z\ as indicated in Fig.71) We now come to the algebra of evaluating the integrals in the above solutions. f R . p.2 The Inverted Double Well Potential 399 to the left of z\ to the solutions of type A. h2 (18. z2h4/A > c2z4/2. I [195].

69)) rzi J<z z dz Z[&-Z*]W = + ^Y /2 m '"U4 1/2 2d2 + 2c2 Since we are interested in determining the proportionality of two solutions in their common domain of validity we require only the dominant 2-dependence contained in this expression.211/2 1 \ /j.2 \ 1/2 . Thus the above factor yields '2c2\l/2 h4 so that (cf. (18.400 i. Eq..40)) E± = exp _/^_2 f ± In A 2c2 . Anharmonic Oscillator Potentials E± = exp = exp .±9/2 = ^ V M z^l2 21 2c2 exp ± i dz 1 4 1 1 V2- 2 (18.2.72) Here the first part is the exponential factor contained in 2/A(Z)) J / A W respectively (cf. We obtain this factor by expanding the expression in powers of zjz\ (since in the integral z < z\).\qh2 + \z2h± .4 In 2c 2 J + 1/2 ^1 2c2 1/2 21 and hence (with use of Eq.^ I 2 2 1/2- zV = F 8c2 Jz [i .73) . (18. .^ ] V 2 and so /^2f 8c2 3 1 2cVl3/2 h* f q fZ\ ck h^y/ E± = exp T 2 jJ Z ^ [ g . (18. In the remaining factor we have (looking up Tables of Integrals) 21 2 dz h 4 2 2c.2. (18. zi 2q 1 c 2 [1 . CHAPTER 18.40)).e.i c V 8c T 2 1/2* ti>_(2 ~ exp 8c \3 2 r< 2cV 2c2 z2 i-24£Y'2u h* 3/2 >. Eq.2^3/2 2_(h^\ll2 2czz 4 N 1/2 8?3\ 2 ~hf ' =F9/2 ^ 2 ^ .^ i ^ ^ c 2 .

*l) (3y^(z) (18.79) +S_(±)exp .77) we can rewrite the even and odd solutions for 9te —• oo as (by separating cosine and sine into their exponential components) -1/4 y±{z) .21 1/2 .*l) (18.(J-*i) VA 0 ) = PVwKB (Z C.77) Now in the domain 2 — oo we have > \qh2 .74) Comparing these solutions now with the solutions (18.76b) apart from factors [ l + 0 ( l / / i 2 ) ] .\z2h4 + J Z\ nl/2 \M fZ .20a) and (18.2 The Inverted Double Well Potential 401 Thus at the left end of the domain of validity of the W K B solutions we have (l. CZ2 JZ1 cz 3^2 V2 h6 12c 2 ' Ac2z2 cz" 3y/2_ (18.21) we now have V±(z) 1 2 \yA{z) ± yA(z)} = \W^(z) WI ± _(J.Zl) ( \ y-WKBV^J \z h* 2 1/4 and ywKB(^) \z h* 2 1/4" (18. In these expressions we have chosen the signs of square roots of h4 so that the conversion symmetry under replacements q — —q. (18.76a) P P r 2/1 2 " 9 ( .18. we see t h a t in their common domain of validity VA (z) = /3ywKB 0)> where . > » Returning to the even and odd solutions defined by Eqs.78) Inserting this into the solutions (18.20b).2Z 4 -C S ' + ( ± ) e x p I i\ cz / cz3 V3\/2 U^2 h6 12c 2 12c2 J (18. h2 — — h2 is maintained.71) and these into (18.75) 9/2 1/2 0= or '(2c )V2 2 and (3 = — 2 2 (-fr2) (2c 2 ) 1 /2_ -9/2 (18.1 ) 9 / 2 _(2c 2 ) 1 /2_ -1 (18.

5.h2) + {q-qo)(^pj = E(q0. (18. (18..82b) This is our second condition along with Eq.85) + --. Eq.. Anharmonic Oscillator Potentials where 5+(±) 5_(±) = = Q / ? ± ^ e x p ( ^ ^ T ^ ) e x p ( . 18. this equation can be rewritten as or as the replacement + / u6 \ 90/2 (-/i 2 )W 2 =^(_) 2®"1 f — J . i.h2) + (q-qQ)h- + (18. (18.3. Inserting this into the latter equation we obtain (the factor "i" arising from the minus sign on the left of Eq.e. (18. E(q.68). (18. i / 3 ± . .65). i.402 CHAPTER 18. (18./ ? = 0.34).84) Expanding about q = qo we obtain £(g.. (18..81) Inserting expressions (18..^ § ( 4 g 2 + 29) + O ( ^ ) .80) Imposing the boundary condition that the even and odd solutions have the asymptotic behaviour given by Eq.6 The complex eigenvalues (1883) We now return to the expansion of the eigenvalues.76a) for (3 and /3./i2) = E{qQ.2.e. we see that we have to demand that 5 + ( ± ) = 0.82b)) /j6 \ 9o/2 («-*) = W? ma-m ''*• with qo — 1.^ j . h2) = \qh2 .^(q2 + 1) . (18.

Since this is not without interest (e. ^ 2 4 y -qh 2H ""{-f' z h H—c z 4 2 1/2 1 ^ 4 + IC2Z4 dz -qh2 4 2 .o 2 + 29) + o ( ^ ft6 \ 2qoh2[ —~ 2 ( ) < • 9o/ 2 ' 2c J -ft 6 /6c 2 '(27r)V2[i(go-l)]r • (18 86) - The imaginary part of this expression agrees with the result of Bender and Wu (see formula (3.36) of their Ref. above) (1.ZQ) . The large order behaviour of the eigenvalue expansion (of Case (1)) which Bender and Wu derived from their result (18.18.2 The Inverted Double Well Potential 403 Clearly the expression for (q — go) has to be inserted here giving in the dominant approximation / h6 \ <?o/2 . i.g ( .-h6/6c2 E = E(q0. ft6 ^ = e. where the superscript {r.yA{z). ZQ (2g)V 2 / h \ 2qc2 he (2q)1/2 In the domain 9ZE > V to the left of ZQ the dominant terms of the WKB solutions are (cf. zo} means "to the right of the turning point ZQ" .l ) ] ! > + = ^ 2 . [19]) for ft = 1 and in their notation 90 = 2 ^ + 1. Example 18. In the above we did not require the matching of WKB solutions at the lower turning point ZQ to the solutions yA(Z). o 2 + l)-^(4. Solution: The turning point at ZQ close to the local minimum is given by -qh2 2* 4 -z2hA -c z ~ 0.3.86) will be dealt with in Chapter 20. h2) of the relations KWKB (Z) =PVA(Z).3: Matching of WKB solutions to others at ZQ Determine the proportionality constants p(q. !*WKB(Z) ~ 2 . S W K B ( 2 ) = PVA(z).h2) The final result is therefore E (-) i ( 2 7 T ) V 2 [ l ( g o . * _ -qh2 -1/4 *• 2 j _ 4 i 1 2 4 -z h H—c z dz -qh2 2 1 2L4 •{/.g.e.-4 1 2 4 1/2 ft + i c V 2 -1/4 + f}' z.~p{q. in comparison with the work of Bender and Wu [19]) we deal with this in Example 18. h2).2 .

zo)/ \ -qh2 2 ZO + -z2hA 4 2 .O. the (approximate) solutions* 1/2 exp • ± i ( .zo) % ' K B (Z> 1 / 4 2 l / 2 h z V Q / 2 7i/2 4 2 e x p ( ^ e .g In other words. On the other hand in the exponentially behaving W K B solutions the quartic interaction term acts as a correction to the harmonic term close to 20.l c y dz 4 1/2 x exp •! / —(r>zo)/ \ 2 -1/4 1 2y x exp 2 1/2 L dz -qh2 2H + ~z2h4 4 - 2 -c2z4 we nave where 2 > 20. N. cf.404 CHAPTER 18. Schwind [247].ic224 2 4 -qh2 + i^fe 4 . Anharmonic Oscillator Potentials where zo > z. and we have 1/2 dz h?_ 2 J /i2 qh2 + iz2h4 _ ^ 4 J dz zn -qh2 + -z2^ 1/2 ..1/2 h? \z{z -zlf' -\z \n\z 2 2 2 + {z2-z2Y/2\ 2q h2 4 • In (V2zh V s/q Then to the same order of approximation . 2 z 2 ) / 2 e \ . In the domain V > SHE to the right of 20 the dominant terms of the WKB solutions which match on to 2 / W K B ( Z ) ' ^ W K B ( Z ) a r e res Pectively -1/4 (r. . Expanding the square root occurring in ^ W K B W ' ^ W K B ^ ) ' Igft2_I22h4+lc2z4 2 4 2 and we see that 1/2 = 7 ^ 1/2 2." / 4 hj (ft2 2 2)( 9 +l)/4 J'W'KB (Z) 2\1/2(fe222)^-1)/4 h) exp(±h2z2) ^2ey/4 * Solutions of this type (which are asymptotic forms for q — 00) have been investigated in t h e > literature. J ^ K B match on to the W K B solutions with exponential behaviour to the right of 20.] { hz combined with the particular constants contained in y ^ ' ^ g . f x Zn -4) h2 2-./ ^ 2 X 1 / 2 exp(ifi.

i.3 The Double Well Potential 405 It may be observed that 3/Y/KB ( z ) corresponds to solution (3. h4 > 0. in fact.13) of Bender and Wu [19].. J/J Hence ^ and h so that \]q] \2"'2 ~ ( 2 7 r ) 1 / 2 e .^ 1 .3.g. The last expression can be made acceptable for q = odd integers by applying Stirling's formula in the denominator of the second of the previous pair of expressions. as in the case of the Mathieu equation." / 4 ( ? « / 4 .(r.5 |"l _ o f .89) . „ / M 1 / 2 i . (18. 4" ) (h?z2)^q+1) for q j£ odd integers. Then . (18. We consider the following equation ^f& with double well potential V(z) = v{z) = -jz2h4 + ^c2z4 for c2 > 0.J ' i J ._ .e.kyil(.zo).yA{z) with expressions 27WKB i J'WKB ' that the factors p.87) The minima of V(z) on either side of the central maximum at z — 0 are located at h2 h8 z± = ± with y(.3 18.18.1 The Double Well Potential Defining t h e p r o b l e m In dealing with the case of the symmetric double well potential. But there are significant differences. In this form the WKB solutions reveal their similarity with the solutions VB i VB > Vc i Vc a n c ' demonstrate that the factors which we inserted (e. (j .1)! = ( 2 7 r ) 1 / 2 e .-Df-J»v we see Comparing now the expressions (18. We can reexpress these relations with the help of Stirling's formula. in Eq.88) + [E-V(z))y{z)=0 (18. we shall employ basically the same procedure as above or.41) for yA(Z).23)) appear quite naturally. ~p requested at the beginning are given by /1y/2[j(g-i)]i2^c-^ h 6 /12e3 /ly^-ih+iwi"-" w 18.z ± ) = . (18.

90) V(i)(^)=0. (18. vW(z±) = ±Qch2. In order to obtain a rough approximation of the eigenvalues we expand the potential about the minima at z± and obtain y dz2 We set + E .V(z±) -l-{z- z±fhA + 0[(z .94) (18.406 CHAPTER 18.z±).i>5.95a) .91) becomes T>q±(w±)y(w±) = o(-^-)y.4 The double well potential.z Fig. (18.93) w± = h±(z . .91) (18. Anharmonic Oscillator Potentials and V{2\z±) V(4)(z±) = = h\ 12c2. 18.92) (thus we sometimes use h4 and sometimes Ii±) and E-V{z±)=l-q±hl With the further substitution + ^. the previous equation (18.z±) y = 0. (18. (18.

(18. (18. we conclude that in the dominant approximation q± is an odd integer..95b) By comparison of Eqs. we obtain another by replacing z by —z.18. (18. given one solution.. (18..96).99) V2\ZZ The equation for A(z) is given by the following equation with upper signs and the equation for A(z) by the following equation with lower signs: A'\z)TV2Lz2-I^\A'(z)TV2czA(z)+(^q±h ±+^r)A(z) = 0.87).e.97b) Our basic equation.3.100) .95a) and (18. i.1.2. qo = 2n + l.2 T h r e e pairs of solutions We define our first pair of solutions y(z) as solutions with the proportionality y(z) = exp \h\ f uV2{z)dz (18.95b) with the equation of parabolic cylinder functions u(z) = D„(z). (18. we obtain d*y (18.i 4 f W » = o.98) Evaluating the exponential we can define these as the pair yA{z) VA(Z) = A(z) eyip = A(z)exp + —z V2\3 4c'' Ac' (18. Inserting the expression (18..3 The Double Well Potential where ^ 407 d2 •« 2- 2 - d8. is again seen to be invariant under a change of sign of z. 18.93) for E into Eq.96) dz2 + 5 « 4 + | . (18. where U{z) = and near a minimum at z± 4_ 4 jp[V(z)-V(z±)]. n = 0. d2u{z dz2 + v+ 2~r u(z) = 0. Eq. Thus again.97a) U(z) = (z-z±)2 + 0[(z-z±)3}.

(18. h2. Aq(z) = Aq(—z) = A-q(z). The result is given by A and E(q.c W + 1) .104) Both solutions yA{z).101b). (18.103) Looking at Eqs.e.101b) To a first approximation for large h2 we can neglect the right hand sides.102) We observe that a change of sign of z in this equation is equivalent to a change of sign of q. (18. Anharmonic Oscillator Potentials and selecting z+ (z2+-z2)A'(z) h z+ = —. We leave the calculation to Example 18.105) "8 + 4 ^ . z).yA(z) are associated with one and the same expansion for A and hence E.z) by either changing the sign of z throughout or — alternatively — the signs of both q and h2 (and/or c). 8hw (18.102) yields the following expression M*) I z2 _ z2 11/2 *+ z + z+ q/2 z+ 1(9-1) \z + z+ 1(9+1)' (18. The dominant approximation to A is then the function Aq given by the solution of the first order differential equation VqAq(z)=0. (18. yA(q. z) = yA{q. these equations can be rewritten as + (qz+-z)A{z) = . i. i. h2 h4 2c' with g+ = q.408 Since CHAPTER 18. we observe that the solution y^iq. ~h2.^ A"{z) + 2c V2 *-A(z) (18. Vq = {z2+-z2)— + (qz+-z).4. n± = V2h2.h2) . -z) = yA{~q.1 6 ) .e. but the solution is a different one. z) may be obtained from the solution yA{q.h2.c W + I ) 2C yfi 2h 5 2 A V2c* g(17g2 + 19) + 0 ( / t . (18.101a) {z2+-z2)A\z)-{qz+ + z)A{z) = ^- A"{z) + £-A{z) (18.^q(l7q2 6 4/i + 19) + (18. Integration of Eq.106) .101a). h2. h2.

We know the first derivative of Aq from Eq. The first such step would be to re-express the right hand side of Eq.102).108) We wish to rewrite this expression as a sum y coefficientiv4. . (18. .105) and (18. A Differentiation yields » = ^^w- (18.Aq 1 + 2^± 2 Aq+2 -(^M^f (18.2 z\) • Aq-$-4 £Aq + Aq—4.109) We observe some properties of the function Aq{z) given by Eq. (18. Thus it is natural to explore analogous steps. Solution: The structure of the solution (18. .(l + + 2 - z y* .106).. 9-2 ZZJf- (18.103): . .z2 ) N -t. + Aq+2 + Aq Aq+2 .114a) . ± 2 .103) for Aq{z) is very similar to that of the corresponding solution in considerations of other potentials. (18. (18. -<4q+2 — z — z+ Aq+2iAq+2j (18.113) 2 .+ 2 ^ ± l + 2^±i + .*l) (* ~ "+> ) A (z)exp q /2 \ 3 4c J (18.4: Calculation of eigenvalues along with solutions of type A Show in conjunction with the derivation of solution y^ that the leading terms of A and hence E are given by Eqs. for instance by componendo z z et dividendo. z • . however.{z2 . as a linear combination of terms Aq+2i. such as periodic potentials.107) KV) ~- Aq(z) {z*-z\y Aq{z) (z 2 z\f (z .112) From these we obtain. we re-express A'q in terms of functions of z multiplied by Aq. . Similarly we obtain _z+_ Z _ Aq + 2 — Aq Aq+2 + Aq 1-2 ^g+2 A ± ± _2A 2 ± 6 + (18.e. J = 0. (18.qz+) [2z2 .101a) with A replaced by A . .3 The Double Well Potential 409 Example 18.j-f-2i(2) We also note at this stage the derivative of the entire solution yA(z) taking into account only the dominant contribution: yA(*) =* -A*2 V2 .18.z\) + 2z(z . ± 1 .4zz+q+ z\{q2 + 1)]. (18.qz+)2 . i.110) Aq+2+Aq-2 9+2 -A. First. = Aq+2i+2jAq.111) {Aq+2 - Aq-2)2 (4zz+)2 {z2 - 2 -.

<? +{q.112). q + 4)Aq+4 + (q. q) = 0.9=F4) = (g=Fl)(?=F3). we obtain 5 K = ( ^ f ) [ ( 9 .116) into Eq.+6 + • • • ] . •••].410 and from this = Hence with Eq. ' T h e reader may observe the similarity with the corresponding coefficients in the simpler case of the cosine potential. (q. (18.4)A„_ 4 + (q.118) 2s/2c ~^j[(?.l ) ( g .34).117) -4(q + l)2Aq+2 Here the first approximation of A. QA?+4 _ 10 A + 6 .z%) 1 n z+z 2 \4z+J [Aq+4 [Aq+4 .112) CHAPTER 18.4 ( g . Eqs. Since VqAq = 0. Vq+2iAq+2i=0 and Vq+2i = Vq + 2iz+.q + 6)A<.4Aq+2 + Aq+A Inserting (18.115) (z*-z%) 2 -i2 (z 2 . (17.119) It is now clear how the calculation of higher order contributions proceeds in our standard way. A(°~) = Aq.q + 2)Aq+2 (18.33) and (17.2\2 (z* . 1_4^±^+8^±±-12^±^ +.2Aq + Aq-4] 1. i.4A ? _2 + 6Aq . (q. where the lowest coefficients have been determined above as (9.q + 2) = -4{q + l ) 2 . -.2A g + A g _ 4 ] . (18.115) and (18. (18. . leaves uncompensated on the right hand side of Eq.q)Aq + (q. (18.4]j. cf.108).3 ) A .q2)Aq-2 + (q. _ 2 + 6 ( 9 2 + l)A ( ] + (q + l)(q + 3)Aq+4.116) [Ag-4 . _ 4 .4 4+* -+ . (18. Anharmonic Oscillator Potentials A 1 . In particular the dominant approximation of A is obtained by setting (q.e.114b) . (18.2 ^9+2 „^9+4 + \4z \4z+J z+ Finally we have also [A q _4 — 2Aq-2 + 2Aq+2 — Aq+4 (18.aAg+S 12+ 16- (18.l ) 2 A . <?) = 6(<j2 + 1) + z: 2 2A +72"' (18.(0) V2 ' 2c K + h4+Ao .101a) the contribution R .z\Y Z2 + {~)\Aq+4-2Aq 1 224 Aq.

9-4)(g-4. Terms jj. h2 2y^ Age? h4 (18.z) Considering only the leading approximations considered explicitly above we have (since Aq{0) = l/z+ = A.-h2.124) . A' (0) = -q/z\) y+(q.h2. where Rq is the sum of terms left uncompensated by AW. The first approximation A(°> = Aq leaves uncompensated on the right hand side of Eq. (18.{^ h2. h2.g) 8z+ (g.g ( 1 7 g 2 + 19). (18. g + 4)(g + 4. y'_(q.e.2 ) (0) (18.4 ) ( 0 ) ( g .Aqjr2i in this may therefore be eliminated by adding to A^0' the contribution A^1' given by (9.101a) the contribution Rq . (g.h2.h2.g) Az.0)=0.9-2) ig-4 ' H ~ 23/X4 iz+ An-2 H + (9. which reduces to 2 \/2c 2 0 = 2(3g 2 + i) + — A + ^ g . 9 + 4) ^+4-8z+ (18.0)=0.121) This coefficient of Aq set equal to zero yields to that order the following equation 0 £) ' {-^) (g.h2. (i) _ / V2c\ [ ( g .(0).yA(z) derived above are valid around z — 0.e. of course.g) -4z+ {q q)+ 2r (9. i.120) The sum A = A^ + A^1) then represents a solution to that order provided the sum of terms in Aq in Rq and Rq is set equal to zero. g . z)\ = 2^A^ h2 '>z)± VA(Q. -*)] (18.0) = y^(q. in the domains away from the minima.3 The Double Well Potential we have T>o 411 -2iz+ J Aq+2i except.0) = ^ . for i = 0.9-4) 8z+ (9.g + 2)(g + 2. \Z-Z±\ > 0 ( -y We can define solutions which are even or odd about z = 0 as y±(z) = = £ fe^te' h2> z) ± y A.123) ^[yA(q.z)±yA(-q. i. y'+(q.28.122) The solutions yA(z). (18. g) -8z 4 .g-2)(g-2. thus yielding the next approximation of A as given in Eq. g . h2.105). 9 + 2) Aq+2-Az+ (9.

91). we have w±(—z±) = —2h±z±. the equation is — with differential operator T>q as defined by Eq.125b) It is clear that correspondingly we have solutions yc(z).2A y(w±). Anharmonic Oscillator Potentials Our second pair of solutions. yB(z). (18.93) and setting w± = h±(z — z±).1 ) D (18. (18. Thus yc(z) Cq[w±(-z)} = = Cq[w±(-z)]+0(h±2).95a) and (18.yc(z) providing a pair of decreasing asymptotic solutions there (or correspondingly y~B(z)iyc(z)).3 M a t c h i n g of solutions Next we consider the proportionality of solutions yA{z) and VB(Z).23) with appropriate change of parameters to those of the present case.yB(z) is obtained around a minimum of the potential. and another VB(Z) ± 25^ch3wl + c2wi . Since w±(—z) = —h±(z + z±).95b) — 1 T>q(w±)y(w±) = j£ Thus we write the first solution yB(z) = Bq[w±(z)] + 0(h±2). This means. but w±(z±) = 0. 18.125a) = VB{~Z) = Bq[w±(z)] + 0(h±2) = Bq[w±(-z)] + 0(h±2) (18. we > . (18.yc(z) with complex variables and Cq(w) given by Eq. 2(9+) .95b) that the solution there is of parabolic cylinder type.Evaluating the exponential factor contained in VA{Z) of Eq.3. (18. Bq[w±{ ^ . in this case we use the Schrodinger equation with the potential V(z) expanded about z± as in Eq.126b) These are solutions again around a minimum and with yB(z).We draw attention to two additional points. (18. Moreover. + 0(hl2). Inserting (18. (18.126a) hjW + l)]! VM = Vc(-z) = Cq[w±(-z)] (18.412 CHAPTER 18. in view of the factor "i" in the argument of Cq the solutions VAJVC have the same exponential behaviour near a minimum. We see already from Eqs.99) for z — z±.^ % [i(?-l)]!24(«.

z±)dz . 2 An+z+e 4w+ (18.-frW.« + ) 2 (9+1) e (18.125a) we see that (considering only dominant contributions) in their common domain of validity e ^+z+ l + O 1 yA{z) = -yB(z).129) .• )2 (^)ito-i) ( 2 z + ) 5 ( 9 + i )e ' yA(z) z+> lft2 2 _ I . Here z± = ±h?/2c are the positions of the minima of the potential.127) Recalling that around arg w± ~ 0. (18. .3 The Double Well Potential (cf.-^(±>(^-*4 ^\h\{z-z±f exp 7>2 r 2 ^n±z± For later reference (after Eq.i*l(«-^ 4'"+^+ 4"'+v ( « .-^/w + ± 1 .• • " .146). Consequently the above integral from z = 0 to z± differs from that from z = 0 to a turning point (as in expression /2(C)) of Eq. We observe that for z — z± the approximation yields exp[±/i? t z 2 t /4]. . Eqs.164)) only in a nonleading contribution and hence implies the equivalence of the exponentials in the relation (18.128) Similarly we obtain in approaching z+: VA(Z) - ( 2 ^) |(9 " 1} e. . a= ^ .97b)) exp 413 '-\hlJQ'uV\z)dz = exp = exp (18. (18. 24(9-D[i(g-i)]! hj (18.97b) ~ exp . (18.172) obtained later. Later we calculate the coordinates of turning points ZQ. ein+z+e in+(z |22+||(<?+i) . (18. . and comparing with Eq.18. (18. Thus for h very large these turning points are very close to the minimum at z+. the dominant term in the power expansion of the parabolic cylinder function is given by Dv(w±) ~ wv±e-w±>4. (18.. Allowing z to approach z+ in the solution IJA(Z). .ZI and find that these are given by z+ + 0(l/h) for finite q (cf. we have \z ~ Z+\^Q ^ l f e 2 2 2 _kh2( ~ i .172)) we add here comments on this approximation.147)).

D. The involvement of these WKB solutions leads to problems. Bhattacharya [23]. We have to impose boundary conditions at the minima and at the origin. and this means at both minima.130) Therefore in their common domain of validity a a = dte+vM' {2z+)^. S. K. 18. The next aspect to be considered is that of boundary conditions.414 and CHAPTER 18.131) We have thus found three pairs of solutions: The two solutions of type A are valid in regions away from the minima.z+)]fr+V ' (18. they assume large quantum numbers. Various investigations^ therefore struggle to overcome this to a good approximation. We achieve the same goal here by demanding our basic perturbation solutions to be interconvertible on the basis of the parameter symmetries of the original equation. since. The solutions in terms of parabolic cylinder functions have a wide range of validity.7r and the solutions of type C around argz = ±7r/2. Since it is more probable to find a particle in the region of a minimum than elsewhere.\-h\)^+l\-\{q l + l)]\ 1+ 0 J_ . K. The pair of solutions of type B is defined around argz = 0.(18. we naturally expect the wave function there to be similar to that of the harmonic oscillator. Anharmonic Oscillator Potentials D_h{q+l)(iw+(z))2i(o+V Vc( ) z 2i(g+1)ei^(z_.3. Thus 1 E . P.+)2 [-1(1 + I)]' [-i(q+ l)}\[ih+(z . g . Rau [22]. Thus it is unavoidable to appeal to other methods such as the WKB method to apply the necessary boundary conditions at that point. de Deus [66] and W. basically. Furry [102] . even above the turning points.4 B o u n d a r y conditions at t h e m i n i m a (A) Formulation of the boundary conditions The present case of the double well potential differs from that of the simple harmonic oscillator potential in having two minima instead of one. R. J. and are both in their parameter dependence asymptotically decreasing there and permit us therefore to define the extensions of the solutions y± which are respectively even and odd about z = 0 to the minima. S. H. but it is clear that none of the above solutions can be used at the top of the central barrier. Bhattacharya and A.

y'_(z±) = 0. (18. and y+(z±) = 0. The odd wave function then exhibits a correspondingly opposite behaviour. as indicated there.133) imply yB(z±) yc(z±) « a' and y'B(z±) y'c(z±) a a (18.18. ^(*±)#0 (18.5.135) Hence the conditions (18. y-(z±)=0.136) . At dtz — ±oo > we require the wave functions to vanish so that they are square integrable.5 Behaviour of fundamental wave functions.3 The Double Well Potential 415 the most basic solution would be even with maxima at z±.132). as indicated in Fig. (18. 18.132) Fig.133) y+(z±)^0.5. We have y±(z±) = -^VA{z)±yA{z)]z^. 18. 18. We have therefore the following two sets of boundary conditions at the local minima of the double-well potential: y'+(z±) = 0. an even wave function can also pass through zero at these points.Zii -yB{z±) a ± a -yciz±) (18.134) y'+{z±) ± 0. and y'±(z±) -V'B(Z±) ± =y'c(z±) (18. y-{z±) £ 0. However. as indicated in Fig.

. Thus sin {^ +1) } = ^^>^30fe-^i. Anharmonic Oscillator Potentials (B) Evaluation of the boundary conditions Inserting into the first of Eqs. .141) and cos<J^( 9 + l)|> 4 ~ cos | ^ ( g o + l) | ~^(q-qo) (_l)*fo>-i)( g _ g o )?[ 4 for sin | | ( g 0 + 1) J + go = i.. (18.m Using Eq.«*{=(.. (18. Eqs. (18. . . also (18. (1.2 we can rewrite the second of (18.54) +l j 2^[I(Q-l)]![I(g-3)]!sin{f(g+l)} ^^ ^ r i r ( T(y+me'm' (18 137) - where we used first the reflection formula and then the duplication formula.9.5. (18. +1)} = 4 m.136) as | M . 1 1 .142) -(<Z-9o)J(-l)^°+1)(-l)1/2- .138) Using formulae derived in Example 18.136) the dominant approximations we obtain (cf. -.134) this equation can be written Now sin | ^ ( 9 + 1 ) 1 oi sinl j(q0 +I) \ + ^(q-q0) ~ (_l)^(9o+i) ( g _ g o )| for cos I ^(q0+ !)[ + ••• 50 = 3 .„*{!<. 7 . _ 8)} s -.416 CHAPTER 18.

18.6 Turning points z$.z\.» ^ B ^ ' . Hence we have to impose at z — 0 the conditions j/_(0) = 0. y'_(0)^0. we matched them to the solutions of types B and C which are valid there and hence permit the imposition of boundary conditions at the minima.144) Thus we require the extension of our solutions to the region around the local maximum at z = 0.18. 18. y'+(0) = 0. we must also demand this behaviour here along with a nonvanishing Wronskian. (18.3. Since the type A solutions are not valid at the minima. We emphasize: We needed the solutions of type A for the definition of even and odd solutions. V(z) — • z E=V Fig. y+(0)^0.1 -"-- <813 1 4) - In Example 18.5 B o u n d a r y c o n d i t i o n s at t h e origin (A) Formulation of the boundary conditions Since our even and odd solutions are defined to be even or odd with respect to the origin.3 The Double Well Potential Thus altogether we obtain 417 ^ • . We do this with the help of WKB solutions.5 (after determination of the turning points) we rederive this relation using the WKB solutions from above the turning points matched (linearly) to their counterparts below the turning points and then evaluated at the minimum. We deduce .

+ £--\h%U(z)=Q. in leading order the integrals to be studied below from ZQ to z = 0 are equal to those from z+ to z = 0. (B) Evaluation of the boundary conditions We now proceed to evaluate the boundary conditions (18. one finds that .24 -C Z O 1/2 ZQ = Ac2 + + °(r A_ 2c 1 2 2q\ 1/2 2 / cq 5 + h± + 0(h~ ) h%) (18. h8 25c2 (18.^1 25c2 + V2 2 Thus again we see that for large values of h2 the turning points are very close to the minima of the potential for nonasymptotically large values of q. (18. Anharmonic Oscillator Potentials from Eq.148) ("f * "The superscript (I.147) Since the minima are at z± — h2 /2c with /i 2 very large. the dominant terms of the WKB solutions are* 1 yWKB\ ) Z 1 1 ~l/A 1/2 ~ -qh\ x exp + -h%U{z) . ZQ) means "to the left of 20" .145) \qh\ + \zW Using z+ = h2 /2c. In the domain 0 < z < ZQ. i. we see that the turning points ZQ and z\ are very close to a minimum for reasonable values of q ~ go> the latter being an odd integer. As a consequence.96) that the two turning points at ZQ and z\ to the right of the origin are given by \qh\ i. to the left of ZQ where V > E.418 CHAPTER 18.e. We also note that the height of the potential at the turning points is h8 qh V(z) 20.144).146) and Iqh* 1/2 z\ Ac" + +o ^ (2g2)V4 2c + /i ' (18.e.\qh\ + \h%U{z) (18.

^ e therefore have to consider the exponential factors occurring in (18. )+^-*+)2-lr} rnD h\z'2)}z Q )\ = .149) and consider both types of solutions in a domain approaching but not reaching the minimum of the potential at z+. Hence we have for ywKB the exponential factor exp ("f * Z+ \*l*e\*(\h% 1 1/2 \qh\ 9/4 + \h\U{z) e-\{z-z+?h\_ (18-152a) . . we have to match yA(z). (18.(29/^)1/2 1 ^(z-z+){(z-z+) 1/2 2 2q_\ 1/2 hJ 2 l l q+ q | ^ ( ^ .149) In order to be able to extend the even and odd solutions to z = 0.^ + ) 2 + igln|2(z-z+)|.18.^ j O e .* 4+ .3 The Double Well Potential and -1/4 419 ywKB\ ) z — \qh\ x exp I / + \h\U{z) 1/2 I.148) and (18.150) Evaluating this we have ±h ~ -h2 GMln 1 In 2g q In +4 1 (z . Thus we consider in the domain \z — z+\ > {2q/h2+)l/2\ h -1.S2 ) dz rzo 1/2 '-Ihl + ^Uiz) I 1/2 / Jz dz (z~z+) - w + 1/2 1 = ±lh+ 1/2 ZQ-Z+ Z — Z+ 2 i^-$) 1/2 ~4 >'^.y^(z) to 2/VVKB(Z) anc ^ ^ W K B ^ ) . (18.151) In identifying the WKB exponentials we recall that y^KB is exponentially increasing and ywKB is exponentially decreasing. 20 dz \qh\ + \h%U{z) .i (2q ln (18.

-|l/2 ~h\U{z \(z-z+?h\ (18. t h a t the Stirling approximation is amazingly good even for small values of the argument).153) This expression is valid to the left of the turning point at ZQ above the minimum at z+. of course. since there is no way to obtain an exact leading order approximation with Stirling's formula for small values of q. Thus using the Stirling formula z\ ~ e~(z+1\z + l)z+2y/2~rr. Stirling's formula is the dominant term of the asymptotic expansion of a factorial or gamma function and thus assumes the argument (oc q ~ 2n + 1) to be large (it is known.126a).154) .ZQ) . However.125a) and (18.125a). This is the aspect investigated by Furry [102]. \ yWKB\ ) z 1 fl(g-3)]! V*(h%)V*\z-z+\h(*+V\2 -q/A h + e* \(z-z+)W+t (18. In a corresponding manner — i. using Stirling's formula (and not the inversion relation) — we have —(1. the results necessarily require adjustment or normalization there in the q—dependence. Anharmonic Oscillator Potentials We observe here t h a t the W K B solution involves unavoidably the quantum number dependent factors exp(g/4) and (q/A)qj/4 which do not appear as such in the perturbation solutions.e.420 CHAPTER 18. we can write the exponential as exp (27T) ("f 1/2 I dz Z+ 1 qh\ + q/A 1.152b) [\(q 1)!] V2 K7: hi Here q/A was assumed to be large but we write [\{q — 1)]! since this is the factor appearing in the solution (18. Since correspondingly -1/4 2 l/2 \*% we obtain + 4/1W*) (*-^)i/'(M)i/*' 27T 1 / 2 {h\)^{\{q-1)]\\2"+ *+ 1(9-1) ~hi l2 q/A e \{z-z+)*hl for \z — z+\ > m 2q \ (1 ' (18. We see therefore. The only way to relate these solutions is with the help of the Stirling formula which converts t h e product or ratio of such factors into factorials such as those inserted from the beginning into the unperturbed solutions (18.

e.3 The Double Well Potential 421 where [\q}\ was written as [\(q — 3)]! for q large.153).154) with the type-A solutions (18. Comparing for z —• z+ the W K B solutions (18.18.156b) Using again the duplication formula^ the ratio of these constants becomes 1 7 (h2+)i/2(2z+)i [fa-W e 2n+z+ — Aft 2 72 (18.159) 7 -i(«-i) r(9^3) ^2~ j(9-l) . we obtain in leading order (i. (18. (18.n('>2o) 3/WKB(Z) =1VA{Z)I (18.155) (2z.123) we have V±(z) 1 -[yA(z)±yA(z)} = ^y^l(z) ± ^j#&(*) (18.K?+1).143) which was obtained with our perturbation solutions from the boundary conditions at the minimum.104). \hl. Returning to the even and odd solutions (18. However.144) we obtain fi(0) Tin the present case as 1 (9-1) 1 — 5 (18.156a) and 7 _ i\(Q-m •K •/2(M )l/4 ^ 2 n« -q/4 (2*+) •|(9-D e ^^4. . . (18. multiplied by (1 + 0(\/h\))) the proportionality constants 7 . and is shown to reproduce correctly the result (18. [\{q — 3)]! are really correct replacements of [\q}\ only for q large.157) is used in Example 18. 7 of t h e matching relations y^NKB\z) i.103) and (18. this result is somewhat imprecise.e. 27T1/2 9/4 1VA{Z).5 for the calculation of the tunneling deviation q — go by using the usual (i.157) Since the factorials [^(q — 1)]!. it is our philosophy here that the factorials with factors occurring in the perturbation solutions are the more natural and hence correct expressions.99).e. as the results also seem to support. linearly matched) W K B solutions. (18.143) T^^-*))- 27T (18. The relation (18.158) Applying the boundary conditions (18.

G2 K [i 4.148).164) / 2 (0) z ca [(1 + k2)E(k) . 213. 60 and 361. n + uY 2c „ ^6 8V2c2 (18. (18.k'/2.. We have 1/2 dz Jz \qh\ + \h\U{z) dz Jz 4 ^-? lfh 2\22c dz ¥+ A4 l 4 3 1/2 1/2 dz Jz — cz h4 CZ qh\ dz4 qh\ <Wc h+ 1/2 Jz zo 2V*\Vc h 4 ) T Jz 2\/h ~~2 /2(^--2 —pz I V2 dz ^?-^-- h+ Z 4? + ^f ^2<«2 (18.162) T h e integral appearing here is an elliptic integral which can be looked up in Tables.iV2 ! i . (18. Friedman [40].161) we can rewrite the integral as r I2{z) = -= dz^{a2 . K(k)} 3 ^ 2 y/T+u[E{k) uK(k)\. Byrd and M. Anharmonic Oscillator Potentials Thus we have to consider the behaviour of the integrals occurring in the solutions (18.3 evaluated at z = 0 is then given by = y/T+u[E(k) uK{k)} (18.149) near z = 0. formulae 220. D.-t The elliptic modulus k (with k' = 1 — k2) and an expression u appearing in the integral are defined by b2 1 U kz = -^ = l + u so t h a t = u=S-^q hi = GV2q. . 3G2 12V2c ^See P.163a) (18.160) Setting 6 =4?-^T"z°' rb Ac2 + VQ—> (18. p..163b) The integral hi?) .05. . F.. p.422 CHAPTER 18.19.z2)(b2 - z2).

.± ' 3G 2 ' T ( ^ (18.3 The Double Well Potential 423 where K(k) and E{k) are the complete elliptic integrals of the first and second kinds respectively.§ Since the integral is to be positive (as required in the WKB solutions) we have to take the upper signs. 1.18.165) in agreement with Ref. which implies small G2. ^ ^ .2 A( L ) 4 s / and z dz yWKB\ J and d^WKBW 2=0 (*?) a ( ..1 " e " O T (18 169) - Y-^-V / 4 tfo. . .e T ^ 5 U W ( 2 „ ) i / 2 ( ^ i W4C • (18 170) (18 1?0) - ^The expansion of I2 used in Ref.( 2V + l)ln V 4 / -4( 2 n + l ) l nV 2 n .. (G\ .3 )l ! . ± . where the result is shown to be (with q ~ q0 = 2 n + l . .1. n. [167]. Here we are interested in the behaviour of the integral in the domain of large /i 6 /c 2 . . n = 0. The nontrivial expansions are derived in Example 5.r1)/10//2 ( f g 1^ yS» /L ( ^ ) l^4 .166 ) ^^-iffil/^(2*)1/affiii£ Correspondingly we find ^ (18-168) S W I 2 = o .. /2n + l 2 ± — .1)]' Gfo.) = 2 \.( ft 7.2. [4] misses an n—dependent power of 2 in the result. It then follows that (again in each case in leading order) '2_hl)9/4 ^ ^ * > l * = ° ^ 23g/S(_g:)t/4(!)g/4e-g/4 "1^Using Stirling's formula we can write this e ( 18 .

.^ W ^ / 4 t o .6 Eigenvalues and level splitting We now insert the replacement (18.424 CHAPTER 18.3 concerning the exponentials.143) the replacement: {hlY'2{2z+fe-^zl -+ 2 9 ( ^ y Z e"A. (18..3. (18. (18. 18. (i*™) We obtain the energy E(q.3. 2 . (18.159) — on using once again the duplication formula in the same form as above — 2"/ 2 (2h% 1 V/2 7 Comparing this result with that of Eq.173). 18.174) .1.172) A corrresponding relation holds for h+ and z+ replaced by /i_ and z-. we can therefore impose the boundary conditions at z = 0 by making in Eq.106). Thus our second condition in the present case of the double well potential turns out to be an identity which confirms our discussion at the beginning of Sec.. (18. we see that the relation is really an identity (the pre-exponential factors on the left and on the right being equal) with the exponential on the left being an approximation of the exponential on the right (which contains the full action of the instanton)..h2) + (q-q0)^=.i ) ] ! e ~ s A ' . Expanding this around an odd integer qo we have ( 8E\ h? —J Inserting here the result (18. and obtain ( * . . .143) with qo — 2n + l..157). we obtain for E(q0.175) ~E0(q0.»-W. Inserting the expressions for h+ and z+ in terms of h.h2) and hence the splitting of asymptotically degenerate energy levels.n = 0. (18.172) into Eq. Anharmonic Oscillator Potentials With these expressions we obtain now from Eqs. with the help of Eq. (18. h2) the split expressions 29o+1/i2(—W2 h6 (18. .

Combining Eqs.176) is described by Bhattacharya [23] as a "modified well and harried result AjfWB.e.V2^\ \n+ i . i n U n / ! Thus AE(q0.7 =( e N n+1 .fr2) is given by Eq..^ 2^ h* .h2) = E_(q0.106). T \ *n ! 1 ' (18. (18.18. which evidently supplies some correction terms.h2)-E+(qQ. 1 u2 c 2 (3 9 2 + l) ^ « b .3 The Double Well Potential where £0(<7o.q2 7.. the pure WKB result of Banerjee and Bhatnagar [14] (i. i. K. in the WKB restricted sense) defined by ^( n +2^KB(0) 2 Here the derivative dE/dn corresponds to the usual oscillator frequency.e. Bhattacharya [23] the "usual WKB approximation" of this expression is — with replacements h4/4 <-» k.^ e 21/^6c2 fe6 ( 1 mass mn = (1 8 . the level splitting.157). (18.1 7 6 ) (2n+9)/4^/ft V7T«! V C / \ e-iI72^.. 2h4' i.h2) / h 6 \ —K />6\n+l/2 q o / 2 2«>+2h2 — —^= m 1. c 2 /2 <-> A — given as E0 = 4\ h(2fc) 1 / 2 o y J H q2 4/c > --^— + q ^ 2s c2 V2 .2 = 2 . .2. for finite h2. the following expression for large q ~ 2n + 1. that without the use of Stirling's formula and so left in terms of e and nn) being this divided by the Furry factor f l / » : = \ .e.1. 1 7 425 2 .n = 0.174) and (18.e.e.. The result (18.178) "In S. i. the difference between the eigenenergies of even and odd states with (here) qo = 2n+l. can be given by In the work of Bhattacharya [23] the WKB level splitting is effectively (i.143) with (18.

these deriva> tions do not exploit the symmetry of the original equation under the interchanges q <> —q. h4 and c2 replaced by 2E.181) "W.426 CHAPTER 18. Then the splitting A ^ W B of Eq.h2 <> — h2. the Furry factor corrected WKB result follows automatically. .3)). Then E^h2) and AE becomes 2^+U 2 (-^)W 2 h6 = Eo{q0. (24) there is k 2 (i0n+i3)/4 //c3/2\n+1/2 V2fc3/2 6XP n\ ft \/nn\ in agreement with AE(q0. that if this symmetry is taken into account from the very beginning." Of course. This. ** Comparing the present work with that of S. Furry [102].180) h8 E0(qo. and A there with c 2 / 2 here.** Had we taken the mass mo of the particle in the symmetric double well potential equal to 1 (instead of 1/2).6) to yield an improvement of WKB results for small values of n.. (18.^ 2 + ^h2 If in addition the potential is written in the form A / . we identity the parameter k there with ft. The Furry factor represents effectively a correction factor to the normalization constants of WKB wave functions (which are normally for mo = 1/2.178). We see therefore. (18. The factor set equal to 1 represents a somewhat "mutilated" Stirling approximation. and harmonic oscillator frequency equal to 1 given by \/y/2~n and independent of n as explained by Furry [102] — see also Example 18.h2) * ^ 2 q o / ^ q l _ m ^ - (18-179) l~o o». h = 1. A>0.4/4 here. Eq. K. (66).2 / 7. h2)\mo=i = . as is also explained by Bhattacharya [23]. (18. we would have obtained the result with E. (18.6 \9o/2 e~h&l^\ with (m 0 = l) (18. supplies the bridge to the perturbation theory results and removes the unnatural appearance of factors e and nn. 2/z4 and 2c2 respectively (see Eq.176) together with Eq. Anharmonic Oscillator Potentials which is unity for n — oo with Stirling's formula.h )/2 2 2 ( — ' 3 6\/2c A h6 n+l/2 h6\ 2{2n+5)/if \ exp 2 \ c / of Eq. Bhattacharya [23]. however. In Chapter 26 we obtain in each case complete agreement of the path integral result with the perturbation theory result with the help of this factor. as we do here. H.2X2 V(z) = --\z2-tL) .

11). M.^ *See E.—Q. these are presumably not of much interest in physics (for reasons explained in Chapter 20).15). Banerjee and S. Jentschura [293]. however.34) and (E. The comparison with Eq. (18. . The asymmetric case can presumably also be dealt with in a similar way since various references point out that the asymmetric case can be transformed into a symmetric one.§ 18. Jentschura [293]. first paper. Mansour and H. §J. K. W. Equation (18. J. Bhatnagar [14]. formula (4. (18.88) h4 = 2/x2. We considered above only the symmetric twominima potential. In principle one could also consider the case of small values of h2 and obtain convergent instead of asymptotic expansions. Gildener and A. J. H. Eqs. (18. Friedberg and T. Miiller-Kirsten [186]. Lee [98]. carried out along the lines of the corresponding calculations for the cosine potential and thus of the wellestablished Mathieu equation. D.W. Patrasciociu [110]. so that by comparison with Eq. provided their result is multiplied by a factor of 2 resulting from a corresponding inclusion of anti-pseudoparticle (anti-instanton) contributions. J. M.182) implies AlE(l.^ Thus for the case of the ground state Eq.c2 = A/2. See R.3. P.3 The Double Well Potential 427 a form frequently used in field theoretic applications. (2. Miiller—Kirsten [167] the potential is written as V2 VW = \ for mo = 1. Zinn-Justin and U. Liang and H. the level splitting is (with h = 1) This result agrees with the ground state (go = 1) result of Gildener and Patrascioiu [110] using the path integral method for the evaluation of pseudoparticle (instanton) contributions. Zinn-Justin and U.h>) * 2V2^^|^y/2e-^3/3A. D.7 General Remarks In the above we have attempted a fairly complete treatment of the largeh2 case of the quartic anharmonic oscillator. 'This will become evident when the perturbation theory result of Chapter 17 is compared with the path integral result in Chapter 26.18. m 2 <-• p 2 / 2 .181) therefore implies the correspondence rj1 <-> A/2. and g2 is given as g2 = r]2/m3.180) agrees similarly also with the result for arbitrary levels obtained with the use of periodic instantons* and with the results of multiinstanton methods. ^To help the comparison note that in J. D.* A similar correspondence is also found in the case of the cosine potential.

A reasonable. Martin. Mahapatra. Simon and Wightman [178]. The ground state splitting of the symmetric double well potential has been considered in a countless number of investigations. Anharmonic Oscillator Potentials Every now and then literature appears which purports to overcome the allegedly ill-natured "divergent perturbation series" of the anharmonic oscillator problem. as he discusses. mostly in connection with instantons. As references in this direction. Sophisticated mathematics — like that of the extensive investigations of Turbiner [273] over several decades — seems to approach the problem from a different angle. B. as some relevant references with their view we cite papers of Harrell and Simon [129]. The double well potential. The immense amount of literature meanwhile accumulated for instance in the case of the Mathieu equation can indicate what else can be achieved along parallel lines in the case of special types of Schrodinger equations. Tables of properties of Special Functions are filled with such expansions derived from differential equations for all the wellknown and less wellknown Special Functions.428 CHAPTER 18.g. in both symmetric and asymmetric form. B. has also been the subject of numerous numerical studies. There is no reason to view the anharmonic oscillator differently. though incomplete list of references in this direction has been given by Garg [105] beginning with the wellknown though nonexplicit (and hence not really useful) ground state formula in the book of Landau and Lifshitz [156]. and numerical studies are presented to support this claim J It will become clear in Chapter 20 — as is also demonstrated by the work of Bender and Wu [18].[131]. Santi and N. Pradhan [182]. [132] and Loeffel. [19] — that the expansions considered above are asymptotic. though not exclusively numerical. in principle its Schrodinger equation is an equation akin to equations like the Mathieu or modified Mathieu equations which lie outside the range of hypergeometric types. in his recent paper on simple uniform approximation of the logarithmic derivative of the ground state wave function of anharmonic oscillator potentials. In fact. . Perturbation theoretical aspects are "See e. Ashbaugh and Harrell [12] and Harrell [130]. The wide publicity given to the work of Bender and Wu made pure mathematicians aware of the subject. like those for anharmonic oscillators. we cite papers of the Uppsala group of Froman and Froman [100]. Very illuminating discussions of double wells and periodic potentials. can be found in an article by Coleman [54]. P. since — as we shall see in Chapter 20 — these nonperturbative contributions are directly related to the large-order behaviour of the perturbation expansion and determine this as asymptotic. There is no way to obtain the exponentially small (real or imaginary) nonperturbative contributions derived above with some convergent expansion. for instance. N.

fc ~ 1 . Datta. Saxena. Cooper and Strottman [221].190) Hence ln|-]=ln : l n ' 7 2 ^ + 2 (18. We have 4 ~k' 2u(l .u)1'2 = V2u{ 1 .^ + • (18.f + • • •) 2u\ .192) 'F 12 16 .185) (1 + u)^2 = (1 + GVW/2 = 1 + GJ±. We obtain the expansions from Ref.184) Hence we obtain for G close to zero: 1-Gy^g l + Gy/2q~ and 1 . (18. Bender. Biswas. 1+ u : Gy/2q. (18.186) (18. X +2 + ' ^ (18. + (G 3G2 3^ + 2lnU 2 + i(2n + l)ln(f)+I(2n ~~ 3G 2 ' 2 l)ln(^±i q 2 1 ^ln q/ 4 2n + l (18.18.191) Our next objective is the evaluation of the elliptic integrals E(k) and K(k) by expanding these in ascending powers of k' .187) We now reexpress various quantities in terms of u. which is assumed to be small./4 E{k) = 1 + • l n In (18. T h e r e f o r e it is convenient t o deal w i t h this here. Wave functions of symmetric and asymmetric double well potentials have been considered in the following reference in which it is demonstrated that actual physical tunneling takes place only into those states which have significant overlap with the false vacuum eigenfunction: Nieto.G2q + --.3 The Double Well Potential 429 employed in papers of Banerjee and Bhatnagar [14]..\G\ + • and 1-k2 =2Gv/2q-4. 2 a. [122] as 13 1. Gutschick.5: Evaluation of WKB exponential with elliptic integrals Show that (l + u)^2[E(k)-uK(k)}'.188) T h e following e x p r e s s i o n a p p e a r s f r e q u e n t l y in t h e e x p a n s i o n s of elliptic i n t e g r a l s .2G^2q + 4G2q . k! (18. Thus k' and hence k' = v ^ ( l .183) S o l u t i o n : We have k2=1-^.189) =2u- 2u2 (18. q ~ 2n + 1. Srivastava and Varma [24] and Bhatacharya and Rau [23]. Example 18.

^ L ) H[(i _ U ){4 + 3u(l . 8 + 6 u 2 . .3u(6« .430 and CHAPTER 18.3u(6u ..u) 2 {24 .194) ln '^J + 2 In i/2u )+H l + 2u(l . (18.3 « .u){24 .u)} + — u ( u . ) J 4 (18.6 in the denominator and in the last step pick out the lowest order terms in u (i.3) .u ( l .2{«(1 .4.1)(1 .a = (l-«)(j-l 2u/ 2 ) .u) 2 {8 . 8 + 9u 2 .193) 2u(l . 4 \ u In —= I + V W 2 1 + ln x(l -u) 31/ +-.3 u .3) = = ~ .u) + 2} + i u ( u .3 u ( l .u){24 .e.195) we obtain E(k) 1 + ln uK(k) .18u 2 + 39M} + 12u 3 (u . (18.u ( l .13)}.u)} u{u(l .u){24 . 1 3 « 2 = .u 3 ( u .3).1)(1 .u ) (^){« + 5^ 1 -)} + 5.u)} .195) With (18. Anharmonic Oscillator Potentials *« = M £ H Consider first E(k): E{k) = 1 + . (18. 8 .196) Now consider the last line here without the factor 2.3u(6u . up to and including u2): u(u .3 .3) 1 + In ^ .u 13 \ 1 ln(± k> fc'2 + - (18.« 2 + 3u] In [ .u) + -uf\ 2 + ju2(l 16 + -(u- W 2 12 4u2(l-u)2 + - l ) u ( l .u) x2 2/1 u2(l-u)2 H K 42l We can rewrite this as — Analogously we have uK(k) 6J ' — {4 + 3u(l .13)} + ±u3(u .3 y .3) .u)(u .u) + 2}] 1 H 1 o u(u .13)} + ~u3(u .1)(1 .{l u ( l -u) / W 2 + 2} .u) + 2u} + i u 2 [ 2 + (1 .3u(6u .« ) ^ { 2 4 .{ M 2 ( 1 .u){4 + 3u(l .2)1 4 uA"(fc) In ( ^ = ) ^{«(1 .«) + 2} + V [ .194) and (18.1)(1 .197) .13)} + 12u 3 (u .— ) .6w2 + 13u} + 12u 3 (« .

/ to + 3GH~ \ 8 .155)) <*VA{Z) = VB{Z) a n d VWKB(Z) = 1VA{Z). Eqs.2 (18.198) 9 «..173) t o be evaluated.^HfH-GH Example 18.3 The Double Well Potential Now consider the bracket [.195) with (18. 2 .uK(fe) ~ 1 + In I — = M-u2(3u.e. V^LJ 4 2 16 ^ . i. i.2 /o 2 431 (18.128) and (18.199) We now return t o the expression on the left of Eq. (18. we obtain: h ^ —^ 1 + 3G 2 V 2 8 2 /„ u 1 . (18.2{w(l -u) + 2}] = u(3u2 . we have.196). where w±(z) — h±(z — z±). [(1 .148) (for particle mass 1/2 and in dominant order) is given by H.3).3) .e.2 l_ln(1 JL\ V -{2. (18. (18.. (18 201) = Thus 3ffl-« (2i/aGi/a2i/V/V"4 3ffl-2 (Gii7»J-4- - '• .«){4 + 3u(l .J? dz^\E -V(z)\ \ 87T2 / lE-Vtz)]1/4 Solution: From Eqs.197) we now obtain 4 1 • E(fc) .-u H I .zo) HWKB normalized _ ( h* y^eM. + iu^Zu2)^~u+ — +0(u3) l . ' 8G 2 q = 4 2 q ln f 26/2 \ o . this becomes _2 2 u 2 / j M 2 / 3^2s 2 ~ 3G 2 2G2 H \ / 2 H / .18.4u .4u . .156a) we obtain a and 7 and hence the ratio (always in the dominant order) T2~ ) w i t h h += ^ h • Since (cf.6: Normalization of WKB solutions Show that the normalized form of the WKB solution (18. l2 = ^ ( l + uf/2[E(k)-uK(k)}. From (18.200) Remembering that u = Gy/2q.. Inserting here from the above expansions the contributions up to and including those of order u2.196) and (18.] in (18.128) and (18. VB(z) = -y\hTB(z) with VB(Z) ~ Bq[w±(z? - ^ [±(q - J l)}\2-^-V .it)} .

normalized = We see that the normalization constant is independent of a quantum number. W.158). -oo .*o)/ ^WWKBOO- (18. Oberhettinger [181]. Anharmonic Oscillator Potentials With the standard normalization of parabolic cylinder functions.e.432 CHAPTER 18. pp.3)]!23<«-D [5(9-1)]' >A. Then ((r.** oo / dwD\. (26.147).Aw) = s/2i 2("-!) V (4-1) and [|(9 . Magnus and F. normalized - V4exp[ ~ f*° dz\/\E~ \E-V(z)\V* V z () With / i 4 / 4 = m 2 and mass 1 (instead of 1/2) the result is ' m2 \ 1'4 e x p [ . Eq.e. Example 18.204) . we obtain (for q reasonably large) 1 |(9-1) and hence J-oo Hence with h+/V4n = (h4/Sn2)1^ W2TT SI {[Uq-IWH"-1)}2 2TT h2+\1/2 ~VWKB \ ~^T f 4^ \V4ir ^g^ J VWKB . 1 7(l. i.143). (18. by using the periodic WKB solutions below the turning points.zo). The constant obtained here will arise in Chapter 26 in the evaluation of the normalization constant of the WKB wave function (cf. i. (18.146) and (18. (18.35)). ° \Edzy/\E-V(z) V(z)|V4 VWKB. 93. We start from Eq. y±(z) = -IVA(Z) ± yA(z)\ = — i M O O ± 1 Ji.l)]![i(0 . Solution: The turning points at zo and z\ on either side of the minimum at 24.7: Recalculation of tunneling deviation using W K B solutions Determine the tunneling deviation q — go of q from an odd integer go.203) Different from above we now continue the solutions (in the sense of linearly matched W K B solutions) across the turning point at zo in the direction of the minimum of the potential at z+.f hi\ VWKB. (18.g. are given by Eqs.hA/2 = h^/4 this implies . and obtained as Eq. 82./ . zo) meaning t o the right of ZQ and note the asymmetric factor of 2) -1-1/4 y±(z) ^ W K B (z) ± ^ = % K B ( Z ) : \qh\ 1/2 \h\U{z) ±— cos 7 **See e.

g. (18. JV = 1..207) (18. 6. 7T\ + -^ 4 7r\ dz -. We could.3.J^Wj +(18.206) !/2 ' + J*+dz(^qh2+-±h%U(z) 1/2 7T' 4 1 =p — sin 7 f*+dz(^qh%-h\U(z) ^ TV + 4 (18.133) at the minimum z+ and obtain the conditions: 0 = — sin 7 and 0 = — cos 7 Hence £ + 1/2 dz(±qh\-\h%U{z) 4_ ± — cos 7 />G^ 2 2 +-i4c/(z)) 4 (18.. i.210) Choosing the point z to be z+.( .-r + ^=co S [r.209) "27 In the present considerations we approach the minimum of the potential at z+ by coming from the left..18. A.2. this implies sin cos sin cos Thus e.Jz0 sin ^o J LJzo 4 Jzo 2 y •••-4-y ••• = ..2. .208) and cot L />G + 1/2 ^(ig4-^+^))"" + J 1/2 2qh%--h%U{z)\ 27 + - :±- (18.. Sec.6.132) and (18.e.i ) N c o s / •••+! provided the Bohr-Sommerfeld-Wilson fZ1 dz^/E J zn quantization 4 condition holds.iK -h\U{z) + -/ 4 (18.3 The Double Well Potential We also note that d dz 433 y±{z) >2+ 1 -F = 7 sln -h%U(z) 1/4 f X —2 cos 7 2 2^ + 1 \ 1 / 2 71 . z{) we expect tt (18.-^-r + + .205) We now apply the boundary conditions (18. Messiah [195]. of course.e.g. i.V(z) = r J zo I ZQ dz[\qh\ \ Z -h%U(z) (27V + 1 ) . Then at any point z € (zo..212) tt See e. from ZQ. approach the minimum also from the right. 1/2 .211) 7 z l . i. from z\. \h\U{z) 1/2 . ^(7(z) r t J z0 •I Zi I V2 dz \qh\ 1.e.

149).209) and (18. 2Sm (2 g /ft2)i/2 in agreement with the right hand side (the last step following from Eq. (18.205).90 — =F— for q0 = 1. (18.e.2 . Bhatnagar [14]. In Eq. where u> is the oscillator frequency of the wells. (18. The formula thus agrees with that in the literature.h2) = = h2 h2 1 -y 1/2 2{E-Eo)=2^{q~q0)=2^±l V2 V2h2 .148) and (18.27 9 . Lifshitz [156]. + l ) j } ^ 1_ 27' (18. We note here incidentally that this agreement demonstrates the significance of the factor of 2 in (18. . (18. M.215) we can insert for 7 / 7 the expression given by Eqs.exp ( \/2h2T V2 7r 7 dz l / -qh\ + \h\U{z) (18. (18. 5. 9._ £/>(£-<•->•) tan{(g + l Since tan ) ^ ^ .3. i. Landau and E. 7r"y in agreement with Eq. and cot< (q + 1) '!}-» for q = 50 = 1.143). Banerjee and S.. 2IO/-K.215). We now see that Eqs.Zi / r. they become c o t { ( .. The prefactor 2\/2h2/n is.X+-+4 7T We rewrite the quantization condition in the present context and in view of the symmetry of the potential in the immediate vicinity of z+ as r J z0 1/2 dz qh [\ +-\h+u^ \ 1/2 •K (18.217) exp ( V [ ° dz^/-E J-ZQ + V{z) This expression represents the WKB result for the level splitting and may therefore be called the WKB level splitting formula. K. One may note that the exponential factor here is not squared as in decay probabilities (squares of transmission coefficients). Bhattacharya [23]. 7 . in fact (with our use of qo). D. . S.215) which results from the factor of 2 in front of the sine in the WKB formula (18. the level splitting is given by AE(q0.216) Inserting this into Eq.214) {«+"i} 0 for q = go = 3. The latter give 7 7 1/2 exp dz dz \qh\ \qh\ + + \h%U{z) 1/2 exp I — / \h%U(z) (18. and L.215) we can expand the left hand sides about these points and thus obtain .210) assume a form as in our perturbation theory.434 CHAPTER 18. (18. . P. ' /^i .213) The integral on the left can be approximated by 1. (18.146)).5. K. Anharmonic Oscillator Potentials Similarly under the same condition (-l)Jvsin J Zn where it is understood that qh2/2 ~ Ey(z±). . 1 1 . W "Cf.

J. Feshbach [198]. The centrifugal potential oc r~2 is generally considered as exceptional and is treated in detail in wellknown texts on quantum mechanics. 435 .* However. Some decades ago — before the discovery of W and Z mesons which mediate weak interactions — and before the advent of quantum chromodynamics. Giittinger and H. Case [45]. n > 2. W. in an early investigation Case^ showed that potentials of the form r~~n. there is no need to have only field 'Generally a potential more singular than the centrifugal term in the (3 + l)-dimensional Schrodinger equation is described as singular.Chapter 19 Singular Potentials 19. the investigation of singular potentials in nonrelativistic quantum theory was motivated by a desire to obtain a better understanding of the (then presumed) singular nature of the nonrenormalizable weak quantum field theory interaction.* The physical analogy between singular field theoretic interactions and singular potential scattering of course breaks down at short distances. also in P. are not as troublesome as one might expect.1 Introductory Remarks Singular potentials have mostly been discarded in studies of quantum mechanics in view of their unboundedness from below and consequently the nonexistence of a ground state. W. In particular Case pointed out that for a repulsive singular potential the study of scattering is mathematically well-defined and useful. H. Morse and H. M. Nonetheless it was thought that a certain formal analogy could be seen if the field theory is supplied with Euclidean spacetime concepts at the expense of sacrificing the interpretation of the interaction in terms of particle exchange. Miiller-Kirsten [9]. since no probabilistic interpretation is available for the field theory matrix elements in virtue of creation and annihilation processes during the interaction. Aly. 'For a review from this perspective with numerous references see H. f K . However. M.

e.2) § C . . our presentation below is deliberately made elementary and detailed so that the reader does not shun away from it. This property singles this case out from many others and attributes it the role of one of very few explicitly solvable cases. J. Maldacena [41]. Tamaryan. Recently.-Q. This chapter therefore gives the first complete solution of a Schrodinger equation with a highly singular potential. but also that the S-matrix can be calculated explicitly in both the weak and strong coupling domains. 19. Park. this particular singular potential plays an exceptional role in view of its relation to the Mathieu equation. Callan and J. We shall see that not only can the radial Schrodinger equation be related to the modified or associated Mathieu equation (i. which are always worth studying in view of the insight they provide into a typical case and the didactic value they possess for this reason. Singular Potentials theories in mind. R. Manvelyan. J.2 19. W.1 T h e P o t e n t i a l 1/r 4 — Case of Small h2 Preliminary considerations We consider in three space dimensions first the repulsive potential V(r) = 4 - (19-1) The radial Schrodinger equation with this potential may then be written 2 z y + k W + l) 92 V = 0. for instance. (19. the attractive singular potential 1/r4 was found to arise in the study of fluctuations about a "brane" (the D3 brane. D. Hashimoto [124]. G. W.2. Gubser and A. In fact. § There one could visualize this scattering off the spherically symmetric potential as a spacetime curvature effect or — with black hole event horizon zero — as that of a potential barrier surrounding the horizon (shrunk to zero at the origin). H. that with the hyperbolic cosine replacing the trigonometric cosine). In view of the unavoidable use of Mathieu functions expanded in series of Special Functions. Singular potentials arise in various other contexts. N. S. It is therefore natural that we study this case here in detail. D referring to Dirichlet boundary conditions) in 10-dimensional string theory. some also permitting further research. S. Liang and Yunbo Zhang [189]. Miiller-Kirsten. K. J. S. roughly speaking a brane is the higher dimensional equivalent of a membrane visualized in two dimensions from which the DS brane derives its name. H.436 CHAPTER 19. In the concluding section we refer to additional related literature and applications. Miiller-Kirsten and Jian-zu Zhang [226].

Manvelyan.4) first for small values of h2 and then for large values of h2. (19. Thus we can expect to find solutions very similar to those of the periodic case (with — for instance — cos replaced by cosh) and with the parameter v given by the expansions we obtained previously. .mo = 1/2 and h = c = 1. J. We begin with the derivation of various types of weak-coupling or smallh2 solutions which we construct again perturbatively. The radial Schrodinger equation then assumes the form 2h2 cosh 2z .7 that we employed also in previous chapters. Eq. This is rarely possible and therefore this case deserves particular attention. J. Our ultimate aim is the derivation of the /S-matrix for scattering off the singular 1/r4 potential. [l+2) ~ * A> ~ the equation converts into the periodic Mathieu equation of Chapter 17. Miiller-Kirsten. Liang and Yunbo Zhang [189]. The following substitutions are advantageous: y = r1/2cf>. \[\ in the correspondence is nonintegral. 8. We observe that with the replacements p2 z^iz.3b) 7 = % j-. It is clear that we draw analogies to our earlier considerations of the periodic potential. and so -=4 5 and n ez = -r = ~ = e-^4fir. h2 = ikg. h = ei^y/k^. (19. (19. This has advantages compared with higher dimensional cases. H.2 The Potential 1/r4 — Case of Small h2 437 where E = k2. 19.^ where \f\ can be an integer and thus leads to singularities in the expansion of v (see e.3a) In the case of the attractive potential (as in the string theory context referred to above. (19. In the following we study the solutions of Eq.3 below) g2 is negative and hence h? is real for E > 0. R.-Q.( I + 0.19. Subsequently we derive the same 5-matrix from the consideration of largeh expansions and calculate the absorptivity in the case of the attractive potential. It may be noticed that since I is an integer. r = 7 e z . and as we have in mind in Sec. using the method of Dingle and Miiller of Sec.4) dz2 + In the literature this equation is known as the modified Mathieu equation or associated Mathieu equation. W.28) below). ig h \j g (19.g.

10) i # > = 2h2 2]^v ~r 2 7 2 ~r 2<\^v o o w w dw w Using the recurrence relations of cylindrical functions. i.5) so that Eq. and H„ (w). (19.Nv(w). First we make the additional substitution w = 2/icosh z.e.2 CHAPTER 19.-. where £>„ : = —_ + i _ + J l . w 2 . (19. dwA w dw { wA ) (lg. Muller-Kirsten and N. Aly. 8.g) This equation is wellknown as Bessel's equation or as the equation of cylindrical functions.438 19. (19. H. —Zu — -(Z„_i + Zu+i). Singular Potentials Small h2 solutions in t e r m s of B e s s e l functions We now develop a perturbative procedure" for solving the associated Mathieu equation in the domain around h2 = 0. Eq.9) leaves uncompensated on the right hand side of Eq.2.7) We shall see that this parameter is given by the same expansion as in the case of the periodic equation. . H.1. Proceeding along the lines of the perturbation method of Sec. Vahedi-Faridi [10].11) We follow here largely H. The solutions Zv(w) are written Jv(w). where these are the Bessel function of the first kind. — Zv+i).6) terms amounting to 1 „ 2 d2Zu A V (19.I .4) becomes dw2 w2 w2 \ dw2 w dw \ j Next we define a parameter v by the relation V2 v2 = (/ + .2 A / i 2 .7. (19. (19. J.i l .Hv (w).^% dw1 1 1 —— = dw = ][Zv-2-2Zv 4 w Zv + Zv-\ = -(Zv-\ 2 + Zv+2]. Neumann or Bessel function of the second kind and Hankel functions of the first and second kinds respectively. W. The zeroth-order approximation = Zv[w) (19.6) becomes to zeroth order </>(°) = <j>v of <f> in h2 Dvcf)v = 0. (19.

I / ± 2 ) = 1.17) Thus a term fiGu+a on the right hand side of Eq.10) in terms of functions G„ defined by Gv+a = —nZ v+a . v . = §. therefore 4>^ leaves uncompensated i#> = / i 2 [ ( i / ^ . However.2. where the starred coefficients are defined by (19. of course. (i/.10) is now particularly simple: i?i°) = h2[(v. ~a(2v + a)~ w2 * a> Zv+a = a(2u + a)Gv+a.13) (19. v + 2)Gv+2]. it is more convenient to use these relations in order to rewrite Eq.15) (19.20) 2)*ZV_2 + {y.2 The Potential 1/r4 — Case of Small h? 439 we can rewrite the expression (19.10) as a linear combination of various Zv.2 ) * ( z / . v . .i/-4)*Z 1 / _ 4 +(u.2 ) * i ^ 2 + (!/.13). z . We assume in the following that 2u + a ^ 0 (the case of 0 has to be treated separately). v-2)*{y-2.6)._2 + {v..12) (19.i/) = 2A.u- (19.18) Now </>(°) = Zv left uncompensated Rv .v + 2)*{u + 2. v)Gv + (y.13) therefore lead to the first order contribution jfU = h2[{v. (19. can be cancelled out by adding to </>(°) the new contribution liZu+aja{2v + a) except. Du+aZu+a but Dv+a — Dv so that DuZv+a = a(2l/ (19.16) ( I / .u + 4fZu+A}. and this means in (19. (19. when a or 2v + a = 0.2)*Zi. v + 2)*Z„ +2 (19. wz The expression (19. where We observe that DUZV = Q. v + 2)*(v + 2.14) (19. .2)G„_ 2 + (y.21) +(i/. The next contribution to (^°) + t^1) therefore becomes 0(2) = /i4[(z.u + 2)*Z„ +2 ].19.!/+ 2)*4°J 2 ]. (19. The terms (19.

in Eq.j^0 (19.2)(v + 2j . . i/ + 2j) + p 2 i _ 2 ( 2 j ) ( ^ + 2j.I/±2)*(I/±2. (I/. v + 2)*{v + 2. j=-i. and p 0 ( 2 j ^ 0) = 0. we obtain 0 = h2{v. ..4)(»/2 . .25)/i 8 32(a-l)3(a-4) .440 CHAPTER 19.17)). . v) +(i/.23) These coefficients may also b e obtained from t h e recurrence relation P2i(2j) = p2i-2(2j .I/±4)*.v + 2)*(i/ + 2. we obtain t h e expansion . Evaluating t h e first few terms.24) P2i-2(2j + 2){v + 2j + 2. p 0 (0) = 1.v-2)*{v-2.l)(z/ 2 .4 + 58.26) iy — 2.2.v) + {v.9) 16 + + ^ j " liy '2° This expansion is seen t o b e familiar from t h e theory of periodic Mathieu functions where (Z + l / 2 ) 2 represents the eigenvalue. v — 2)*(v — 2. Ri. Adding these terms and setting the coefficient of Gv equal t o zero. . i/)] + • • • . v ( 4 5 a . etc. (19. i/ + 2j). .25) Finally we have to consider the terms in Gv which were left unaccounted for in Rv . .v + 2y{v + 2.455a + 1291a . i/ ± 2)*. i/ ± 2)* (i/ ± 2. Zv + Y^h2i i=l J2 p2i{2j)Zv+2j.v)] (19.i/±2)*. (19. (19. subject to the boundary conditions P2i(2j) = 0 for | j | > i.22) where p2(±2) p4(±4) = = (i/. P2*o(0) = 0.2 + 2 9 ) ^ 64(^ 2 . Reversing the expansion and setting a = (I + 1/2) 2 (not to be confused with a e.1169)/i 12 .g. 1 1 \ 2 + 2) ~ -) s X 2 A = v 2 hA (5v2 + 7)h8 v .v) + hA[{v. Singular Potentials Proceeding in this manner we obtain the solution ^ = 0(0) + 0 (1) + ^(2) + . v + 2j) (19. we obtain 9 h4 = a 2(a-l) 3 2 (13a . p4(±2) = (i/. + 2(y2-\) — + 32(i/ 2 -l) 3 (^ 2 -4) —^ (9. ^ 64(a_1)5(a_4)4(a_9) + 0 ^ )• (19-28) 1 6 .

we may say that the first approximation <^°) leaves uncompensated terms amounting to R^ = 2ft2 (A .7) into Eq. and is therefore real for both cases g2 > 0 and g2 < 0.cosh 2z)4>.An(u + n). (19. Thus to O(0) in h we have (jr ' = 4>v — cosh vz.i . 19. ±2.2.29) sinh vz or e±uz.34) .31) It follows that Dv+2n4>v+2n = 0. ±{ v+ z e = sinh(v + 2)z + sinh(z/ — 2)z. so that J2 (19. (19. 2 .33) = = cosh(^ + 2)z + cosh(^ — 2)z. — We return to this question later but mention here that this problem has already been dealt with in Chapter 17. Du+2n so that D„(j)v+2n = ^n(u + n)<\>v+2nSince 2 cosh 2z cosh vz 2 cosh 2z sinh vz 2 cosh 2z e ±vz = Dv .2)<t>„-2 + (v.4).3 Small h2 solutions in t e r m s of hyperbolic functions For later purposes we require yet another type of solutions.2 The Potential 1/r4 — Case of Small h? 441 We note that this is an expansion in ascending powers of h4. (19.^ + e±(-v~2)z. we can rewrite the latter as j ~ . The solutions 4> of the modified Mathieu equation are now completely determined. We are still left with the question of what will happen if 2v + a = 0 or v — ± 1 . v . (19.19. (19.v2(j> = 2/i 2 (A . Thus v is real for small values of | h21. £>„ = — . = 0. Substituting (19. v)(f>v + {v.cosh 2z)4>v = = 2h2A(/>„ .30) £>„&.32) (19.v + 2)<t>v+2]. . apart from a normalization factor which we have chosen (so far) such that the coefficient of Zu in <f) is 1.h2[(j)u+2 + 4>v_2] h2 [{u.

h2i *=1 J2 j=-i. W.4 Notation and properties of solutions We now introduce standard notation as in established literature.J7t0 p2i(2j)<f>u+2j.* The solutions of the modified Mathieu equation which we are considering here are written with a first capital letter. (19. Singular Potentials (I/.h). we prefer to adhere to one equation with different solutions. in the present context should not be confused with the same symbols having a different meaning in the case of solutions in terms of cylindrical functions since it is generally clear which type of solutions and hence coefficients is being considered. F. For rigorous convergence and validity discussions of any of these solutions — our's here differ only in the method of derivation with the perturbation method of Sec. 8.h). In fact we could have obtained the same Ru by starting with the modified Mathieu equation for h2 replaced by — h2. (19. > These solutions correspond in the periodic case respectively to the solutions and me^.1 . > (f> with </v = sinh vz — Seu(z. 19.442 where (I/. In order to avoid confusion arising from the use of different equations.h) = <(>„ +Y.38) (f> with (\>v = exp(vz) — Meu(z. Meixner and F. • (19. Arscott [11]. M.35) The form of i ? ^ is seen to be almost identical with that of the corresponding expression for solutions in terms of cylindrical functions. In particular the following notation is used for the solutions obtained above which we characterize here by their first terms: <f> with (j>v — cosh vz —• Ceu(z.2. CHAPTER 19. Schafke [193].h). etc.I/±2)*. The solutions in terms of cylindrical functions are written *J.7 — we refer to Meixner and Schafke [193].J/±2) = . v ± 2) etc.I/) = 2 A . .37) where p M (±2) = (I/. Denning (19 36) (">" + <*) =o^7Ta)> we now have the solution - p(z. The use of the symbols (v.

Mi. this normalized solution possesses contributions exp(i/z) in higher order terms. (19.h). i.43) and hence similarly M^\z + inir. Oberhettinger [181]. see this reference p. h) = av{h)M^ (z. Magnus and F.h) and oo i M^\z. h). h) = exp(inwn)MP(z. Me„(z. {2h cosh z) • MJ?\z. For the relation of Hankel functions used below. (19. (19.2 The Potential 1/r4 — Case of Small h? similarly: <fi 4> <f> (p with with with with <j>v = J v (2/icosh z) .e. h).39) Writing two solutions out explicitly we have for example — apart from an overall normalization constant. 4>v = Hi. 17. h) are therefore proportional to each other as a comparison of Eqs.41) i=l We see immediately that Meu(z + niri.h). Also from the expansion of Jv{z) in rising powers of z. we obtain$ Ju(2hcosh(z + inir)) — Jl/(2hcosh z exp(m7r)) — exp(inuTr)Jv. Meixner and F.h). introduces the dominant order function into higher order contributions''' — oo i 2i Meu(z.h) = Jv{2hcosh z) + ^h 2i J^ j=-iJjLQ p2i(2j)Ju+2j(2hcosh z). 4>v = Hv (2h cosh 2) * M^\z.h = £ p2i(2j)(-l)j exp[(^ + 2j)z] (19.MP(z. i. 16.40) does not. h). (19. h) = exp{vniri)Me„(z. As emphasized in Chapter 8. .19. as Mev{z.44) implies. (19.h) = exp(vz) + Y. ' (z. p. whereas the — as yet — unnormalized solution (19. W.g. which. (19. — oo cv2r{h2)e^+2r>. Schafke [193]. as we remarked earlier.e. <f)v = N„(2hcosh z) MJ?\z. 'See e.42) The solutions Meu(z.40) Cev{z. 443 (19.44) (19.h2) oo = Y.(2hcosh z).42). W.45) T Below we frequently write the normalized solution as in J.h).h)±Sey{z. h).

h) on the other hand can be shown (cf.h) Using further properties like (19. 4) = exp(±^7r)i\4 3 ' 4 \ M^3'4) = M™ ± iMJ?\ (19. h).h) . 130) to converge uniformly for all finite complex values of z. we use the notation of Meixner and Schafke [193] and so write the expansion of the function Mev(z. The functions Me±v{z. 2M™ = eiuvMi3) + e~ivlrM^\ . These points are important in our derivation of the S'-matrix below. H™(z) = e-^HfXz). h) = M™(z.47) With this equation we obtain for nonintegral values of i/:§ ±isin WK MJ)3A\z.e. Meixner and Schafke [193]. Equating the coefficients to zero we obtain [-{y + 2r)2 + \]cv2r = h2(cv2r+2 + c£ r _ 2 ). As mentioned earlier.g.exp(TM"r)MW(. For later essential requirements (i. Mi 3 .e. we have the following for a change from v to — v. ^ .e. Meixner and Schafke [193]. the explicit evaluation of the S'matrix element) we elaborate here a little on the computation of coefficients of normalized Mathieu functions.49) Inserting this expansion into the modified Mathieu equation (19. i.46) H^z) = e^H^(z).h2): oo Mev(z.50) E. .43).444 CHAPTER 19. MJ>>(0. r=—oo (19. p.\c\r + h2(c^2 + 4 r + 2 ) ] e ^ + 2 ^ = 0. (19.8.h) can be shown (cf.48) The series expansions of the associated Mathieu functions M„ (z.h2) = ] T cv2r(h2)e{y+2r)z. oiv{h) = (19. s (19.4) we obtain oo £ r=—oo [(i/ + 2r) 2 4. the first relation is obtained as follows: 2M<£1 = M^l+M^l. h) . Singular Potentials where clearly MeJQ. p. 178) to be convergent for | cosh z\ > 1 but uniformly convergent only when | cosh z\ > 1 for otherwise complex values of z. i.

^ p 2 2 ^ or c 2r+2 2r-2 1 (19.52) in agreement with Meixner and Schafke [193]. 122.(/^. 117.^ c 2r 2r 2T~2 or -2r ^-2 /l-2[A_(I/ + 2r)2]-£2l± 2 . h~2[\ . W.^ C 2r+2 (19. Thus now c 2r .2 The Potential 1/r4 — Case of Small h? which we can rewrite as fc2^±2 c 445 = [A _ („ + 2r) 2 ] . 2r (19>51) 2r For ease of reading we give the steps in rewriting this. Alternatively taking the inverse of Eq.53) /i-2[A-(^ + 2 r ) ] .19. + o2„r 2 l2 ] - C 2r 1 -2r c 2r-2 or 1 c J =^[A-(. \) /.* For 'Actually. .[X-(u 2 + 2r-2) } " c 2r-2 This continued fraction equation can again be used to obtain the explicit expressions of coefficients of normalized modified Mathieu functions. + 2 r ) 2 ] . Meixner and F. p. Schafke [193]. see J.54) 2 C2r 4 ^ h. \ .51) we have -2r (19./ i 2 C ^ . (19.(v + 2r)2} / z .^ + 2r + 2 ) 2 ] .2 [ A . Eq. .T22r[\ A . p. eg = c~v = 1. (39).

+4)!ll_. W. (19..( " + 2)2] ~ h2 h-. J . but follow our earlier reference H. Vahedi-Faridi [10] (this paper contains several misprints which we correct here). Aly. Manvelyan. J. Miiller-Kirsten and N.55). Singular Potentials nonintegral values of v examples obtained in this way are (see Example 19. H. + i).5 D e r i v a t i o n of t h e S .52) and obtain c% 1 eg fc-"[A-fr + 2)»)-h_a[A_(.2.+0^">- < 19 .55 ' Example 19. . Spector [257]. W.12) ^ 384(^ + l)(z/ + 2)(z/ + 3) | " ^rW^hy.1 for the evaluation of a typical case): hb + 0(hw). H. (19.8 ^^> 1 hi(v2+4i>+7) 32(^+2)(i/-l)(^+l)2 4 ( i / + 1) /l 2 4 ( i / + 1) which is the result expected. M. Then di 2 1 t" + 2(^1) + • • • . .^ ^ + 2)(i/ + o> l)(i/ 3) 0(^10). + 0(/.C + l) 3 (i/ .27) and truncate the continued fraction after the second step.l ) 19.1: Evaluation of a coefficient Evaluate explicitly the first few terms of the coefficient Cj/cfJ given in Eq. H.._{u+m h2 M ( " + D + 55^iy] + i ^ 5 j /i 2 4(. J.Q . i/2 + 41/ + 7 7i6 + 0 ( / i 1 0 ) . Miiller-Kirsten. and also R. Liang and Yunbo Zhang [189]. t For this purpose *We do this in the manner of the original derivation of R.m a t r i x Our next step is the derivation of the explicit form of the ^-matrix for scattering off the potential 1/r 4 since this is possibly the only singular potential permitting such a derivation in terms of known functions. Solution: We put r = 1 in Eq. Here we insert for A the expansion (19. 8 4 ( i / + 1) 128(i/ + l)2(v + 2)(v-l) + 5 10 32(i/ + ^ l)(i/ + 2) + 768(^ „„ .[v^. 128(i/+ l) 2 (i/ + 2 ) ( ! / .446 CHAPTER 19.

Then y reg = = r rV2M®(z. W . 2 ) 2 1/2 ex Z/7T IT H=i —J TTW P l+O 10 (19.3a) and (19.. and considering the propagation of this wave-front. Oberhettinger [181].5) w — 2h cosh z = [ kr -\ r Thus r —> 0 implies \w\ —> oo.19.59) which tends to zero with r. \w\ S> 1 and — IT < arg w < IT is known to be given by* (19. and then the continuation of this to infinity.h) 1/2 w) (19. we have W r = ip + ut — 7r/4' . The asymptotic behaviour of the Hankel functions H„ ' (w) for \w\ ^> \u\.56) where by Eqs. this wave function near the origin is the wave transmitted into this region (as distinct from the reflected and ingoing waves). We obtain the regular solution by choosing Zv(w) = H„ (w) for 5ft(z) < 0. In the case of the repulsive singular potential here.58) The behaviour of y reg near r sa 0 is therefore given by 2 1/2 2/reg "\ 5 exp 1\TT r\ —r- exp i . (19. .57) ^ 1 . I / + 2 2 l\TT +E ^ i-\ E j=-i.2 The Potential 1/r4 — Case of Small h2 447 we require first the regular solution yreg of the radial Schrodinger equation at the origin. 22. (19. The time-dependent wave function with this asymptotic behaviour is proportional to (here with ui — k) —iuit—g/r+iTr/4 Fixing the wave-front by setting <p = —lot + ig/r + TT/4 = const.j7L0 P2i(2j)exp -i[v + 2j + 2 2 . Magnus and F. p.

2h and the square of this expression implies 2 2 ~ ' A. these solutions are seen to have the desired asymptotic behaviour for r —• oo: » 1/2 y± e±ikr e x p ml 1 nk ) 1 + ^2h2i i=\ Yl -—i.h) can be shown (cf. '(z. In a similar manner we can define solutions y± by setting for 9fJ(z) > 0: y± = rl'2Ml?>*>(z.j¥=o Ki(2j)exp(Tmj) (19. but uniformly convergent only when | cosh z\ > 1 for otherwise complex values of z. Meixner and Schafke [193].j^Q p2.448 CHAPTER 19. (19.A1-4A. Magnus and F. AV . we can derive y_ from y+ since one can show from the circuit relations of Hankel functions^ that M^\z + m.3a) and (19. ^ 1 ) (e i 7 r z) = -Hi2!>(2) e-^H^iz). p./i). (19.h) = -exp(-Mr^)il4 4 ) (z.3b) ln • 7T Q"*4' r fg ° Vk- (19.61) In fact. 17. h) through the entire range of !&{z). '-2-"4 .h) oo j 2 — rV flp^ + ^fc* £ i=l j=-i. Now from Eqs.V + § 9 -2 > 4hh* . The series M^ (z.60) Using the above asymptotic expressions for the Hankel functions. . Singular Potentials so that when t — oo : r — 0.62) Also with Eq. We now require the continuation of the regular solution y reg to solutions behaving like y± at r — oo.57) the condition | cosh z\ > 1 implies I cosh z\ kr + ig/r >1. Prom the relation r = (ig/h)ez we see that r — 0 > corresponds to $l(z) — —oo and r — oo to $l(z) —> oo. We require therefore > > the continuation of Mi./l ~ 2 ~ 2z + " 4 > 0.A1"—2 z '+"2 A~~ n or rA .(2j)<gH • (19. p.#. 4%r g >0 4r^r ' rg See W. Oberhettinger [181]. This means that the origin of coordinates acts • > as a sink. 178) to be convergent for | cosh z\ > 1.

z = In h i— r0 4 and 7T z = In • ro { 4 for ro < r < oo. 449 This condition.49).e. Originally we chose the sign of m / 4 as in Eqs. p.38) or (19. i.r2_) > 0 with r£ = rg(2 ± >/3).19. 19. the domain r_ < r < r+ — as illustrated in Fig. which has to be bridged by using another set of solutions.e.3b) and (19.3a). Since the real part of z changes sign at r = ro. Meixner and Schafke [193]. (r — r+)(r + r+)(r — r_)(r + r_) > 0.1. i Im z z = In r/r0+i7t/4 . These solutions converge uniformly for all finite complex values of z (cf. . is satisfied only for r < r_ < r_|_ or r_ < r + < r. The one further point to observe is that the variable w = 2/icosh z = (kr + ig/r) is even under the interchange z — —z. r 7r for 0 < r < ro.-ft/4 _Z z = In r/r-m/4 0 r 0 =(9/k) 1/2 Fig.rl)(r2 . Thus we choose . and in order to maintain this symmetry.1 The domains of solutions.63) Thus there is a gap between the two regions of validity — i.2 The Potential 1/r4 — Case of Small h2 Hence (r2 . (19. 19. This interchange in the > solution My interchanges y reg with y+.ijt/4 Rez . we have to assign different signs to the imaginary part of z in the two distant regions. 130). (19. (19.62). A suitable set is the pair of fundamental solutions Me±v defined by (19.

h) = Me^u(—z. as just explained. i. these relations can be re-expressed for one and the same point z = —iir/A: rl/2[aMe-„(z) rl/2[a'Meu(z) + + /3Mev(z)]z=_in/A p'Me^(z)]z=_i7r/4. we express in the domain r_ < r < r+ the regular solution as a linear combination of solutions Me±v with coefficients a and /3. Singular Potentials Then starting from the region r ~ 0.l/2Me_i/(T. and we determine these coefficients from this equality and the additional equality obtained from the derivatives.67) Next we have to continue the solution beyond the point r+ to a linear combination of solutions y+ and y_. + ^[r'^e^z)] dr z=—iw/4 + (31 ^[rll2Me. At r = ro the real part of z vanishes and then switches from negative to positive.(2)]z=w/4 + 0Me.e. Hence we match the right hand sides of Eqs. a4-[r1/2Me-v(z)] dr a'^-lr^MeAz)} It follows that (19. (19.v{z)\ . h) (as one can check or look up in Meixner and Schafke [193]. (3 = a . (19. i.)]-(19-64) The right hand side of the first equation now represents the regular solution in the domain r_ < r < r+.[ r l/2 M. Thus. (3) (r)] aA[rl/2Mei/(r)]+/3i:[r. 131). Thus we set rl/2M^\r) = = rll2[aMev(r) + (5Me-v{r)}. p.^)] dr z=in/4 dr (19.B^0. to r^2[AM^ + BM^\ A.66) a = (3 .z=—in/4.65) -in/A a>d[rl/2Mei/{z)]+f3>d[rl/2Me_Az)] Since Mev(z.v(z)\ —m/4) + ^[r^Me. at this junction we require also the imaginary part of z to change sign. .450 CHAPTER 19.64) there to another combination of solutions Me±v by setting: rl/2[aMey{z) = rl0/2[a'Meu(z) a^{rl'2Mev{z)} dr dr + ^Me_.

70) by Me'_u (the prime meaning derivative). (19.3a) and (19. (19.68) \j3Me„(z) + aMe-v(z)] P±[Me„(z)}+afz[Me-„(z)] = = [AM^\z) + BM^(z)}. (19. we use the notation W\f. (19.68) the derivatives with respect to r by this relation. (19.) d_ _ 1 d dr r dz Hence for any of the functions Mv: | [ . (19.19.70) = a±[Meu(z)}+P-^{Me_„(z)} A^[M^{z)] +B^[M^(z)}. (19.3b) we infer that (z = l n r + const. (3. with the replacements of Eq.69) Thus if we replace in the derivative relations of Eqs.71) We now determine the coefficients a.65) and (19.67): rl/2[(5Meu{r) + aMe_u(r)} pd[ri.2 The Potential 1/r4 — Case of Small h2 451 This solution can be continued into the domain below r = r+ by matching to the right hand side of Eq. multiplying the first of Eqs. (19.70) by Me-U and subtracting .g]:=f(z)*&--g(z)WZ) dz dz Thus. For the Wronskian W of two independent solutions f(z) and g(z) which is constant and can be evaluated at any point. (19.2Mei/{r)]+d[rl/2M] dr dr = = rxl2[AM^\r) + Ad[ri/2M(z)(r)] dr +B^[rl'2M^{r)]. dr BM^(r)]. for instance.65). and the second of Eqs. (19.64) the relations M&Hz) j-z[M^\z)] and from Eqs. A and B. Thus we obtain from Eqs. (19.68) From Eqs. (19.69)) cancel out in view of the nonderivative relations. ^ ] = ^ + ^ ^ . the nonderivative parts (from the first term on the right of Eq. Then. (19. and we are left with relations expressible only in terms of z. = aMev(z) + 0Me-v(z).

B = J 2 L _ e . or obtainable by substituting e. (19.^ ^ a „ Q _ .Me-vy W[M?\Mev] W[Me-v. using Eqs.M^] --J™a-V.\ \ ( av -2iwKa-v ^ v . and the second is obtained similarly: W[M?\Me-v\ W[Mev..e.M^]= IT -e~iv*a-v. i. 7T W[Me-v.4] = .46).. 7T [1. (19.3] = -[1.* « * r r z H .48) (or cf. 170/171).452 CHAPTER 19.g. we obtain the first of the following two equations. With these expressions the quantities A and B are found to be A= f au ^ 2isin V-K \ a_u 1 a.77) av I 2ism V-K \ ct-v av ^Thus.73) We now use Eq. (19.75) With these relations and Eq.58) and considering large values of \z\. 7T (19. (19.u}W[Me. Me. (19. h)/M^\o. we use the following circuit relation which can be derived like Eq.j] given in Meixner and Schafke [193] (p. we obtain immediately W[Z. and Wronskians W[M^\MIJ)] = [i. W[M^\Mlf)]W[Meu.] W[Mev. (19.^ 47 9? MJ>4)] Me-V] [3.Me^} (19.71) we obtain in a similar way A B = = -W\MP\ Me.76) .M^} W[Me„. Meixner and Schafke [193].(0.Mev]' P l ' From Eqs. (19.A] = W[Mi3\z).4] = .. the leading terms of the respective cylindrical functions. .M^] W[Mev.74) Moreover. p.u. = IT --<*„.e.46)) W[Mev. TV W[Me-v.MiA\z)} = 2(-i)(2/Trw)(dw/dz) = -(4i/Tr)(2hsinhz/2hcoshz) ~ -4i/>.47) we obtain (av defined by Eq. . 169) M0) = e«^Ma) _ isin U1X M(A) _ (19. W[MJ?\ M^] W[Mev. Singular Potentials the second resulting equation from the first resulting equation.57) and (19. h). i.Me-v] = = = -av. M^4)] + W[MJ>3). (19. av{h) = Me.

f c . in Chapter 16).h) = Me_u(-z.82) The S'-matrix is defined in the following way in the partial wave expansion of the scattering amplitude /(#). (19. Z ) e ^ ] .h)] ( m / 9 \ 1/2 {-L\ {Aeikre-i(V+l/2)*/2+Be-ikrei{v+l/2)*/2^ g) We set fl = ^ = M|M). e-ii/7r/2 (19. Z ) e .h) + BM^\z.(-l)le-^}Pi(cos 9). (19.85a) .(ite™)" 1 { 8 J ' f(-k. The superposition of an incoming plane wave and an outgoing radial wave is written and re-expressed in terms of partial waves for r — oo » with z = rcos 6 as: 1 tCT" "1 OO e *« + /(*)!_ ~ 5 > ' c < f c P . 2zitsm z/7T zzitsin I/TT Inserting these expressions into Eq. (19.79) where the second expression follows from Eq. f c r . (19. Then f-' . B=tfD-'"".19. (19./) = (l)1/2R2-^-2i^)ei^)U^.78) we obtain Ay reg « _ ._ [ / ( f c .46) together with the relation Mev(z.83) The S-matrix element is therefore given by ifc*"* ./ ( .l) ' We note already here that with the substitution R = ei7r7.h).80) -2^fc (19-81) where \7r/ i t sin VK /(-*. (19.2 The Potential 1/r4 — Case of Small h? The regular solution in this way continued to r ~ oo is then y reg ~ ~ 453 Tl'2[AMiz\z.g. as we recall from earlier considerations (e.

z—>oo 2r \l/2 -i(is+l/2)ir/2 2hir cosh z r> \ „2ih cosh z R--U + i [Re —2i/icosh z R Thus.58)) 1/2 2r e~ 2hir cosh z.56) to (19. 9?.2 The repulsive potential and the various waves. (19.S-matrix in the form s i n 7T7 Si = CHAPTER 19. 19. V(r)=g2/r Fig. the reflected wave and the transmitted wave the quantities: A Ar At = = Re1' R 2zsin 7r(7 + u). 19. and with the left hand side following from Eqs. h) (see Eqs. Thus in terms of the (3) variable z. Fig. we can write this (multiplied > by 2ismvTT. R — — = 2i sin ivy. Singular Potentials r(i+k) sin 7r(7 + u) (19. (19. (19.81) in another form from which we can deduce the amplitudes of reflected and transmitted waves.56).86) .454 we can rewrite the .85b) We shall obtain the 5-matrix in this form in the case of large values of h2 later. we can define respectively as amplitudes of the incident wave. R 2i sin TXV. and recalling that yreg is proportional to a function Mc> {z.58)) in the limit r — 0. cf. (19.2. We can rewrite Eq. (19.

e. A derivation is beyond the scope of our objectives here. if R becomes a pure phase factor and eil/7r a real exponential.e. This implies basically the evaluation of the expression R of Eq. " S e e S. (19. unitarity is preserved. 178).h) is (cf.e-™*/R\2 ' ' . therefore we cite it from Meixner and Schafke [193].2COS2VTT = (R.89) .2.39)) the associated Mathieu function expanded in terms of Bessel functions of the first kind.( i J e ^ ) " 1 ! 2 = R2 + -^ . Eq. / i ) = Y.In Meixner and Schafke [193] (p. S. i. unity minus reflection probability = transmission probability. (p. (19. Hashimoto [124] and R. this expansion is given as 1 oo MJ> Hz. Jv+2r. (19.84) of the ^-matrix.e. (19.88) where the factor Me„(0.-Q.m a t r i x Our next task is to evaluate the expression (19. Meixner and Schafke [193] therefore developed an expansion in terms of products of Bessel functions which is more useful in practice owing to its rapid convergence. and then exploit it for the evaluation of the quantity R. in which v is complex. Eqs. however not here. By inserting in Eq. The latter is what happens in the S'-wave case of the attractive potential. l=—oo 1 1 lite*"* .** 19.h) The function Mo (0.e. The expansion is: 4 r ( / i 2 ) M S ( z .„„„X (1 Q Q'T'l l \Reiun -e~iuw/R\2 ~ \Reiv* . W. (105) to (108).28)) and R = ey real. Liang and Yunbo Zhang [189]. J.±)2 + (2sinj/7r) 2 .. i. i. (-^)l4lu{h2)Ji-r(he-z)J±u+i+r(hez)." \R~Ji\2 1 | tO _ | 2 i s i n VK\2 [ | .h) = 1 Me V\ 5 / r J2 __ 0 O c»2r(h2)J„+2r(2hcoShz).79). but in the 10-dimensional string theory context. H. 180).2 The Potential 1/r4 — Case of Small h2 455 One can verify that for v real (which it is here in both cases of attractive and repulsive potentials in view of Eq.6 E v a l u a t i o n of t h e S . (19. J. Gubser and A.19. i. Muller-Kirsten. M^l)(0. y We observe that this relation remains valid if the real quantity R = e and the pure phase factor eW7r exchange their roles. (19. h) serves the use of the same coefficients as in the other expansions.88) the power expansion of the Bessel function one realizes soon that the expansion is inconvenient in view of its slow convergence. Manvelyan.

h) = Jo(h)J±Ah)-^^Ji(h)J±„+i(h) c0 {n ) ^ ' C^r^J-2{h)J±l/-2{h) cr(h2) + •••. Next we evaluate M^1}(0. We require the power expansion of the Bessel function of the first kind which we obtain from Tables of Functions as I)EH^(-T)Thus.6 2 ) m 6 (^_i)3(z. h2 is complex but v2 is real (cf. (19. (19.g. Eq.h) v 2 = ^2 (-l)lc%ih2)Ji~r(h)JM+r{h).90) and choose r = 0 (one can choose e.g. As a matter of introduction we recall the definition of the coefficients with Eq. (19. (19.. /n ^ ./i) with the help of Eq. Singular Potentials 4r (h )M$(0. setting r = 0 in Eq. (19.55).49).28)). /=—oo (19./>) = Co> 2 ) £ 7 " = — OO j ± > . u»-^.3 9 ^ . we obtain M&(0.2_4)(l/2_9)iv2y) + <AA.90) Here the coefficients c2"{h2) are the same as those we introduced earlier for the normalized modified Mathieu functions in Eq.456 so that in particular for z = 0 oo CHAPTER 19.49) from which we obtain for z = 0: Me„(0. one obtains *.e.90) is in some sense amazing: It permits the evaluation of one and the same quantity M±J (0. (19. h) in many different ways. The formula (19. e. i. Here in the case of the repulsive potential.90). ^ (19. r = 2 as a check).94) .91) Inserting here the coefficients given in (19. r = 0 and 2. by allocating different values to r. + 2 u -l\2j 2 fh\2 "2+2 2 MX4 {v -l)(v -4)\2 2 2(^ + 4 ^ .

.96) we extract for later reference the relation sin7rz/ R IT h h 2\ 2 Av [v\(v-iy...97) + This expansion will be used below in the low order approximation of the absorptivity of partial waves. From expansion (19.7) F " (i/±l)2(^^l)(i/2-4)V2.12i/2 + 64u .2 _ ^ 4 ( ^ 2 _ 4)2 (19.1) 2u(4u + 15i/ .9) V2 1 fh ±v 1+ 2 + • (19.93) and the coefficient expansions (19.32i/ .23) (hxQ 2 3 2 2 V ± l) (i/ T 1) (^ .2 The Potential 1/r4 — Case of Small h2 457 Inserting the power expansion (19. .!' or with the help of the inversion formula of factorials R = v\{y — 1)! sm-Kv f h 7T -2v (^ . 4(z/ T H ^ 3 .1 ) 2 ( I / + 1 ) ( V 2 . one obtains M£!(O. + # "+" ( I / .4 ) V 2 > 1 (-I/)!V2/ [1+ 2_Mx2 2(i/2-3«/-7) (h)4 (iy+l)2(i/-l)(^2-4)' ./I) h u .96) With this explicit expansion we can evaluate the S'-matrix.l) 5 (19. We observe that this approximation involves v and h4 and hence is real. _2_(h\2 . In this procedure our attention is focussed particularly on the quantity called absorptivity in the case of the attractive potential.IV 2 2 2(v =F 3i/ .4)(i/ .111) / / i 5 4 3 1+ 4i/ 2 + +' (z. 2 (h\2 ^ l/2_1l2^ 2(i/ 2 +3f~7) /feu .] 2 \v i + {v2 .19.95) These expansions imply for the quantity R: I/! -2v ri1 i .55) and collecting terms in ascending powers of h? (here for the case of nonintegral values of v).2^ 2 ± 59^ .

28) that v is real.^ i 4 2+^ # 1 / •.28)). + ( i + s w (E 4 .9 / .458 19.103) Then SiSt (1+5) i . and sin 2TT/3 = ft/.99) which can be rewritten as SiST „2ira 1 1 1 R2)V 1 1 R*2 16 1 R2 Ana „2na + i 2 leZ7rQcos27r/3 R* +1 )e 2 ™sin27r/? }]"' (19.102) (19. 9 / « i(2yr/3)2 ss i(sin 2vr/3)2 ~ 2 sin2 TT/3. where n is an integer and a and (3 are real and of 0(h4) We have with Eq. (19. (19. (19.e-iun/R2)(eiu*n ~ (19. Thus we set v = n + i(ot + i(3) = (n — 0) + ia.104) We now consider the case of a — 0 implying that g of Eq.101) is > zero. Singular Potentials Calculation of t h e a b s o r p t i v i t y We consider the absorptivity in a general case. (e iun e~iv*^/R*2)' (19.R*2 ^ * #2 \ -1-|-1 tf2 R* (19. (19. In this case R — R* and so 1/i?2 ~ Ofji4). Then cos 2TT/? = 1 .7 CHAPTER 19.101) where / is complex and g is real. Eq. (19. and hence allow for the general case of complex parameters v (the Floquet exponent in the case of the periodic Mathieu equation) although here in the case of small \h2\ we know from Eq.2.98) (cf. This is the case of real .84): Si ST 1 (1-1/R2)(1-1/R*2) . _ i?* 2 2+ g # 2 i?* 2 2 -3/(i + s)(4 + i?*2).100) Here we set 2^/3 = 1 + j / ? e 2™ = l + gt (19.

107c) Here h2 is actually Vh4.* ) - 1-4(^V = 1 T h e absorptivity A is therefore given by A = 1 .107b) = Ah2[l + 0(h4)}.61n/i} + ~9~ 0(h8) (19. and t h e values of ip(\/2) and ip{2>/2) as given in Tables^ with Euler's constant C ~ 0. J. 2 /l4 '•i Al=0 2[(l + 1 / 2 ) 2 .97) this can be written A 4TT2 [v\{v . (19. In all comparisons with these papers — which consider predominantly the 10dimensional string theory context — one has to keep in mind that many formulas there do not apply in the three-dimensional case we are considering here. W. J. one can calculate the next term of the expansion so t h a t At=Q = 4 ^ 4h4 {7 .28).106) and evidently violates the unitarity of S. Using the derivative of t h e gamma function expressed as t h e psi function ip(z) = T'(z)/T(z). Oberhettinger [181].2 The Potential 1/r4 — Case of Small h2 parameters v = n — (3. Here in t h e case of the attractive potential (as remarked after Eq.1)\} 2 h \2 iv 1 + Au {y . Hashimoto [124]. . We can also evaluate now the amplitudes (19. (19. p. Muller-Kirsten. With t h e help of Eq.6C . (19. = 1 . (19. i.-Q.SiSf » 4 sin7rz/ R (19. S. Magnus and F.105) sin7r/3 (19. Liang and Yunbo Zhang [189] (there Eq. Hence with Eq.^ ) ~ 2 2 459 1+ 9 / • & ( ! .61n2 . ( I l l ) ) in agreement with a result in S.3b)) h2 is real for E = k2 > 0.577. Gubser and A.86) and obtain in leading order for I — 0: Ai = *[l + 0(h% Ar = ±[l + 0(h2)].l ) 2 2 \) +0{h«) (19. f W .19.3.1 ] Oih* For S waves this implies an absorptivity given by ((—1/2)! = 1/7F) (19.107a) In this result* we now have to insert the expansion of the parameter v given by Eq.e. Manvelyan. 1 29/ ^ (1 . At = 2i[l + 0(h2)}. H.102) we obtain Sis. "This is the result effectively contained in R.

completely different solutions. N. Park. J.1. S.460 CHAPTER 19.H. 2 2 a = [l + 2 (19. W.h) (19.a }tp = 0. h) for large h and upwards I = 0. 2. Miiller-Kirsten and Jian-zu Zhang [226]. we again make use of the replacement (17. H. H. where the part of interest here is based on H.108) For the large values of h2 that we wish to consider here. 19. i.4) which — for some distinction from the small-h2 treatment — we rewrite here as 2 + [2h cosh 2z . Singular Potentials 19.$ q(l. K. Vahedi-Faridi [8] and G.3. Tamaryan. H.109) ^We follow here mainly D.3 19. (19. Wannier [278]. Miiller and N.e.h) 10. This is therefore a highly interesting case which found its complete solution only recently.3 The function q(l. J.1 The Potential 1/r 4 — Case of Large h2 P r e l i m i n a r y remarks In the following we have in mind particularly the case of the attractive potential (h2 real) and rederive the 5-matrix obtained above by using the asymptotic expansions for large values of h2. W. Aly.4 Fig. We are again concerned with the associated Mathieu equation given by Eq. . In addition this solvability is another aspect which singles the l/r 4 -potential out as very exceptional and like all explicitly solvable cases it therefore deserves particular attention.3.26) and hence set a1 = -2hl + 2hq + A{q.

(17. i. Therefore the second term on the right is nonzero. With the boundary conditions y+(0) = l. The replacement (19. (17.110) a where y+(7r) is the solution evaluated at z = TV.e.109) in this way enables us to obtain asymptotic expansions of solutions very analogous to those of the periodic case.2 T h e F l o q u e t e x p o n e n t for large h2 In our treatment of the periodic Mathieu equation in Chapter 17.yJ\ — 1 and cos TTV . 19.21b). and a is its normalization constant. which is even about z = 0. i. (19. .19. since it is clear that large h2 considerations require a knowledge of the large-/i2 behaviour of the Floquet exponent v. The behaviour of q as a function of h for some values of / is shown in Fig. we defined the Floquet exponent v and observed with Eq.§ a W[y+.111) we have W[y+. to which we have to return in this subsection. y± oc A{z)e2h sm * ± A(z)e-2n sm z .l + 2y+(l. which therefore cancels out or can be taken to be 1. y'_(0) = 1. We begin.5) that this is given by the relation cos iris = ^ M . with an essential mathematical step. (19. 19.h*y_(l. (19. We also found the boundary condition (17. however.y-} where VF[y+. and A / 8 is the remainder of the laige-h2 expansion as determined perturbatively. y-(0) = 0.112) In Chapter 17 the unnormalized even and odd large-/i2 solutions of the Mathieu equation were given by Eq.e. (17. J4(0)=0.113) 'We are considering nonintegral values of v.3.h*y (19.63).3 The Potential 1/r4 — Case of Large h2 461 where q is a parameter to be determined as the solution of this equation.21b).3.y^] = ab\f\ is the Wronskian (in leading order) of the solutions which are even and odd about z = 0 respectively. See also Eq.

39 2 14 /i 2 + (19. in the second expression 2h is obtained in addition from differentiation of the exponential factor).e.116) (in the first expression we have A(z) ~ Aq(z) with Aq(z) given by Eq. (19. J.118a) "We emphasize again: For integral values of v the normalization constants are different. Kin B y± oc y+(z) = N0 A(z)e2h A(z)e2h sin z + A{z)e~2h . i. (19. W. This paper contains the normalization constants of large-/i 2 periodic Mathieu functions. See R. B. from y+(z) y-(z) Nn BMz)] a B W oh e"" s m z + C[w{z)]~hl -_• -e a C[w{z)} 2h . J. Miiller [73]. (17. . K [ (z)}2hsinz (19.114) at z = 7r/2. and hence with Eqs.43) we see that z = 7r/2 implies W(TT/2) = 0.462 CHAPTER 19. We now obtain the expressions needed in Eq.32). Dingle and H. Singular Potentials and their extension to around z — TT/2 by the relations C[w{z)\ ^_2h sir Mz)}„2h. (17.112) from evaluation of Eqs.117) a a Referring back to Eq. which is \/2 at z = 0. Dingle and H.A(z)e-2h sin 2 sin z sin z (19. Hence we require (obtained from Eq.115) Setting z = 0 we obtain in leading order y+(0) = 2iV0A(0) and y'_(Q) = 4hN0A{0). (19.114) a a Our first step is therefore the determination of the normalization constants No.111) we have^ iVo = 2 3 /2 1 + °{k and K = ¥frh l +O (19. (19. B.46) together with the expansion of the Hermite function from Tables of Special Functions" ) B[w = 0] 7T 4 [lfo-l)]![l(g+l)]! 2 /i 5 82q2 . Miiller [73]. (17. W. 1 1 Or see R.

(19. Thus in particular (8h) i(?-i) a (3q2-2q + S) 27h + l\(q-W " ' Inserting these various expressions now into Eq.[_ l ( g 2h and hence 5 + 3)]! '-All Ah 1+2 (3g2 .i ( g +3)]! 1- 24/i {-2hl>2).19.118b) The factors a.2g + 3) 27/i ( 3 g 2 .117) are known from the matching of solutions in Eqs.112) we obtain COS TTU + 1 7T '2Wa[\(q-l)]l[-\(q+l)]\ 02h 1 -¥h + 2TT ' 2^ha [\(q .119b) COS TTV + l ire This result gives the leading contribution of the lavge-h2 expansion determining the Floquet exponent v.! ) ] ! [ .8 g + 3) +0 ^2 (8h)^[~l(q + l)]\[-l(q + 3)}\ 2eh ' ~ (19. Presumably they extracted it from "between the lines" of a sophisticated paper of Langer's [159] which they cite together with others. . With the help of the duplication and inversion formulas of factorials the result can be reexpressed as cos( g 7r/2)[£(g-l)]! .3 The Potential 1/r4 — Case of Large h2 and 463 as dz )z=n/2 dw (dw\ o\dzJz=n/2 1C%2 . Tamaryan. 210.1Q11QM cos irv + 1 ~ •==— e .\ { q + 3)]! 7re 4/i 24/i + (-2/i 1 /2) ( 8 / l ) 9 / 2 [ _ l ( g + !)]. (17. (19.3).62a) and (17. p. a in Eq.119a) This formula — derived and rediscovered in Park. (19. Miiller-Kirsten and Zhang [226] — is cited without proof by Meixner and Schafke [193].62b) (see also the explicit calculation in Example 17.87 2 14 /i 2 [ i ( g .! ) ] ! [ .h . (19.

. —h.3. V— 2in (19. „ 1 A „ —A cosh z——H . h. We define these solutions in terms of the function Ke(q.z) = A(q.h.? s m h z) Vcosh z (19.121) We let Aq(z) be the solution of this equation when the right hand side is replaced by zero (i. of course.464 CHAPTER 19.120) The resulting equation for the function A(q. Then straightforward integration yields > inq 4 „.122) Correspondingly the various solutions ip(q. z). h.2 we insert the expression (19. the expression (19.h.109) remaining unchanged.z) or as tp(q. —h.h. h. h. —z). i = 1.-z).e.-h.123) We again make the important observation by looking at Eqs. h(q..h) =-?==. With the solutions as they stand.s i n n z ± iq)A = ± K qj dz 2 Aih d2A dz2 (19. for h — oo). z) = ip(-q.3 Following the procedure of Sec. z) = ip(—q. each with a specific asymptotic behaviour.120) to (19. h. Vcosh z \ 1 .h. (19.3.h. 1 / l + isinhz\T9/45R^oo e^ / Ag(z) = -===[r^— ~ . .122) that given one solution ip(q. we can obtain the linearly independent one either as ip(—q.124) Since this function differs from a solution ip by a factor k(q.z) = Aq(z)exp[±2hiswh z] ^~°° exp ±z/te ( ')eT^/4> V cosh z e x = Aq(z) exp[±2hi sinh z] ** ~~°° ^ h e ^ V cosh z e ^q/\ (19. dA 1.2.z)exp[±2hismh z\.z) := = exp[i7rg/4] j===^Aq{z) exp[2/wsinh z i(77r/4 V-2ih k(q.z).h)iP(q.h). Singular Potentials C o n s t r u c t i o n of large-fa 2 solutions 19. In the following we require solutions H e ^ ( z ) . 17. h. .109) into the equation (19. z) are TP(q.3. it is still a solution but not with the symmetry property ip(q.4.z) i/>(q. —z). (19.108) and set ip{q. tp(q. h. z) can then be written as .

3 The Potential 1/r4 — Case of Large h2 465 Instead. i.125) is t h u s seen t o b e i d e n t i c a l w i t h t h e p r o p o r t i o n ality factor in E q .45). (44) of W a n n i e r . : 7r/2. In order to be able to obtain the S-matrix. we h a v e 77 = 0—> 6 = 0. we h a v e ( u s i n g cosh 2rj = 2 c o s h 2 77 — 1) ry Jo ' = dr. . This is.e. (z. It follows t h a t S e t t i n g s i n h 77 = t a n 0 . the Bessel functions contained in the expansion of the associated Mathieu function M{. IT a n d 7 = 00 — * 7 > : 2fcsinh y + i<j7T h Aw 26/? T h e factor e " r ( ? + 1 ) / 2 o n t h e r i g h t of E q .127) sin z T /'1 M TT\ — <>{ 4 ./h).^ s i n h 7?)]g. i. o u r h2 is W a n n i e r ' s — k2 a n d o u r a 2 is his a w h i c h in t e r m s of o u r p a r a m e t e r q b e c o m e s a = 2k2 + 2iqk + A / 8 . (0. (19.e. ->• * F $ 0 * e".fi2) can be reexpressed in terms of Hankel functions./i 2 ) M (19.126) For large values of the argument 2/icosh 2.WJ " C o n v e r t i n g t o n o t a t i o n of W a n n i e r [278].fc 2lqk A + 1/2 + 4fc2 c o s h 2 77 A / 8 P .<5±1) (19 125) - (the expression on the far right in leading order for h2 large).h2). / 2i<jfc + A / 8 \ ~ 2fc / drj cosh 77 1 + * _' Jo V 8fc2 c o s h 2 77 7 2fcsinh y + [ t a n . i. we have to match a solution valid at Kz = — oo to a combination of solutions valid at $tz = oo.19. 2j<. Ju(z) TTZ VK 7T cos z T~ 4 1 + 0| ^V. * ** = > rojKe(- -".h2) = OvMPfrh2). as we saw earlier. achieved with the help of the Floquet solutions Me±v(z. One can show that the quantity < o of Wannier* (see above) is related to q by $o = & iqir/2 + OO. . or — equivalently — by the appropriate expansion of the Bessel functions as given in Tables of Functions. av{h2) = ^(0J ^ Mi i .e. Me„(z. [ \ . w h e r e Iv := I drj\f a + 2k2 cosh 2*7" J0 I n s e r t i n g t h e e x p r e s s i o n for a. We observed that these satisfy the relation (19. (19. in performing this cycle of replacements the function picks up a factor. W a n n i e r ' s p a r a m e t e r $ Q is defined b y 1 -*0 = 2 K = °o fv I l i m [/„ — 2fc s i n h y].

129) The solutions so defined have the following asymptotic behaviour (where e(z) = (2/icosh z)~ll2 and h2 = ikg): H. z): Re(2\z.3./i) Be^(z.vr .q.-q.q.g.q. He (-z.h) He (z.h) = e(z)exp exp ihe^ 4 + i\ 12 i4 3^-cO. (19. In this way we obtain in the domain .h) (3) = = = = Ke(q.g.q.q.h) = e(z) exp ihez .h) Be^(z.-q. 3te»0.130b) r_0 rll2exp\-g/r (ig) ' 19.4 T h e c o n n e c t i o n formulas We now return to Eq.h. (19.-h).466 CHAPTER 19.128) We now define the following set of solutions of the associated Mathieu equation in terms of the function Ke(q.q. (19. Singular Potentials Note that the sine part is nonleading! Retaining only the dominant term of this expansion we have (with z — 2h cosh z) > Me±l/(z. (2) (19. (19.h) = e{z) exp ihez + iexp[ikr + i-7r/4] kr 7T .-h).130a) Re^(z.130b).z). h. 3te»0. HeW(z.130a). exp[—ikr — in/4\ kr Re{2)(z. (19./i).128) and consider the cosine there as composed of two exponentials whose asymptotic behaviour we identify with that of solutions of Eqs.h )~ \ -ot±v cos(2/icosh z = vir/2 — 7r/4) F V2/tcosh z (19.eW(z. KeW(-z.

(19./i) — sin 7T7 Re^(z.h2) = 2vr i a_ 2TT />) _ e ^ / 2 H e ( 4 ) ( Z j ^ Q (19.i ^ / 2 H e ( l ) (z> ?) (19.125) the proportionality factor can be seen to imply the relation exp •ffa + 1) sin7r(7 + "} (19. h) = Ke(-q. e x ^ r / 2 H e ( l ) ( 2 > g> />) _ e .134) Considering now the first of relations (19.132) These relations are now valid over the entire range of z.(-z. h) = sin 7r(7 + v) He (1) (z. (19.134) in the region where the function He^(z. h) — sin 7T7 He^(z. we obtain in the domain tfcz <C 0 the relations: Me. we see that the solutions (compare with Eqs.q. (19. / i ) + sin7r(7-^)He(4)(2.li 2 ) Oil.h) is asymptotically small (cf. (19. . Changing the sign of z. q./i).19.h).h2) a.3 The Potential 1/r 4 — Case of Large h2 fc>0 the relations: = 2TT 467 Msv{z. t With Eq. these are exponentially increasing there.* ^ / 2 H e ( 3 ) ( Z ) g? Mev(z. -z) are proportional there. q.« * r / 2 H e ( 2 ) ( ^ g> ^ ft) _ e W 2 H e ( 2 ) ^ ^ ^ = a_ 2TT e . (19. h.131) where the second relation was obtained by changing the sign of v in the first.(z.135) sin irv 'This means.h2) Me-„(z. (19. -h. and thus can be matched in this region./i 2 ) = Me_j.q.g.130a).132) into Eqs. (19.133) In a similar way one obtains the connection formulas — sin -nv HeW(z. q.q. q.g.h) sin irv = sin7r(7 + ^)He (3) (2.g./i). Eqs. e W 2 H e ( 3 ) ( ^ ? j ^ _ e -M/*/2jj e (4) ( ^ q j ^ e .130b)). — sin 777 H e ( 3 ) ( 2 . h) = Ke(q. (19.131) and eliminating He") and setting again exp[m7] := we obtain the first connection formula ozwr — sin irv He (4) (z. and He(*> exponentially decreasing. Substituting the Eqs. z) and He (3) (z. g .129)) He (2) (z.

is only the dominant term for large h2} 19. which in our case is the solution He^4^. The latter is defined by the following large-r behaviour of the solution chosen at r = 0. q. Squaring Eq. .137) We see that this expression agrees with that of Eq. (76): "It is not likely at this stage that an analytic relation will ever be found connecting (what we call) v and 7 to (what we call) (I + 1/2) 2 and h2".e'i7ri 1.119a) or (19. (19. (19.138) TXVICOS TXV We obtain the behaviour of the Floquet exponent for large values of h2 from Eqs. (19. Wannier did not have the relation (19.135) is the equation corresponding to Eq.e. (19.l ) ' s i n 7 T 7 ^VK /2 ikr sin 7r(7 + v) _l)le-ikr -i5i„—ilir/2 SxelKr .5 D e r i v a t i o n of t h e ^ . Singular Potentials The factor on the left.1 .119b).l ) (l„—ikr e From this we can deduce that Si (19. The quantity 7 is now to be determined from Eq.133)) — sin TXV He^4' (z.135). (19. exp 2^(9 + 1) and solving for sin TXJ one obtains the expression sm 7T7 sm TXV = ~iei7rq/2sin le iq-K • COS 7T7.119b) as „4/i 1 + COS TXV ~ rrl ( g . (60) in the work of G. h) ~ — sin TXV -(-1) sin 7r(7 + v)e kr ikr in 4 rl/2e-g/r+m/A / (ig)1/2 ( .468 CHAPTER 19.( .133) we can now deduce the 5-matrix Si — e2t51.m a t r i x From Eq. cos TXV + sin TXV. and therefore remarks after his Eq. of course. where 8i is the phase shift. Wannier [278].(19.3.Sin TXV. ' cos TXV ± v 1 + elin sin2 TXV =p v'cos 2 TXV . Thus here the S'-matrix is defined by (using Eq.119a) and (19.139) ''Equation (19. (19.136) ^ jrfl+l/2) (19:I35) sin 7r(7 + v) sm WKljrtl-a/2) S m TXV (19.l ) ] ! c o s ( W 2 ) 2 2 (8h)i/ 2TT29/ 2 +0 (19.85b) obtained earlier. sin TXJ.135). i. H.

e. .4 The absorptivity of the S wave (attractive case).140) From this we obtain the absorptivity A(l.e. Using Stirling's formula in the form z\ ~ e~zzz+1/2V2^. we can approximate the equation for q ~ h (i.28) shows. i. irrespective of what the value of I is) by cos™ + l = c o s ( T j^-J y/e1. T h i s is r e a l for UR a n integer.137) together with (19. Si xe ~ te ilir COS -KV V cos2 irv — 1 — e (COS^Q COS ITU ig7r 1 + e'iqn ielln cos TXV 2 2 C O S TTU (l-p/2)n (19.138) we obtain. (19. as a comparison with Eq. 19. A{l.3 The Potential 1/r 4 — Case of Large h2 469 Since the right hand side is real for real h. since cos TVU is large. h) of the Z-th partial wave.h):=l-\Stf ^cos rr(uji + ivi) = cos Tn/jicosh COS ^7TQ N 2 1- (19. 9/2 (9-1) with the approximation q ~ h obtainable from Eq. the real part of v must be an integer. (19. i.141) COS TTU irvj — i s i n WR s i n h TTUJ.8h COS From Eq.e. This is different from the behaviour for small values of \h2\.139) could actually be neglected) implying | cos m>\ = cosh nvi. (19. for the attractive potential. imaginary part uj (so that 1 on the left hand side of Eq.109).§ Since the right hand side grows exponentially with increasing h the Floquet exponent must have a large 1=0 1 large h approximation small h approximation Fig.19. (19.

^ A highly mathematical study of the potential ~ r~2 as an emitting or absorbing centre has been given recently. Handelsman.470 CHAPTER 19. Eden [46]. 19.e. Cvetic. Partial or other aspects of the weak coupling (i. W Some of these sources can be helpful in further investigations. With the exploitation of perturbation solutions of both the periodic Mathieu equation and its associated hyperbolic form for both weak and strong coupling (more precisely h2) together with corresponding expansions of the Floquet exponent it was possible to go as far as the explicit calculation of the S'-matrix and the absorptivity. Vazquez-Poritz [61]. The diminishing fluctuations of A(l. Rosendorf [225]. Shabad [248]. S . Bertocchi. one may expect corresponding results for other singular power potentials. Apart from the papers already referred to. Fubini and G. D. Pope and T.k) in the approach to unity are too small to become evident here. Masson [191]. small h2) case have been considered by some other authors. Lew [128]. . Paliov and S.142) we can see the behaviour of the S'-wave absorptivity as a function of h. N.107b) and (19. Pao and J.139) 2Tr(16h)q /32\ f e Thus with Eqs.4 . "G. Treiman [271]. **A. and G. (19. A. Singular Potentials with near asymptotic behaviour (for h2 — oo) > Afi . H.** Finally we should mention that the potential r~ 4 together with the associated Mathieu equation have also been studied in interesting contexts of string theory. (19. Tiktopoulos and S. Limic [177]. Jones [78]. C. it seems it is not possible to guess from the above the result for the absorptivity of a singular potential with an arbitrary negative integral power. Lii and J. Esposito [88]. **R. L. see also M. in fact. Challifour and R. S. A. Cvetic. D. Apparently this is one of the very rare cases which permits such complete treatment." as well as other aspects. 11 J. I r a n [61] and M. in further contexts beyond those already mentioned. H. A. Lii. This is sketched schematically in Fig.-P.4 (there are tiny fluctuations in the rapid approach to 1). N. In essence. and. J. N. F. Yuan-Ben [285]. Furlan [20]. such as phase shifts. H. Dombey and R. E.4 Concluding Remarks In the above we have considered the solution of the Schrodinger equation for the potential r . B. Y. 19.' Rudimentary aspects of a singular potential with a general power n have also been considered previously.

*R. e. Muller [74]. This is indeed a remarkable connection. B.g. the cosine potential is more suitable for such studies than the anharmonic oscillator since its case is simpler. f R . Dingle [71]. the subject is much older and had been explored earlier in great detail in particular by Dingle^ with the subsequent investigation of the behaviour of the late terms of asymptotic expansions of the eigenvalues of the Mathieu equation and others.* In fact. Wu [18]. [19].1 Introductory Remarks The subject of the large order behaviour of perturbation expansions — meaning the study of the late terms of the asymptotic expansion of some function with a view to extracting information about the exact properties of the function — received wide publicity with the publication of the anharmonic oscillator studies of Bender and Wu. We shall also see explicitly. also convergent expansions are known and a lot of literature exists on the equation. We concentrate in this chapter on the large order behaviour of asymptotic *C. T. Dingle and H. Thus in the following we shall not only obtain the large order behaviour of the coefficients of the eigenvalue expansion but also that of the coefficients of the wave functions as well as the connection between these. and how this is related to the level splitting — in fact we shall see this in both cases of the cosine potential and that of the anharmonic oscillator. M. 471 . how the large order behaviour of the expansion of the eigenvalue is related to the discontinuity across the latter's cut. W.Chapter 20 Large Order Behaviour of Perturbation Expansions 20. Bender and T. B.* However. the asymptotic solutions in different domains all have the same coefficients. J.

19.2.2 The function ur = r\/pr vs. 20. r for p = 4. r for p = 4. Since the method is also applicable to convergent expansions. 20. one could naturally also explore the large order behaviour of these in a similar way and expect the behaviour discussed in Sec.1 The function ur = pr jr\ vs. It is a feature of a physicist's approach to a problem that he employs a method of approximation which is such that the first term of the corresponding expansion yields a rough value of the desired answer.6. Large Order Behaviour of Perturbation Expansions expansions using in particular the recurrence relation of the perturbation coefficients obtainable with the perturbation method of Sec.6.2. 8.472 CHAPTER 20. 8. as evident from Example 17.7.2 and below.1 and cases in Sec. Presumably this has never been done so far. Fig. 5. 10 12 14 16 18 20 Fig. This approach leads .8.

whereas in the case of the asymptotic expansion it is the opposite: a factorial in r divided by a power. The formal perturbation expansion of a quantity E(g ) . even if each En is known. the function E(g2) is given only approximately by Yln=o En92n. in the power expansion of the exponential function).1 Introductory Remarks 473 automatically to asymptotic series. and the expansion is said to be Borel summable. For an authoritative discussion of these aspects we refer to Dingle [70]. T V lim h2N 2 n EQ?) . If the function E(g2) can be written as the Laplace transform E{g2)= dze'z/g F{z)=g2 dte^FigH).1 and 20.4) Jo Jo its Laplace transform F(z) is called the Borel transform of E. (20. In some cases the answer is affirmative. that the maximum of the absolute value of terms in a convergent expansion and the minimum of the absolute value of terms in an asymptotic expansion are reached approximately when the value of the variable is approximately equal to the number of the term. i. this term represents the size of the error. A considerable amount of work has recently been devoted to the question whether it is possible to reconstruct a function exactly from its asymptotic expansion. The behaviour of such terms is illustrated in Figs. for instance. 20. The traditional method of using an asymptotic expansion is to truncate the series at the least term. as we discussed earlier in Chapter 8. equivalently for g = l/h.3) n=0 Thus for any given value of g . The behaviour of the r-th term (r large) of the convergent expansion is generally of the form of a power divided by a factorial (as.1) is an asymptotic series if for any N lim 92^o 1 g 2N N E(g )-^2En92n 71=0 2 = 0 (20. . It can be seen from these.20.e. and the integral E(g2) is called the Borel sum. E(g2) = f2 Eng2n = Eo + J2 n=0 n=0 E 2(ra+l) ^9 (20.2) or.2. The question arises: What is the information about F(z) that can be obtained from the asymptotic expansion of E.e. i.J2 2n h E (20.

. . that of F(z)) over a domain that is larger than its region of convergence (see Chapter 8 and Dingle [70]). We observe that a function E(h) with asymptotic series ^ r=\ A has an essential singularity at h — 0. Large Order Behaviour of Perturbation Expansions Eq.1)? Asymptotic expansions originate from integrating a series (i. (20.5) then with the integral representation of the gamma function F(z) = (z — 1)!. (and E0 = 0). . (20. In this ideal case the integral of Eq.n _ En+\ n! En+1 ^ (-!)"«!.4). n=0 (20. many asymptotic series do not even alternate in sign (as in the case of the series in Eq. J we obtain Now if />oo dte-Hz-\ °° oo ' 0 dte-H*-\ TEn+1g^V . In general this is not so easy.474 CHAPTER 20.6) The power series of F(z) converges inside the unit circle about the origin at z = 0. We can therefore write down a dispersion relation representation choosing the cut from 0 to +oo: . 1 we obtain oo F(z)-£(-*)" = — . . it is important to know the large order behaviour of the asymptotic series so that the question of its (exact or only approximate) Borel summability can be decided. (20. El . .n E(9 )~ 2 n=0 ^ E0 +1) = Ea^2( °° oo /»oo = J2ang2^ „-n -'O n=Q dte~H\ a_i = E0. In fact. n=0 (20. . (20.e. Q. In any case.4) can be evaluated and E(g2) thus recovered from the asymptotic expansion by Borel summation. Thus if CO F(z) = a-iS(z) + J2 anzn. The function F = 1/(1 + z) is an analytic function which can be continued beyond this circle to the entire positive z-axis as required by Eq.6)).

. (20. .3 and g < 0. 20. Weinberg [228]..g./ \p\dr\. § where 7(ff) = To exp | .. for < ^ 0. if the potential or a boundary condition is complex such that the problem is nonselfadjoint.2.11) V. (20.1 Introductory Remarks 475 except for possible subtractions. N = l. we have a quantum mechanical problem where probability can leak away. Comparing the last two equations we obtain (by expanding in the latter the denominator in powers of h'/h) -I j-QO Ar = 7T JO tir~l%E{ti)dti. $s E can be calculated by the WKB procedure. (20. and E represents an eigenenergy.2. Popov and V. This is the case if the real potential has a hump of some kind. or can we determine Ar independently? Fig.10) r as shown in Fig. S.N = l. ? When ?s E ^ 0. 20.20.. e.. i.e.. M..10) is considered in detail in this work. and with g = 1/h and it is found that§ En = 4 n ) ~ \lis)..9) The question is therefore: How can we determine 9 E(h).3 The potential V(r) gr». The potential (20. if the potential is V(r) = --+grN..

113)). (20. In the first such investigation it was sensible to consider the Schrodinger equation with the cosine potential.476 CHAPTER 20.76) (see also the comment on the comparison with the Kepler problem before Eq.73) in the W K B or semiclassical approximation.^ The logarithmic factor in Er is a novel feature of this case. Evaluating the integral show that 00 r=0 v ' where n is the principal quantum number of the Coulomb problem as in Eq. (11.2 Cosine Potential: Large Order Behaviour It is instructive to investigate various methods for obtaining the large order behaviour of an asymptotic expansion. L. the cosine potential being t h e most completely investigated will also be seen in the following. M.11) is. Large Order Behaviour of Perturbation Expansions where the frequency of particle bounces against the potential barrier is 70 ~ ^classical/27r and w c i assica i = 2ir/Tkepier oc 1/n 3 . An interesting application is provided by Example 20. Solution: The solution can be found in a paper of Eletsky. (20. as was (and still is) not the case for t h e equation 1 V . However. Eletsky. the eigenvalue expansions of the cosine potential and of anharmonic oscillators. Popov and V.1. Weinberg [83].9) for Ar. The most direct way to approach the problem of determining the behaviour of the late terms of an asymptotic expansion is to consider the equation determining these coefficients. . the Breit-Wigner formula (10. V. Straightforward Rayleigh-Schrodinger perturbation theory usually does not enable this in view of the unwieldy form of coefficients after very few iterations. Thus in the following we apply these methods to our typical examples. (11. since this is effectively the Mathieu equation for which — for comparison purposes — extensive literat u r e was already available. of course. (14. Example 20. S. Here n is the principal quantum number of the Coulomb problem (see Eq.7 with its focus on the structure of coefficients of perturbation expansions was seen to permit even the formulation of the recurrence relation of its coefficients. the perturbation method of Sec. 20.11).1: Application to the Yukawa Potential Evaluate 9 £ = —7/2 for the Yukawa potential V(r) = -g— r / \ r and insert the result into Eq.134)). as we expect on the basis of Eq. Popov and Weinberg. 8. Equation (20.

61)) that the coefficients M2i of the expansion of the eigenvalue A = E(q. . With some simplifications for large values of i this equation can be solved approximately and the large order behaviour of the coefficients Pi(q.j) are known.13) represents effectively a partial difference equation. 3)? . (20. obey the recurrence relation jPi(q.j + 1). (17.13) In the special case j = i the boundary conditions stated after Eq.j) of functions ipq+4j. Eqs. The recurrence relation (20.2 Cosine Potential: Large Order Behaviour 477 with an anharmonic oscillator potential. This has been done" and it was found that (we cite the results here since we rederive them below in detail by other methods. Miiller [74]. arising in the i-th order of the perturbation expansion of the wave function oo 1 i ip — const. (17.h) = -2h2 + 2hq+j.26) and (17. i.J) = (g + 4 j .58) and (17. (17.j) +(q + 4j + l)(q + 4j + 3)i^_ x (g.e.34)) that the coefficients Pi(q. are given by i . h). Dingle and H. j .j) a n d M2i can be deduced. % = Aq. B. (cf.l ) ( g + 4 j .58) eliminate the last two contributions and one obtains the exact expression ^ ' Z j _ 4<[2» + ( g .20.55)) X = E(q.l(q.2 A = Mx + ~M2 + 7^WM3 + •• • .l ) / 2 ] ! i![(g-l)/2]! • We have seen earlier (cf.l ) +2[(q + Aj)2 + l + A]Pl.60).14) Thus these coefficients can be obtained once the coefficients Pi(q.12) (20. W.3 ) P i _ 1 ( g .Bq.M2i+1 for large i. (20. Eqs. or equivalently of the quantity A.Cq. J. M2i = 2j2 J[Pi(q. Eqs. to avoid discussion of difference equations here) 2 l{q j) ^ (16)«(»-l)!(2i)«/° / _ 2q -2q + 3\ f i \ ' -[\{q-mw-m\ sr ){j)> (2(U5a) R. Thus in the case of the Schrodinger equation with periodic cosine potential we saw (cf. (17. (17.

•9-1 {[-l(? + i)]![-|(? + 3)]!}2 (20.15b) The z-th term in the expansion of the eigenvalues A of the Schrodinger equation with cosine potential (Mathieu equation) is therefore given by ith term i\ (8h) i-l 1 2g2 + 3 2i )[{[\{q- 7.*. see also below) .l (20. also written lCj. In the following we concentrate on the cosine potential and rederive the large order behaviour obtained above by other methods.19) ^nOW =2(9"1) We observe that successive terms do not alternate in sign.l ( 9 + l)]![-|(Q + 3)]!}2 i!^" 1 1 or (with the help of the duplication formula.16b) With the help of Stirling's formula one can derive the following relation: i large 1 (i + m)\ \-o (20. Hence exact Borel summation is not possible. In the first method we do this by first obtaining the imaginary part of the eigenvalue with the help of nonselfadjoint boundary conditions. and from this for i even or odd it is found that Mi ~ (16)^! 1 2g2 + 3 1i . Large Order Behaviour of Perturbation Expansions where the bracketed expression at the end is the binomial coefficient.17) We see that assuming i is very large compared with m is here equivalent to assuming m is approximately zero.18. .16a) (20.th term iW~^q 1 {[i(g-l)]!} 2 2 7 r(8/ l ) i .9-1 l)]![i(g-3)]!} 2 { [ .1 22 the result becomes i t h term (i + 2n)!2 2 " 1 where n 1 (20.478 CHAPTER 20. (20. Using this relation and the duplication formula l-w('-\W±^.

we emphasize a few points *M. W. We retain the previous boundary condition at one minimum. t To enable comparison with the work of Stone and Reeve we shift the cosine potential considered previously by 7r/2.e. A means to achieve this was found by Stone and Reeve. Miiller-Kirsten [200]. 20. From our previous considerations it is clear that the formal perturbation expansion of the eigenvalue X(q) in terms of the parameter g.2 solutions obtained earlier. when this is g0 = 2n + l . Reeve [262]. . in the calculation of the level splitting as a result of tunneling.20. The potential we consider here then has the shape shown in Fig.* However.5- \ \ / \A Fig. . but now change the boundary condition at the neighbouring minimum in such a way that the difference q — qo becomes imaginary.4. i. -\ 1- \ \ \ \ \ 0. 20. J.3.5- A i 1 1 r f \ \ M\ \ \ / ! \ \ -1 \ \ I 3 \ \\ \ /41 i \ \ -0. Stone and J. . we follow here a formulation in the context of our earlier considerations of the Mathieu equation. which means that the potential is there approximated by an infinitely high harmonic oscillator well with levels enumerated by the quantum number qo or (equivalently) n. f H . this difference was found to be real.4 The cosine potential shifted by TT/2. remains unchanged. but with different boundary conditions. the difference q — q$.3 Cosine Potential: Complex Eigenvalues 479 20. At the risk of repeating what we explained earlier. This is a very instructive exercise which shows how the eigenvalues change with a change in boundary conditions. 2 .1 Cosine Potential: Complex Eigenvalues The decaying ground state Our intention here is to consider again the cosine potential and hence the Mathieu equation with the same large-/?. n = 0 . . Previously. What changes as a result of different boundary conditions is the deviation of q from the odd integer qo. .3 20. l .

Thus around z = 0 the potential behaves like t h a t of an oscillator well with minimum there. we shift the argument by n/2. a boundary condition which at the face looks somewhat abstract. Effectively we consider the tunneling from the left well in Fig. and thus obtain ip"{z) + [A + 2h2 cos 1z\i>{z) = 0 (20.4. 20.2h2 cos 2z]ip{z) = 0. . Large Order Behaviour of Perturbation Expansions for reasons of clarity (since some considerations here may otherwise not be sufficiently clear).21) <u> + ( i « .4. Thus one chooses not the solution with real argument at the minimum.4 (around z — 0) to the one on the right (around z ~ ir). Around z ~ 0 w e can expand and obtain i)"{z) + [A + 2h2 . (20. Now we change this situation at the minimum on the right.20a) As stated. X + 2h2 = 2hq+—. as we found these for this potential. 8 the equation can be approximated by d2ip(w) (20.Ah2z2 • • • }^{z) = 0. (1 w < 2a22) cylinder T h e solutions of this equation are the real and complex parabolic functions D ±(q~i)(±w)> In the following we find it convenient to select solutions of this type in the domains around the minima of the potential in Fig. The physical wave functions there are therefore the solutions which are real and exponentially decreasing. but achieves exactly what we want. 20. i.480 CHAPTER 20. Setting w = 2h1/2z. but t h a t with complex argument there. the degeneracy of eigenvalues of neighbouring oscillators being lifted by the finite height of the barrier separating them. which means replacing z by z±ir/2.T K ^ a D_i_{q+1){±iw). To be more precise we recall the Mathieu equation we had earlier: ip"{z) + [A .20b) with potential V(z) = — 2h2 cos 2z as shown in Fig. 20. and allows its imaginary part to tend to infinity. We can distort the potential there or alternatively impose a different behaviour there on the wave function. Here we choose t h e second way. Previously this situation was the same at the neighbouring minimum. and correspondingly we assume alternately even and odd oscillator eigenfunctions there.e.

This somewhat abstract looking boundary condition is effectively simply one imposed on a complex solution of the equation giving the oscillator eigenfunctions. The calculation for the general case is nontrivial and so will first only be written down and will then be verified by specialization to the case of the ground state. 2. The proper harmonic oscillator solution there would be that for infinitely high barriers to the left and to the right with a dependence on an integer n which enumerates the quantum states in the well from the even or symmetric ground state with n = 0 through alternately then odd and even states upwards with n = 1. 3 . (where for the ground state q = go — 1) D_i{q+1)(±i2h1/2z) ' ± zoo (20. however. . A derivation is given subsequently. Eq. A (which we can loosely dub solutions of WKB-type). whereas here we start off with a solution even or odd about a minimum. (20. . that at z = 7r. We consider here explicitly only the case of the ground state around z = 0. Since the large-/i2 solutions valid away from the minima were the solutions of types A.24) Considering t m O i n Eq. 17.23) .63). we naturally constructed the even and odd solutions with these (cf. i.e.> 0 for z -> ±ioo. we see that ips ~ cos \j2qhz for z ~ 0. so that the deviation q — qo results from the finite height of the barrier taken into account with boundary conditions at the next minimum. Thus we write the boundary condition at Kz = ir: tp(±iw) — 0 for > w — 2h}/2z -K z— > i. .4. One should note the difference to our earlier case of the calculation of the level splitting: There we were seeking solutions which are even or odd about the central maximum of the potential. .e. we now choose an harmonic-oscillator type of real solution around the left minimum of Fig. whereas here we construct these from oscillator-type solutions. 20.22). Here — again as before — we take the finite height of the barrier to the right of z = 0 into account by taking q only approximately equal to go = 2n + 1.20. (20. the even or symmetric solution about z = 0 can be taken as *• = liDfr-riVhWz) +Dh{q_qo)(-2h^z)}.3 Cosine Potential: Complex Eigenvalues 481 As explained above. Thus. by a different method. such that this deviation becomes imaginary. . considering around z = 0 the ground state of the harmonic oscillator for which go = 15 and replacing go by g.

482 CHAPTER 20. Oberhettinger [181]. From Tables of Special Functions-'.25) qo) E2 0: (20.h(-2h1/2 J £2=0 (20. p. Magnus and F. and then evaluation at u = 0. Thus the only nonvanishing contribution comes from differentiation of Y[—v).28) Extracting Du{—w) we have A.-w) = e T Dv{w) 2TT n-») Av-ViD-v-^iw).•w) + 2ir T{-u) <"+!)% D-v-i(iw). 92.we obtain the following circuit formula of parabolic cylinder functions Dv(w): Du{w) T DV{.30) 'See e.27) requires the differentiation of this expression with respect to the index v. Large Order Behaviour of Perturbation Expansions Although the two parabolic cylinder functions in Eq.g.27) The differentiation of the parabolic cylinder functions with respect to their indices is nontrivial. Setting v = 0 in T(—v) = (—v— 1)! gives infinity.26) Thus here we can approximate ips by expanding it about E2 i)s~DQ{2hll2z) + E2 ~{DE2/4h(2h ^ z) 1 2 + DE2/.3. or logarithmic derivative of T(^) for which:§ ^(1/) r» (20. . tp(v). We see this as follows. The differentiated function Y{y) is in Tables of Functions expressed in terms of the so-called psi function. Magnus and F. This is the feature which gives rise to an exponentially increasing contribution in the neighbourhood of z — 0 and hence provides the dominant contribution in going away from the origin. we know from the solutions of types B and C in Sec. 2. 17.27) differ by an angle IT.24) are linearly independent around z = 0 for q ^ qo. (20. §See e.g. Oberhettinger [181].2 that one parabolic cylinder function is exponentially increasing in h and the other exponentially decreasing. The arguments of the functions appearing in Eq. W. p.(. this can be handled in a few lines. (20. As before we set A = -2hz + 2hqA where in the present case Ei = -2hz + 2hq0 + — and A & = Ex + E2 (20.29) Equation (20. However. (20. W. (20.

Magnus and F.4.)e" -w /4 2 l ^D-x(iw) (20. 1E2 + --± 0(e~hz2) 2 Ah Ihir ehz* E2V'< 16/i2 V2^ehz2~ 2hM2z (20. Magnus and F. 3 . .. (20. we obtain 4>s D0(2h^z) D0(2h1/2. Oberhettinger [181]. r(-i/) We now return to Eq. . . . i/^0 483 T(v) It follows that d dv [T(-u)\ (20.g.e. For u — 0 one obtains the result (20.+w2/i IW Inserting this result into ips and retaining only the dominant contributions (i. 5 7 7 . with the following expression for the large-z behaviour'! (here v = E2/AK) Dv{w) ~ wue~w2/4 = evlriWe~w / 4 for | arg w\ < ^TT.31) dT(-u) r*(-v) dv 1 i/>(-v) ^ o -l. W.33) = IV2~TT. (20.32) which we can use here since all other problems have been resolved: d D {-w) dv v (20. dropping from the derivative part of Eq. Oberhettinger [181].31).29). (20. pp.29) v=0 (in + In w)D0(w) + v2ire (m + lnu. Thus calculations for qo ^ 1 are not given by M. This result is » not easily generalized to <jo > 1. 92. i. "See e. which are valid around z ~ 7r/2.3 Cosine Potential: Complex Eigenvalues From formulas given in the literature we deduce the limiting behaviour^ -1. we obtain the formulas (X) / \ 1 OO I » J ] (l+-)e-.e. Differentiation of this equation and subsequently setting v — 0 yields.//n and ij>{y) = -C . p. those of type A. We now go to the WKBtype solutions.20. Reeve [262].34) This is the behaviour of the symmetric ground state solution away from the origin towards the limit of its domain of validity.27) exponentially decreasing contributions. including the derivative of DE2ufl(2h1/2z)). and "Prom W.+ v ^ v k^x k ^ + ^ where C = Euler's constant = 0 . Stone and J.

36) cos | . Then V'WKB becomes (with cos z ~ — 1 + z' /2. (20. Recalling that we have shifted the potential by 7r/2. by there replacing z by z + ir/2.. We now return to the WKB-like solution (20.34) and (20. where we can then match these to the solution ips. To this end we set z' = z — ix. we obtain the required solutions from the earlier ones.e.e. with expansion of cos z in the exponential) D0{2hll2z) = e~hz ~ e~2h / „2/i cos z \ — ).38) in its limiting domain towards 2 = 0. i.34). (20.36) we can identify the dominant large-/i2 behaviour of the constant C\ as c\ = e~2h. y COS c\ J (20.40).23) demands (with constant A) ip = AD_1(±i2h1/2z') -> 0 for z' -+ ±ioo. cos z/2 ~ (20.cos z — 2/icos z ipA^ Tcos | and i n A —-. i. that in approaching z = ir.e. We observe that in the matching domain (i. Eqs. Large Order Behaviour of Perturbation Expansions extend these to their limit of validity to the left. (20.37) Comparing Eqs. 2 c2e2he-hz UJ2 . Thus the WKB-like solution becomes p — 2/lCOS 2 V'WKB ~ D0(2h1/2z) D0{2h ' z) l2 + c2 + c2 ^-T-sinf -2/i hz2 _z_ 2 (20.484 CHAPTER 20.35) We construct with constants c\. (20.sm I We now want to match this solution to that of Eq.41) 2TT .39) Now. (17.29) and (17.40) From Magnus and Oberhettinger [181] (p. 92) we take the formula: Dv{w) = T { y ^ \eiv^l2D^v^{iw) + e-'^D-^-iw)] .36) and consider this at the other end of its domain of validity. Thus in the present case and with q ~ qo = 1 we obtain „2ft. c2 the WKB-like linear combination e2hcos ^ W K B = ClipA + C2lpA ~ Ci z _—2/icos z — + C2 r-T-- (20.— 721 .36) V'WKB cie~2hehz' 9h - . near z = n the boundary condition (20. -smf (20. x . (20.

.34) we obtain E2 = i%\ = ±ih22Qe~8h.e.complex conjugate] (2Q42) (2032) y / | Z e _ ^ + 2(.44) = DQ{2h / z) ± -(2h7r) / e.3 Cosine Potential: Complex Eigenvalues Thus for v = 0 we have DQ(w) = —j=[D-\(iw) V 27T 485 + D-i(-iw)] = real.20. J Hence Eq.1/V)_le^_ Z With this result Eq. We derive the general result in the next subsection using our own method. (20.^-.e. for both even and odd harmonic oscillator wave functions around z — 0 with qo > 1) is nontrivial. ^'ml-iw <2a47) For qo = 1 we recover the result (20.43) we obtain.40) can be written ^ ~ AJ^e~hz'2 + A~(i2hl^2zTlehz'2. i. (20. 1 2 4 1 2 8h ^^2 2 (20. (20. A = ±i4/l1/2e-4h ^| t c2 = T i2(2/ivr) 1 /2 e -6/ l .43) Comparing now Eqs.45) Comparing this result with tps of Eq.2/.39) and (20. The general result stated by Stone and Reeve [262] is therefore presumably guessed.46). (20. (20. with c\ = exp(—2h) from above. (20. and hence (adding and subtracting equal terms) D-i(i2hV2z') = ^[D^h^z') + D_1{-i2h1/2z')} Z + hD_1(i2hl''1z') = \V2^D0(2h^2z') Z D_l{-i2hll2z1)] D^(-i2hl/2z')} + hD-iWh^z1) + -[D^i(i2hl/2z') Z ^D0(2hl/2z') Z .46) Repeating these calculations for the general case (i.38) can be written TAWKB (20.

486 20.5.(|)#0. 20. H.z. .5.*. (20. On the other side of the barrier at z = —TT/2 we consider different boundary conditions. Fig.Ah2 ( z + .E curre it \ j. harmonic osc -71/2 E>V / E<V z+ O \ E>V TC/2 \ \ . Thus in the immediate neighbourhood of z behaves as 2 1 2 ip ~ e ±i(E+2h ) / z ~ e -7r/2 the wave function i^{z) ±i(2hq)1/2z **We follow P. W. •+(=)-* V-(f) -0.3.5 The cosine potential from — TT/2 to n/2. V(z) . the result is formula (334) there. (20. Considering the original equation. Miiller-Kirsten and A. (20. this equation may be approximated there as d2ip ~dz^ + E + 2h2 . Large Order Behaviour of Perturbation Decaying excited states Expansions For our derivation of the generalization** of Eq.e. i. Wiedemann [4]. •_(:)*<>.46) we consider a half period of the cosine potential as illustrated in Fig. Achuthan. We have to impose two sets of boundary conditions. at z = TT/2.e.48) These conditions provide the required quantum number qo = 2n + 1. Eq. 20. i. we impose the usual type of boundary conditions as for the minimum of a simple harmonic oscillator there.20a).) + 7T V> = o. J.e. i. and expanding cos 2z around the point z = — TT/2. 20. At the minimum to the right of the barrier shown in Fig.2 CHAPTER 20.

three pairs of solutions of types named A.20. (20.z). For the continuation we use again the WKB formalism. Recapitulating these from Chapter 17.' 4 / 1 1 / 2 COS ( -Z + -7T \2 4 . i. (20.51a) where H Bq[w(z)} w(z) = h(«-v (w) Bq[w(z)} = Bq[w(-z)}.r/2 ~ e -i(2hq) V 2 .-z)=tpA(-q.49) This is a complex boundary condition which requires the coefficient of the other exponential in ip±(z) (continued to and beyond — 7r/2) to vanish.21)).50a) V'AfeM) = where (the second expression is needed for later comparison) Aq(z) cos§(9-i)(iz + lyr) + W) [ t a n (±z + W)]-«/ 2 [sin(i* + ±TT) C O S ( ^ + ±vr)] V' ' sinfr+Vaz (20. for COS I -Z ± -7T 1 1 2 4 >o^\ The second pair of solutions is ipB(z) -J2hsmz B. (A) Boundary conditions at the right minimum We begin with the evaluation of the boundary conditions at z = 7r/2. as shown earlier.50b) which are valid as asymptotically decreasing expansions in the domain away from a minimum.-h.Hz^+Ol^ Bq[w(z)}+0 h1'2 (20. B and C.3 Cosine Potential: Complex Eigenvalues 487 (using Eq. For particles or probability to leak through the potential barrier from the trough at z = —ir/2 in the direction of negative values of z where we choose the potential to be zero. we have IPA(Z) „2hsinz Aq{z) + O il>A(q. we demand that ^(*)L<-.e. (20. (20. Proceeding as in our previous cases and substituting the expression (20.21) into the Mathieu equation we obtain.51b) 2i(«-1)[i(9-l)]'.h.

=F2 2\1/2{28h2y/4e~Ah l + O (20. (20.52b) The first solution is valid asymptotically around z = —vr/2. These solutions are valid asymptotically with the first in the domain around z = ir/2 and the second in the domain around z = —ir/2. (20.48) and proceeding as in Chapter 17. We saw in Chapter 17 that some of these solutions can be matched in regions of common validity. Thus we have IPB(Z) = aipA(z) and VcO*) = O ^ A O 2 ) .57) where go = 1) 3. ^ W H . i.488 CHAPTER 20.54) The even and odd solutions are defined as iP±(z) = -tyA(z)±iPA(z)}. Finally the third pair of solutions is Cq[w(-z)}+0 Cq[w(-z)}+0 Cq[w(-z)} where Cq[w(z)} = 22(9+1) •((7-3) 1 h1/2 1 hW> (20. and the second around z = 7r/2. the relation q .e.^ i . we obtain our first transcendental equation.55) Extending these solutions to the domain around the minimum at z = 7r/2.53) where a (8M («-!)]« 1 + 0 { \ a = 1 (8/i) K9+ ) l + O (20. (20.56) Applying the boundary conditions (20.52a) = Cq[w(z)}. 5. we obtain 1>±W = \ a a (20. Here the upper sign refers to the even states and the lower to the odd states. Large Order Behaviour of Perturbation Expansions the function H!/{w) being a Hermite function. \vw). We return to this condition after derivation of the .qo .

D. Jz+ ' T In principle the integral could be evaluated exactly — see P.5 9 ) The WKB solutions to the right of z+ (where V > $IE to the left of the maximum of the barrier) are given by ttw . and we obtain the proportionality by appropriate expansion.Ah2 cos2 z ip = 0. formula 280.e. F. p.3 Cosine Potential: Complex Eigenvalues 489 second transcendental equation from the boundary condition at the other minimum. Friedman [40]. r ywKBlZJ ~ ~ — [4/i2cos2 -2/i ]i/4 2 9 z 2 2 e x p ( £ [Ah2 cos2 z - 2hq]ll2dz) ' 1 2 .2hq] / dz) We know that the solutions I/JA{Z). (B) Boundary condition at the left minimum The solutions of type A have been matched to the oscillator-type solutions at the right minimum. 163 — but here we are interested in the WKB solution in a domain where it is proportional to a solution of type A. (20.67).[ [Ah cos z . Byrd and M. C O S 1/2 ^" (l) ± ' -2<2+<2- (2 °. Thereafter we extend the latter across the turning point down to periodic WKB solutions. We begin therefore with the matching of V'WKB ( Z ) to IPA{Z) and integ rait* V'WKB (Z) to I=f V'AW- Hence it is necessary to consider the elliptic (20.58) Ignoring the correction A / 8 . e x p ( . . Inserting the approximate form of the eigenvalue into the Schrodinger equation we obtain dz2 + 2hq+ — .01. The solutions of type A are valid to the right as well as to the left of the maximum of the barrier.20. I^A(Z) overlap parts of the domains of validity of these solutions. i. we see that the turning points at z± are given by 2hq . Our procedure now is to match these to exponential WKB solutions. Some nontrivial manipulations are again required to uncover the generation of important factorials.61) [Ah2 cos2 z-2hq]1/2dz. (20.49) and we can apply our boundary condition. as we shall see with the result of Eq. These then provide complex exponentials like (20.Ah2 cos2 z ~ 0.

59) 2h COS Z+ - ~ 0.64) .» _ i y v .sin2 z = l . 2/i In the same spirit we can set h[z + l 2h 1 — cosi I z + — 7r 2/i[l + sin z\. we obtain f ~ I = = 2 1/2 Ah [z + 1 2hq + o\h \z+- 2 dz J Z+ 2h I /MH)' dz 1/2 1/2 2h 2/i /•«"(. (20.62) as follows.62) where we set — in accordance with our expansion *+ + 2 IT 1/2 <? + (20. We have for the following expression contained in (20.490 CHAPTER 20.) +0 z+ IT Inserting this into the integral I.62): L: = In ~ and hence L = ln In * f 2(z + f) {q/2hyn + + (* I In + 1/2 1/2 2hj 2 In In 2h 4(1 + sin z) (g/2/i)V2 cos z' 2(z + f) (qphy/^z+Z) !sin2(f + f) ( 9 /2fc) /22sin(§ + f)cos(§ + ^2 ' 1 4- = ln 4tan(f + f (9/2MV2 (20. (20. using the relation 1 + sin z = 2sin2(2i/2 + 7r/4). Jz++7r/2 \ 1/2 2 V Integrating this becomes z+(ir/2) 2/i h 2 Z+ V IT 2h 4/i In ^ + 1/2 zA 2h 2 2h I ln z+ i) ( I 1/2 2 + 2+ 2 1/2 2/i J (20.cos2 ( z+^-) = (z+ . we write cos2 z = l .63) We handle the logarithmic terms in Eq. Large Order Behaviour of Perturbation Expansions Since cos(z + n/2) = — sin z.

^ Proceeding similarly with the other WKB solution we obtain rWKB (z) ~ e x p ( .64) into Eq.67) (27r) /22«+5(4/i2)V4V^y The last expression is again obtained with the help of Stirling's approximation. (20. . (4^2)i/4 [ i( g _3)] ! V/ l (20.65) [Z[4h 2cos2z-2hq]1/2d< e2he2hsin ^e-g/4(g/2/t)g/4 2<?[tan(§ + f )}i/2 (20.3 Cosine Potential: Complex Eigenvalues Inserting the relations (20. .20.50b). (20.63) and (20.66) The WKB solution therefore becomes (approximating the denominator of (20. .68) ' T h e use of Stirling's formula here and below — essential to obtain factorials — assumes q to be large so that corresponding corrections would have to be calculated.50a) and (20. We ignore this here.62).60) by (4h2 cos2 z)1!4). on using in the second step Eqs. 2*4(4tf)V« W ^ e-[l( V D]! 1 /2y^ (20. 1/2 (20. we obtain I ~ 2h + 2hsm z Hence exp 491 V^m 2h\~'I z TT 4 — tan . <$&?(*) e2he2hsinze-i/4(q/2h)i/4[tan(% + l)]'q/2 2?(4/i 2 ) 1 /4{2sin(f + f)cos(f + f ) } 1 / 2 e2he-q/A(q/4y/4 /2\q/4 ./ * [Ah2 cos2 z IZ+ 2hq\l/2dz) + f )]«/ 2 (4/i 2 COS2 z)V4 e-2he-2hsm zeq/4(q/2h)-i/4[tan(% /2^~Q/4 (4/i 2 )V42-9{2sin(f + f ) c o s ( f + f)}V2 ^-27*29-1 (4/l )V4 e -?/4( g / 4 ) 9 /4 ^ J e" ^^(2TT) /2 2 1 2 ^ ) / 4 /2\-" - .+ qj V2 4 A. but mention that for q an odd integer: j(9-D l + O and 1 r(9-3) -4(9+1) ! = ±7I .

69) l(8M9/4e-2^ \2 [ i ( g .—-^ (2^)^(16^2)1/4^ ^ _ _ [T+(±)exp{i(2/ig)1/2(z+_z)} 2(2/ig)V4 + r _ ( ± ) e x p { .68) we obtain ^±(z) = obpA{z)±t/JA{z) l/^y/4e-2fe(27r)1/22^^(4/i2)1/4 1 « llfh\-^e^[l(q-3)]\2-^ ± /(r. (14. (14.2 [ ? ) Imposing the boundary condition (20.e 2[J(«-l)]l[j(?-3)]! i l[|(g-3)]!e (20.+) + B^) HAh^/\ir.. We now return to the even and odd wave functions defined by Eq.T 7 77 r )V2 /4(2 .U ^K^.KW4^8 4 . S1H (2/ KZ ) / (z f 2 (2^).™) .67) and (20.49). i. io(8/*)-g/V^i6ft2)V4[±(g-3)]! .z+)/ VWKBW 2\2j (2^72 l(8^/4e-2fe(27r)i/22V2(4fr2)V4 2 < nz+) ^WKBW [\{q-l)]\ 1 (8/i).1 ) ] ! 1 (8/Qg/ e4 2fe l[|(g-3)]!e^ . (20.492 CHAPTER 20.55) and match these to the oscillatory W K B solutions V'WKE:( Z ) > V'WKB (Z) t o the left of the turning point at z+. we obtain T _ ( ± ) = 0. (20. Large Order Behaviour of Perturbation Expansions again using the Stirling formula.?/ e^2 /2(4/ l 2)i/4 [ i ±2 (^)V^ 4 1 (g _ 3)]! (/]Z+) ^WKBW- Inserting here the oscillatory W K B expressions (cf. where _ 1 + [±) -*) + £ (20. Eqs.i ( 2 / i g ) 1 / 2 ( z + .z)}]. Thus with IPA(Z) and TPA(Z) taken from Eqs. 1 2 1 ±-2 .50)).l^l !T ^^* T2 7 r ( ^^ /. i 27r(8^/ 4 2/t /4 ' r-(±) .49). we obtain l(8/i)g/ 4 e.2 f e (27r) 1 / 2 (16/t 2 ) 1 / 4 2 (2/wz)V4[i(9_i)]! cos (2M1/2(^-^)+4 7T .

in fact. (20. (20.72) We observe that with the square root as on the right hand side both sides are complex. we obtain E(q.3 ) I = }(»-« !7rl/22-I(. (20. Eqs.h)+i%Ego.20.73) Inserting this expression into A = E(q. In both cases we wrote (cf.47). (17. (20.3 Cosine Potential: Complex Eigenvalues Using the duplication formula \{1-l) this becomes 7rV 493 \ ^ . (20.e.h) = = E(q0.h) + i l + O '4{[^o-l)]!}2 E(q0. We now have the two equations which have to be satisfied. Equation (20.h) + (q-q0)( BE — 90 E{qQ.26).72).75) . (20. h) of Eq.72) is our second transcendental equation. Ex = -2h2 + 2hq0 + A 1 (<? .3 R e l a t i n g t h e level splitting t o imaginary E We can compare the results of the two problems with the cosine potential and the same solutions but with different boundary conditions.4h (16 2 /» 2 )9/ 4 (9-1)]! = (16 2 /i 2 )«/ 4 = (-16/i) 9 / 2 . Eqs.21) expanded about q = qo.-l)j /2 e-4fe(162/t2)^4_±l or ±i 7r\ 1/2 (16 2 /i 2 )9/ 4 e.57) and (20. (20.3. Combining these equations by replacing (28/i2)<?/4 in Eq.57) by the left hand side of Eq. 20. who derived it for qo = 1 and guessed the form for general values of qo.qo) El Ah' (20.h)+2h(q-q0) (16h)i°+le'Sh E(qQ. for q an integer both sides are imaginary. the result (20. i.72) we obtain q-qo q0fiSh 2(16/i)*>e {[!(<?<)-I)]!} 2 1+ 0 (20.74) This result agrees with that of Stone and Reeve [262].25)) E = A = Ei + E2.

25Eqo=8h[ 2\1/2 .2 below. H.4 R e c a l c u l a t i o n of large order behaviour Now that we have obtained QE.3. The same formula can be derived in the case of the double well potential.e. Miiller-Kirsten and A.W [£(90-1)]! < (20.+ | . A = . J. we can obtain the imaginary part calculated above from the formula 4 (H) A ^ o = 27ri(<5£90)2. 7 8 ) i=l V .* The existence of a formula of this type was conjectured without an explicit derivation like that given here and only for the ground state.I g | ± l ( 2 „.74b).) {I6h)™/2e-4h 1 + . * It would be interesting to see a derivation of the formula — reminiscent of an optical theorem — from first principles. cf. B.9) the coefficients Ai of its perturbation expansion. by others. . 20. W. Eq. i=l We can calculate Ai for large i from the relation Ai = * Jo 1 P°° ti^QE^dti (20. For definiteness we recall the appropriate expansions together with the definition of coefficients Aj. its tunneling properties and the large order behaviour. we can recalculate with the help of Eq. i. We had A= or _2ft2 + 2ft. Fateyev [26]. f E . Achuthan. we recall that we obtained there for the difference 25Eqo between levels with the same oscillator quantum number qo. the discontinuity across its cut from zero to infinity. (17.494 CHAPTER 20.80) *P. Large Order Behaviour of Perturbation Expansions Referring back to the calculation of the level splitting with selfadjoint boundary conditions. (20. (20. Setting ng AEqo:=2iSfEqo. Bogomol'nyi and V.76) he Thus SEqo is the deviation of one of these levels from the harmonic oscillator of level. Wiedemann [4] and Example 20.77) One can say that this simple relation summarizes the intricate connections between the discontinuity of the eigenenergy across its cut. A.

3 ) P i _ 1 ( c ? . and the method can be applied in many other cases. Dingle [74]. Muller-Kirsten [200]. (20. We also observe that these terms do not alternate in sign.14). i ^ = ^ + E?27MT E ^ifoj^W (20-82) The coefficients Pi(q. The integral is recognized to be of the type of the integral representation of the gamma function.13) and the coefficients M2i or M2i+i of the eigenvalue expansion can be expressed in terms of these as in Eq.t We consider again the periodic cosine potential with (unnormalized) eigenfunction expansion oo . For convenience we write Eq. J. j . J).81) We observe that this result agrees with the result of Eq. (20. a = J + \<1> (20-84) .. (20. 20.j) + {q + 4j + l)(q + 4j + 3)Pi-1(q.20.4 Cosine Potential: A Different Calculation A further very instructive method of obtaining the large order behaviour of a perturbation expansion was found by Dingle* in application to the case of the cosine potential or Mathieu equation. W. Using again n = -(go .1). the integration is seen to give (20. But as Dingle remarks.74).j) satisfy the recurrence relation (20. (20. f H . there is nothing special about this case. We sketch this method here slightly modified and extend it to the final formula for comparison with our previous results. B. (20. *R.13) out again: jPiiQj) = (g + 4 j .l ) +2[(q + 4j)2 + l + A}Pi_1(q. so that an exact Borel summation is not possible.l ) ( < ? + 4 j .19) for the large values of i under consideration here.83) One now sets ftH = ( 2 % P * ( 9 .j + 1).4 Cosine Potential: Another Method of Calculation 495 with $$E(h) given by Eq.

85).1/2(z) (a-1)! (20.z)WaA/2(z) Setting V.h) 2+ 8h a p(a. 1 | 1 / 2 (4 (20. the recurrence relation (20. (20.(a) oc --. Large Order Behaviour of Perturbation Expansions and considers large values of j for large values of i. (with a = j + g/2 in the coefficients) 8/? —Pi(a) = P i _x(a .a R .4 ) ^ . Abramowitz and I.e. .e.89) Replacing here K by a and \x by 1/2. (20.496 CHAPTER 20.^(Z). i. p(a. Now. we obtain (2a .91) + Wa+lil/2(z) = (1 . z2 (20.h) = ^2pi(a) with p. (20.87) Taking terms of 0(h~l+1) of this equation.83) can be written j .1) + 2 P i _i(a) + Pi_i(a + 1).90) See M.e.27h pi(a) a = 4j(4j + 2 g . g < 4 j . 4j(4j + 2g + 4) + / . Taking dominant terms.ll{z)= L .l/2l Wa.g. e. Whittaker functions WKjfl.1) + 2 (a — 1) a . Stegun [1].86) one can convince oneself that Eq. which are solutions of the differential equation known as Whittaker's equation. 0 4j(4j + 2g) pi_i(a .l. Pi(a) i.h) +p(a + l. (20.4. (q + 4j) 2 ~ 4j(4j + 2g). (20. a. we regain Eq. formula 13.h). A.31. i.85) Considering now the generating function of the perturbation coefficients.85) implies p(a .K + ^ ] L + K-^)WK.88) satisfy among other relations the following recurrence relation which is of significance in the present context-t {2K-z)WK>li{z) + WK+1. (20. d2WKh dz2 + 1 4 K z z-M2n WKtll = 0. nx pi_i(a + l).

88) which reminds us immediately of a radial Schrodinger equation with Coulomb potential. a (20. > > ^See W. (20. p. Magnus and F .(« .Tl ss\{-a)\{-a-l)\ (-z) s=0 s — a)\(s — a — 1)! for |z| — oo.j — —j. Dingle discovered the solutions^ p(a.88) is known.h) = \~\Pi(&) with pi oc h l. h) is expressed in descending powers of z = 8/i.92) with Eq.(« -1) 2 ].86) have to be chosen such that p(a. h) is expressed in descending powers of h.(« .20.g(h) are functions of h which in view of Eq. g(h)(20. > (20.93) (-a-1)! ' where f(h). In the present case this means the replacements q — —q. h —-> —h.\)V .h) ~/(fe) ' (a-1)! . we have to remove a factor (recalling a = j + (7/2 of Eq. another solution is obtained from this by changing the signs of K and z since this leaves the equation unchanged. i.. Thus the approximate large-z asymptotic behaviour is WKtp(z) ~ exp •FH V -dz z e ^' 2 exp(/clnz).„ +1)2] 7 n\z \n which with K = a and /z = 1/2 is: WAz)~e-^z«Y. v . such that p(a.4 Cosine Potential: Another Method of Calculation we obtain Va-i.94) Hence to insure that p(a.i/2(z).92) Comparing Eq. ^Observe that if one solution of Whittaker's equation (20. (20. We can infer the approximate behaviour of the Whittaker function for large values of z from the differential equation (20. (20. Oberhettinger [181]. The more precise form obtained from Tables of Special Functions' is w*A* z 2 e / z" n=l [f .84)) {8h)q/2( -Ah for j < 0. (20.87). .i/2(z) + Va+lA/2(z) = 497 2+ Va.e. 89.

1/2(8^) ( .. x 4i[2» + i(« — 1)1! il! [ ^ ( 9 .95b) Then for j > 0 we obtain from Eq. Muller [74].8 / t ) for j > 0.). we obtain the result of Dingle: (20. This expression can be obtained from the recurrence relation (20. . W.. Large Order Behaviour of Perturbation Expansions and write / i \ pl<x = j+ 2<7>h) « (8h)~q/2e4h v ^-+9/2.1. B. (20. Correspondingly we set for j > 0 / 1 A pla = j + -q. J. h\ oc (-8h)q/2e~4h ( _ .94) here we have a series in descending powers of h. (20. ^ .j _ g / 2 | 1 / 2 ( . _ i s.498 CHAPTER 20.96) and a similar expression for j < 0 with /i — —/i.84) we conclude that the original coefficients Piiflij) a r e f° r large i given by the following expression — apart from an as yet undetermined proportionality factor cq — We determine cq by imposing as boundary condition the expression for Pi(q. this becomes / 1 \ (-1)5'--" ^ U + iq)<(s + iq-1)\ 1 Picking out the term in hrl. Dingle and H. j — —j. 2.83) and is given by 1 1 .+ i f or 3 < 0- (20. n = 0.l ) ] ! 11 R..95a) Inserting (20. (20.94) with s — s — j : > x = J + y u ( 1)J ~ f (*+&!(*+*?-pi i_ Using the reflection formula (for q = 2n + 1. i). g — —q. Comparing > > > this result now with Eq.

14).j) (i + lg)[2i + i ( g . Thus ° 9 (i + \q) 2«(i+Ig-l)! p . Dingle and H.1 )/ 2 2 2i ' ^i\{\{q-l)]\ [2i + g . 5.J)} -2 i=i 2 -[2i+l(q-l)]!22il2 i![i(g-l)]! i (20. S.e.97d) Then the coefficient M<n of the eigenvalue expansion. .97a) we obtain (observe that as required this expression vanishes for j > i) Pi(q. Miiller [74]. A special case of the formula can be found in I. Gradshteyn and I. formula 0.e.20. W. J. B. note the misprint there: (r — s) ought to be (r — s)\. p.97c) We assume that in a leading approximation we can cancel the factors (i + \q). (20.4 Cosine Potential: Another Method of Calculation 499 We observe that Po(q. is M2l = 2j2j[Pi(<l. l{Q. (20.l ] ! (20.l) (i+y) 2*%(q-l)}\ [2i + §(g-l)]l i\[i+\q-l]\ ' Inserting this expression into Eq. i.3)]!2 2 '+^.{j + \<l).0) = 1 as desired (although not really enforcible for large values of i). M.l ] ! {[. (20. E* J'=I i+M 2 [2i + 9 .98) the duplication formula 22z-l / y P*-l>'--7J-<*-l)'(»-j»and obtain M-i% + \{q-^)W + \{q.l ) ] ! 2 M [i + \q)\ i\\l{q-l)]\(J + \q) [j + lq]\[i-j]\- (20. formula (30).157(4).99) In the next step we use for both factorials [2i + ^(q — 1)]! in Eq.98) We evaluate the sum by approximating this to a formula given in the literature ** r \ f R\ __{r + R-l)\ J (r-l)\(R-l)\' E j=0 i. Ryzhik [122]. of Eq.100) **R.-+i?_l]!}2- (20.Then [2i + ±(q-l)}\2* Pi{qJ) [i + \q]\ (20.

86).5 20.o i • .. i. i. We refer back to Eq. The imaginary part we obtained is interpreted as effectively the discontinuity of the eigenvalue across its cut from h2 = 0 to infinity. Eq.1. [19]..101) This formula allows us to cancel all factorials [i.500 CHAPTER 20.l l ! with the large i behaviour in agreement with the result (20.105 ) the latter because we see from the eigenvalue expansion (18. (18.17).5. n = 0. [^ + i ( g . i. Eq.. replacing 2i by i: 2 ^2 8 ^ % + *"*}' (20.3 ) ] ! _ 1 Hence M or..2.e.86) and set (cf. since \{q — l) + \{q — 3) = \q..-' 1 i • (20.1 Anharmonic Oscillators The inverted double well In Chapter 18 we obtained the imaginary part of the eigenenergy E of the Schrodinger equation for the inverted double well potential. i. the result of Bender and Wu [18]. i. 20.e.i ) ] ! [ ^ + | ( g .e. i\ M i large 1 .102) Thus the result is M m 1 2«Fi + g .34) that £ is an expansion in ascending powers of h 6 /c 2 .e. (18. (18. .l. Large Order Behaviour of Perturbation Expansions In the following step we use the approximate formula (20..e.]\.4)) £ = ^ = ^+ w and y= ^ > ( 20 .Am i .81) for q — qo = 2 n + l . With this we can now derive the behaviour of the late terms in the largeh2 perturbation expansion of the eigenvalue of the Schrodinger equation for the proper quartic anharmonic oscillator.

20. (20. h2 _ h222K+llHr+K+ll2{-l)r+1 r+ K . one obtains the result Mr 2io+hr+1+f (-l)r[r + f ] ! 7T3/2[1((?0-1)]! (20. in terms of K where go = 2-ftT + 1. in fact as one with Borel summability in view of the factor ( .77) can be verified. i.e. and hence inserting here the imaginary part from Eq.l ) r .86). (with E = h2£/2) my') = 290+1 (y')9o/2 e -2/'/3 (27r)V2[i(go_i)]!' and integrating with the help of the definition of the factorial or gamma function. 20.4) of Bender and Wu [19].110) This is the result in agreement with the large order formula (4.108) Proceeding in this way. We leave this to the following Example. Once this has been obtained one can insert the imaginary part into the Borel transform and obtain the behaviour of the late terms in the eigenvalue expansion.106) With £= E r=0 Mr (20.107) we obtain (-i)r r 71" JO yirZ£{y')dy'.e.5 Anharmonic Oscillator The Borel sum can now be formally written 501 (20.5.(20.2 The double well It is clear that one can now proceed and apply a non-selfadjoint boundary condition also in the case of the double well potential and thus obtain the imaginary part of the eigenvalue. The result establishes the series unequivocally as an asymptotic series. and then the relation (20. (18. the discontinuity across its cut. This can then be compared with the level splitting we obtained for this potential. .109) or. i.

t p . E.Q . Miiller-Kirsten and A. Harrell. " P . S. Grecchi. E. Silverstone [52]. Nakai. J. G. R. . of International Workshop on Perturbation Theory at Large Order (Florida. J. M. Gram. Paldus and H. . Achuthan. M.77) was conjectured without explicit derivation and only for the ground state by Bogomol'nyi and Fateyev [26].2: Late term behaviour of double well eigenvalue expansion Derive for the case of the double well potential — by application of a nonselfadjoint boundary condition (as in the case of the cosine potential) — the imaginary part of its eigenvalue and hence the large order behaviour of the asymptotic expansion of the double well eigenvalue. *Some of the literature in this field following the work of C. J. S.77) has been obtained in the Schrodinger theory of the molecular hydrogen ion H^ . Large Order Behaviour of Perturbation Expansions Example 20. R. it can be used to obtain the other. Achuthan. Liang and H. in t h a t when the quantity on one side is known. H.77) and its possible generality has been referred to by Brezin and Zinn-Justin [36]. W. Wiedemann [4]. Silverstone [62]. J. J. M. Harrell.* A more recent status evaluation is provided by the book of LeGuillou and Zinn-Justin [161]. J . J. Propin. Cizek. Damburg. t The relation (20. J. V. Bender and T. Graffi. Wiedemann [4]. and is there shown to have a deep meaning (cancellation of the imaginary part of the Borel sum by explicitly imaginary terms in the perturbation expansion).6 General Remarks The study of the large order behaviour of perturbation theory has become an individual direction of research. Cizek. J. It was later explicitly established for the level splittings and imaginary parts of arbitrary states in the cases of the cosine potential and the double well potential.502 CHAPTER 20. [19] can be traced back from articles in the Proc. (357) to (371). R. K. Damburg. W. Harris. J. Miiller-Kirsten [164]. Its usefulness has also been demonstrated there. from p a t h integrals with instanton methods) and the large order behaviour of the perturbation expansions which establishes the latter clearly as asymptotic expansions. Eqs. An approach to arrive at the relation via instanton considerations has been examined by Zinn-Justin [292]. H. Miiller-Kirsten and A. Propin and H. T.^ 20. Solution: For the solution we refer to the literature. J. Grecchi. J. *R. A relation similar t o (20. S. J. 1982) [139]. K. The relation (20. V. + The above investigation of the large order behaviour demonstrates explicitly the intimate connection between the exponentially small nonperturbative effects derivable from the perturbation expansions with boundary conditions (or. W. Paldus. as in later chapters. Wu [18].

Chapter 21 The P a t h Integral Formalism 21. P. Thus we are in particular interested in rederiving the level splitting formulas obtained in earlier chapters with this method. S. however. P. in general. The path integral is. However. A further wellknown reference is the book of L. since in such cases the probability of this system choosing a path far away may be considered to be small. This formalism is particularly useful for evaluating scattering or transition amplitudes. R. The formalism is not so useful for bound-state problems (in this case the initial wave function is. Schulman [245]. We illustrate the method here by rederiving the Rutherford scattering formula with this method. Kleinert [151]. Hibbs [95]. Felsager [91].* This formalism deals with an ensemble of paths {x(t)} rather than with wave functions and constitutes an alternative to canonical quantization in quantizing a theory.1 Introductory Remarks In this chapter we introduce the path integral formalism developed by Feynman. For the application of path integrals to the hydrogen atom (i. However. Chapters 2 and 5. which explains also difficulties. *R.e. ^ Our interest here focusses on quantum mechanics. we digress later a little and consider the path integral also in simple contexts of scalar field theories which can be treated like models in quantum mechanics. can be found in the book of B. 'See in particular H. our main interest in path integrals and their uses will focus thereafter on their evaluation about solutions of classical equations. 503 . unknown). P. Probably the most readable introduction. A brief introduction is also contained in R. A standard reference is the book of R. since this illustrates the most important and most frequent use of path integrals in a wide spectrum of applications ranging from field theory to condensed matter physics. Feynman [94]. particularly useful in cases where the behaviour of a system is to be investigated close to its classical path. Coulomb problem) we refer to existing literature. Feynman and A. Feynman [93].

2 P a t h Integrals and Green's Functions The one-dimensional Schrodinger equation for a particle of mass mo moving in a potential V(x) is (with mostly h = c = 1) i d2 + V(x) tff(x.e—»0 Sn.x.x fc _i) 2 2e2 V{xk) Thus: Given xk(tk). we wish to know* ip(xf. =J dx(xf. we can compute the action S1.tf.ti)ip(x.x).tf) = / dxK(xf. Obviously K(xf. We are interested to know the wave function tp at a final time tf > ti and position coordinate x = Xf.ti).t).1) Let the wave function i/> at time t = ti. The Feynman Path Integral Formalism 21.tf\x.e—»C lim V e -m0xk fc=l - V(xk) = lim n—>oo.tf\ip) . */ = U + (n + l)e. 2mo dx2 (21.504 CHAPTER 21.ti\tp).2) where K is a Green's function. (21. x. (21.ti){x. ti). be ip{x. i. (21. The Lagrangian L of a particle of mass mo moving in a potential V(x) in one-time and one-space dimensions is given by L = TV= -m0x2-V(x).3) We divide the time interval tf — ti into n + 1 equal infinitesimal elements e such that £() ti = = hi ti + e. the initial time.5) Sn = X / fc=l mo (zfe .4) so that the time-sliced action S becomes rtf «+i S = / 'ti n+1 dtL = d£L = n—»oo. ti) = 5(xf . U. (21.e. *In Dirac's notation we write this equation later (xf.

Here and in the following we need one or the other of the following integrals (cf. We now consider intervals in position coordinates and in particular the following integral over intermediate position coordinates Xk which corresponds to the sum of exp(iS/h) over the uncountable number of possible paths from Xi — XQ to Xf = xn+i.2 The Path Integral and Green's Functions 505 Above we considered time intervals of length e.1 A path in (x.tf). as will be shown to be possible below.1 (for convenience we write in the following in the argument of the exponential frequently S instead of S/h): oo poo po> / -oo dxTl JS/h dx\ I J—oo dx2 ••• J— c X f~ x n+l Fig.28)): f dr\ exp ia a) ' (21.6a) . Thus there is a countless number of possible paths that the system may choose in propagating from its initial position at (xi. 21. The multiple integral involves individual integrals like f dxk exp Xk)2 + {xk -Zfc-i) 2 } with Xk contained in the step up to this point and in the step away from it.ti) to its final position at (xf. The repetition of the quadratic factors in the argument of the exponential with indices differing by 1 suggests the construction of a recurrence relation.t). the Fresnel integral (13.21. 21. of which one is illustrated in Fig. The usefulness of the method therefore depends on the fact that some paths are more probable than others.

m0 . (21.2 + (x + x f c + 1 . for example.-ir 1/2 £dxeXp[i^{2(3 dxexp\i—<2lx+ 72 iV(e) exp Z7TT.s f c _i) + (xk+i .8) Note t h a t in spite of the n integrations.xk_i) dx exp i — { 2 a .xk+x oo —oo oo .6b) —c ia o drjr] e x p—V € (21. with f™ dze"™ 2 * 2 / 2 = s/2ir/w2. tf > U.z f c _i) 2 } } . 2 + 2a.tf. dx exp z—{a. the power n + 1 in the denominator has t o be seen in conjunction with the n + 1 terms in the sum of Sn in "Thus.506 CHAPTER 21.6c) In the evaluation of these integrals we replace a by a + ifi. one has j ^ dr1eif-a+i^ /£ = . dxneiSn/n.t ~Xk+l> + xk+i 2 ) +{xk . We then integrate and finally let /x go to zero.7) Thus we expect t h a t — in view of the multiplicative y/e here in front of the phase factor — in the limit e — 0 > oo />oo /•oo / dxi -oo / J—oo dx2.(x fc+ i . n > 0. ie f me\ 2a\a~) 1/2 (21..mQf Xk L ~2e^ ^ .xk+i) exp .5)): G(xf.Zfc-i) 2 } I +-{xk+i -x*. (21.Xi.* W i t h the first of these integrals we obtain (compare the result with the Green's function (7.ti) '•= f g m ) (N(€\]n+i dxL.37)) dxk exp d(xk . The Feynman ia Path Integral Formalism f J—c J dr) r\ exp -v = 0.x f c _ i ) 5 \m0 J (21. J—oo dxneiS/h — • 0.. One therefore removes the troublesome factor by defining (Sn being the sum in Eq..0 ^7 (x f c + i .

10) can be looked at as the product of a succession of free particle Green's functions (or propagators). (21. It is neither a variable of integration in T){x} —> Ylndxn. § See e. 11 For discussions see e. G(xf.21.ti) For tf > ti we now write* G(xf. nor the limit of integration of some variable Xi. The additional factor is needed later to combine with the number n contained in the extra factor y/n + 1 (in Eq. The consideration of the denumerable number of possible paths connecting the particle's initial position with its final position may be handled in a number of different ways. and will therefore not be attempted here.5). the Green's function K of Eq. In the expression rx" / ?){x}eiI{-^ Jx the quantity x" is only a symbol to indicate the endpoint of the path.9) T>{x{t)}eiS/h = ^ww^Idxl---IdXnelSn/h-(2L10) This is the formula originally given by Feynman. we arrive at a path integral without the measure factor N(e) above.g. (21. i. 183. (21.5).e. it may be noted that with a different philosophy concerning the summation over paths. and so on.e. See the square root in front of the exponential in Eq. . however prefixed with a new factor.U) = / JXi=x(ti) = K(xf. We shall see later (with Eq. B. Felsager [91]. (21. i.tf. (21.7) for n — 1) to give a time interval. 21.tf.2 The Path Integral and Green's Functions 507 Eq.§ The factor [iV(e)]_(n+1) associated with the multiple integration is described as its measure. Felsager [91]. p.26)) that — ignoring the potential — the multiple integral (21.Xi.2). Thus one first integrates over x\ from — oo to oo and obtains an exponential involving (x2 — xo)2. p. as in our discretization in Eq.ti) for tf > U.g.24).' These factors N(e) were introduced to cancel corresponding factors arising from the Fresnel (or Gaussian) integration. which all vary from —oo to oo. However. (21. A mathematically rigorous definition of the path integral is difficult and beyond the scope of our present aims. 183. Different constructions of these families of paths usually lead to different measures. summation over all possible paths.1. in fact. These integrations are indicated by the horizontal arrows in Fig.tf.Xi. (21.t In the following we demonstrate that the function constructed in this way is.Xi. B. Then one integrates over X2 from —oo to oo and obtains an exponential involving (x^ — XQ)2.

It is for these reasons that — unless required — the measure is frequently ignored. so that the piecewise linear paths fill the entire space of paths.Xi. pp.ti) = 1 j^p: f°° cte n +iexp } • fm0.tf + e.11) For reasons of transparency in the next few steps we replace xn+2 by x and suppress temporarily tf.xi.xi.tf. (23. the other with that of the action of the free particle SQ. however. we have See. Thus consider an arbitrary continuous path connecting the endpoints and use the time divison as above but now connect the intermediate points with straight lines.Xi.ti).1 and allowing n to go to infinity. I We employ this method later in Example 23. The Feynman Path Integral Formalism The factors N(e) can be avoided by instead summing over piecewise linear paths. Then we have on the right the factor G{x+rj).Then we can construct a recurrence relation as follows. we obtain the same as before except that there is no factor N(e). 187 . we have G(xn+2. One way to avoid this problem with the integration measure is to consider the ratio of two path integrals.188. .3 in the evaluation of I a specific path integral (see Eqs. Each of these linear pieces is completely characterized by its endpoints. 21.72) to (23.74)). the discussion in B.508 CHAPTER 21. (21. Expanding this in a Taylor series around x.ti.U) imo 2 G(xn+i W)Idr.tf + e. one with the full action S. Felsager [91].exp 2e n -ieV(xn+2) G(xn+2 + V. Consider a time t = tf + e for position Xf = xn+2. and one is calculating the Feynman amplitude or kernel relative to that of the free theory. 16 x2 -Sn+l) \ 7T2~(Xn+2 -V(xn+2) Setting now xn+\ — xn+2 = n. We next determine the equation satisfied by G in order to verify that this is indeed the Green's function for tf > t{. We have G(xn+2. Performing the sum now in the sense of summing and hence integrating over the vertex coordinates in Fig. Then the measure drops out.

15) We see that G(x. we obtain: G(x. ti) satisfies the time-dependent Schrodinger equation.tf. U) + .U -ieV(x) 2i7re\ 1 / 2 m0 ie d2 G(x.6c).t Thus + 0(e2 (21.tf + e. tf. 2mo dx2 (21.Xi. .6a).tf.ti) for tf > ti (21.— p G ( j .ti) +V(x)G(x.ie 1 d2 2nV0dx^G{X.ieV(x) + ri< —G(x. Retaining only terms up to those linear in e.tf.ti) = 1 d2 + V(x) G(x.Xi.U + 0(e ).tf)= dxG(xf.16) .tf.2 The Path Integral and Green's Functions so that G(x.Xi.ti) = G(x.ti) +\if(^G{x.xi.tf.21. U) = e-^-G + 0(e 2 ) dtf 1 d2G(x.U). Xi. U) + Q(e2 = (l-ieV(x) + 0(e2)) G{x. tf + e. It follows that ip(xf. tf.ti) = j^rr dnexp I -^-rj1 .xi.ti) N(e) or G(x.ti) .6b) and (21.ti) 2 —l€ — ^2 + V(x)G(x.Xi.Xi.xi.x. tf.xi.14) = G(x.xi. tf.tf + e.U) +••• \ 509 G(x.ti)ip(x.tf.G(x.(21.Xi.tf.Xi. U) .tf.xi.Xi.Xi.xi.ti)\ .tf + e. (21.tf.tf''Xi.tf.tf.13) We integrate with the help of the integrals (21. d and this means d i—G(x.xi.

It follows therefore that {{W ~V+2m~0~^) K{Xh t/.tf.1)).U)G0. Hence our original function K is the same as G.ti) = = 9{tf-ti)—G + G5(tf-ti) .Xi.3.16). now called So.U.20) . (21.x. or K(xf. as this is also called.510 CHAPTER 21. K0 = 9(tf .*<)' ( 2 L 1 8 ) 0(tf ~ li)-QfG + 5(Xf ~ X)5^f since G(xf. We have therefore established that the Feynman integral (21.x. is the Fourier transform of the nonrelativistic propagator.tf.x. The Feynman Path Integral Formalism also satisfies the time-dependent Schrodinger equation. Thus we consider here the zeroth order of the latter case and begin with the configuration space representation of this free particle Green's function.17) Differentiating K with respect to tf we obtain —K(xf. as may be verified by inserting this into Eq. Eq. X U) = i6{Xf ' ~ X)6{tf " U)' (2L19) Thus K is the Green's function for the problem of the Schrodinger equation. (21.15) is a wave function (cf. (21.1 Configuration space representation In the case of V = 0 we have from the previous section for the action. In a problem where V is small it is convenient to calculate K with a perturbation method.tf. Eq.10) is indeed to be identified with the timedependent Green's function.x.U)6(tf . The result.U). (21. The function G which satisfies the time-dependent Schrodinger equation (21. 21.ti) = 5(xj — x). and the Green's function KQ\ S0= Here (cf. 21. denoted by KQ.8)) Go(xf.tf. (21.U) = G(xf.3 The Green's Function for Potential V=0 We now calculate the Green's function K for potential V = 0.U) = / T){x} eiS\ [fdt \mx2.

Xi.tf.Xi. We obtained the same expression previously — see Eq.7) for here n = 1 (n+l)/2 lim ( —^~ K2meJ m /n A = 0 V 2e (21.oo /*oo / -oo cfcci / J—oo cfcc2 • • • / J—oo + (x.23) We now insert our expression for In into Eq. 2t Go(xf.z / c .ti) = = G0(x.Xi.24) ( s / .x 0 )" n + 1 (in agreement with Eq.Xi..21.Xi. (21.ti) = V = 0 511 m0 lim n—>oo \ 2irie oo (n+l)/2 /-oo oo /*o / dx2. -oo dxi 71+1 / J—c /_ dxn exp oo irriQ -oo ~27 ^](a.t) is the configuration space representation of the free particle (V — 0) nonrelativistic Green's function.tf.ti) lim n^oo lim n->oo m0\(?1+1)/V27rie\"/2 1 z —— exp 2e(n + l )-(a.22) • \ n/2 -.3 The Green's Function for Potential and hence Go{xf. But t = tf — ti = (n + l)e.t) = ( m0 \2mt 1/2 irriQ and writing x = Xf — Xi.tf. f c .. exp 2 (21.23) for Go and obtain Go(xf. (21.ti)= iX (o. (21.Xi) exp \2-Kie{n + 1) 2e(n + 1) Note the square root factor in front which originates as discussed after Eq. n + i .8). In Example 21. exp where Xj = xo and Xf = xn+\ and there k = 1).t) K0(x./ .tf. TJ+1 dxn exp[i\{(xi 2 — XQ) + • • ^n) }] (21.Z i ) \2-KieJ \m0 J y/n + 1 1/2 m0 «mo (21.1 we show t h a t oo /.25) T h e quantity Ko(x.Xi.ti) Ko(xf.37) — by explicit solution of the differential equation which the Green's function satisfies. we have Go(xf.tf.ti)9(t).i ) 2 . .tf. (7. (21.21) fe=l and t = (n + l)e = tf — U and h = 1. Then Go(xf.

x0.54). (7.x „ + i ) 2 = (x„ + 2 . we have instead of 5*0 and KQ expressions S and K. (21.x ^ 2 + y2 .. n+l = lim / n ^{a^jexp fc=i ^°° Jxi = lim / • • • / dx\dxi • • • dxn I —1 n-Kx J J \2mneJ / \ 1/2 .6a). (21.10) in a product form.^(-Xn+1~X0^2+iX(-Xn+2~Xn+1^'i.t0) (21.f_^y e.tn)••• K0(xi.x „ + i ) 2 = ( x n + 2 -Xi + Xi .e. ••• 2 / \ 1/2 \2iriheJ dxldx2•••dxnK0(xn+l. i.an) Then n I 1/2 (n + 1)A" "I 1/2 fX r i-oo ^n + dxn+1eiX{^+v2-y2Hx"+2--x"+1 l ) 2 } J —OO (n + l)A n «ix„+1ea(Sl»2-2»lI«^-I. the latter being a Green's function with interaction like that for the harmonic oscillator.ti. The Feynman Path Integral Formalism We can now see the group property of the Green's function or propagator by observing that with the result (21.512 CHAPTER 21. For n = 1 (see below) h This result can be verified like Eq. or else by induction.25) we can rewrite the Feynman path integral or kernel (21.)+(1"+2-.22) -oo Indxn+leiX<-x"+*-x»+^ 1/2 f°° dxn+ie^:. (21.)!}| J — OO .2y(xn+2 . Solution: The result can be obtained by cumbersome integration.22). e 2he ^{Xn+l-Xn)2 = \2iriheJ lim ••• n^ooJ J x(^*-) e^(---) . as / JXi f r D{x{t)}eiS°lh ._*R±HY+l_{xo_X2f J —! r dXieiM(xl-xo)2 + (x2~xi)2 ei^(x2-x0) and using Eq. Eq.xi)2 = 21 xi ^2-o(„. Example 21.1: Evaluation of a path integral Verify Eq. In our proof of induction we assume the result is correct for n.. ^ (xi .7) by direct integration with ^ . If the potential V(x) of Eq.26) with n integrations and n + l factors KQ. (21.xo)2 + (x2 .5) is carried along.tn+l•xn. Then oo 2 / (21. (n + l)A"_ We set y := x n +i — Xi = x n +i — XQ and (x n + 2 .

2 M o m e n t u m space represenation Our next task is to derive the momentum space representation of the free particle Green's function KQ. (21. A(n + 2) 21.21. (21.Q .25): Ko{xJ) = = (S) e{t)(^X'2 y J .n+2-a. n + 2 .27) we have (with a = -t/2m0) . a dp exp [ipx + iap2] = exp — i(21. i. (n+l)A n n j(a.30a) —6(t) / dp exp ipx — to J-oo 1 i-—t 2m. e > 0 . n+l T h u s as r e q u i r e d 1/2 Jn+l ^ e J — oo 1/2 . (21.2.i) H —-3/ . We pass over to momentum space with the help of the following two integrals.i j ) H n +— z 1 .max exp i 2t 6(t) 1/2 \2mtJ f°° 2-KimQ J dp exp J—c ipx P -t 2mo (21. ) H — (xn+2-Xi) n+2 n+l n+2 -y2 . it -. rc+2 rl 1 / s2 513 so that n dz = dy = d i n + i . The second integral is an integral representation of the step function. This quantity is also described as the nonrelativistic free particle propagator.2y(xn+2 .a .3 The Green's Function for Potential V = 0 Next we set z : re+1 (x -)-2 — Xj). (21.x2rriQ 1/2 toij_.2 y ( a .28) exp (— V 2irimo J —( dp exp ipx — i p* 2mo (21.3. The first.Xi) + ( x n + 2 Xi)2.i) . and + 2 2 l n + 2 ( x n + 2 . iir(n + 1) 1/2 jn+1 n+l L(n + 2)A'n + l I 1/2 4 (n + l ) A . 0(t) Ti: i r Art alr- This integral can be verified by applying Cauchy's residue theorem in the plane of complex r and using one or the other of the contours shown in Fig.e. 21.6a). Using Eq.27) '4a in can be verified by completing the square in the argument of the exponential in the integrand to ia(p + x/2a)2 and using Eq.29) Hence with KQ from Eq.. s 2 n + 2 2 „ / ^ n + 1 / \2 -(x„+2-a.

t) i(2- h?L L E iv 00 dr exnlipx T '2mo t + itr] — te ~U2^LdPJ ^-E+£. This is the type of expansion most people describe as perturbation theory.514 CHAPTER 21. an expansion in rising powers of the coupling constant (contained in the potential).31) is the nonrelativistic propagator which describes the propagation of the free particle (V — 0) wave function.Q (21.28) this becomes K0(x. Thus in first . The generalization to three space dimensions (required below in Eq. 21.51)) is fairly self-evident. ( 3 b) ° The expression E+ V — it 2m. The Feynman Path Integral Formalism Im t \t>0 Rex Fig. i. The first order correction contains one factor of the potential V. 21. (21. the second order two factors of V and so on.2 Contours of integration.4 Including V in First Order Perturbation We now proceed to calculate the first order correction in a perturbation expansion.e. Replacing 9(t) by its integral representation (21.-*' where 2mo — r.

_i By regrouping factors this becomes Gl n+l lim ^ e n—>oo V{xhle) dxn / i -c-i) n+l dx z+i / * * / d / _ i . .xi.t)\.tk) J_( x exp fc=l = ie(—2(xk-xk- 1>{x(t)}exp iSo-i dtV(x.+i 6XP L — {xk -Xk-iY {-iV(xhti)} 1/2 X Em fc=l — 0 (xk-Xk-i) W i t h Eq. .i.^){-iV(a.21) this can b e written as Coo oo /•oo />oo / -oo dt / J —oo ^Go(cc/.21.tf. .Xi.tf. dxn exp i ^ —.t)] n+l .U)..5) a n d (21.U) = — lim fv{x(t)}el —— js (n+l)/2 I dxi dx2.^ J_00 V(xk.ti. .i V e / D { a .ti)= n+l ri_> f V{x(t)}eiSo „ iSo f ' dt[-iV(x.xfc_ -OO J— OO J—OO L J. (21. (21..20)) and with elS = exp i S i o / hi dtV(x.t/.^.Xi. dxi ( m0 V V27rie/ m0 \^He \ exp fe=.33) xG0(xi.t) the first order correction is (G\ ~ G — Go) Gi(xf.i..t) JSo dtV(x.4 Including V to First Order: Perturbation Expansion h=l.)} (21. (21. } e °° f l f •/ oo /"oo F ( a . x (n+l)/2 \ lim .32) Since — we recall this for convenience G(xf. l-i f f 515 order (cf. dxr.(xfc . Eqs. ^ = k ) = lim -(S^ A H*-\ n-KX) *—J \2irieJ r "+1 />oo X / dsi / dx2. \2meJ n+l J..

516 CHAPTER 21.tf.37) Jti The last two contributions do not describe the time evolution from ti to tf.x2.t2)V(x2.t)V(x.ti) t2)K0(x2.x2.t) ! Jti r*f dt'V(x.I dtV(x. . They must therefore be deleted. It can now be surmized that the next order contribution to KQ + K\ is K2 = (~i)2 / dti / dt2 I dx\ I ti.t)K0(x. t2.t2. (21.xi.35) We also observe here that the problem of time-ordering arises only in perturbation theory.e.ti).xi.t.ti)K0(x.t) =l-i dtV(x.U) xV(xi. / Vdt+ JIfVdt)( fti = Jti Jti ftf ) \Jti pti Jti I' Vdt' + f f Vdt' Jti rtf Vdt' J t\ j Hi Vdt rt\ Vdt' Jti Vdt Jt\ Vdt'+ Vdt+ Jti + \ S Vdt I Jti f Vdt'. i.ti). = -i dt2 / dx2Ki(xf. The first two terms lead to identical contributions and therefore cancel the factor 1/2! in the expansion.xi.36) {-if •j.i / dtV(x. (21.e. in writing down the expansion exp i dtV(x. Xi. i. The factors of "V" originate from the series expansion of the exponential exp ftf .tf.t2)V(x2. Jti Consider the term of second order. t) Ju Thus the term containing n times V contains a factor 1/n!.x. This is cancelled by n! different time orderings. t') + • • • .t2) xKo(x2. (21.34) We observe that we can extend the ^-integration to — oo since KQ contains the step function 9{t).tf. OO /"OO / -oo dt / J—oo dxKo(xf.t) (21. The Feynman Path Integral Formalism and hence correspondingly for the first order contribution Ky. dx2Ko(xf.

tf. For various reasons — such as illustration.ti) = K0(xf. we are concerned with quantum mechanics. of course. The diagrams represent mathematical quantities and hence are designed on the basis of rules.Xi.tf.4 Including V to First Order: Perturbation Expansion 517 The full perturbation expansion of the Green's function K is now seen to be given by K(xf.tf. These may be formulated in configuration space as below. Such diagrams in the context of field theory are called Feynman diagrams. x"^ X x"2.ti) (21. called Feynman rules. This can also be written in the form of an integral equation which when iterated yields this expansion: K(xj. Here. tf. transparency or brevity — it is useful to introduce a diagramatic representation of individual perturbation terms.xi.tf.t)Ko(x.t)V(x.t. (21. E) which is given by (in one-time plus one-space dimensions) V(x. x r l i -'V(x. and hence the diagrams representing perturbation terms here are Feynman diagrams corresponding to those in field theory.t.t)K0(x.39).40) The rules for the present case are given in Fig. 21.t)V(x.t2 K o (x 2'l2 .t) ) K K K0 -'V X K o Fig.tf.x.Xi.E).38) .ti) -i dt = K0(xf.39) dxK(xf.3 Feynman rules and representation of the Green's function. The Fourier transform of V(x. 21.i ] P I dt J dxKn(xf.3 along with the diagramatic representation of Eq. (21.t) is V(p. U) (21.21.ti).t)= f dpdEeipx-iEtV(p.x.Xi.ti).Xi.Xi. or in momentum space as mostly in field theory. .

Xi. * k 1 • . (2) The incident flux (number of particles passing through unit area in unit time) is equal to the incident velocity times the density of incident particles and this is = n—-7 particles per cm second. and dn = dnxdnydnz. The transition amplitude A describing this process is defined by the Green's function sandwiched between the inand out-state wave functions. for the allowed values of k: kx = 2imx/L.tf. . to one particle in all of space. i. (21.ipf have to be properly normalized.= l. ^(x. There the wave functions ipi.42) VV dn dp Jv We are interested in the total cross section defined by transition probability per second incident flux / Here (1) dn/dp = density of (plane wave) states in the box (of volume V — L 3 ) per unit three-momentum interval** = V/(2irh)3. dn = (L/27r) 3 dk. The plane wave function normalized to one over a box of volume V = L3 is. We recall for convenience from **Note that from the wave function we obtain.tj)K{xf. 2 The expression |A| 2 /T can be related to first-order time-dependent perturbation theory in quantum mechanics. mo V since ip is normalized to one particle in volume V. (21.518 CHAPTER 21.k = 2wn/L. with p = Kk.41) We have to pass over to three space dimensions.5 Rederivation of the Rutherford Formula As an application of the path integral method we now consider scattering of a particle off a Coulomb potential.e. (3) The transition probability per second is = \A\2 . in one-space plus one-time dimensions by A= dxf dxiil)*f{xf. T = duration of time for which potential is switched on. and this means the transition of a nonrelativistic particle from an initial state 'i' to a final state ' / ' with wave functions ipi (or ipin) and ipf (or if>out).ti)il)i(xi.t) = .e. The Feynman Path Integral Formalism 21. here with E — k 2 /2mo.ti). i.7 = e i k ' x _ i : l i with f i/}*i/jdx.

-. The transition probability per unit time is . «= — .44) = u(Ek.47) . 4-7T 1 3 7 (21. (21. Then H'ki can be taken out of the integrand in Eq. u = ^-\H'kfp(k). H = H0 + \H'.21. in which T is a large but finite time. # = ^an(i)VnM) n = a^(t) + \a^(t) + .r)e-iE^\ H'ki(t) = J dv u*k(r)H' Ui(r) Thus the amplitude ak is effectively the spacetime Fourier transform of the interaction or potential XH' contained in the Hamiltonian.5 Rederivation of the Rutherford Formula 519 Chapter 10 the Schrodinger equation with the subdivision of its Hamiltonian into an unperturbed part and a perturbation part.117) and Example 10.e. (10. P(*) = j r (21-45) (cf..T)\2 f • (2L46) We take the following expression for the time-regularized Coulomb potential. and with these the following perturbation ansatz: ih^* with an(t) = HV. Eq. Then for stationary states M^t) and itm^ = f HUt'V^t'dt'. (21. i.= t7Ei4 1} wi 2 k It is frequently assumed that the only time dependence of H'(t) is that it is switsched on at t = 0 and switched off at time t > 0. V(x.44) and f_oo can be replaced by fQ.k. Hence (h = 1) atot = f J j2^~k dp V Vm0\A(p.— .t) = -e~^lT\ r where r = Ixl. Proceeding in this manner one obtains Fermi's "golden rule".4).

tf)ipi(xf. (21. dt dxip*f(xf.t.44) (here in the scattering problem with initial and final energies equal). we obtain A1 = clS(ti -tf) / dxf(-i)V(xf.x.39) OO /"00 / dt / -OO J —OO dxK0(xf.48) where from Eq.t)K0(x.). The lowest order approximation of A. This now has the form of the amplitude (21. by threedimensional dxf and so on. (21. (21.520 CHAPTER 21.i x (-i)V(x. (21. i. We can relate this amplitude to the amplitude ark of Eq.^ / .50) Here x — xj — Xi and t = tf — ti.tf-. is now in three space dimensions A = Ai= / dxf / d^ty*f{'x.49) together with the following wave functions (of asymptotically free particles) 1>i(*i. x. ti) = co5(t — U)5(x — Xj) etc. so that 0(t) can be dropped) K0(xf.tf.t)V(x. t.>/>. t)K0(x.51) Inserting this into Eq.30a) K0(xf. The Feynman Path Integral Formalism In calculating the Rutherford formula with the help of our previous formulas.* / .Xi.t) = ^ J*ie*-<*'-*>-<<*'-*>. In our previous one-dimensional considerations we had with Eq.yLi.tf-x. The three-dimensional generalization is (taking tf » T and ti <C —T.e.f. (21. also called Born approximation.. Xj.f.x.ti). (21. T a large but finite time. t.44) by contracting the propagators to instantaneous point interactions.U) = ~ e ^ l ^ \ ^(*/>*/) = _ L e * P .tf)Ki(x.49) i.tf)ip*(xf.ti). (21. £.tf.tf).t) (21..tf. (x*.tf)K0(xf. we have to replace the one-dimensional quantities dxf etc. (21.ti)'ipi(xi. Ai = dxf / dx.52) . Thus with the ansatz Ko(x.xuti) = ^ J°° d p e i p x _ i ^ * .

f f dxi f dt f dx f dq / d q ' p V 2 (2TT)6 2 +iq' • (x — Xj) — i- 2mo (t — U) + ik • x 2mo . which is evaluated in Example 21. 2 „ x -v~ m o (21. Then we require the following integral.21. At2 dteiat~^ a2T2 f°° =e " J dre-^lT2 p2-k2 2mo = (~)l'2e-a2T'' / 1 6 .). Then integrating over q and q' we obtain .t)exp A(p.56) (21. (21.59) . (21.T) = JdtJdx(-^\V(x.55) We evaluate this integral with the help of the following integral in which we complete the square in the argument of the exponential and set r = t .r) = = -y .58) In doing the integration over x we introduce a temporary cutoff r = 1/M by inserting the factor e~Mr in the integrand.57) we have A(p.5 Rederivation of the Rutherford Formula we have (for ingoing momentum k and outgoing momentum p A(p.k.k.54) Now inserting the expression for the potential V(x.53) The integrations with respect to Xf and Xj (including in each case a factor (27r)3) yield delta functions 5(q — p) and <5(k — q') respectively.p ) .k.~ .T) = x exp L z/ q/2 k2 521 f dx. this becomes A(p.2: piq-x / Y iukawa •" dx- e~Mr 47T q + M2' 2 (21. .p 2. k2 -ip-x + i-—t + ik-x — i 1 2mo 2mo (21.k. P2 .i ( k 2 .iaT2/8: L With 0 0 • .fdx-e^-rt* dx-e< p) x fdte1 ' (-4-) 2m0 T2 e "^^J2" 16 . (21. £).T) = -^JdXJdt%-*t2/T2exp i ( k .

\ 2?7lo / 1 (21. k2 .59).„.64) Inserting the result now into our expression for the total cross section. Using Eq. dE = PjV = ^ E ± . e (21.59) we can re-express A as I /7rT2\1/2 e (P 2 -k 2 )T 2 f A-jr ~) p * P . write down the probability per unit time which is ( 8 \1'2 _a2 16^ .61) We can now Hence in our case \A\2/T has to be replaced by \A\2/(nT2/8)1/2.a ( — ) ^ .65) But E=*-.63) We establish this result for T — 1/e in Example 21.3. . ( 21 . In our case this has not been done.t o t= y dp ( 2 v r ) 3 "V 3 I W: 2mo y m0V k 32^a2S(^) y ( k y 22 [[ ( k _ p ))2 + M 2 ] 2 ' k-P 2 (21.„ 2. (21.P ) 2 + M2]2 4 VvrT2.63) we have: Probability per unit time = 7-7777.522 CHAPTER 21.k.e. / / P 2 _ . 66) o . The Feynman Path Integral Formalism This means: The result of Fourier transforming the Yukawa potential ~ exp(—Mr)/r is the propagator (21. (21. otot contains the factor \A(p. e 8{k) = lim -7=*—^ (21. For this we require the following • representation of the delta function: —k2T2 rp —k2T2/8 S(k) = lim T . where M has the meaning of mass (in the case of the Yukawa potential the mass of the exchanged meson). k .60) Now.< ^ £ \M VTTT2. This expression assumes that the integration over time t is normalized to 1. i. Using the result (21. we obtain .51 — 2 /w2 y ~ —. y 2 [ ( k .62) We want to consider the limit T — oo. <rtot. integrating over the square of the ^-dependent factor in V we have /"°° dt(e-4*/"1*}2 = (^p) ' 2 instead of T. 2 with 2mo J 2ir a2 Ct 16TT2 -LU7I F [(k .T)|2/T.^ . { I i _ F T l } . mo m (21 .T2f 8 V/ 2 ." + M ] p) ' "•"• ^2 -. T ) ~ .

with k = y 2moE / we obtain the differential o?m\ 4fc sin 4 (0/2)' 4 cross section (21. i.3: A representation of the delta function Determine the Fourier transformation of e~ax of the delta function <J(fe) = lim — / dxe~ax a-*o 2-K J-oo for a > 0 and hence the following representation eikx = lim — .e.= e " f c a ^ o 2v^fa /4a = lim £^o —. p 2 = k 2 .^ Jo 4-7T q 1 M — iq 47T i q2 + M 2 Jo G— 9 Example 21. —Mr 4w = .70) . (21.e.p ) 2 = 2 k 2 ( l .^ . (21. setting M2 = 0. r\ poo Yukawa = = ^ Jo 2TT / r e~Mr y_i -dre~Mr d(cos 9)e^r sin qr = — G / cos e = 2TT / J0 —-e~Mr iqr { eiqr .j .21.68) Inserting (21. we obtain f mlV2 dQ v / 2 m ^ 3 2 7 T 3 a 2 atot r) 3 fcV 2 4fc 4 4sin 4 (0/2) ' -JW) da _ d& = i.j .^ . and inserting the expression thus obtained into the final result of Example 10.cos 9) = 4k 2 sin 2 | .69) This is the wellknown Rutherford formula. esfn (21.67) T h e delta function here implies energy conservation.2: Fourier transform of the Yukawa potential Show that [ j / j eiqx dx—-e . 2mo / dft.e .68) into Eq. /Yukawa = ) Solution: We choose the 2-axis along the vector q.59) by replacing there q 2 by ( k — p ) 2 .65) with the cutoff M -> 0. Then (x = r) foo „2jj..67) and (21. so t h a t (21. Example 21.* 9 ' d r e . (21.M £ M * = / m0kdn = m0^2m0E (k .4. giving the potential a coupling constant. This result may also be obtained from lyukawa of Eq.5 Rederivation so t h a t of the Rutherford Formula 523 M^f) .

(21.6 P a t h Integrals in Dirac's Notation It is instructive to devote a little more time on the Feynman formalism by rewriting it in terms of Dirac's bra and ket notation. a. a simpler method suffices. '<*> = . Since dxe" 1 1 1 = « / . (y|x) = 6(y .524 The function required is CHAPTER 21. We write the state vector of a particle of three-momentum p in the momentum space representation: |p).* . ' i a = lim e —0 p-k 2 /e2 e^TT 21.^0 a^C 2^/TTO. Direct integration /4a . The two representations are related to one another as we saw in Chapter 4 by the following expressions or Fourier integrals: |p> = | d x | x ) ( x | p ) ..71) we obtain a representation of the delta function: <5(fc) = lim — / 1 /"oo a ^ O 27T J _ 0 0 dxe-ax 2 eikx = lim —g(k) o ^ 0 2?r ^ O 2-K 1 i 1 = lim k 9 .x) (21. (21. In the present case.q).71) with respect to k yields the same as °° 2 .73) 5{p) = (W / dxe"P'x' 5{x) = J^rI dpePx ''- (2L74) The normalizations (q|p) = (27r)3<5(p . The Feynman Path Integral Formalism o° 2 / -oo dxe~ax eikx. Va we obtain g(k) = J-e~k V a = 0.70) and (21. -co -oo dx -co (eikx)e~ax dx J = — / ikelkxe~ax 2a 7 _ 0 0 dx = ia^ Thus g(fc) satisfies the first order differential equation g'{k) + (k/2a)g(k) yields g(k) = Ce~k2/4a with the constant C = g(0). however.75) . We consider again nonrelativistic quantum mechanics. We have |x> = I J ^ l p X p l x ) .71) Solution: In general one uses for the evaluation of such an integral the method of contour integration.72) -co From (21. In configuration space the state of the particle is written |x). 7 r°° d 9 xdxe~ax elkx = -— dx — (e~ax -oo 2a J-oo dx Partial integration of the right hand side implies oo /.oo J / )eikx. Differentiation of (21. (21.

81) Comments on signs in Eq. i.t) = +iH\x. Then the states {|x.If '/'(x. (with ft = 1) d/dt\ip) = —iH\tp). Hence for the infinitesimal transition (x'.t) = —iHip(x.t\ = -iH{x. x' — x infinitesimal. at initial time ti to its final position xy at time tf. i.i) = K(x'.21.t).78) P|p) = p|p) is to be interpreted to say: The quantity p is the eigenvalue of operator P with eigenvector |p).e.80) (21. pp.82) (21.t\..t'.p). so that \if>t=o)s = H>)H. ) H = (x|</>o)s. so thattt d — |x. i.82): The general Schrodinger equation has the state vector to the right. we must have d/dt{x. (21. d/dt\x. t' — t.83) (21.t> =iH\x.t)} constitute a "moving reference frame" in the sense that ip{x.^ l x ) .e.x.e. H.79) Feynman's principle is expressed in terms of the propagator or transition function (x'. The relation between Heisenberg picture states and Schrodinger picture states is \ipt)s = e~lHt\ip)H.The time development in the Schrodinger picture is given by \tpt)s = e~lHt\4>o)s. which describes the propagation of the particle from its initial position x.6 Path Integrals in Dirac's Notation therefore imply 525 (x|p) = e ipoc .t). See also L..t'|x.t) = e* H t |x).e.t). as had to be shown. (21. Then ( X | ^ .We define |x. The time development is given by the Hamilton operator H. We consider first an infinitesimal transition. Similarly we have a position operator X with X|x) = x|x). -M^ip(x.i'|x. i.76) These relations imply the completeness relations 1 = J dx|x><x|. Ryder [241].t).e.i) = e iH4 |x. i. . The equation 1 = J ^|p)(p|. 159 . |x.We now introduce the Dirac bra and ket formalism. 4) is to satisfy the Schrodinger equation.£) = ( x V ^ ' . t\ipo)s. (p|x) = e~ipx. (21. Thus (q|P|p) = p(q|p) = p(27r)3<5(q . (21.77) (21. (21.160. t) = (x.0) = e""|x).

2mo Since (21.y<&(p'ix')(x'ivix)<X|p) = f dxe. P|q) = J 7^3P|p)(p|q) = J dpp|p>*(p .V 2 < 5 ( x .85) The momentum operator P satisfies the relation P = /(2^3P|P)(P|' (2L86) so that it projects out the momentum q of a momentum state |q)..89) Using this and previous relations we obtain for the Hamiltonian (21.P V<P |e l ) | P > e P X <2184 = /(w/(^ ' ' *" '"'' *=£Uv«.i p '.88) Thus P has the operator representation P= /dx|x)-V(x|.q) = q|q>We also have by mapping onto configuration space P|p) = = P /*dx|x)(x|p) = P /"dx|x)e i p x = p /"dx|x)e i p x f dx\x)—eipx= /"dx|x)-V(x|p).e.x y(x)e i p .x). (21.91) . " - » We suppose that for a single particle of mass mo moving in a potential V(x) the Hamiltonian is (21.90) (P'IVIP) = y<&. (21-87) (21. (21.x') + V{x)5(x' .* .x = V(p . i.p').85) the amplitude or expectation value or position space representation (x'|#|x) = . The Feynman Path Integral Formalism We can rewrite this relation as <X t M ' ' = /(^/(l]3 < X ' I P ' > < P ' | e ""'"'' ) | P ) < P | X > e.526 CHAPTER 21.

ti to x j .tj+i\x.i'|x.t'f / w? eip''x'^(p ~p/) = / l r ip''x' / "x"e~JP'x"v{x")e" = = Hence + y(x') 2771Q d /dx"<5(x'-x")F(x"KP'x' y(x')e i p ' x '. EQ — Hi .p'). the signs are reversed when these operators are applied to the ket vector \x.t> d f dp' P' r_dp f dV rivf-*!ini\rmt-f) J (27T)3y (27T p e |p) e -ipx +i(t-t'){p'\H\p) + ---}e-ip-x .t') \^-(27rf5(p 2mo +V(p-p' where we used (21. for the amplitude of an infinitesimal transition. t).93) + -ipx P r ip-(x'-x) -iH(t'-t) (2TT) 3 (21.f|x.e.82)).t) (cf.92) We now return to the expression (21. (x'.tj).p) + i{t . (21. tf subdivided into infinitesimal transitions (xj+i.94) In the above we have chosen the signs such that p = —iV and H = i— ot (21.j. (21.p') • • • j W < 5 ( p ' .95) when applied to the bra vector (x. i. Thus with •"•/ — ^ n > *•/ — ^ni ^ 0 — Xj. We now consider the transition of the particle from Xj. We have <x'.21.6 Path Integrals in Dirac's Notation we have in momentum space the representation (p'|ff |p> = ^-(27r) 3 <5(p . 527 (21.p') + V(p . i|.92). But e'ipx + 0[{t .84).

zmo T h e n t h e integration over all p ^ can be performed by using Eq. x exp [ . .i/|xj. .tn respectively. (21.ti) rx/=x(i/) d x i dx2 .. then corresponds to the sum over all paths. x i .. .e.e. dx.x i ) H h pn • ( x n .xic) by fc — Xfe-j -H(pk. i.xn)(tn -in_i)]. . The resulting relation is m0 2irieJ 1/2 m0x| exp ze ) " / 2TT exp Pk 2 mo 1 (21.97) H hH(pn. T h e integration over the n—1 intermediate positions.i /Xj=x(*i) f J [ ^ 1 dp2 3 3 J (2TT)3 (2TT) ..528 CHAPTER 21.99) .94) and thus obtain a phase space expression.X n _i)] h) (21..a — —e/2mo and divide b o t h > sides by 2TT.ti|xo.. x i ) ( t i .27) in which we make the replacements x —> eifc. (Xf.. The Feynman Path Integral Formalism we obtain (it will be seen below why we do not introduce normalization factors here) (x/.x 0 ) + p 2 • (x 2 . .2. t n _ i ) t„-2>---<xi.t0>(21-96) We define a p a t h in configuration space by a succession of n + 1 points x o . .98) tk — *fc-l # ( p f c . d x n .tj) = / ••• / d x „ _ i d x n _ 2 . x f c ) = ^ .xk (21.tf\Xi.ti) = x 0=Xi n^n fc=1 ? <^Pi (2TT)3_ exp / ^fa . t n | x n _ i . d x i ( x n .i { f l " ( p i .tfe-i) fe=i . .tf\Xi.x2)(t2 We can rearrange the factors here in the following form c„=x/ n—1 (Xf. .+ V(x f c ). We now insert for each of the infinitesimal intermediate transitions the amplitude (21. d x i .ti. d x „ _ i .. x n which the particle passes at times tQ. = 1 X X<jPfc We now replace H(pfc..t 0 ) + H(p2.dpn-l dpn (2TT) 3 (2TT) 3 exp [i{pi • (xi . i.

-H(x.102) We see t h a t the factor (with h = 1) m0 2m(tk 3/2 3/2 [N(e)}-< = tk-i) \2irie) is precisely the normalization factor inserted earlier for the one-dimensional case in Eq. Tokarev and A. / {xf. to / m0 \ 3 / 2 las.100) over V{p}= L J ^ rwooll V (2-7T)3 1=1 ' lim T T T ^ *The reader interested in a rigorous treatment of representations of solutions of Schrodinger equations in terms of Hamiltonian Feynman path integrals may like to consult O. . (21.e. Smolyanov.] ex H ^I (.98) can be rewritten as 3n/2 2^") x / n dxfe exp * ^o** ~tfcV rn-l <i 2 m ° x fc ~ ^ (21.8) as N(e) so t h a t the integral does not vanish for e — 0.98).21.103) is another form of the path-integral for which the integration with the help of Eq. (21. i. (21. > Thus the phase space representation of the path-integral* (21.6 Path Integrals in Dirac's Notation 529 Forming products of expressions of this type we see t h a t this relation can be generalized to the three-dimensional case.e. m& ze ) J (2vr)3 exp it p fc • xfe IP* 2 mo (21.tfc-i) fe=i £(**-'*-*){ p*-**-^} —tk-i m0±2k (21.ti) = J D{x(t)}D{p(t)}exp|i [' dt\p-i.100) Hence with all tk — ifc-i = e we obtain /n dpi i _(2vr)3 exp •. Truman [254]. A.p)]\. (21. G. G.tf\xi.101) Then Eq..fc=i 3n/2 r n m0 exp 2_^i{tk 2m(tk . i.

530 CHAPTER 21. x. Note that we have n factors [iV(e)]3 but only n — 1 integrations dxk. The Feynman Path Integral Formalism yields automatically the correct normalization factor or metric in the configuration space path integral^ in agreement with Eq. (21. Consider the following expectation value with primed and doubled-primed states denoting some intermediate states between initial and final states denoted by indices % and / : (xV'l x(<) \x'. xfc).x) . (21. We check below that this is given by x = / ( ^ | p > i | ( p | - <2L107) See also C.106) In order to be able to deal with this expression we first require the momentum operator representation of the operator x(t).8).e. p .ti) where n-l D{x(£)}exp '. i. t to x. £ / | X J .8). (21. I hi dtL(x. £ J ) with no operator in between. Thus finally we have Feynman's principle: (Xf. pfc) = Pfc • xfc .iJ(x fc . .104) where for the particularly simple Lagrangian L under consideration here -m0±l .t'}. n-l (xf. t.x) (21.tf\xi.tf\xi.105) D{x} n ^ o o [7V(e)]3n Udx kk=\ In this notation the measure factor is contained in ©{x}.ti) = ^lim^ N(e)3n / * JJcbc fc exp i / fc=i * dtL(x. operator (21. pk) = L(xk. We know from classical mechanics that this relation is a Legendre transform which transforms from variables x.H(xfe. this agrees with the corresponding discrepancy in the one-dimensional case of Eq.V(xk) = m0±l . Clearly for a better understanding one also wants to explore the case of sandwiched operators in the path integral representation. Garrod [106]. We have now obtained the path integral representation of the matrix element ( X / .

.t'}. so that we obtain a c-number position coordinate which we can shifted across other c-numbers.t"\x5){x5\x'.109) .5 l^ t 'lp / > e .i p '• x .* * f dp = j (2^ |p) ^ d . (21. Thus <xV'|x|x'.107).*') = / I F / (2?FeiP"'x" / ^ 5x5 (p // K^> 5 )(.0 = J'-^ j'^e^''(p"\e-^''xe^'\p')e-^'. .77) and thus have 531 *!*') = *M = * 7 ^ I P > « . With the wave functions (21. ••• y (2^ x= f dp |p) ^(p|- d We now consider an expectation value in which we replace the operator x (which is sandwiched in between exponentials containing Hamilton operators) by the expression (21. (21.6 Path Integrals in Dirac's Notation For the verification we use Eq.t"\x\x'.v ' x ' and thus {x". .21.76) we can rewrite the factor (x4|p)^(p|x5) as (x 4 |p)x 5 (p|x 5 )./ ^ I < « .t'} = = fdx5x5(x"\e-im"\x5)(x5\eim'\x'} f dx5x5{x".108) (x 4 |p)^^-(p|x 5 )(x 5 |e^ t '|x 6 )(x 6 |p / )e. x fdxG | ^ P <p"|x3)<x3|e-^V) x (21.* * . (p|x). With contractions and integrations over delta functions the expression for the matrix element then becomes <xV|x|x'.

.£j to Xf.tf — tn and x$ = XQ. (21.0 = and hence /dx 5 x 5 [(x"|x 5 )(x 5 |x') .85) (xV|x|x'. Thus with x^ = xn. We now consider the transition of the expectation value of operator x from Xj.112) where x" is a c-number.tf subdivided into n transitions (xj+i.f ) i f e i p . £ n _i|x|x n _ 2 . ] . = x"6{x" .ti = to as before. (21.90) we can rewrite this expression as (x".t') dp .£ n |x|x n _i.x') .i t " ( x " | # | x 5 ) ( x V ) +. we obtain (here and in the following the limit n —> oo at the end being understood) (x/. tn-2) • • • (xi.• dx1 (x n . This is the expression corresponding to that of Eq.x ' ) (2^ i^+vM + D «p-(x"-x') (21.x') . we obtain with (21. . The Feynman Path Integral Formalism In the second last expression here we expand the exponentials and obtain (xV'|xK.t'){x"\H\x') = X + •••} (21. x / / e .t"\x\x'.t n _i) x ( x n _ i .tj+i\x\xj. t0). Inserting here the integral representation (21.i(t" .113) Inserting for each of the infinitesimal intermediate transitions the amplitude . (21.t') (xV'|x|x'.it / / x / (x"|ff|x / ) + it'x"{x"\H\x>) + • • • .£'> = x"[«5(x" .iit!' -1') 2mo + vtf) + <5(x"-x').i ( t " .i{t" . We now proceed from there along parallel lines.0 dp x„ / (27T): 1 .*/<x"|x5)(x5|tf|x'> + .tj) at equal time intervals e..94) for the case with no operator in between.74) of the delta function.£/|x|xj.532 CHAPTER 21. ti|x|x 0 .111) 1 . (21.110) Recalling Eq. ( x " .£j) = / ••• / dx n _idx n _2 .

U).x 0 ) + p 2 • (x 2 .3 " n d x f c .iy|x|xj.. dyi -l xi x 2 .t') . Xn _i X„ xi dx.tf\x.ti) 533 rxf=x(tf) = / • • • / dXidx.x 1 )(t 1 -Z & ) +H(p2.10) or (21.x n _ i ) ] .t"\x'.{^( P i. x (21. tj | xj1 x»..105).2 . ..x2)(t2 ~ h) + • • • + H(p ••• + p „ .y (27F (27F •" (2^(2^)3 exp[-..2 .7 Canonical Quantization from P a t h Integrals We saw above that in the path integral method the evolution of a system is expressed by a basic functional integral like (21. For simplicity we consider the case of one spatial dimension.115) where (there is no integration over x n ) n-l 0){x(t)}= l i m [ i V ( e ) ] .t') = (x" + 5x". the string of factors xi X2 . n—>oo -*•-»• 21.114) x exp [i{pi • (xi .t"\x'.114).xi) + f f dpi dp2 dp n _i dp n The momentum integrations can be carried out as before and we obtain finally (x/..112).89) is consistent with the conjugate expression dL (2L116 * . . tj \ xt. (21.x ..t"\x'. dxnn _i Xi X2 . we obtain* (xf.and hence may be put in front of the integrals to yield (xf.t') sider the variation S(x". tj) = xf (x/.7 Canonical Quantization from Path Integrals (21.t') = and con- 5x"Vx„{x".tj) = / JXi 2){x(i)}x(i) exp i I I Jti dtL{. (21.t"\x'.( x „ . We demonstrate first that the operator representation of P given by Eq.21.« • > We reconsider therefore the transition amplitude (x". x n _ i x n i JXi=x(ti) /Xj=x(tj) 7 . x n _ i x n reduces to the single factor x^.117) "•"One may observe here that if x = x^ = x„ in the integrand of Eq. Before we can consider canonical quantization in the context of path integrals we have to investigate in more detail the role played by momenta.\Xi. (21.. ..t"\x'.{x".t') (21.

V°{x}[i6I{x)/h]e* Jx' The following steps are familiar from classical mechanics: yV 51{x) S /1 dtL(x. (21. Then. For the purpose of enabling some formulations below. jt/ 37 ( -£T.l&cdi. dL r /•*" .t i_(dL_ x't' h\dx J t„ (21.119) The classical equation of motion follows from Hamilton's principle for which 5x = 0 at t — t'. t". (21. dx dt \ dx _ (21. The Feynman Path Integral Formalism with 8x" = 5x(t") ^ 0 and 5x' = 8x(t') = 0. dL d5x / oft—5x + / dt dx dt Jt' ox Jt t' and hence <9L —-oxdt + ox ox nt t.(x".tt). . we separate the metric factors from T){x} in Eq. dt\Ox w(x) = r dt d^_dL(dL dx dt \ dx °L 8x + ^ T!5 X 1 ox (21. (21. <9L' dt —dx + — 5 x ox ox f .534 CHAPTER 21.120) However: In our case here 8x ^ 0 for t — t" (although 5x — 0 at t = t').t') = - 6— / 1 V°{x}t .t"\x'.t"\x'.122a) = 17) 5x"Vx/. Hence w « = (t). " (t).118) — = (21 5x"(x".t'). Thus in this case of classical mechanics dL_ d fdV. and we demand the validity of the equation of motion. with N = N(t" .118) JI(x)/h .10) and set D{x} = D°{x}/JV(i/ . x) Jt Jt dL.i/(x)/fi ) j ^ ) dtL(x. " M{x)/h &(t ) = fa (L2) 2 11 and — using Eq.t'): 1 fx" 6(x".x).

Xf = x(tf).125c) Corresponding to our relation (21. / Jxi V°{x} / Jx' V°{x}.. 6 (21. The following properties are assumed to hold: (1) / Jti V°{x} = ^ x. (21. ^2\x'.1>) = ^{x".124) where with A = tf — ti. 535 (21. . Streater and A.1 (A / ).x). Here f*f D°{x} indicates that the integral is to be taken over all x with boundary conditions Xi = x(ti).21.115) in three space dimensions. (21.t'). § The evolution of the system is expressed by the Feynman functional integral (xf.122b) By comparison with Eq. we now have in one spatial dimension {x^tflxlxuU) = J^ff'DQ{x}x{t)eiIWh. F.89) we see that Having explored the appearance of canonical momentum in path integrals we can proceed to extract the canonical quantization. Wightman [264].tf\xi.U) = T777TT / &{x}eaW\ IXi=x(ti) ) iV(A) JXi=x(U (21. x' (21. We consider this here in quantum mechanics in one spatial dimension but parallel to field theory.125a) (2) the completeness relation holds.1 (A)iV.7 Canonical Quantization from Path Integrals or -JL(x»y\x. (21. S. I(x) = / JA dtL(x.125b) and the measure property (3) N~\A + A') = A^.126) R.t'\ = l.t"\p(t")\x'.t')(x'.

tf\T(xix2)\xi.t') = -ih(x..t"\x'.t"\x"\x'.ti) = Xi(xf. (21.tf\xi\x'.t') \ i ox" i ox"J or in commutator form [P. (2) and (3) above.130) ~i5x7jX"{x". (21. The Feynman Path Integral Formalism The time t is in general a time between ti and tf.114) (xf. Thus (xf.t').ti) f ' V0{x}xlX2eu^lh.t"\x'. Taking the functional derivative 5/iSx" of this equation we obtain 'ilx7'^' '^"^x"\x ''*') = (21.. tn lie correspondingly in consecutive order between U and tf.t')=x"(x"..132) The generalization to higher spatial dimensions proceeds along parallel lines.ti) = x' 'Y^{xf.t"\x'.tf\x(ti)\xi. (21.t') (21. . we have the relation (xf. (21.t>).ti).t'){x\t'\x2\xi.x](x.122b) we can deduce the nontrivial canonical commutation relation arising here.t"\x'. Thus if T denotes such ordering. We have (cf.t\x'.tf\xi.t"\x'.129) = ^ y With Eq.t"\X'.131) = -ih(x". Then the relation follows with the help of the properties (1). Eq. (21.127)) (x". We can rewrite this expression as the commutator relation (-ITU*" -x"-^-](x". (21.tf\T(Xl • • • xn)\xi.t').127) We can now write down an expression for a time-ordered product of operators.t\x'..ti) = —!— p V0{x}xx • • • xneu^h.t'). We consider the case n = 2. (21. (21.128) where tf is later than ti and t\.t') + x"^J^{x".133) = -ih{x". But for t = ti we have (and equivalently for t = tf) as we can see from Eq.536 CHAPTER 21.

although the first few terms indicate a convergent behaviour. in the context of quantum mechanics are not convergent. T. z = a + i/3 with a — oo or f3 —> oo or both. the series converges.e. as we have seen in Chapter 8. > 537 . An essential singularity is really an ambiguity rather than an infinity. Such an expansion parameter is usually known to be or assumed to be small in some sense. Besides. W = e x p ( l / z ) .e. it has some finite. In physical applications such cases are rare. 102. divergent. i. N. Expansions of this type arise. in the sense of rigorous convergence tests. but asymptotic. It was therefore natural to search for alternative methods of expansion.g.g. E. Consider e. In field theory it has been known for a long time that expansions in ascending powers of a large coupling constant are fairly meaningless. i. If the expansion is mathematically well defined. Such expansions are strictly speaking. p. so that successive contributions to the first approximation do not invalidate this approximation. integrated over) beyond its radius of convergence.1 Introductory Remarks In field theory a perturbation expansion is generally understood to be an expansion in ascending powers of a coupling constant.Chapter 22 Classical Field Configurations 22. Whittaker and G. The vast majority of perturbation expansions discussed e. it was known from quantum mechanics that Rayleigh-Schrodinger perturbation theory by itself does not yield e. the modulus of the ratio of the (r + l)-th term to the r-th term is larger than one.g. for sufficiently large r. if a function is expanded in the neighbourhood of an essential singularity* or if an expansion of the integrand of an integral representation is used (i. i. the exponentially small level splitting which occurs in "The expansion involves both positive and negative powers of the deviation.e.g. such as the fine structure constant of quantum electrodynamics. Watson [283]. nonzero radius of convergence in the domain of the coupling parameter.e. see e.

and in general this change is accompanied by certain constraints. the latter's topological properties if any. the dominant contribution is singular (has an essential singularity) for vanishing semiclassical expansion parameter (h — 0). although this terminology is not precise. However. it is an expansion in rising powers of a semiclassical expansion parameter which plays the role of Planck's constant h in the quantum mechanical WKB approximation.538 CHAPTER 22. specific boundary conditions have to be implemented in order to yield these effects which are therefore frequently termed "nonperturbative". In the following we consider various typical examples. higher order corrections are of orders ±. and has led to insights which previously seemed unimaginable. i. One therefore faces the problem of quantizing a system which is subject to constraints. The classical.h2. such as soliton theory. c-number first approximations.) l/h2..h.h3. The study of expansions of this type has turned out to be extremely fruitful.e. such > a series begins with a contribution proportional to (e.h°. Many of these aspects are interesting by themselves . In particular it enabled nonlinear problems to be studied and led to a consideration of topological properties. one considered in particular an expansion which is such that the first approximation is purely classical in a certain sense and such that this ignores quantum effects. On the contrary. before we reach the stage at which quantization can be considered we have to deal with numerous other aspects such as the stability of the classical approximation. and the corresponding expansion is called the "loop expansion". c-number. the procedure discussed thus far) to systems with constraints was developed by Dirac [76] and is introduced in Chapter 27. A further challenging task was the development of methods of quantization of theories which incorporate classical. A method which achieves this is in particular Feynman's path integral procedure.g.... In searching for other means of expansion. nature of the dominant approximation does not imply that this describes classical motion.e. This type of expansion is decsribed as "semiclassicar. There.e. i. The fundamental extension of the method of canonical quantization (i. The procedure requires a change of variables from the original ones to collective and fluctuation variables (in analogy to centre of mass and relative coordinates). the consideration of conditions which insure the existence of Green's functions of this new type of expansion procedure (which again leads to asymptotic expansions) and so on. These methods are often described as methods of collective coordinates. Classical Field Configurations the case of the symmetric double well potential or in the case of a periodic potential.

22. to keep this case in mind. . or even a solution to a modified field equation (e. modified by going to imaginary time). monopoles. may like to consult the article of D. The important aspect we want to draw attention to here is that of "spontaneous symmetry breaking".2 The Constant Classical Field We consider first the case of a constant classical field in a scalar field model called the complex Higgs model or Goldstone theory. (22. vortices and kinks. A main aim of this chapter is to generate some appreciation of the distinction between so-called topological and nontopological configurations (later referred to as instantons and periodic instantons and bounces respectively).V(4>). In the following we trespass somewhat and only at very few points the sharp dividing line between quantum mechanics (which is effectively a onedimensional field theory) and simple scalar field theories. t The classical first approximation in the procedure outlined above may be simply a constant (time and space independent) quantity.22. it is important for a better understanding of the considerations which follow. We write the complex scalar field <f)(x) in four spacetime dimensions The spacetime Lagrangian density of the specific theory we consider here is given by C[4>.1) ' T h e reader who wants a highly advanced overview of instantons. It is evident that symmetries and their violation by the classical approximation play a significant role in the entire consideration. d^] = d^d^ . a static (time independent but space-dependent) solution of the classical equations of motion.g. and also to see these in a somewhat broader context (by comparison with higher dimensional cases). We begin therefore with a brief recapitulation of the simple Higgs model which exemplifies this case and exhibits the phenomenon known as "spontaneous symmetry violation" or the Goldstone phenomenon.* assuming that this is acceptable to a reader with some familiarity of the basics of the more complicated field theory of electrodynamics. However. We shall not be concerned so much with the case of a constant first approximation. This means theories with field densities which are therefore infinite-dimensional. Thus references to field theory will be exceptional and hopefully do not irritate the reader. Tong [272].2 The Constant Classical Field 539 and in any case deserve detailed study. V{</>) = m§0V + |A0(<^)2.

Felsager [91]. for m§ < 0 the potential is a double well potential with two minima at \<f)\ ^ 0. i. An even more familiar example is the Ginzburg-Landau theory [111] of superconductivity § in which the phase transition to the superconducting state implies the transition to the double well potential shape). and hence here [d^ + m20}(l) = --Xo(f (22. . m l + -\04>*4> )$ = <). § (22. c-number) solutions <j)(x) = 4>c = const. 22.£). of the equation of motion? For these we must have all derivatives of <j) zero.540 CHAPTER 22. V[|<t>|] m 2>o o Fig.1 Different potentials for different signs of m§.1.2) We ask: Are there classical (i. Classical Field Configurations where d^ — (—d/d(ct). The Euler-Lagrange equation is d„ dC d(d. (In models of the early universe one often considers V(|0|) with m\ > 0 at the early stage.3) See e. 22. For UIQ > 0 the potential V((f>) has a single well with minimum at |</>| = 0.g. B. £)) is bounded from below we must have Ao > 0. V ) . p.e. as illustrated in Fig. 433.S£/<90(x.e. To insure that the Hamiltonian H = f dxH with Hamiltonian density 7Y(0(x. then after cooling with a phase transition one considers expectation values |0| 7^ 0 corresponding to minima at \<$>\ ^ 0.

i.4) Here (3 is a spontaneously chosen phase like a stick held upright along H in Fig. whereas the field (bc becomes (a) A oi(0+<x) V~2 + Every new phase defines a different solution. and so ft[<M=m6:(^* 1 + -Ao(0*0)2. 22. we have 541 0.22. But for Ao > 0.</>2)-plane. mg < 0. w A iP 3ml 5* "XT" (22. TT = —J. the only such solution is the case we wish to consider.e. with 0 = 0. H = —C = V.ir]=7r<fi-C.2 The Constant Classical Field For Ao > 0. The U(l) phase transformation <b — exp(ia)((> leaves both £ and the Euler-Lagrange equa> tion unaffected.2 when allowed to fall chooses an unpredictable phase parallel to a radius in the (0>i. aq> For (b = const. It is important to observe that <bc does not possess the rotational symmetry of the Lagrangian density C or of the Euler-Lagrange equation in the plane of complex fields <f>. the equation has this invariance. 22.e. not the solution x(t)). Thus the classical solution 4>c violates the U{\) symmetry of C and of the equation of motion. Fig. (22.2 The spontaneously chosen phase. We can convince ourselves that <fic is the field associated with a state of minimized energy. like the solution x(t) of the simple Newton equation mx(i) — —V'(x) violates the invariance of this equation under time translations t —> t + 5t (i. The Hamiltonian density H is defined by the Legendre transform H[<f>. TTIQ > 0.5) .

(22. . one obtains the equations (d^ and (d^ + ml + ^ A 0 A 2 V = 0 ( ^ 2 .3). For every value of the phase (3 we have a different constant c-number field configuration 4>c.3) obtained above.e. 22. (22. (22.7b) In Eq.6) Inserting this into Eq. e. We examine this in more detail by setting <f>(x) equal to the classical c-number configuration plus a fluctuation field n(x) which is again complex like <f>(x). and we wish to investigate the behaviour of the field cf> in its neighbourhood.e. we set cf>(x) = <f)c + ri(x). >° for m 0 < °> .7a) (22. Classical Field Configurations the density Ti is minimized (i. V?) (22.e. ^ ) o + m 2 + 1A 0 A 2 ) ^1 = -A (± A0A2 + m 2 ) + 0 ( ^ 2 . Suppose we choose one such configuration.g. i.7b) the constant on the left and the coefficient of A on the right of Eq. i.2. along the trough of V (which we can call a longitudinal direction).3ml = -2ml = ml + TA 0 A 2 = 0. (22.542 CHAPTER 22. i. &H/d(fi = 0) if [m20 + ^>]</> = 0.7a) vanish on account of Eq. where (f>c = — T](x) = —= [V-l (x) + ilp2 (x)} • (22. We identify the coefficients of the linear terms on the left hand sides with those of the masses of the fields ipi. as indicated in Fig.tp2. Then we can reach some point in the neighbourhood of 4>c{(3 = 0) by travelling from 4>c partly along the direction of minimum configurations. the one for (3 = 0. and partly by climbing up the parabolic wall on either side. in a transverse direction.Then m\ m\ = ml + -A0A2 = m2.2) and separating real and imaginary parts. This is the condition (22. Clearly these are the field configurations which trace out the circular bottom of the trough of the double well potential.e.

It is known from there that the vanishing mass of the photon together with the transversality of the electromagnetic field implies that only two of the four components of the four-vector potential A^ are independent. has to be removed in order to allow a well defined perturbation procedure (i. the component which climbs up radially outward along the profile of the potential implying a tp2(x) term with the potential V(</>) and so in the Lagrangian density £. tangential to the classical path. t. then a massless boson exists. whereas ipi(x) is y ' " " " " " ^ . (22. Like the Goldstone mode in the above Higgs model it leads to a divergence in the Green's function of the theory. This wave function is in the quantum mechanics constructed about the classical configuration (j)c at this point or collective coordinate of the classical path a vector in the Hilbert space pointing in the longitudinal direction. ^ = h eip / \ ^2 l .7b). We can see the problem here by looking at Eq. of course. irrespective of the question whether the perturbation series as such converges or not.r')S(t . 22. The elimination of the . There the wave function with an associated vanishing eigenvalue is called a "zero mode". which. (3=0 Fig.3 The spontaneously chosen phase seen from above. We have here an example of the Goldstone theorem which says: If the solutions of the equation of motion do not possess a continuous symmetry of the Lagrangian. r'. the existence of individual perturbation contributions).e.t'). The appropriate Green's function G is the inhomogeneous solution of the the equation [8^ + m20]G(r. t') = d(r . though not identical phenomenon. Thus the Green's function is similar to that in electrodynamics. Goldstone's theorem applies to fully relativistic field theories.22. i. In our later discussion of theories with nonconstant classical field configurations we shall encounter a similar.e.2 The Constant Classical Field 543 We observe: The field ipi(x) has acquired a real and positive mass whereas the field ^2 0*0 is massless! We observe that the field ^2(2) is that component of ip which is directed along the trough of the potential.

To insure that the energy (see below) is positive definite we require V[$] > 0. W. spinless field $(x.3 Soliton Theories in One Spatial Dimension We consider here the two basic soliton models known as <J>4 and sine-Gordon theories.e.* Potentials of higher polynomial order in $ than $ 4 can also be considered. and we choose the metric ?7oo = + 1 .544 CHAPTER 22. F. Muller-Kirsten [291]. so that we are interested to have a parameter *For a collection of many informative papers on solitons etc. The principal aspects will always be very similar so that it really pays to study simple models in considerable detail. We see therefore that the classical c-number field configuration and the attempt to develop a perturbation series in its neighbourhood leads to intricate connections between symmetry properties. Lohe [179]. equates to zero the coefficient of this would-be-divergence. Later we wish to develop a perturbation series about a classical configuration of $. 22.8) where V[<&] is a self-interaction or potential of the field (to be specified later). Behera and A. Classical Field Configurations other components results from the vanishing mass and the constraint called the gauge-fixing condition which can also be looked at as the condition which removes from the Green's function G the divergent contribution.£) in one spatial dimension. Zimmerschied and H. 7711 = — 1 ( 1 6 Minkowski manifold). J. t Potentials of this type have been discussed by M. . see e. N. we shall need only at a later stage. Khare [17]. (22. The Lagrangian density is taken to be C[^.t In particular the sextic potential can be considered along parallel lines.* We consider the theory (later: quantum theory) of a real. We make one more assumption concerning V which. C. *S. zero mass configurations and constraints even before the question of quantization of the fluctuation field T){x) can be considered. instead we shall consider static and time-varying solutions in important models of one spatial dimension. In the following we will not be concerned with classical c-number solutions of the Euler-Lagrange equations which are simply constants. however. Rebbi and G.g.d^} = ~d^d^-V[^) = ^2-^(^j -V[$]. A. i. Soliani [237].

The field $(x. but in such solutions which are subject to (ignoring the dimensional aspect) reasonable physical conditions.13a) We write such fields <j>(x). (22.9) with quartic potential is defined by ^V\g$\ = ^V\g*. . 9 9 The Euler-Lagrange equation is seen to be (with c = 1) D * + V'[*] = 0. Since $(x. From Eq. In a quantum theory the field $ is an operator defined on a space of states.l]. such that J > — — — = 0.22.11) (22. The other condition which we impose (and study in detail) is that of stability.«{x) = V'(<p).12) where the prime denotes differentiation with respect to $ . In fact to begin with we restrict ourselves further and consider c-number fields <> —• (f) which are static. (22. i. Before we consider quantum aspects we study classical c-number fields which satisfy the Euler-Lagrange equation. The first such condition to be imposed is that the energy of the solution of interest must be finite. — m-$ 1 — cos Im 1 I ^ $ (22.g] = \ v m For instance.t) has an explicit time-dependence it is a Heisenberg-picture operator.§ Unstable classical configurations are also important and will be considered in later chapters.13b) We are not interested in just any solution of this equation.e.12) we see that they satisfy the Newtonian-like equation <j. (22. (22.3 Soliton Theories in One Spatial Dimension 545 which serves as the expansion parameter of the series. t) is an observable (which in general — in field theory as compared with quantum mechanics — does not have to be hermitian although we assume this in the present example).l}. For this purpose we assume that V[$] depends on a scaling parameter g such that Vm = V[t>. the so-called ^-theory = ^V[g$.e. 1]. i.10) n*] - m4 r 25 i- m2 and the sine-Gordon theory with cosine potential by ±V[g$} = \v[g$. ^ ^ * + V'[*]=0. the reason for this condition being that we visualize the classical solution as representative of a lump of energy. (22. at least in the present case.

The energymomentum tensor T^ and the conservation law it satisfies are given by the equations dC T^ = -n^C + Q^^d^. dC • Too = -r?oo£ + -=r$ = -C + vr$ = H. HI tanh m(x->k) 9 Fig.e.T = <I>. the classical solution will therefore have to be such that this integral is finite.e.14) where the overdot implies differentiation with respect to time t.2\dx 2 ' 1 fd& + V[$].546 CHAPTER 22. Classical Field Configurations The energy JE?[$] of the system is defined to be the spatial integral of the Hamiltonian density 7f($.17) is reminiscent of classical mechanics if we interpret x as time and V(<f>) as the potential energy of a particle at location d>. n = & .17) Since we have to integrate over all space.15) The zero-zero component is equivalent to the Hamiltonian density H.4 The wellknown soliton of $4-theory. .7r) where IT is the momentum conjugate to $ defined by dC ? r = . (22. We observe here already that the integrand of the integral in Eq. i. (22.c = -<£z + . (22.T^ = 0. 22. <9$ i. (22.16) In the static limit we obtain therefore and write this as Em = H[<t>\ -+ E(4>) = fdx 1 2 \dx 2 + V(</>) (22. d.

the $*-theory. "The name sine-Gordon is derived from the analogous Klein-Gordon equation or theory. See J. in the case of the first. Rubinstein [240].18) Here XQ is an integration constant. 4m . 22. 22.x n ).13b) for the specific potentials (22. 22.21) This sine-Gordon configuration is depicted in Fig. one obtains by straightforward integration (but see also below) the soliton configuration m 4>c{x) = ± — t a n h 777(2. < = — tan [e > | ° ] 9 2 rrm/g — Fig. The sine-Gordon theory differs from the Klein-Gordon theory in that it possesses invariance under the shift rf> — <f> + 2-n".3 Soliton Theories in One Spatial Dimension 547 As we shall see below. Again XQ is an integration constant. (22. In the sine-Gordon theory^ the classical or soliton configuration is (see below) found to be (x) = ±4— tan" 1 with energy 777 .777.4.1 .e. O±m(x-xo) (22. around <j> = 0 the sine-Gordon potential behaves like the Klein-Gordon potential proportional to 2 4> . The energy is correspondingly obtained — as will be shown — as 4 m & 9 The typical shape of the configuration given by Eq. it is not difficult to find classical c-number solutions 4>c to Eq. In fact.18) is depicted in Fig. .5 The soliton of sine-Gordon theory. m ( x . (22.5. (22.20) (22.10) and (22. > . i.22.11). — XQ).

± 1[tanh. . and hence 4/71 9<t>{x.10) we obtain f<t>{x) x — XQ i 9 ! -1 = ± / J<j>(x0) dtf>- 1 .13b) which we can write d \W? = !-"(•». 4m In tan tan 5 l/e±m(*-xo)tan^£^)U.9 2 4>{x) = i. With we obtain |g(/>(a. Here the constant is zero for V(<p) and <j>' zero at x = ± o o .* ( m' m 2 r)] ^2" . (22. or \(<P'? = V(4>)+ const. 0(x) = ±4—tan" 1 j e * " 1 ^ " * 0 ) ! .e. Classical Field Configurations The solutions (22. Inserting the potential (22.o)| =m7r.^^! m ^m <p(x0) m 4>(x) = ±—tanhm(x — XQ). In the case of the sine-Gordon theory defined by the potential (22.18) and (22.1 "»JtKxo) 9 1 .20) are obtained from Eq. Thus J<t>(xa) \/2V(6) JXOJxn U(x0) y/2V{</>) dx = x • XQ. <^>(x) = ± dx x=g4>/m \ / 2 ( l — COS X -I Am dx lx x=g4>/m A / 4 s i n 2 ±1 In tan .548 CHAPTER 22.11) we obtain r<Kx) Jd> 4>{x0) m d{g<t)/m) 9 ml g V h(l-cos&) m or m(x — XQ) I.

Jackiw [141]. however. (22. However. in the one-dimensional Ising model. A state of stability is therefore associated with minimized action and/or energy. Suppose now that we demand stability in the sense of minimized energy. Naively stability means that a system does not deviate appreciably from a state of stability (or equilibrium) if it is allowed to fluctuate between neighbouring states. the curves rising monotonically from negative to positive values are known as kinks. the discussion of Bogomol'nyi equations) in a simpler way.19). The relation (22. for instance. instead of doing this now.22) and that its second variational derivative be positive semi-definite.17) implies dx Jdx dx n£)' + ™ ( ' • # \ dx J S(f>(y) \dx and with partial integration we obtain f dx d fd<f>\ dx\dx J | 5V(</>) 5(x -y) + 5(j)(x) Tx5{x~y) .21) for the energies can also be obtained from Eq. The kink solutions are also frequently described as "domain walls" in view of their analogy with the domain separating upward spins from downward spins as.17).23) See for instance R. The expressions (22.22. (22. This is called more precisely "classical stability J The Euler-Lagrange equation is obtained by extremizing the action or Lagrangian. the others as antikinks. (22. this does not require the action to be minimized. Then we have to demand that the functional derivative of the energy E(<p) be zero. SE((f>) 5cb(y) 0.4 Stability of Classical Configurations The concept of stability can be complex.e. 22. we obtain these expressions later (cf. (22.4 Stability of Classical Configurations 549 Prom their monotonic behaviour the solutions <f)c in these cases derive their name as "kinks". . i.

i. A theorem on zero modes Let S[U] be the action of some field U defined on Minkowski space. written down in Lorentz-invariant form. dx (22. in particular the invariance under spatial translations. Thus we have to investigate the eigenvalue spectrum of the Schrodinger-like equation d2 + v"{<t>c ipk(x) = wkipk(x) dx2 (22.25). in fact ip0(x) const. Assume that S[U] is invariant under the transformations of some symmetry group G with elements g = exp(i\T) € G.28) . if we apply the generator of spatial translations d/dx to the classical equation. (22. the integrated contribution vanishes. namely (p'c.24) For so-called "classical stability" the differential operator of this expression has to be positive semi-definite. (22.e. dx2 + V"{4>) (22.13b). its eigenvalues w\ > 0. it is also called a "translational mode". in fact. Proceeding to the second variation we obtain at 0 = <fic the relation 52E((j>) 5<j){x)5(t)(y) S(x-y). and the Euler-Lagrange equation retains this property. with eigenvalue w2. The eigenfunction 4>'c{x) is for this reason called a "zero mode". (22.27) shows that the zero mode results from application of the generator of translations to the classical c-number field configuration. Eq.e.e.550 CHAPTER 22. (22. the function 4>c(x + a). a very general phenomenon which can be established as follows. The Newton-like equation (22. (22. i.26) Comparing this equation with Eq. Thus the vanishing of the first variational derivative yields again Eq. i. Thus for every translational shift "a" we obtain a new solution </>c in much the same way as we obtained new solutions by a change of phase in the Higgs model which we discussed before. Since it arises from the violation of translational invariance by 4>c.a / 0 is not the same.13b) with classical solution 4>c. if 4>c{x) is a solution. we see that the latter has one eigenfunction. of course. In fact. The original Lagrangian was. This is. we obtain 4'(x) = v"(4>c ^ or d2 dx2 4>'c(x) = 0.25) Before we begin to study this equation in detail for specific potentials we can make a very important observation on very general grounds. Classical Field Configurations Since y ^ ±00.-A. = 0.13b) retains the invariance although the solution (f>c does not.

(22.17) we obtain oo dx[V(4>c) + const. i.30) Differentiating this with respect to A (applying —id/dX) and setting A = 0. (22.e.4 Stability of Classical Configurations 551 where T is a generator and A a parameter.20) are such configurations in two particular models. Invariance of S[U] under transformations of this group implies S[U} = S[gU]. (22.22. We shall see later that these zero modes lead to undefined Green's functions for the semi-classical perturbation expansion unless one or more suitable constraints are imposed. lim [V(<f>c) + const. 0 = S'[UC] = S'[gUc}. (22.31) Inserting this into Eq.18) and (22.32) = V'(</>). i. / -oo Thus for E(4>c) to be finite.33) x—>±oo . We return to our considerations of stability. and that S[UC] is finite (or equivalently the energy). we obtain 0 = -i^S'[exp(iXT)Uc]\x=o = S"[UC](TUC).27). We can choose the potential V such that the constant is zero.e. Then the vanishing of the first functional variation implies that 5S = 0. In the one-dimensional case under discussion we can rewrite Eq. and we have seen that Eqs. We see therefore that the occurrence of these zero modes is a very general phenomenon.]. it is an expression like (22. we are interested in classical c-number field configurations of finite energy. The right hand side can be interpreted as meaning: The application of the second functional derivative of the action (at U = Uc) to (TUC) is zero. this integral has to be finite. (22. As stated above.] = 0 or x—>±oo lim V{(/)c) = 0 for const. in fact.29) Suppose now the Euler-Lagrange equation possesses the classical c-number solution Uc. = 0.13b) as ~d~4> > ' The first integral of this expression is \(tfc? = V(<l>c) + const. Thus TUC is a zero mode. (22. (22. (22.

22. This number is defined like a charge in field theory by the spatial integral of the time component of a current. the potentials have degenerate minima (i.10) we must therefore have that for x — ±oo: 4>c — ±m/g as indicated in Fig. (22. The charge Q is therefore given by dxk° = / dxe01di(bc = [ ^ ( s ) ] ^ = ±—. .552 CHAPTER 22. In the sine-Gordon theory > • defined by Eq. ± 2 . (22. the classical field (bc must approach one minimum for x — . ± l .1). . .11) we must have correspondingly for x — ±oo: » 771 9 -2vrn. Thus the classical energy or vacuum configuration is not unique. » In the <fr4-theory defined by Eq. . (22. V{4>) = 0 at different values of </>). The solutions can be characterized by an integral number called the "topological (quantum) numbcf^ "topological charge" or "winding numbed (for an illustration see Example 22. Solutions which approach > > the same minimum for +oo and —oo are the constant solutions <f>c = ±m/g in the $ 4 -theory. In the case of the 3>4-theory we can define a conserved current k^ by (22. in view of the antisymmetry of eM^. Classical Field Configurations V Fig. oo J—oo y The topological quantum number q is defined by 0.34) jfc" = e^dv(l since d^k^ = 0. n = 0 . For a configuration to have finite energy. 22.6.35) and it is seen that in this case q can be ± 1 or 0.e.c o and another for x — oo.6 The minima of V in $4-theory at x — ±oo. .

cos(27rnA)]. Tfl \ (22. .11) are given by ( so that for x — ±oo: > — J. Such a smooth deformation (in the sense that exp(i%((2))<^c depends smoothly on 6.7r) does not change a boundary condition lim [\<t>c{x)\ = (/>c(±oo). in general. x—>±oo so that under such deformations the configuration remains in the same topological sector defined by the boundary condition.7T (22. A property is. and exp(ix(0))<pc reduces to 4>C when x(#) = 0. if it remains unchanged under continuous deformations of the (field) configuration while preserving finiteness of the energy.A2)2 + 0 except for A2 = 1. Then for x — ±oo: > > > V(\4>c) (2 ^ 0 ) ? 4z( 1 . one may ask how topology comes into the picture. First. . For such configurations we have the energy oo 2 2 dx ^ A ( ^ ) + y ( A ^ / (22.4 Stability of Classical Configurations 553 Since we call q a topological quantum number.22. The difference Tfl 4>c(oo) — 4>c(—oo) = 2-7T—An . We can see this as follows. Thus we consider the family {X(pc(x)} of (field) configurations where A is some parameter which assumes real. ± l ± 2 . Again we see that E(\<f>c) is finite only for A = ± 1 .36) -oo Consider the $ -theory with 4>c — ±m/g for x — ±oo. zero is not a meaningful topological number.38) 4 V(\<t>c) ->m ~2-[l . termed topological. . Thus we expect the constant solutions (which are characterized by topological number zero) to be nontopological in some sense. n = 0 . In the case of the sine-Gordon theory the minima of the potential (22. and we call this "topological stability. Thus the condition of finiteness of energy does not permit A</>c deformations in this case other than those with \ = eix(-e) for X (0) = O. 2g Thus even without varying E with respect to A we can see that E(X(f)c) is finite only for A = ± 1 . continuous values.37) for the one-dimensional case here. .

Eq. 22.31) to (22. W.33)) that the classical configuration cf)c is the solution of the nonlinear second order differential equation (22.9 the foregoing arguments in a higher dimensional context. .43) **See e. J. Miiller-Kirsten [215]. (22.40) (22.17)) f 00 > 0.42) Thus the energy has a lower bound given by the right hand side of this inequality. From this we construct the inequality [</>'{x) = Vm^)}2 F Since (cf. if oo / dxy/2V{<l>)<f>'{x).554 CHAPTER 22. the factor A = exp[ix(#)] varies over a complete circle). (22. Thus with one space dimension higher there. (22. We observe (a) the topological current k^ of Eq.** (b) the time component k° depends only on <p but not on momenta (again in contrast to the case of Noether currents). -00 (22.34) is conserved independently of the equations of motion (other than Noether currents whose conservation follows from the equations of motion).13b) or ^ ' ( * ) ] 2 = V(</>). Eqs.5 Bogomol'nyi Equations and Bounds We have seen previously (cf.g.428. (c) the time component k° is a spatial divergence. (d) the topological quantum number iV arises in a completely classical context. where n±<XJ are the integers n corresponding to the minima of the potential at x = ±00. Classical Field Configurations involves the difference An = rioo — n. H. its integral is nonzero only on account of nontrivial boundary conditions. pp. -00 (22.-^ = N. The inequality is saturated. (For a broader view we consider briefly in Sec. 22.41) dxy/2V{$)4>'{x).e. the energy is minimized. 425 . i.39) E(cp) = J — 0 0 we can write / dx oo ~1 2(0'(x)) 2 + y(^) (22. (22.

Here (V(c/)) > 0. . for instance. 555 (22. B. Bogomol'nyi [27].40) is called the "Bogomol'nyi equation" since it was first introduced by Bogomol'nyi.19). In the sine-Gordon theory we can consider.n > 1. We discuss this in the following Example. y/V(ft) > 0) oo =2nnm/g r<p=2imm/ g / dxy/2V{<f>)</>'(x) -oo /•rj)=2Trm/g = / J<j>=0 d</>y/2V(<f>) = = n Jd>=0 rt=Mmlg n / J<p=o d<f>y/2V(<l>) 2 / ^\ — # W 2 1-cos— .5 Bogomol'nyi Equations and Bounds We observe that (cf.39). Eq.^ Clearly it solves the second order equation or (22. m 9 V V / and with x = g4>/m this becomes P IA\ -CTnin(0) = = /^ J /^ V ™3 f2W A I A • 2X n—7T I ax\/2ll — cos x) = n—7T / ax* Asm — Z 3Jo 2ir r 9 Jo V 2 2m 2 cos — n — n^-[-2(-l)+2] 2 ~ m3 8m 3 = n—rrin agreement with Eq. (22. (22. the set of configurations (fic which interpolate between 0(—oo) = 0 and <f>(oo) = 2-nnm/g. In the case of the <fr4-theory we obtain 9 9^£ 3m? <t>{c°)=rn/9 <f>(-oo)=-m/g m 3 / j \ 4 m 3 9 2 2 9V 37 3^' which is the expression given by Eq.21). In order to obtain a better understanding of how topology comes into the picture we recall that the sine-Gordon theory has an interesting analogue in classical mechanics. (22.22. n E .44) This first order equation which saturates the inequality (22.40)) this occurs when 4>'(x) T V2V(4>) = 0.

i)) The continuum limit is obtained by t a k i n g ^ (since xn = na. } n= — oo roc —> / L J . .1: Classical analogy to sine—Gordon theory Consider a string along the x-axis. * / dx -OO + 2^ V dt 1 K (d<t>\ 2 \7T I + ^9^(1-cos 4>) (22.g. rotate) only in the (y. z)-plane and only such that their elastic strings remain taut the position of any one of the pendulums is determined completely by an angle <p(xn.t) 4>{xn.7 The pendulum analogy. t)). Assuming that the pendulums can move (i. H.ij\2 Thus the The potential energy due to gravitation is (maximal for <j> = 7r) mgr(l — COS <f>{Xji.e.t) measured e.t)\2 +mgr{\ .2 V dt + -k[(f>(xn+i. V dt „ 1 (d4>_ dx • i^gr(l — cos(j>) (22. all being identical. We can now write down an expression for the energy of such a system of (say) n pendulums. Goldstein [114].46) **See e. rotate) only in the (y.45) We can therefore write the Lagrangian L=T-V• r dx -fir P 1 2 . Solution: Consider a string along the x-axis as described.cos</>(x„. from the y-axis as shown in Fig.e. ka —> K. Assume that the pendulums can move (i. The kinetic energy T of the bob at position xn at time t is (recall " m r 2 r / 2 " ) 1 —mr 2(d<j>{xn. with mass m and connected with short strings to neighbouring bobs. Chapter XI.t) -<j>(xn. At equidistant points xn = na along this string we attach strings with pendulum bobs. so that oo > u. xn+\ — xn = a) 00 a —> dx. Initially the pendulums are suspended freely in the gravitational field. Construct an expression for the energy of such a classical system and its Lagrangian. 22. z)-plane and only such that their elastic strings remain taut.t) 2 V dt The energy of interaction with neighbouring bobs is (recall "fcx 2 /2") k[(f>(xn+i. Classical Field Configurations Example 22.g.7. total energy of the system is with the number of pendulums allowed to become infinite *= £ i ''( "' ) .556 CHAPTER 22. Fig. 22.

t) of Eq. We are interested in studying (field) configurations $ in the neighbourhood of the classical solution cpc. (22. (22. (22. wk would be imaginary and so the factor exp(—iwkt) in rj(x. (22.13b) we have for small fluctuations rj(x.50) This equation is identical with Eq.e.t) = <l>c(x) + r}(x. 22. which represents an expansion in terms of normal modes. (22. ' Equating coefficients of exp(—iw k t). Hence we set ${x. Inserting (22. is therefore also called "winding numbed. we see that ipk(x) has to obey the equation 92 dx2 +V"{(t>c{x)) ^ fc (x) = w^k{x).7 linking the pendulum bobs).47) where n(x.12) we obtain d2 d2 \ Since (j)c{x) obeys Eqs. t) = ^ k ~ iL2)71^ + V "^c{x))'q[x. the topological quantum number.13a) and (22. (22.t) = °- (22'48) (22.48) becomes ( d2 \ X I [-Q^2 + wl)Tpk(x) e*v(-iwkt) k ^ X = k ^2v"{4>c{x))iljk(x)exp(-iwkt). and so would invalidate the procedure which assumes that rj(x. Eq.47) into (22.25) which we obtained as the condition of classical stability for eigenvalues w\ > 0. It is instructive to obtain this equation by yet another method. t) is the fluctuation (field).49) exp(-iwkt)ipk(x). each finite energy configuration beginning and ending with a classical vacuum corresponds to an integral number of rotations about the i-axis (i. We can understand what stability means in the present case. This number.6 The Small Fluctuation Equation 557 Thus if the classical vacuum corresponds to the case when all pendulums are pointing downwards.t) is a small . If some w\ were negative. 22.t): {w With the ansatz r)(x.6 The Small Fluctuation Equation We now return to the Schrodinger-like equation (22.25) obtained earlier.22. of the continuous curve in Fig. The equation is called stability equation or equation of small fluctuations.t).49) would imply an exponential growth in the future t > 0 or in the past t < 0.

i.50) has the general form of a timeindependent one-dimensional Schrodinger equation.558 CHAPTER 22. is a vanishing eigenvalue w\ = 0 acceptable for stability? Apparently for w\ = 0 the energy is not minimized in the direction of its associated eigenfunction.e.A \ Inserting <f)c of Eq. V"(<b) = -2m2 + 6g2(f>2. the zero mode.1. and need 9 = tan" 1 j e±m^Xo) 1. This means that the stability is not universal but only local in a certain sense. i. cos46» = 2cos2(2(9) .2 m 2 + 6m 2 tanh 2 m(x .51) (b) Sine-Gordon theory: Here 1 — cos m m 9 V"(<f>) = m 2 c o s f ^ . . (22. Classical Field Configurations fluctuation.x0) = —2m + 6m 2 (l — sech 2 m(x — XQ)) 6m Am' cosh m{x — XQ) = (22.52) tan 6 := e±m^xo). we obtain V'^c We have m cos ± 4tan" 3±m(:r—xo) (22.18). In fact. approaches a constant value which is nonzero in both the <54-theory and the sine-Gordon theory. The equation of small fluctuations (22.e. In fact. V"((/)c(x)).20). we shall see later that the fluctuations n(x. since we shall need the explicit form of V"{<f>c(x)) we calculate this for the two cases: (a) ^-theory: Here "<*> = ^ | i n*) = -2m 2 (l-£^)*. we obtain V"(<t>c) . t) have to be orthogonal to the zero mode (which is thereby circumvented) so that the Green's function of the expansion procedure exists. For x — ±oo the quan> tity acting as "potential". Inserting (j)c of (22. Thus for stability w\ must not be negative. If that is necessary.

i. We can investigate the spectrum as follows.6 The Small Fluctuation Using the formula cos 20 we obtain cos 40 1 .22. Those of the other case are complex and periodic at infinity.53) = m(x — (22. even and odd solutions. and as a consequence a limiting form of the small fluctuation equation of all basic potentials.54) Eckhardt Hence in either case the stability equation is of the form (setting z 1(1 + 1) cosh z TP(Z) = w -^iPiz) m* Regarding —1(1 + 1)/ cosh 2 z as the potential or Poschl-Teller potential — and setting it is known as the A = 4-*2 as the eigenvalue. (22. Consider the equation fd2 dz2 . we see t h a t the equation represents a Schrodinger equation with a potential which vanishes exponentially at ±oo.* For such a potential the spectrum can be b o t h discrete and continuous. (22.tan20 1 + tan 2 # Equation 559 1 — e x p { ± 2 m ( x — XQ)} I + exp{±2m(x — XQ)} sinh[=Fm(a.e. — XQ)] cosh[±m(x — XQ)]: 1 .56) *We shall see later that the Poschl-Teller equation is a limiting case of the Lame equation. which satisfy the boundary conditions V>-(o) = o. The eigenfunctions of the discrete case vanish at ± o o and are square-integrable.tan26> l + tan 2 i9 / 2 (cosh 2 m(x — XQ) 1) — cosh m(x — XQ) cosh m{x — XQ) 1cosh m{x — XQ) 2m 2 cosh m(x — XQ) Thus V'fa) = m2 xo)) d2 . -^+(0) d 0. We can therefore construct solutions of definite parity.55) The differential operator is even in z. . ip+.2 (22.ijj^ respectively. A | *(* + !) cosh 2 z tp(z) = 0.

i. when in the even case n ( n .57) Then the equation for x{z) 1S (22. .Z ) + |(A + Z2) "n+1 (n + l)(n + §) &n- (22. + \{\ ) 2 ~Q"m 1/2 oo «n+l (n + i ) ( n + l ) ( n + ^ ) ( n + ± .I) .61) Discrete eigenvalues are obtained when the series (22.4 n ( n .60) n=0 n=0 oo we find the recursion formulas x+(o = E ^ " > n(n-l) = x-(o+=l £ £&»£n.59) Setting (22. i.60) terminate after a finite number of terms.( 2 n . (22.l2 = . + l2) = 0.Z ) + ^(A + Z2) = 0. A -* An = .560 CHAPTER 22. and in the odd case n+\)U+\-l)+\(\ i.e.e.e.Z ) 2 .e.Z)2.( 2 n + l . 2n < Z. (22. Classical Field Configurations We now set (with ip either ip+ or ip-) tp(z) = (cosh z)~lx(z) and change to the variable £ = sinh2 z. i. A ^ An = . Since for £ — oo: * the function ^>+ is normalizable provided -Z > n.58) = (1 + sinh2 z)~l/2x(z).

We can see the continuous spectrum of Eq.e.10) and (22..N<1.63) we see that the continuum starts at A= 2 2^ = 0.2.e. for N = 0.51) into (22.11). 3m 2 . (22.e.65a) _dz2 cosh 2 z l=2. (22.63) d + K(2ik - 2LO)—V ' dijj' + 1(1 + l)V = 0.\=K-l2 From (22. N = 0. We can now derive the spectra of the stability equation for the potentials (22.is normalizable provided 2(n + . (22. From (22.Ny.1: w2 = 0.62) we see that the equation possesses discrete eigenvalues w —2 . i.u2)—^V 2 dio and ip(z) = elkzV{to)...( 2 . °2 i. (22. w2 = Am2. We can summarize the results in the statement that the discrete eigenvalues are A = -(Z-AT) 2 . 2 . the continuum is w > Am2.62) where even (odd) iV are associated with even (odd) eigenfunctions. (22.. d w = tanhz.54) by setting X = k2 Then 1 . m i.22. m N<2.64) This is the equation of Jacobi polynomials V\a' (u/) with a = —ik of degree I in tanhz.2Z = .1.j < I .50) we obtain with z = m(x — xo) tp(z) = 0.6 The Small Fluctuation Equation 561 and the function ip. (a) ^-theory: the equation Inserting (22.

.. d . It is now particularly interesting to look at their wave functions. %(x) Fig. We know that 4>c is a monotonic function.53) into (22.8 The zero mode as typical ground state. of course. N<1.e. .8 and we see it has the shape of a typical ground state wave function.2 The discrete eigenvalues are now given by w -^-l m^ = -(l-N)2. the zero modes. Thus either spectrum contains the expected eigenvalue zero. . (22. Eq.65b) dz2 cosh^J. Thus 4>'c(x) has no node. i.=1A=^_. . . 22. Thus it has no node and so represents the eigenfunction of the lowest eigenvalue. so that its derivative is nowhere zero (except at ±oo). and — as expected — this eigenfunction is even.50) we obtain with z = m(x — XQ) the equation (22. m m ipo{x) oc —<fic{x) = ——tanhm(x — XQ) = g cosh m(x — XQ) dx g dx (22..562 CHAPTER 22. This property can. In the $ 4 -theory we saw that (cf.66) We see that this is a nonvanishing function for any finite value of x. 22. m d . The zero mode is depicted in Fig. Classical Field Configurations (b) Sine-Gordon theory: Inserting (22. also be deduced from d(j)c/dx.27)) . for N = 0: to2 = 0 and the continuum starts at w2 = m2.

t a n h x . 1 o rrr tanh x — . (22. (22.-oc (22. .69) The quantity MQ here is called the mass of the kink (it is the mass of the soliton solution only in the classical approximation.22. Finally we consider the normalization of the zero modes. .o) ±4 m g l _|_ e ±2m(x-x 0 ) 1 cosh m(a. without quantum corrections). ipo{x) oc —<bc(x) ax oc 4 m d _i Q±rn(x-XQ) —tan e g dx m D±m(z-a. Imposing the condition F we have in the <I>4-theory: /•oo dx[^o(x)}2 = 1. .e.67) J —( Mx)= and hence m w0i^ 1 2 4 m ±—tanhm(x — XQ).3 . (22.= E(<f>c). .z 0 )]-oo = 8 . Similarly in the sine—Gordon theory we write so t h a t 4 /-oo * " / > & • ) -W171 dx oo cosh m(x — XQ) TYl = 4 — [ t a n h m O r .e.70) . d .68) d{mx) m(x-x0) 4 m3 3 ^ g Ml J cosh / i.6 The Small Fluctuation Equation 563 In the sine-Gordon theory the zero mode is proportional to . i. — XQ) OC Hence the conclusion is similar to that in the previous case.

41) for the energy of the classical configuration but now with respect to a d-dimensional position space. Olive [113].72) with T{$) = I' ddx]-{V(j))2 > 0. This question has important consequences. ddx {V4>f + v{4>) (22. Goddard and D. I. The arguments of Derrick were later rephrased by several authors whose line of reasoning we shall be using here. H.4) <j>(x) —• a<f>(ax). to underline the fundamentally different nature of topological and nontopological finite energy configurations.g.73) n<(>a) = j ' d^V^)2 = a2~d / dd(ax = J' ddXl-{V4>(ax)f d 4>{ax) 5ax . (22. Affleck [5] who uses in his Eq. however.7 Existence of Finite-Energy Solutions So far we have been considering the case of one space and one time dimensions. ^For basic aspects of scale transformations see e. f G . Kastrup [145]. i.564 CHAPTER 22. (2.> 4>a{x) = 4>(ax). J As a first example we consider the expression (22. They may help. P. *See e. P(</>) = f ddxV{(f)) > 0.* The general scaling argument used for this purpose was first introduced by Derrick t and is therefore referred to as Derrick's theorem.2-d. Classical Field Configurations 22. it is important to understand their physical implications. A. This difference may not always be appreciated later in our very analogous treatment of (topological) instantons and (nontopological) bounces. H. One can have different types of scale transformations — see e. We now wish to inquire whether static finite-energy configurations could exist in more than one spatial dimension. Derrick [68]. for if such configurations exist in a realistic theory with three spatial dimensions. Under this transformation (the verification is given below) £(</>) ^ E(</>a) = T(cf>a) + P(</>a) = a^T^) since for instance + a" d P(0).e.T(4>).71) We now consider the scale transformation (with 'a' some number € R) 4>{x) . we consider E(<P) which we also write as (22.g. *The considerations of this section extend in part beyond quantum mechanics and into field theory. I. .g.

x = (x1. (22.79) In order to emphasize that (f> — (p^ in any direction. Next we consider more complicated cases. we see that the stationarization of E with respect to a is possible only for d = 1 as in the examples we have been considering so far. In fact.7 Existence of Finite Energy Solutions 565 We saw that the static configuration with a = 1.V(<P) (22. to be finite the field <P ~^ 4>oo £ -Mo for |x| — oo.xd).\ eMo(22.e.71). It is convenient to discuss this limit in terms * of the directions of unit vectors from the origin to the (d — l)-dimensional surface in d-dimensional Euclidean space defined as the unit sphere S*-1 = {x|x 2 = 1}.80) .73) we see that this implies (2 . Consider first the Lagrangian density £[&M = \{d^)(d^) . Thus this scaling argument excludes the existence of static finite-energy configurations of the given theory in more than one spatial dimension. we set 0 dE(<pa) da J a = i (22. (22. (22. Can the energy also be minimized for more than one spatial dimension? To see this we stationarize E(<f>a) with respect to a.e.d)T(<f>) = dP{4>). since d2E(<pa) da2 2(2-d)T(<f>). It is also a virial argument in view of the virial theorem relation T{<j>) = P(<f>).77) with V(4>) > 0 and x G Rd and the set of minimum configurations Mo := {4>\V{4>) = 0}. we write • = lim J r .75) Since both T and P are > 0 and d is integral. i... i.78) For the energy of a static configuration.e.x2..22.76) we see that E(<f>) can be minimized (i. a=l (22. the second derivative is positive) only for d = 1..74) From (22. d = 1 minimizes the energy. (22.

566 CHAPTER 22.1 (A) Topologically nontrivial mapping x = +00 : S = +1 x = -00 : 5 = . A correspondingly topologically nontrivial field theory would therefore have to possess a!7(l) symmetry as in the complex Higgs model (recall that in our considerations of $ 4 . x = 00. Thus in order to obtain a topologically nontrivial case we must have a different classical vacuum in association with each point at infinity. If we go to two spatial dimensions (d = 2). is a connected region. in fact. therefore. (n — n = 0) (B) Topologically trivial mapping x = +00 : 5 = +1 <—> <> o = 27rn^ /o a. —1.81) and (since d = 1) S = ± 1 . (ptx = <f>-oo. i. We see.e. ( with — = lim <pc(x) 9J *->±oo TYl \ (22. — 1). Classical Field Configurations In the sine-Gordon theory (d = 1) the equivalent statement is (22. the Lagrangian of these theories therefore does not possess the U(l) symmetry of the complex Higgs model but instead invariance under replacements $ — — $ and $ —• $ + 2TT > > respectively). = —00 : S = — 1 <—> ^-oo = 27rn^ The topologically trivial case is given when the classical vacua associated with S1 = + 1 .and sine-Gordon theories we assumed $ to be real. the easiest way to visualize points at infinity is to take a circle Sl of radius r and let r — 00. S1.1 manifold Sd~1\d=i <-• manifold Mo <—> 0 O = 27rn^ Q <—> 0_oo = 27rn'^ kink limits (n-n'/ 0) const. —00. that we have a set of disconnected classical vacua (</>oo><^-oo.38). Then each point at infinity is characterized by a different value of the polar angle 9. of • course. one set can be mapped into the other. Table 22. —00 or equivalently S = + 1 . and. Thus in this case S^1 is a disconnected set consisting of two points. .e. 0oo 7^ <>< > in the case of the kink solutions) and /-x associated with this a set of disconnected points in space (00. i.e. i. are the same. as is the case for the constant solutions.1. These observations are summarized in Table 22.

On the other hand. Similarly.• M0. the field must map position space into field space at infinity. * n reach any other <^(oo) by simple phase shifts. (22.e.e.e. the one point S = + 1 . —00 or S = 1 . i. the manifold of classical vacua would have to possess a similar geometry if the theory is to be topologically nontrivial. Then the entire set of configurations (f)^ = \4>oo\ exp(i#) would correspond to. to have topologically distinct classical configurations. Thus. i. 00.e.1 . by allowing <fi to have three components in some "internal" (isospin) space.e. we have w e ca {Mx)}> and 1 3 1. given any (00) = |«Koo)|e* = -<?[<p(xMx)-a2]2 567 say with 9 — 9$. if cj> were real and position space two-dimensional. therefore. .g.22. i. assuming invariance of C under the transformations of the group SO(3) in internal space.3 (^real) v{4>) = -/ (22.e. we would have only two points at infinity.2. 9) would depend only on r = |x| and hence the problem would be effectively one-dimensional. This will be "nontrivial" in the above sense only if 0oo is independent of 9. i. map into. if we consider a theory in three spatial dimensions and so the unit sphere S2 = {x| x{ + xl + xl = 1}. i. for instance.83) . phase transformations 6Q ^> 9Q + 59. i. cj)(r. if <p is real. that if the problem is to be topologically nontrivial. Thus.7 Existence of Finite Energy Solutions Suppose. If position space were one-dimensional. 9)) Mo = {</>(oo)|</>*(oo)0(oo) = a2}. VMx)) with (0(oo) = linv-xx) (p(r. We see. This can be achieved e. $(r) : Sl .82) Now Mo := \ 4>(oc) ^24>2 = a2\.

In the following Example the use of Derrick's theorem is demonstrated by application to some gauged theories. we cover <S^_1 n times. a = 1. The symmetry of the potential (and Lagrangian) — i. if SO (3) nonabelian) are two models defined by the following Lagrangians which we cite here solely for the purpose of being explicit.„ where G r = 9 M £ . invariance under replacements $ — —$ in $ 4 -theory.e. A) = T(4>..V{4>a]. (b) The Georgi-Glashow model I! with nonabelian gauge field in 3 spatial dimensions with Lagrangian density (V as under (a)) £[AM. Nielsen and P. . going once around S%~1 (i. Any 4> £ MQ can be reached from any other cf> e Mo by application of an element of SO(3).e. depending on d) and correspondingly (with n complete windings) the transformation of one classical vacuum into another. B. Summarizing we see that </>(x) is a mapping from S^T1 to <S^-1. + i ( D " f t W ) a . Example 22. a theory being described as 'gauged' if a gauge field like the electromagnetic field is involved. so that details of these models are irrelevant.eeabcA*Avc and (V^U = d»<t>a eeabcA^oWrite the energy E(<j>) of such models (the functionals F and V containing no derivatives) £(</>. $ — $ + 27 > • -T in sine-Gordon theory. . it is effectively a surface (and so a continuous set) S2. 11 H. the appropriate rotation. sphere. If. through each point at infinity). Glashow [108]. the winding number is n. The theories are: (a) The Nielsen—Olesen (vortex) model (also called Abelian Higgs model) " with Lagrangian density V(4>) = Ug2 2A 9 2 FM„ = du.Ta} in an 50(3) invariant theory and so on — determines the transformation from one point at infinity to another (on a line. circle.3. In order to preserve lim y ^ i a\ such rotations must tend to the identity at r = oo.2: Derrick's theorem applied to gauged theories Wellknown and important theories involving a gauge field (AM if abelian and -A". the subsequent discussion requires only the general form of the Lagrangians.e.e.Av - dvAf i where 0 is a complex scalar field and the spatial dimensions are 2 or 3. Olesen [220]. $ — $exp(ia) in a U(l) symmetric theory. L. Georgi and S.2.568 CHAPTER 22. A) + T{A) + P{<j>).a M J ..</>] = -~G^Gai. i. $ — > > $ exp{i Xlo=i Pa. i. Classical Field Configurations The manifold A4Q is a sphere in a 3-dimensional Euclidean space. % .

84) is positive semi-definite and so must be finite by itself.85) Here the difference in the transformations results from their difference as scalar and vector quantities which are defined by the transformations ^V) = 0(x'>. (22. i.A) P{4>) 569 j / i F ( « ( P ^ ) t ( P ^ ) „ > 0. T(A) = f ddxV(4>) > 0. ^ A ) = A 2 . (22. Thus we set dE(<fix.d T ( A ) + A-dP(</>). Solution: Each contribution in (22.84) and investigate the existence of finite energy classical field configurations. ->• A„ x ( x ) = AAM(Ax). A A ) = A2-dT(^. P(tf>A) = A " d P ( 0 ) If £ is to allow a static finite-energy solution. A.T{A).. Mx') = E with a scale transformation x — x' = Ax. 4>Y We have T«(4>A) = J ddx(d„4>x)2 = J dd .A). we have {^-/»)A^ (22.d /" d d x ' a</)(x') <9x' d</>(Ax) 3x M A 2 /" ^ ( A z ) d<p(Xx) A / A^ d(Ax M ).87a) C?M„(x) — \2G^{\x).89) P{4>) are positive semi-definite and not all zero. (22. (22.87b) We consider explicitly the example of <j> — </>\(x) = <p(\x) for > T ( V(4>):= Jddx(d. negative. (2 .dP(4>) = 0.86) (22. A) + (4 .22.d)T(4>. J 4 ) + A 4 . <0 ->• <£A(X) = <KAx).d T ( ^ . l j ddxGijGai] > 0. some contribution must be . it must be stationary with respect to arbitrary field variations and so with respect to the above scale transformations also.Ax) dX Since T((f>.e. i. = 0.e.d)T(A) . A). > Since T>IJ.7 Existence of Finite Energy Solutions where T{<j>. We now consider scale transformations with A £ R.i T(AX) = X4~dT(A). (i)^ 2-d T Considering the contributions of (22.84) in this way one finds that the quantities involved transform T ( ^ .<^[x) transforms like a vector. and so B ( ^ A . (22.

is temperature dependent and changes sign at the critical temperature Tc. R. Tchrakian and H. (b) For d = 2: A theory with all terms would be a candidate — in fact the Nielsen-Olesen theory is an example.570 CHAPTER 22. (d) For d = 4: This is possible only if the sign of V is reversed — this is exemplified by a so-called pure gauge theory with so-called instanton solution (which we shall encounter in a simpler context later). tt The — now established — microscopic BCS theory of superconductivity was formulated some eight or so years after the macroscopic Ginzburg-Landau theory.** e. of course. if we allow a term T^ 4 ' {<f>) with fourth powers of derivatives.89). A) or P{4>) for d = 3. From this development one knows that in the superconducting state of the metal the electrons combine to form pairs called "Cooper pairs'" which then have bosonic properties. exemplified by soliton and sine—Gordon theories. Skyrme [253]. L. a = a(T) = a T —T c . H. a > 0.t) such that | ^ | 2 describes the density of Cooper pairs. J. Miiller-Kirsten [268]. (22. An explicit and solvable Skyrmion model in 2 + 1 dimensions is considered in D.8 Ginzburg—Landau Vortices We have already referred to the Nielsen-Olesen theory.e. D.89) the contribution (4 — d)T(4>). Landau [111]. time-independent) field configurations is equivalent to the Ginzburg-Landau theory of superconductivity. We can therefore use our understanding of the BCS theory in interpreting the Ginzburg-Landau theory. . The Ginzburg-Landau theory assumes that the static energy density is then given by £W0 = ^ W ( x ) | 2 + 7 + \aMx)\2 + \m*)t- (22-91) The parameter (22. W. Thus such a contribution could counterbalance T{4>. (22.89). One can thus define a scalar field or wave function ty(x. In the following we consider a related theory. Equilibrium states of the superconductor are assumed to be described by the time-independent static wave functions ip(x).92) The expression ^ ( x ) ! is known as the "order parameter. (22. (c) For d = 3: All three terms must be present as in the Georgi-Glashow model. This case is.90) and we would have to add to Eq. ft V . We first demonstrate — as a matter of interest — that the Nielsen-Olesen theory for static equilibrium (i. Ginzburg and L. We then have a situation of the potential like that described in the context **T. T^{<t>x) = \4-dT(</>). (e) For d > 5: In this case it is not possible to stabilize Eq.g. There are other posssibilities if we permit higher powers of derivatives as in Skyrmion models. 22. H. Classical Field Configurations (a) Thus for d = 1: P(4>) must be present but T[A] and so the vector field would not be required to satisfy Eq. (22.

(22.2) £(^.94) to check the dimension) 1 = I—a. Sec. i.^ A n (this is called "minimal coupling of the electromagnetic field^ with vector potential A to the Cooper pair field ip in a nonrelativistic treatment) where g — 2e~ is the charge of the Cooper pair.93) J' (22. from x < 0) means — as is familiar from electrodynamics — replacing V by V . (22. which we recognize as one half of the kink solution we obtained previously as the topological instanton solution of the Schrodinger equation with double well potential (in the present case ip(x) = 0 for x < 0.e. setting S£. is a measure of the distance from the surface of the superconductor to where .94) the solution is (in the domain 0 < x < 00) i>{x) = ^/^tanh(^x\ . The parameter 7 serves to put the minima of the double well at V = 0. Example 22.8 Ginzburg-Landau Vortices 571 of the complex Higgs model (cf. leads in a one-dimensional case (considering this briefly) to the equation ^ | = aV + /?^ 3 axz Imposing the boundary conditions (in the domain 0 < x < 00) ^(0) = 0. assuming there is no superconducting material in this domain). Variation as before. 22.2): For T > Tc the overall potential has a single well (minimum). Clearly the length (see Eq.e.95) This expression can be shown to be equivalent to the integrand in the Nielsen-Olesen theory.(tp) = 0. ^(00) = -a (22. Then the static energy density becomes (cf.A) = i|B| 2 + i iXQ + 7 + ^ H 2 + |/^|4. i.22. Returning to the three dimensional case and switching on a magnetic field (applied from outside of the superconducting material. but for T < Tc it has a double well shape.

J. as indicated in Fig. In the Maxwell equation See e. 450.e. The magnetic field on the other hand.98) where M is the magnetization resulting from atomic currents. The density of the Cooper or super current is given by the typical expression of a current. Classical Field Configurations superconductor (x>0) Fig. 22.572 no superconductor CHAPTER 22.M.e. We can see this as follows. (22. V x M = j s . the order parameter \ip\ attains (approximately) its asymptotic value. i.96) ipH. . 22. inside the superconductor it falls off exponentially away from the surface (this is the Meissner-Ochsenfeld effect generally described as the expulsion of lines of force from the superconductor below the critical temperature T c ).e. (22.9 Behaviour of ip and B in the superconductor. W. p. i. 2mi Setting < / > \P\e vp B = V-i|A.g. h [^DV'-V'DV]. yJ—a/[3. (22.9. exhibits a completely different behaviour — in fact. i. Miiller-Kirsten [215]. H.97) we can rewrite j s as m \ h In macroscopic electrodynamics the magnetic field strength H is (in SI units) defined by H := — B .

22.99) Inserting (22.M i. Using the relation "curl curl = grad div . Superconductors of type II are characterized by the fact that when the magnetic field is again increased and beyond a first critical value.e. Also dD/dt = 0 in a static situation like the one here. (22.L B = O. Approximating p by its equilibrium value. this becomes A B = -no V x j 5 . the normal state of the metal begins to re-appear in thin vortices and not uniformly thoughout the metal as in the case of superconductors of type I. V x B = [i0V x M = /lois. and V x (V x B) = / J 0 V x j s . m xi A B . m Here A^ is known as the (London) penetration depth (or length) (in practice this is of the order of 1 0 . These vortices therefore carry magnetic flux which is necessarily quantized .div grad" and V • B = 0.8 Ginzburg-Landau Vortices 573 the quantity j is the density of a current applied from outside. This is zero in the present case. There are two types of superconductors. K i ™*. Hence 0 = V x H = V x ( — B . (22. More and more vortices are formed as the magnetic field is increased further and further until only small superconducting domains remain which then disappear completely beyond a second critical value of the magnetic field.97) this becomes (since curl grad = 0) AB = Mo—Vx m A = /x 0 —B. those of the so-called type I and those of type II. In the one-dimensional case we have B{x) = B(0)e-x/XL.100) we obtain 2 m ( 0\ The ratio XL/XQ is called the Ginzburg-Landau parameter.5 to 10~6 cm)..e. i.

Schaposnik [69] after their equation (3. 22.9 Introduction to Homotopy Classes In this section we consider topological properties of classical finite energy configurations and introduce the concept of homotopy classes.g) two-dimensional space as being bounded by a circle at r = oo. Finiteness of E requires \<t>\ = \<fioo\ on this circle. Classical Field Configurations (see below).36).574 CHAPTER 22.2 — identified as above with the Ginzburg-Landau theory — type II superconductivity occurs^ for A > 1/4 and type I superconductivity with the complete Meissner-Ochsenfeld effect for A < 1/4. The existence of these vortices with quantized magnetic flux was predicted by Abrikosov* and are therefore frequently referred to as Abrikosov vortices (the idea was originally rejected by Landau).2 we saw that finiteness of P(V) puts no restriction on the phase of (f> so that r e (22. (22. De Gennes [67]. i. in fact a periodic function of the polar angle with period 27r. 'See the remark in H. |x|—>oo Thus at spatial infinity |</»| has the value of a zero of the potential. We can think of (e. de Vega and F. In all the models we considered so far we observed that the energy E contains a contribution P(V) = j ' afxVM). In the two models of Example 22.9). See also P. The existence of these vortices has been confirmed experimentally. Abrikosov [2]. Thus the asymptotic field </>(r. . 6')|r_>00 represents a mapping from the circle at spatial infinity to the circle defined by x(#) (or. A.102) In the one-dimensional soliton case of Eqs.101) (p(r. In the Nielsen-Olesen model defined in Example 22.9) ^e^ '>\<poo\. A.. i. of two points on a line. *A. where exp(ix(0)) is an arbitrary phase factor. If the energy E is to be finite — a basic requirement for the field configurations of interest to us — this integral must also be finite. we must have that lim V ( H ) = 0. by the element exp[i%(0)] of the internal group A(s) = e * W : SI-+SI. in other words. (22. the space consists of two diametrically opposite points on the circle. (22.e. G. J.e.37).

X ( 0 = O)].1. Misner [96]. In mathematical language S1 is therefore a compact. 0 < 0 < 2TT.* We can think of this as a fixed line in space which is provided at each point with an infinitesimal direction element that is free to rotate in the plane perpendicular to the line (like the set of pendulums considered in Example 22. topological group with the usual multiplication as group operation.1).e. Naturally an integer cannot change continuously. The 1-sphere is the unit circle in the space of all complex numbers. abelian (i.10. .102) is called the exponential map of S1 onto S1. defines the orientation of this line element.9 Introduction to Homotopy Classes 575 By a map or mapping / : X —> Y of a space X into a space y we mean a single-valued continuous function from X to 1". Suppose <j>.103) Later we demonstrate that this winding number n remains unchanged under continuous deformations (those of Eq. Then a continuous function of (j> on [0. W.10 States of winding number 0. The map satisfying X(x + y) = \{x)\{y) is said to be a homomorphism.2 imply that such a map from a circle to a circle is associated with an integer which we called the winding number n. 22. i. Fig.e. 22.118)). Sl := {zeC\\z\ = 1}. States of the strip corresponding to winding numbers 0 and 1 are illustrated in Fig. Thus smooth deformations *See D. X is called the domain of / and y its range. We may define this as « = ^[X(0 = 2 T T ) . (22. Finkelstein and C. The map A : S1 -> S1 defined on S1 C R2 by (22. (22. We can illustrate the simple low-dimensional example under discussion by reference to a strip with Mobius structure. commutative).22. 2-n] remains continuous if the function assumes the same value at the two endpoints. The two theories we considered above in Example 22.

The winding numbers n are such topological invariants. The continuous deformation or distortion of a set of points does not alter the topological structure of the set (e. Thus two homeomorphic spaces have equivalent topological structures. (22.i'ldx(0) 9A Thus for (22.. the winding number must be a constant of the motion. it cannot be deformed continuously into the so-called "classical vacuum" (the latter is a constant configuration in our soliton example). i. We observe that since time evolution is continuous. For that reason n is called a "topological invariant?. we required that lim r—>oo w o.g.105) the corrections falling off faster than 1/r. one hole. In our consideration of the Nielsen-Olesen theory we also required the gauge field A^ to behave such that the energy is finite... it retains this twist or hole under the continuous distortion) and is described as a homeomorphism. The gauge field required in the asymptotic limit (22.105) is described as "pure gauge" since it is determined entirely by the phase x(#).g. The different kinds of connectivity of these topological spaces (e. i.104) so that d2x hJ Since A has direction e#. we can write 1 d h J ee rde w At h r^oo . Homeomorphism subdivides the possible spaces into disjoint equivalence classes (to be explained below).576 CHAPTER 22. since otherwise the energy integral would behave like f d x f dr dr 2 J r J r lnr. finite. connectivity. r^oo 1 dx(9) n r o0 (22. elements in the same class have the same topological characteristic.104) to be finite we must demand that g . if it has one twist as above or a hole.e. no hole. i.) remain the same under homeomorphic transformations and are therefore called topological invariants. Classical Field Configurations of the fields which preserve the fmiteness of the energy must preserve the winding number n.e.e. A field configuration with n ^ 0 therefore cannot be deformed continuously into one with winding number zero. In other .

9):=t(f>£\9) + (l-t)(f)W(9) with te[0. 9){^ and hence the functions (22. However. In terms of the phase factor \{x) = eixi-e\ we can rewrite the relation (22.105) as «A(I)--'AWV*<I> \(x) = J*V\ ( =• VxW)) (22. is also zero at r = oo. the gauge field.9 Introduction to Homotopy Classes words. for B / 0 . Of course.105) insure that B ^ O . i. Suppose two such configurations are given by O) W(9) = ei&)V\ n fixed. i.107) The fields <f>(r. It follows that the field Fij.e.106) J g j r d d g Thus the magnetic flux is quantized. into S i . (22.22. with n = 1. the group space of (the phase transformations of) U(l) continuously. We can calculate the magnetic flux $ through the region with $ = * rd9A0 = .e. (22. the higher order contributions in (22. i. cannot be deformed continuously into one with two loops (n = 2). Therefore. going once around S^. the number of flux quanta 2irh/g being the winding number n. if A = V x .e. An example is the mapping H(t. gauge transformation of a static abelian field is A -> A' = A + Vx- 577 and we know that the The field which is determined entirely by V x is therefore termed "pure gauge". However. A field configuration with one loop (or twist). the vector potential A.108) map the spatial circle S^.110) . Ffiu. it is determined by a gradient (l/r)d/dd. (22.109) A W(0) = e^ W. cannot be pure gauge everywhere.<b rd9—4^ = n. we can consider transformations which describe continuous transformations of a configuration with fixed n. we have B = V x A = curlgradx = 0. we can go n times around S^.l]. Then we can construct a one-parameter family of maps or transformations which transform </>n (0) continuously into (j>h (6).

In fact. it is clear that this is also a homotopy. 8) = cfinl\9).8) itself must be a configuration with definite winding number. time cannot make a field jump from one homotopy class (see below) to another. Thus by varying t from 0 to 1 we deform (fin (9) continuously into (fin '(8). i. n. n' ^ n. (b) symmetry and (c) transitivity. to the internal group space of £7(1).112) are homotopic iff there exists a homotopy connecting (fin (9) and (fin (9). We verify these in the following example. these two maps (fin from S^.3: Homotopy an equivalence relation Verify that the homotopy map H(t. Classical Field Configurations which is continuous in the parameter t and is such that H(0.e. H(1. Still on the level of a mathematical text is (in spite of its title) the Argonne National Laboratory Report of G.578 CHAPTER 22.9) is called a homotopy (emphasis on the second syllable).g.-T. the unit interval I = [0. Reuter [77].111) as a map from the space described by 9.1]. The map H(t. e. (22. Since the passage of time is continuous. The relation (22. are said to be homotopic (emphasis on the third syllable) to each other and one writes ^(0)^^(9). Thomas [270]. f T h e standard texts on homotopy theory are N. Solution: As stated.9) = <fiW(9) and H(l. i.e. i.e.113) . H-.111) In principle an expression like (22. Steenrod [261] and S. S%. Two curves which can be related to each other in this way. But this type of combination is excluded sind H(t.e.112) is an equivalence relation — and as such it is defined by the three properties of (a) identity. — S\ > (22. Example 22. H. (22. (fin (9) in the above example. t The homotopy relates a representative function or curve (fin ' (0) to another function or curve 4>n' (9) by giving a precise meaning to the idea that the one can be deformed continuously into the other. (b) symmetry and (c) transitivity. i.e.lQSl^S^ with H(0. the phase factors of (f)(9). and the space of t. we have to show that the map satisfies the three properties of (a) identity.110) could be written down for configurations with different winding numbers. A superposition of classical configurations with different winding numbers does not minimize the energy. Dittrich and M. 9) is an equivalence relation. A brief introduction can be found in the very readable book of W. i. Hu [134].9) = ^°\0).0) = $\6). like <pn (0). We can consider (22.

9) = H(l. 0). Thus 7Ti[C/(l)] = vrifS'1] corresponds to (i. Returning to the maps \(x) : Si -^ S1^ (i. so that H is a homotopy connecting <jy-l>{0) with <f>^>(9).e. On the other hand H(0.ffi(O. . since and tf (1. ' . these groups are then called homotopy groups (cf. (c) We can demonstrate the property of transitivity as follows. we see that of the set of these maps some are equivalent and others are not.0) = </>(0) is continuous there.* 22. as they are called. Two elements of 7ri [f7(l)] which belong to different classes are homotopically inequivalent.9) = . 0) = tf2(l. (b) Next we check that if 0<°) ~ <p^. \ M)-\ H!(2t. U(l)). i<t<l.10 The Fundamental Homotopy Group 579 (a) Evidently the homotopy H(t.t. below). is isomorphic to) the set of integers Z (positive. 9) = <f>W (0). 9) := H(l . Moreover. Then H(t.9) H(0. Hence <f>(0^ ~ <p 2 ).0) for for 0<t<|.22. Then there exist homotopies Hi and Hi connecting <j>(0> with (j>W and t^ 1 ' with </>(2) respectively. i? is a homotopy connecting </>(°) with <^>(2'.0).10 The Fundamental Homotopy Group The set of disjoint equivalence classes in the case considered previously is denoted by the symbol Here the subscript indicates the dimensionality of the one-dimensional sphere S1.9) = <f>(0\6) and H(l. is such that H(0. (22 1U) At *=2!: But ^ ( ^ ) = ^ i ( M ) = #2(0. <f>W .e. Hence the set of all such maps breaks up into equivalence classes or homotopy classes. 0) = ^ (0) and 6(1. Homotopy classes are sometimes also called Chern—Pontryagin classes. In some cases the set of homotopy classes possesses group structure (satisfying the group properties). <^(2).e. 0) = <£{2). which therefore demonstrates that the homotopy relationship is transitive.e) tf2(2i-l. In the above case of 7ri[[/(l)] we know from our earlier considerations that every equivalence class is characterized by a specific value of the winding number. negative or zero). . a class being a set of equivalent maps. being the continuous map connecting </>(°) with <j>(-1'>. 9) = H(0. at t = 1/2 both i?i and H 2 give c/^1'. 0). i. Assume <p( ' ~ (j> and <p( ' ~ <> ' /( for continuous maps <^(0'. 9) connects <pn (0) with itself continuously. then </>W ~ <t>{0)• We define an H(t. so that H(t. We now define the function H ( f 0 ff . 0) = 4>W (0).

115) Varying t € [0. Fig. In the diagrams^ of Fig. S 2 the surface of a sphere in a three-dimensional Euclidean space. T (22. 52.580 CHAPTER 22.g.) why 7i"i[S2] = 0. .1] continuously to 0 one regains xo(#) = 0(b) In the cases n = 1 and n = 2 we have Xi(0) = 9 and xi(0) = 2e for al1 e - Similar considerations are given in the wellknown book of R. This trivial map into a single point is described as a "degenerate map" and is illustrated in Fig. characterized by winding numbers n are for instance (Xn(#) being a continuous function modulo 27r): (a) For n = 0: Xo(0) = 0 for all 9. Rajaraman [235]. to make the result plausible. T < 9 < 2TT. Sketch of idea demonstrating that nilS1] = Z.11 for the special values of t = 1 and t = 1/2 with Xo(0) = to t(2vr -9) for for 0 < 9 < 7T. p. and then introduce some of the related mathematics. In order to really appreciate how topology comes into the picture here one should understand (e.e.11 the circle drawn with the continuous line is the circle S%. Classical Field Configurations Thus there is a denumerable infinity of such homotopic classes or sectors. n=0: t=1: t=1/2: i. 22. Appropriate maps. but TTI[S2] = 0: S 1 is a circle. which is to be mapped into another circle Si indicated by dashed lines.11 The trivial n = 0 map allowing shrinkage to a point. 22. 22. How can we see this? We first sketch the idea.

is invariant under infinitesimal (continuous) deformations.h.116) In fact from this we obtain (see below) 2 - n and this means r.12 The n = 1 and n = 2 maps not allowing shrinkage to a point.s. It is seen t h a t in these cases the continuous curve can be distorted into t h a t of case (a) only by cutting the dashed circle — thus these cases do not allow a continuous shrinkage to a point.117) 2^ d6X(x)-\-1(x). and hence the winding number.118) is an infinitesimal (continuous) real function on the circle with (5f)o=o = (Sf)0=2n- 1 See S. In the cases n = 1. 0 < X(P) < 2vr. of (22. The maps for these cases are illustrated in Fig.e. i. i. (22. (22.s. We make another observation at this point. 22.e. 2 the maps are wound around the dashed circle.I** dOJn*W±e {8) X 2vr J0 — / 2Wo dO(-iri) dO \ d6 l. One can prove that the expression (22. . d (22. phase factor or element of U(l)): A(x) : Xn(x) = em*W = [Xl(x)]n. 22.103)) A(x) -> \'{x) where 5f(x) = \{x) + i\(x)5f(x) = A(x) + 6\(x). to the case of winding number zero.h. Coleman [55].22.117).12.e.117) -in ±. For winding number n we have the mapping (i.10 The Fundamental Homotopy Group 581 n=l: cutting of dashed circle n=2: Fig. the transformations^ (the result may be looked at as 8n of expression (22.

e. only one homotopy class. the identity class (the group space S2 is simply connected. It is intuitively clear that any circle drawn on the surface of a sphere can be shrunk to a point. Considering iri(S2) we recall that this symbol represents the set of disjoint equivalence classes into which the maps A (re) from S1 to S2 (circle to sphere) can be subdivided. Steenrod [261]. which means that a curve connecting any two points of S2 can be continuously deformed into every other curve connecting the two points) and one writes 7T1(S2)=0. il**( ^')- r-2-Tr A ( 21 9 2 1) ' de\x2 -*|(«/W). there is a theorem which says thatH TTg(Sn) = 0 for q < n. whereas 7ri(50(3)) = Z2 (i. One can show that the group space of SU(2) which is the sphere 5 3 C R4 is simply connected and that therefore TTI(S3 = 517(2)) = 0.e. Thus there is only the trivial map into a point. . Hence i /"27r ^ i Thus we have shown explicitly that the winding number remains unchanged under continuous deformations. In fact. an integer modulo 2) i. 15. Sec.582 CHAPTER 22. Classical Field Configurations Under this transformation n changes by 5n Now. i. 1 1 See N.e. the cyclic group with two elements (since 50(3) is doubly connected).8.

and in particular the exponentially small level splittings of the double well potential. Together with the infinite Euclidean time limit this requires for the calculation of the tunneling contribution to the ground state energy the Faddeev-Popov method for the elimination of the zero mode. finite Euclidean time) classical configurations (which are nontopological). our treatment in this chapter is restricted to the consideration of the asymptotically degenerate ground state of the double well potential which is based on the use of the topological instanton configuration.2 Instantons and Anti-Instantons A one-dimensional field theory is a quantum mechanical problem. An analogous procedure will be adopted in Chapter 24 in the use of the (nontopological) bounce configuration for the calculation of the imaginary part of the ground state energy of a particle trapped in the inverted double well potential. in such a way that the factors appearing in the result are exhibited in a transparent way. different from literature. This is an important standard example of the path integral method which serves as a prototype for numerous other applications and hence will be treated in detail and. Since we have not yet introduced periodic (i.e. In each of the cases considered we obtained the result by solving the appropriate Schrodinger equation.1 Introductory Remarks In Chapter 18 we considered anharmonic oscillators and calculated their eigenenergies.Chapter 23 P a t h Integrals and Instantons 23. 23. In the following we employ the path integral in order to derive the same result for the ground state splitting (only that in this chapter!). We consider such a theory now with action S((p) depending on the real scalar field 583 .

n = 0. 23.v{4>) <?VV m4 __ 1 .2) Fig. Fig. in particular E.e. Patrascioiu [110] and Chapter 18. "For comparison with literature.1.2 = 2m 2 (here m is the parameter in the potential V{4>) of Eq.1 The double well'potential.* /•tf CHAPTER 23. 23. 2 2 4 ™ = ~.m ^ + §^24 ' ^2 > a (23-J) The potential V(</>) has — cf. note that here the mass of the equivalent classical particle is taken to be mo = 1. 2mo has as the normalized ground state wave function </>o(?) = <<?|0) Thus for mo = 1 we have in our notation here a.2.584 <)>{t). (23.1)) and hence E0 = mh/y/2. and ^2(0) : m* . The Hamiltonian of the simple harmonic oscillator (cf. i. ! " ^ j = ^ 2 . Chapter 6) 1 2 . (23.1 — degenerate minima (positions of classical stability or "perturbation theory vacua") at positions 4>{t) = ±m = ± 0 O . Gildener and A. Path Integrals and Instantons S(<f>) / dt \i>2 .

This tunneling affects the eigenvalues._(0) = 0.2 — and we can deduce from this the boundary conditions. but there is tunneling between the two minima if the central hump is not infinitely high. we have the Lagrangian L(x. </. and in the above 4> corresponds to x. Fig.2 Instantons and Anti-Instantons 585 Classically the position <f> = 0 is a point of instability.23.In mechanics of a particle of mass 1 with ~ position coordinate x(t). The states of the system here must be even or odd under x — ±x. 23. We can develop a perturbation theory around either of the minima. i.x) = -x2 -V(x). and so the question is: What sort of boundary conditions do we have to impose on the wave function other than its exponential fall-off at 4> — ±oo? We can decide this if we observe that S = J dtL is invariant • under the exchanges 4> ~ * ±</>. We know from the shape of the potential that the quantum mechanical energy spectrum is entirely discrete (no scattering).2 The wave functions of the lowest states.e. The corresponding wave functions ip±(x) then must be peaked at the extrema of stability — as in the examples illustrated in Fig. These boundary conditions imply nonperturbative contributions to the eigenvalues which yield the splitting of the asymptotically degenerate oscillator . must have definite parity as in a physical > situation. this alone does not yield the level splitting. 23. e.g. W(0) = 0. as we did in Chapter 18 — but as we saw.

that given by Eq.e.7) so that (23. d(f> d ' dr2 dr It follows that the equation of motion is now given by 55. r .3) for tf — oo.tf\ . Then (0o.7. -SE.e. i.5) and iS = dLE d dr[d(d(f)/dT)} i. i.9) . Equation (23. we consider the amplitude for transitions between the vacua or minima at 0/^ — ±</>0 = ±rn/g over a large period of time. (23.U) = J'D{<f>}exp[iS(<l>)/h] (23.£j — —oo (we assume the normalization factor or metric to > > be handled as explained in Chapter 21).586 CHAPTER 23.£'(0) = 0. For small h we can attempt a stationary phase method. classically forbidden domains then become "classically" accessible.e. (23.3).8) V(4>) + const.0o. The transition amplitude for this.2 m 2 ( l m z (23. Euclidean time T = it. dr m+™ (23.6) d*l dr 2 .2. n) = f with the Euclidean action rTf SE(<i>) = / J TV f V{<f>}exp[-SE(<P)/h} (23. is finite if we go to imaginary. (23.7) is to be solved for 0 with the boundary conditions m 0(r) = — for Tf — oo and 0(r) = > m for n -oo. dr We observe that Eq.7) resembles a Newton equation of motion with reversed sign of the potential as a result of our passage to Euclidean time.4) J <bi =—(An rTf (ITLE = I JT. But quantum mechanically it can tunnel through the hump from one well to the other. | .2 V'(<l>) = . (23.e.<h. i. In the semiclassical path integral procedure that we want to use here. dLE = 0. We are therefore interested in the amplitude — called Feynman amplitude or kernel as we learned in Chapter 21 — (</>o. Path Integrals and Instantons approximations into an even ground state and an odd first excited state. A classical particle starting from rest (0 = 0) at —m/g cannot overcome the central hump and move to +m/g.

and having just enough energy to reach the other extremum at 4>f = m/g where it will again be at rest. this means the total energy of the particle in Euclidean motion is zero.e. 23.3 The instanton path. we see that the equation is the same as the classical equation for a static soliton (i. Thus we can picture the particle as starting from rest at one minimum of V(</>) or maximum of —V(<p). in the present case -(T — To) = m tanh m(r — TO). The solution (J>C(T) can also be looked at as describing the motion of a Newtonian particle of mass 1 in the potential —V (</>)• This is a very useful description. 23. time-independent) in the 1 + 1 dimensional <J>4-theory which we considered previously. is depicted in Fig.18).e.23.e. — (23. Its trajectory. and hence kinetic energy gained is equal to potential energy lost. . from fa — —rn/g.3.2 Instantons and Anti-Instantons 587 In fact. i. In the present context of Euclidean time the configuration (f)c is generally described as an "instanton". Thus we know the solution from there which is the kink solution (22. i. If the constant arising in the integration of the Newton-like equation is chosen to be zero (see below). the classical path. Fig.10) where TO is an integration constant.

Also note that we can re-express SE in the form 2 SE f0 c (oo) T J-oo \ dr j J0 C (- where ^iw = .8) we obtain T f- SE dr[2V((j)c) + const. thus indicating the nonperturbative nature of the expression.Tj) oc exp 4 m3 m^2 (23. The classical equation (23. Inserting the result (23.' . For a nonzero constant we obtain the periodic instantons discussed in Chapters 25 and 26.13). We also observe that the expression is singular for g2 —> 0.4.12) dimr) cosh m(T — TO) m tanh x tanh x 00 _ 4m*_ (23.10) which rises monotonically from left to right is the pseudoparticle configuration called kink in the context of soliton theory and instanton in one-dimensional field theory or quantum mechanics with topological charge 1 and Euclidean action given by (23.4).13) We observe that this expression is identical with the energy of the static soliton of the 1 + 1 dimensional soliton theory which we considered in Chapter 22.5). we obtain lim f+ OO. (23. the constant must be zero. Path Integrals and Instantons The Euclidean action of the instanton solution (23. We also need to know how we can calculate the difference AE of the energies of the two lowest levels from this expression. The configuration (23.TJ- T fro.^ . 23.10) and is called the antikink or anti-instanton configuration and obviously (following the arguments of Chapter 22) carries topological charge — 1.13) into Eq. Tf = oo.].7"/! ).11) For this quantity to be finite for TJ = —oo.r 0 ) ' (23.14) But this is not yet enough.588 CHAPTER 23. 'Note that this integral may be evaluated for finite limits r = ± T and presents no problem in the limit T oo. (23. Using Eq.* With this choice and inserting mo = j dr J = m we obtain* SE = m T lm4 2 g2 cosh4 m{r . (23. but with the same Euclidean action (23.10) is given by (23.13). This anti-instanton configuration rises monotonically from right to left as shown in Fig.7) admits another configuration which is the negative of (23.

(23.g. .Writing ••• + L ••• and performing the variation in the usual way with 5<j) = 0 at r = ± 0 0 . so correspondingly one defines the instanton or kink with topological charge + 1 and the antiinstanton or antikink with topological charge — 1.15a) or # ( r ) = 0(r-ro)^c(r-ro)0(ri-r)-e(T-Ti)^c(r-^)0(T2-r). from the sketch or from the variation of SE.15b) has the shape shown in Fig.TO far apart and T'Q 3> TQ. / > Looking at the instanton (23. 23. b u t if TQ. e.4 The anti-instanton. 0c(r) = -&(T). The sum (23. The terminology is motivated by analogy with the phenomenology of particles: Thus like a proton has electric charge + 1 (in units of the fundamental charge e) and the antiproton has electric charge —1.23.15b) a mending together of an instanton and an anti-instanton at r = T\ with —oo = To < T\ < T2 — 00 for TQ.10) we can ask ourselves: Does it make sense to consider the sum of an instanton localized at TQ as indicated in Fig.TQ are very far apart their sum differs from an exact solution only by an exponentially small quantity.5.2 Instantons and Anti-Instantons •<l> 589 ^ ^ ^ 0 ^ ^ ^ _ _ ^ Fig.15b) a solution of the classical equation? Is it a topological configuration — if so.15b) since we require configurations from vacuum to vacuum. 23. Both configurations comunicate between the two wells of the potential at < = —m/g and (ft = m/g. #(r) = </>C(T . in the case of (23.15b) is not exactly a solution.r0). 23. and an anti-instanton localized at (say) TQ. what is its associated topological charge? W h a t is its Euclidean action? The sum of 4>c and (j)c = —4>c defined by (23.15b) In the case of (23. Clearly we are here interested in configurations of type (23. (23.5. Questions which arise at this point are: Is the nonmonotonic configuration (23. as we can see e.TO) + ^ c ( r .15a) we have a simple superposition.g.

In fact.15a) is a configuration which starts from 4> = —i^/g and ends there. m/g / o / o ^ 0 1 \ -c'\ o\ T2 ' ^^__ -m/g Fig.r 0 ) At r = Ti the variation Sep is finite but nonzero so that we are left with an exponentially small contribution which. Finally. at oo the variation is <<> = 0). g cosh m(r — TQ) (23. Thus it belongs to the so-called vacuum sector of solutions and its topological charge (determined by the boundary conditions at plus and minus infinity) is zero. vanishes as T\ —> ±oo (of course.T 0 ) } ] . since (23. T\] with LECIT = 0. J —oo the classical equation is valid only up to a contribution dL E d(d<t>/dT) Ti oc <ty(Ti) coshr m(T~i .15a) can be argued to be zero. 23.CHAPTER 23.16) . #(r) dr 1 1 m 9 cosh m(r — ro) cosh m{r — T'Q)_ cosh2 m(r — TQ m cosh2 m(r — TQ) g cosh m{r — ro) m 1 [1 . Ignoring exponentially small contributions the Euclidean action of the configuration (23.2 m ( T o .5 Superposition of an instanton and a widely separated anti-instanton. it also does not exactly stationarize the Euclidean action. 5^ In order to answer the second question we observe that (23. we see that in the domain [—oo. Path Integrals and Instantons '.15a) is not exactly a classical solution. of course.e x p { .

23.2 Instantons

and

Anti-Instantons

591

and looking at Eq. (23.13) a n d ignoring exponentially small contributions, we obtain (note t h e same values of t h e limits)
<Koo)=(/>(-oo) SE dA

Jd> oo)

#A*T

For t h e configuration (23.15b) we have with a similar argument
f-Ti SE \J-oo roo JTx \ / /
rTi

( [ * • • • + [°° •••)LEdr=

( f
\J-oo

1

...-

f

i— Ti 1

...)LE
J

= 0.

J-oo

As a further example we consider a configuration consisting (approximately) of two widely separated instantons (or kinks) and one anti-instanton (or antikink) localized at TQ , TQ and TQ respectively, § # ( r ) = ^ T - T o ^ T - T f M T i - T )
-0(r - T I ) ^ ( T - r^VCTa - r)

+^(r-r2)^c(r-r(S3))0(T3-r) with - o o = T 0 < r^ 1} < 2 \ < r^ 2) < T 2 < r ^ < T 3 = oo.

(23.17a)

Fig. 23.6 Two widely separated instantons and one anti-instanton. Again t h e configuration is non-monotonic and as in t h e previous case one can show t h a t it is only approximately (with exponentially small deviations) a solution of the classical equation. In order t o answer t h e question concerning the associated topological charge of </>(3)(r) we recall t h a t this is determined by t h e boundary conditions of the configuration; thus in t h e present case t h e configuration belongs to t h e instanton or kink sector with topological charge
3
r (3) r

0

( E.g. in the overall range of r = 2T we can choose the locations TQ i ) : 2T/3.

2 T / 3 , T ^ 2 ) = 0 and

592

CHAPTER 23. Path Integrals and Instantons

+1 (like the single instanton or kink). Thus for each of the four charge sectors (corresponding to the two vacua and the kink and antikink sectors) we have a multitude of approximate solutions^ of the classical equation as illustrated in Fig. 23.7 in which (a) and (b) depict the two constant vacuum sectors and (b) and (c) the instanton and anti-instanton sectors respectively.
(a) (b) (c) (d)

Fig. 23.7 Classical solutions in their topological sectors. In evaluating later the Feynman amplitude about such configurations we use the completeness relation of states. Thus in the case of </>c (r) we write
oo poo

/
-oo

d<t>2 /
J —oo

d<j>1(<t>f,Tf\(j>2,T2){<l>2,T2\(f>l,Ti)(<l>l,T1\<l)i,Ti).

(23.17b) Evidently the evaluation of this amplitude requires two intermediate integrations.

23.3

T h e Level Difference

The nonrelativistic Schrodinger wave function ip of a discrete eigenvalue problem for a potential V satisfies a homogeneous integral equation which we can write ip(x',t')= dxK(x',t';x,t)tp(x,t), t' > t. (23.18) Here K(x',t';x,t) is a Green's function satisfying the differential equation DK(x', t'; x, t) = iS(x' - x)5(t' - t) with differential operator d Id
2

(23.19)

(23.20)

" B . Felsager [91], p. 143, therefore calls these solutions "quasi-stationary configurations". Further discussion may be found there.

23.3 The Level Difference

593

(One can convince oneself by differentiation that the integral equation is equivalent to a time-dependent Schrodinger equation). In Eq. (23.20) V0(x) = V(x) - SV(x), V = V0 + SV,

where 5V{x) is the deviation of the exact potential V from an approximation VQ. (Since we cannot — in general — solve a problem exactly, we have to resort to some approximation; thus we have to distinguish between the exact problem and the approximate problem which we can solve exactly; the Green's function is constructed from the eigenfunctions of the exactly solvable problem). The Green's function K(x', t'; x, t) can be expanded in terms of the eigenfunctions T/4 (x) of the Schrodinger equation for the potential VQ(X), i.e. (to be verified below)

K(x',t';x,t) =
We have (cf. Chapter 7)

^Y,f -

dE

'•i0)*(x')^0)(x)exp{-iE(tl-t)} E-E&0)

(23.21)

" 1 d2 - V0(x>) 2mo dxf2

^<?V) = - i « V )

so that DK(x',t';x,t)

•JL
dt1

V0(x') +

d2 2m 0 dx'2 E-E, dE 2^
(0)

^h(x/)^n0)(x)exV[-iE(t'-t)]
n
J

-iE(t'-t)

=

i6(x-x')5(t'

-t),

(23.22)

as claimed (note that the first delta function results from the completeness relation of the eigenfunctions of the exactly solvable problem). We can go one step further and integrate out the independence of Eq. (23.21). Replacing En in the denominator by (En —ie),e>0,so that the integrand of (23.21) has a simple pole at E = En —ie, and integrating along the contour shown in Fig. 23.8, Eq. (23.21) becomes (with the help of Cauchy's residue theorem) H K(x', t';x,t) = Y, Tp{n}* (x')^n0) (x) exp[-i£4°) (*' - t)}9(t' - t).
"This formula is discussed, for instance, in R. J. Crewther, D. Olive and S. Sciuto [60].

(23.23)

594

CHAPTER 23. Path Integrals and Instantons

This is our earlier expression (7.9).

Im E t - t>0
-ReE

Fig. 23.8 Integration contour. We now recall from Eqs. (21.10), (21.17), that the Green's function K can be written as a path integral, i.e. that K(xf, tf;xi,U) = J cD{x}eiS/H{tf - ti). (23.24)

Evidently we want to relate Eq. (23.23) to Eq. (23.24) in the case of our one-dimensional $ 4 -theory. In the derivation of Eq. (23.23) we distinguished between the exact problem with potential V and an approximate problem with potential VQ which can be solved exactly. In an actual application such a subdivision can be a matter of convenience. Thus we could try to approximate V by taking as VQ the harmonic oscillator part of the potential and 5V oc <5 4 then the set {T/4 } would be the set of oscillator eigenfunctions ?>; which is wellknown. Alternatively, we could be interested in a completely different problem, e.g. one with Vo given by Eq. (23.1) and 8V oc <^6. In this case ipn would be the exact eigenfunctions of the Hamiltonian for the (/>4-potential (23.1). It is shown for instance in scattering theory that the exact transition amplitude is characterized by poles at the exact eigenvalues (e.g. in the case of the hydrogen atom all discrete eigenvalues are simple poles of the scattering amplitude). Thus if we want to relate (23.24) to the path integral representation of the transition amplitude for the (/)4-potential (23.1), we must take in (23.24) the exact (real) eigenfunctions ipn{x) and eigenvalues En for the potential (23.1) which, of course, we do not know. We proceed as follows. From the above equations we obtain K(fi,t';4>,t) = Y,M<i>')Tpn(<t>)eM-iEn(t-t')/h}
n

= f %>{<!>} exp[iS/h].
•*

(23.25)

23.3 The Level Difference

595

This equivalence of the standard decomposition as a sum on the one hand, and on the other hand with the Feynman integral on the right is the reason for describing the Feynman amplitude also as kernel. We now consider the left hand side of this expression for Euclidean time r = it. Since r' > r, the expression is dominated by the lowest and next to lowest eigenvalues E+,E- associated with the ground and first excited state wave functions ip+(<fi) (even) and i/j-(<fi) (odd) respectively. Thus the left hand side can then be approximated by (with h = 1)
K(<//,T';0,T)

~

^+(cf>')^+(4>)eM-E+(T'-T)} +^_(0')^-(<£) exp[-E-(r' - r)].
(23.26)

We are interested in the transition with oo and Moreover, V>+W>o) = V+(-<fo), so that K(<f>',T';4>,r) ~ {^ + (</>o)} 2 exp[-E + (T'-T)] -{</>_(0o)} 2 exp[-£_(T'-T)]. We define AE, E0 by** AE == E--E+ so that E+ = E0Then
K{<J)Q,T'\-<J)Q,T) -<t>Q-

(23.27)

^-(0o) = - ^ - ( - 0 o ) ,

(23.28)

(23.29)

(level splitting),

E 0 = - ( E + + E_),

(23.30)

-AE,

E_ = E0 + -AE.

(23.31)

~

{V'+(0o)}2exp

- ^ - l A ^ ( r ' - r ) (r' - r . (23.32)

-{V>-(<fo)}2exp

- \E0 + -AE\

Here the wave functions at the minimum position (fro are comparable to (i.e. approximately given by) the corresponding ground state eigenf unctions of the harmonic oscillator and so (apart from overall normalization) (cf. Chapter 6) tp±{4>) oc < exp 1

(<£ + <

±exp

-{<!>-<t>o?

= ±ip±{-<f>) (23.33)

**Here AJE denotes the splitting of the originally degenerate oscillator level. This is therefore twice the deviation of a split level from the originally degenerate oscillator level.

596

CHAPTER 23. Path Integrals and Instantons

with E0 = mh/y/2 ( as we noted at the beginning of this chapter). Hence we can set 1 ± e _2 *° V#o)(23.34) l^±(0o)l = e-2*§ ± I Then
K{(J)Q,T'\-(J)Q,T)

« =

{V(^o)}2exp[-JE;o(r'-r)]|eA^'-)/2-e-AE(^)/2| 2{V(^o)}2exp[-JB0(r,-r)/^]sinh|^:AJB(T,-r)|, (23.35a)

where we re-inserted h for dimensional completeness (observe that when AE(T' — T) = AEAT ~ h and small, the hyperbolic sine may be approximated by its argument and one returns to (23.25) for Euclidean time). The relation (23.35a) assumes fixed endpoints 4>o, i.e.

d<j>" j #'(V|</>X</>" V

foMW.rW

+ 0o)<^'|^>
(23.35b)

= <^|0o) (</>0,r'| - 0o, r) (cp0\ip).
v '

Clearly without the delta functions the end points are not fixed. (This situation will be required in Chapter 26 in the case of certain bounce configurations). We now want to deduce the explicit expression obtained with the path integral formula and then extract AE by comparison. In this comparison we require r' — r to be a large Euclidean time interval (—T, T) with T — oo. It is clear from Eq. (23.35a) that we cannot put T = co rightaway; > thus we also require the study of a gentle approach of T to infinity, which necessarily makes the calculation more involved. We also note that since the above expression assumes quantization (discrete eigenvalues), the comparable formula must also involve quantization. Thus we must go beyond the purely classical contribution, i.e. we have to consider field fluctuations about the classical configuration.

23.4
23.4.1

Field Fluctuations
The fluctuation equation

We consider fluctuations n(r) about the instanton (or correspondingly static kink) configuration (pc: </>(T) = MT) + V(T). (23.36)

23.4 Field Fluctuations

597

Of course, the level splitting AE (to be extracted later) must be based on a symmetric treatment of instanton and anti-instanton. Allowing for fluctuations which are not tangential to the path one necessarily introduces more degrees of freedom — in much the same way as in allowing relative motion together with collective motion. Thus the consideration of fluctuations leads to the consideration of a larger (in fact infinite) number of degrees of freedom, i.e. to the continuum. We naturally impose on the fluctuations the boundary conditions 77(T = ± T ) | T - O O = 0. (23.37) Clearly we have to expand the action into (23.5) we obtain
SE((J))

about

4>C{T).

Inserting (23.36)

««- />[(K
(ITLE = /

i(g+ £) + m + „ /«-/>[K* + £,
dr
-oo

1 / d0c \ ~ m"* 2 2 V dr J ' 2g
2

m

2J,2

1 2J,4 -re+2*

+

WlA{t) -*»-**-w+tf*
(23.38)

,2J2 J +Sg*<%rjz + 2<?2</>cr?3 + - 5 2 T ? 4

Here the first pair of curly brackets {• • • } yields as before the contribution 4 m3
SE(<PC) = V~2

3?

(cf. Eq. (23.13)). In the second pair of curly brackets in Eq. (23.38) we use the identity

s:Am - MW.
dr where we used Eq. (23.37). Hence
—oo •2,

-oo

dr2
(23.39)

dr2

•V(j),

+r,2{^924>l ~ m2} + 2</V0C + -g2rf

(23.40)

598

CHAPTER 23. Path Integrals and Instantons

The term linear in 7 vanishes as a result of Eq. (23.7). Hence 7 SEW and SE(<j>) = = Here
oo

= SE(<l>c) + SSE(ri)

(23.41a)

JdrLE^c

+ v) + --(23.41b)

SE{ci>c)+l-JdTr]\T)L'E{4>)W

dr
/
-00

\ it) ^^ ~ ^
jvi^eMSE^/h]
T){<p} eM-SE(4>c)/h]

+

+2ffW+ .9W (23.42)

\

Hence, since the Jacobian from </>(T) to T?(T) of Eq. (23.36) is unity (with iS = —SE and subscript 1 in the following indicating that the single instanton case is considered),

-I

exp[-6SE(v)/fi] (23.43)

exp

4 m6 Jv{r)}exp[-8SE(ri)/h}, 3g^h,

where we used Eq. (23.13). We consider the approximation in which we restrict ourselves in Eq. (23.42) to terms quadratic in n. This approximation is called the Gaussian or '•'•one-loop''' approximation.* Then
oo r 1 / j„ \ 2

dT
/ -oo oo
(23J.0) f 00

\{tr) +'W*S-»'>
2\dTj

dT

i/^y

2

+ rf m 2 {3tanh J m(T - r 0 ) - 1}
(23.44)

1

(7?, Mr?),

where M is an operator.t We set (observe that r acts only as a parameter which parametrizes the classical path; the dynamical — real time-dependent — quantity is £n) V(r) = ^2^nrjn(T),
71=0

(23.45)

"The word "loop" refers to internal integrations which this non-classical contribution requires. ^The factor 1/2 is extracted so that the fluctuation equation assumes the form (23.54b).

23.4 Field Fluctuations

599

where {r/ra(r)} is at every point r of the classical path a complete set of orthonormal eigenfunctions of M with eigenvalues {w^}: Mrtn(T) = wlrin(r), dTrjn{r)r]m{T) = p25nm.
(23.46)

Here we have in mind p 2 = 1. However, for ease of comparison with the literature* we drag the parameter p2 along. Then

(T?, Mr,) (2 = 5) E [°° dr(^Vn, M^Vm) = E l & l W -

(23.47)

Formally, if we insert (23.36) and (23.44) into h of (23.43), we obtain h = I 0 exp with 4 m3' (23.48a)

I0 = Jv{r,} exp
where

2^M??)

(23

^7%
fc=0

exp
n

(23.48b) D{77} = Urn FT dr](rk)

(23.49) (in our earlier discussion (cf. Sec. 21.6): 77(T^) — %, and r?(r) is given by • Eq. (23.45)). With v(Tk) = Y^nVniTk)
n

we have dr](rk) = E d CnVn(r k ),
n

(23.50)

and (the determinant being the Jacobian of this transformation) (23.51)
fc=0 fc=0

Hence I\ = exp 4 m3 3 ^

d

K^)/(n^) e -[4? le " 12 ^
•fc=0

(23.52a)
•"•Note that a different normalization of {rj n } as in E. Gildener and A. Patrascioiu [110] (they have in our notation p2 = h/fj,2,fj,2 = 2m2) introduces additional factors on the right hand side of Eq. (23.47) and hence changes some intermediate quantities.

600

CHAPTER 23. Path Integrals and Instantons

Now,
2 2 d^e -w P e/2
J

2TT

9 9'

(23.52b)

—c

so that the integral in Eq. (23.52a) can be evaluated formally to give I\ = exp 4 m3 3 ^

•*(%>) fl V C
d
k=Q

2TT

1/2

(23.52c)

provided no w^ is zero. We now show that this condition is not satisfied, i.e. we have one eigenvalue which is zero corresponding to the zero mode associated with the violation of translation invariance by the instanton configuration. This is the case if the zero mode, like all fluctuation modes, is normalized over the infinite time interval, not — however — if a finite interval of T from — T to T is considered, in which case the associated eigenvalue is nonzero (as will be shown below) and Eq. (23.52c) is well defined. Hence first of all we show that the differential operator M is the differential operator of the stability equation or small fluctuation equation. The operator M was defined by (23.44), i.e.
i r<x>

^{V,MV)

2

dr nMrj = \ drr/Mr]
-OO oo

\f-JTn
'dr]\

K£) + 7 ^ > (23.53) £ + ""<*•> V,
dr
J —(

where in the last step we performed a partial integration and used Eq. (23.37), i.e. n(dtoo) = 0:
dr]
°°

f°°

,

d2

From Eqs. (23.44) and (23.53) we obtain

M v"(4>c
= 6g2(f>c2 — 2m 2 = 6m 2 tanh 2 m(r — TQ) — 2m2 (23.54a) 4m 2 — 6m 2 sech 2 m(r — To).

Thus Eq. (23.46) becomes d2 1 (23.54b)

23.4 Field

Fluctuations

601

We have considered the spectrum {w^} and the eigenfunctions {r]n} of this equation previously; cf. Sec. 22.6. There we found t h a t the spectrum consists of two discrete states (one being the zero mode) and a continuum. In Eq. (23.45) we have to sum over all of these states. We also observed earlier t h a t the existence of the zero mode d4>c/dr follows by differentiation of the classical equation, i.e.

A.

dT2

dr'

:[V

(</>)], implying

dr2

V'itc

dr'

°'

Thus we know t h a t the set of eigenfunctions { ^ ( T ) } of M includes one eigenfunction T)Q{T) with eigenvalue zero. Before we can evaluate the integral (23.52a) we therefore have to find a way to circumvent this difficulty. First, however, we consider in Example 23.1 a gentle approach to t h e eigenvalue zero of the zero mode by demanding the eigenfunction ipo (r) to vanish not at r = oo but at a large Euclidean time T, as remarked earlier. § Some part of the evaluation of the p a t h integral will then be performed at a large but finite Euclidean time T, and it will be seen t h a t this large T dependence permits a clean passage to infinity. Example 23.1: Eigenvalues for finite range normalization
Derive the particular nonvanishing eigenvalue of the fluctuation equation which replaces the zero eigenvalue of the zero mode if the latter is required to vanish not at infinity but at a large but finite Euclidean time T. Solution: The operator of the small fluctuation equation is given by Eq. (23.54b). We change to the variable z = m{r — TO) and consider the following set of two eigenvalue problems: (a) Our ultimate consideration is concerned with the equation ( — j + 6sech 2 z - n 2 J ^0(2) = 0 with However, we consider now the related equation (b) ( —.r+6sech22-n2)ipo(z) V dzz J = -^-^(z) m2 with ^ n ( ± m T ) = 0, n 2 = 4 and Vo(±°°) = 0.

and our aim is to obtain ui 2 . We proceed as follows. The unnormalized solution of case (a) is (cf. Eq. (23.12)) dt/>c m 2 1 ipo[z) = —— = -=—, z = m(T -T0). orr g cosh z (Note that this is the zero mode which normalized to p 2 , i.e. / drr/^ = p 2 , is 770 = p(d(pc/dT)/^/SE, where SE is the action (23.13)). We multiply the equation of case (a) by i>o{z) and the equation of case (b) by ipoiz), subtract one equation from the other and integrate from —mT to mT (i.e. over a total T—range of 2T). Then

fmT
J-mT

, f . dHo
V

j d 2 Vo\
dz2

t5g fmT
mz

, . r

dz2

j

J -mT

This corresponds to a box normalization. trascioiu [110].

See also Appendix A of E. Gildener and A. Pa-

602

CHAPTER 23. Path Integrals and Instantons

On the left hand side we perform a partial integration; on the right hand side we can replace (in the dominant approximation) the integral by that integrated from — oo to oo, and since this is the integral over twice the kinetic energy (and hence S B ( ^ I C ) of Eq. (23.13)) we obtain (remembering the change of variable from T to z) , dipo Vo — dz dj)o\ Wo . i -mT dz J . diPo
dz

-mT

™2 39 2 '

(23.55a)

In order to be able to evaluate the left hand side we consider W K B approximations of the solutions of the equations of cases (a) and (b) above. The W K B solution of case (a) is (c being a normalization constant and the ZQ occurring here a turning point)

^°( z )

=

~TJl e x p ( ~ /

dz 1/2

v J,

v(z)=:

- 6sech 2 z

and

V(ZQ) = 0.

Thus for z —• oo we have v1/2{z) — n and tpo(z) —> 0, as required. The corresponding WKB > solution of case (b) which satisfies its required boundary condition, i.e. vanishing at z = mT, is similarly seen to be

*>(*> =

^

exp

Z**1/2)-">(-* L
/ dzv1/2\ + exp ( - 1 I

f,V2 dzv

exp

["dzv1/2
J zn

where v = n 2 — 6sech 2 z — (u> 2 /m 2 ). Differentiating the last expression we obtain d ipo(z) = —cv1'4 exp ( dz and so dz But (see ipo(z) above)
•>Po\ z=mT ^ -CV 1/4 ' exp rmT

dzv1/2\

exp ( /

,f, dzv1 / 2

1

4'o\z=mT — - 2 c 6 1 / 4 exp

rmT

/ Jz0

dzv1!-2

dz

/ J z0

dzv1/2).

Hence (with n2 ~ n 2 — (uig/m2)

for T —> oo) and analogously —^o(—mT) ~ 2 — i>o(—mT). dz dz

dz

•4>o(mT) ~ 2 — ipoljnT), dz

It follows that the left hand side of Eq. (23.55a) is

4>o
But
^0(2)
:

mT dj>o mT ~ 2ip0—-ip0 dz dz -mT -mT 4m 2

g cosh z

^

z

for z —> ±00,

so that 8m-* e = F 2 i — ^ 0 ( 2 ) — =F — =F2V>o(z) for z —> ±00. dz g Hence ip0(±mT) — i>0(±mT) az = 2 ^ 0 ( ± m T ) — ^ o ( ± m T ) ~ q=4( — dz \ g
2\ 2 -4mT

23.4 Field Fluctuations
Equation (23.55a) therefore becomes

603

-if,
\

2 c

-4mT „

^0 4m4

/

m2 Sg2

We thus obtain the result

wl z, 9 6 m 2 e - 4 m T ,

wo ~ voe~2rnT,

vo • 4\/6«

(23.55b)

We observe that this eigenvalue is positive; the configuration associated with the zero mode is approximated by a classically stable configuration. For T — co the eigenvalue vanishes. Note that > here we used a Euclidean time interval of total length IT. This boundary perturbation method may be repeated for higher eigenstates which are then found to possess T-dependent eigenvalues.

23.4.2

Evaluation of t h e functional integral

It is clear from Eqs. (23.52a) and (23.52c) that the infinite product they contain provides a particular difficulty. It is the evaluation of this quantity that we are concerned with in this section. The highly nontrivial method was devised by Dashen, Hasslacher and Neveu* with reference to other sources. ^ We set Jo := det
drj(Tm

din

/(nIW)^',
^
dr

d£k exp

?£i« n\

2

2 2 Wnp

(23.56)

One introduces the following mapping or so-called Volterra transformation for large but finite T at the lower limit
y(T)

=

V

(

T

)

-

/

Z(~T)=V(-T)

T

-^ 0,

(23.57)

where N(T) is the zero mode, i.e.
N{T) =
=

HL

1

.

(23.58)

g cosh m(T — r 0 )

One can then show

fvu
*R. F. Dashen, T Integration in I. M. Gel'fand and and W. T. Martin

— see Examples 23.2, 23.3 below — that Vz } Dry exp 1 \1/2
2ith)

lJTAr2(T)]
-1/2 T

[N(T)N(-T)\

dr

-1/2

/

TN*(T)\

(23.59)

B. Hasslacher and A. Neveu [63]. functional spaces and its application in quantum physics is lucidly described in A. M. Yaglom [107]. In particular this paper utilizes results of R. H. Cameron [43], [44] which are required in our context below.

604

CHAPTER 23. Path Integrals and Instantons

Observe that this expression is independent of the normalization of the zero mode N(T). Also observe that the transformation^ (23.57) transforms effectively to a free particle Lagrangian. After evaluation in Example 23.2 of the functional determinant occurring here, the integration requires the introduction of a Lagrange multiplier which enforces the endpoint conditions, i.e. boundary conditions, on the fluctuation rj{r) and hence on 0(T) itself; this latter calculation is done in Example 23.3. The evaluation of IQ with the above formula is then straightforward. We have for T — oo >
m
V K

'

'

1 g2 cosh4 mT

m -24e

-4mT

and
T

dr

i
71 m4 m

rmT

/

J-mT mT sinh 4z
32

dz cosh z

+

sinh 2z

3z

mT -mT

+—

ill
m 32
5

AmT
(

We observe how the large-T dependence cancels out in In and obtain the simple result In = det dr)(rm)
d£n

allfc

n

2n
w

1/2

m \ 1/2

kP2

(23.60a)

Note the variable T at the upper limit of Eq. (23.57) which implies that the transformation is a Volterra integral equation. This is an important point and should be compared with a Fredholm integral equation which has constants at both limits of integration. Consider a differential equation with second order differential operator M and Green's function K(x,x'), e.g. (M - X)u(x) = f(x), MK(x,x') = 5(x-x'), a < x < b,

and with boundary conditions B\ [u] = 0, -B2M = 0. The solution u(x) of the differential equation can be written u(x) = h(x) + \f
J a,

K(x,x')u(x')dx'

with

h(x) = f
J a

K(x,x')f(x')dx'.

For f(x) 7^ 0 this integral equation is called an inhomogeneous Fredholm integral equation; for f(x) = 0 the equation is called a homogeneous Fredholm integral equation. If one has an integral equation where K(x,x') is a function of either of the following types, K(x,x') = k(x,x')0(x-x') or K(x,x') = k(x,x')0(x' - x),

the inhomogeneous Fredholm equation becomes a Volterra equation, i.e.
rx rb rb

u(x) = h(x) + A /
J a

k(x,x')u(x')dx'

or u(x) = h{x) + A /
J x

k(x,x')u(x')dx'.

In the book of B. Felsager [91] this transformation can be found in his Eq. (5.42), p. 190.

23.4 Field Fluctuations

605

Inserting this result into Eq. (23.52c) we obtain for the kernel h the simple expression ' m\ ' 4 m3 (23.60b) irh) 3 hg2 We see that this result is no-where near to allowing a comparison with our earlier result Eq. (23.35a) for the determination of AE. The reason is that the zero mode has not really been removed, so that all fluctuations, i.e. perturbations, are still present. This has to be changed and is achieved with the Faddeev-Popov constraint insertion that we deal with in the next subsection. Example 23.2: Evaluation of the functional determinant
Apply the shift transformation (23.57) to the fluctuation integral (23.43) and evaluate the functional determinant occurring in Eq. (23.59), i.e.

Thj

Solution: We first derive the inverse of the transformation (23.57) by differentiating this which yields

JV(r)
We can rewrite this equation as
Z(T) N(T)

f,{r)
N(T)

N{r)
N2(T)

T)(T)

:

d ( r](r) dr\N(T)

Thus

?j(r) r
N(T)

i= r
J^

T

dr '

Z(T) N(T)

' hand side of this equation yields

When multiplied by N(T),

r) J-j partial integration of the right
r)-JV(r) I" K ' ' J-T

V(T)

dr'—(—^—)z<T'), dr'\N{T')) y h N(r)£ ,
dr' N(T') 'V \Z{T').

r,{r)

=

z(r) + f(r),

f(r) =

(23.61)

For further manipulations we observe that with Z2(T) (obtained by squaring the expression of the first equation above) we have by differentiating out the following expression:

\ i, M^
dr~

2/ ^NM
JV(r) r,2(r)V"(0c)

,

N

, ,JV(T)
N(T)

, , ,N2M
'N2(T)'

so that since N(r) — V"{tj>c)N{r) = 0 (by definition N(T) being the zero mode)

V2{T)

+ V"(0c)r/2(r) = z2(r) +

^[^(r)

Mr)'
N(T)

606

CHAPTER 23. Path Integrals and Instantons

The derivative term will from now on be ignored since when inserted into the action integral and integrated it yields zero on account of the vanishing of the fluctuation r)(r) at the endpoints T = r 0 = - T , T. Thus 5S,

i

r1

dr

+ v"(<pcW

£

dr

-z2(r) 2 v

;

The functional integral to be evaluated is To of Eq. (23.48b) which excludes the classical factor. Thus

h

/^ } exp[-/; /T {i(J) 2 + i^(, c )}
jv{z}
D{V} Dr]

f.A\i2{T)]
D{z},
Dz Dr] Dr] Dz

T>z Dr]

Our next step is the evaluation of the functional determinant. Prom Eq. (23.57) we deduce

dr]{r") ) 7

^L=S(T-T")+

(T

K(T,T')5(T'

-T")CLT'

with

K(T,T')

.

J-T

N{r'Y

(23.62)

where we allow for a general r dependence