•
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 401402, Hackensack, NJ 07601 UK office: 57 Shelton Street, Covent Garden, London WC2H 9HE
British Library CataloguinginPublication 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 9812566910 ISBN 9812566929 (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 ParticleWave Dualism 1.4 ParticleWave 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 BraFormalism 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 OneDimensional 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 Timedependent and Timeindependent 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 TimeIndependent 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 RayleighSchrodinger Perturbation Theory 8.6 Degenerate Perturbation Theory 8.7 DingleMiiller 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 OneDimensional 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 TimeDependent Perturbation Theory 10.8 Transitions into the Continuum 10.9 General TimeDependent 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 Hydrogenlike 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 Hydrogenlike 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 ShortRange 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 BohrSommerfeldWilson Quantization 14.5 Further Examples 15 Power Potentials 15.1 Introductory Remarks 15.2 The Power Potential 15.3 The ThreeDimensional Wave Function 16 Screened Coulomb Potentials 16.1 Introductory Remarks 16.2 Regge Trajectories 16.3 The SMatrix 16.4 The Energy Expansion 16.5 The SommerfeldWatson 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 Smatrix 19.2.6 Evaluation of the Smatrix 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 largeh 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 Smatrix 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 FiniteEnergy Solutions 22.8 GinzburgLandau 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 AntiInstantons 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 FaddeevPopov constraint insertion 23.4.4 The single instanton contribution 23.4.5 Instantonantiinstanton 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 SineGordon 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 instantonantiinstanton 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 instantonantiinstanton 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 FaddeevJackiw Canonical Quantization 27.9.1 The method of Faddeev and Jackiw 28 The 28.1 28.2 28.3 QuantumClassical 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
but arise also in recently studied models of large spins. screened Coulomb potentials and maybe singular potentials. basic postulates and standard applications. 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. any problem more complicated is frequently dispensed with by referring to cumbersome perturbation methods. With the growing importance of models in statistical mechanics and in field theory. its mathematical foundations.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. The introduction to quantum mechanics we attempt here could be subdivided into essentially four consecutive parts. again point the way: For scattering problems the path integral seems particularly convenient. Chapters 1 to 14. that is the Coulomb potential and the harmonic oscillator. Thus this first part . in fact. numerous problems could. 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. one problem being that there are few nontrivial models which permit a deeper insight into their connection. Excluding spin. These basic cases will be dealt with in detail by both methods in this text. whereas for the calculation of discrete eigenvalues the Schrodinger equation. even though the expansions are mostly asymptotic. 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. However.e. like periodic potentials. To what extent the two methods are actually equivalent. we recapitulate the origin of quantum mechanics. has not always been understood well. be treated to a considerable degree of satisfaction perturbatively.g. In the first part. but also about complex energies that he encounters in a parallel course on nuclear physics. 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. in Feynman diagrams in quantum electrodynamics). Thus important level splitting formulas for periodic and anharmonic oscillator potentials (i. the aforementioned exactly solvable cases. with degenerate vacua) were first and more easily derived from the Schrodinger equation. and it will be seen in the final chapter that potentials with degenerate vacua are not exclusively of general interest. With various techniques and deeper studies.
Thereafter the concepts of instantons. we deal mostly with applications depending on perturbation theory. the chapter thereafter deals with Yukawa potentials.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 longestablished and wellknown textbooks. The following Chapter then deals with Schrodinger potentials which represent essentially anharmonic oscillators. periodic instantons. i. and their eigenvalues. In the majority of the cases that we treat we do not use the standard RayleighSchrodinger perturbation method but the systematic perturbation procedure of Dingle and Muller which is introduced in Chapter 8. The earlier Chapter 17 also contains a brief description of a similar treatment of the elliptic or Lame potential. The most prominent examples here are the double well potential and its inverted form. In the second part. 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. which is presumably the only such singular case permitting complete solution and was achieved only recently. we derive respectively the levelsplitting formula and the imaginary energy part for these cases for arbitrary states. and the behaviour of the eigenvalues is discussed in both weak and strong coupling domains with formation of bands and their asymptotic limits. The following chapters deal with the derivation of level splitting formulas (including excited states) for periodic potentials and anharmonic oscillators and — in the oneloop approximation considered — are shown to agree with those obtained by perturbation theory with associated boundary conditions. The potentials with degenerate minima will be seen to reappear throughout the text. Chapters 15 to 20. 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. 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). the method of matched asymptotic expansions with boundary conditions (the latter providing the socalled nonperturbative effects). 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. We also consider inverted double wells and calculate with the path integral the imaginary part of the energy (or decay width). Using perturbation theory. This is followed by the important case of the cosine or Mathieu potential for which the perturbation method was originally developed. After a treatment of power potentials.e.
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. These considerations demonstrate (also with reference to the topic of spintunneling and largespin behaviour) the basic nature also of the classical configurations in a vast area of applications.g. The application of path integrals to the same problems with the same aims is seen to involve a number of subtle steps. This method is therefore more complicated. from our precise reference to unavoidable elliptic integrals taken from Tables). The particular solutions and eigenvalues of interest in physics are — as a rule — those which are asymptotic expansions. 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. This puts Schrodinger equations with e. The introduction of collective coordinates of classical configurations and the fluctuations about these leads to constraints. Employing anharmonic oscillator and periodic potentials and reobtaining these in the context of a simple spin model. this comparison on a transparent level being one of the main aims of this text. cubic). quartic. and whenever available also with the results of WKB calculations. however. 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.xvm tion of the Mathieu potential — reappears as the potential in the equations of small fluctuations about the classical configurations in each of the basic cases (cosine. we can make the following observations. an approximation whose higher . except that these are no longer of hypergeometric type. such as limiting procedures. for instance. 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. it is. the Mathieu equation. All results are compared with those obtained by perturbation theory. The physical behaviour there (in the transition region between quantum and thermal physics) is no longer controlled by the Schrodinger equation. for instance. Comparing the Schrodinger equation method with that of the path integral as applied to identical or similar problems. With a fully systematic perturbation method and with applied boundary conditions. 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. anharmonic oscillator potentials on a comparable level with.
Nonetheless. like E. the references referred to are never cited by mere numbers which have to be identified e. which is particularly important in the case of elliptic integrals which require a relative ordering of integration limits and parameter domains. Park (Masan). The author is deeply indebted to his onetime supervisor Professor R. with the source given in the bibliography at the end. Watson [283].XIX order contributions are difficult to obtain.g. D. B. . MiillerKirsten *In the running text references are cited like e.* H. 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). at the end of a chapter (after troublesome turning of pages). 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. Throughout the text some calculations which require special attention. are relegated to separate subsections which — lacking a better name — we refer to as Examples.Q. This endeavour developed into an unforeseen task leading to periodic instantons and the exploration of quantumclassical transitions. Tchrakian (Dublin) and Jianzu Zhang (Shanghai). N. H. Their deep involvement in the attempt described here is evident from the cited bibliography. The author has to thank several of his colleagues for their highly devoted collaboration in this latter part of the work over many years. D. Instead a glance at a nearby footnote provides the reader immediately the names of authors. whose research into asymptotic expansions laid the ground for detailed explorations into perturbation theory and large order behaviour. Andrews). K. as well as applications and illustrations. For ease of reading. B. so that the reader is spared difficult and considerably timeconsuming searches in a source (and besides. also for the verification of results. Whittaker and Watson [283]. As a rule. J. in particular Professors J. an additional motivation was that a sufficient understanding of the more complicated of these problems had been achieved only in recent years. shows him that each such formula here has been properly looked up). e.g.g. we also consider at various points of the text comparisons with WKB approximations. 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. The line of thinking underlying this text grew out of the author's association with Professor R. 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. Since this comparison was the guideline in writing the text. Whittaker and G. W. Liang (Taiyuan). formulas taken from Tables or elsewhere are referred to by number and/or page number in the source. T. not the least for permitting a more detailed and hopefully comprehensible presentation here. Dingle (then University of Western Australia.
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). Kirchhoff's law in thermodynamics says that in the case of equilibrium. very few explain in this context what he really did in view of involvement with statistical mechanics. Planck had arrived at his formula with the assumption of a distribution of a countable number of infinitely many oscillators. Thermal radiation (with wavelengths A ~ 10~ 5 to 10 .and RayleighJeans laws for the limiting cases of small and large values of the wavelength A (or AT) respectively is generally considered as the discovery of quantum mechanics. We do not enter here into detailed considerations of Planck. the amount of radiation absorbed by a body is equal to the amount the body 1 . A "perfectly black body" is defined to be one that absorbs all (thermal) radiation incident on it. 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. Although practically every book on quantum mechanics refers at the beginning to Planck's discovery. we want to single out the vital aspect which can be considered as the discovery of quantum mechanics. Instead.1 Origin and Discovery of Quantum Mechanics The observation made by Planck towards the end of 1900. that the formula he had established for the energy distribution of electromagnetic black body radiation was in agreement with the experimentally confirmed Wien.
T) = dv3eC2U/T. T) were known and tested experimentally. radiators. energy per unit volume) of the radiation (i.1) containing the constant h by treating the radiation in the cavity as something like a gas? By 1900 two theoreticallymotivated (but from today's point of view incorrectly derived) expressions for u(u. Introduction emits.2 CHAPTER 1. u(u. of the photons or photon gas) in the cavity with both possible directions of polarization (hence the factor "2") in the frequency domain v.1) ' y Here u(v.2) and the . and then measuring the increase in temperature of the heat bath. T)du is the mean energy density (i.e.1) c is the velocity of light with c = u\.l ) where x = ^ = ^ . v + dv in equilibrium with the black body at temperature T. y J c 3 \ex . the formula (to be explained) u(u.1 Absorption in a cavity. A being the wavelength of the radiation. The parameters k and h are the constants of Boltzmann and Planck: k = 1. 1.e.3 4 J s. These expressions are: (1) Wien's law. Fig. Black bodies as good absorbers are therefore also good emitters. kT kXT (1. h = 6.e. 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.626 x 10 .T) = 2*?£(?)kT. (1.e. Let us look at the final result of Planck. In Eq. How did Planck arrive at the expression (1.38 x 1(T 23 J K'1. and the other in the region of large A (or AT). i. It was found that one expression agreed well with observations in the region of small A (or AT).. i. (1.
in the first place Planck had tried to find an expression linking both.1. exp(x) ~ 1+x) and "large" (exp(—x) < 1). (1. Moreover Planck assumed that these oscillators do not absorb or emit energy continuously. . We see. he searched for a derivation. discretization begins to enter. that the formulas (1. T) u(i/.47TZ/ 2 {x small).1 Origin and Discovery of Quantum Mechanics (2) RayleighJeans law: u(i>. When Planck had found this expression. which are indistinguishable) among the N indistinguishable oscillators at . To this end he considered Boltzmann's formula S — klnW for the entropy S. Considering Eq.T) = av e6"/Ti' where a and b are constants. every oscillator corresponding to an eigenmode or eigenvibration or standing wave in the cavity and with mean energy U. Planck now imagined a number T of oscillators V or iV oscillating degrees of freedom. (x large).e. C3 being constants. (1. Indeed. e xhv. C2.2) and (1. and he had succeeded in finding such an expression of the form u(v. so that W represents the number of possible ways of distributing the number P := NU/e of energyquanta ("photons". but — here the discreteness appears properly — only in elements (quanta) e.1) as approximations. 1. This is the point.T) 2^^kT.2 Distributing quanta (dots) among oscillators (boxes). and thus over a discrete number of admissible states.1) in regions of a. Here W is a number which determines the distribution of the energy among a discrete number of objects. where the Fig. "small" (i.T) = 2^C3T. we obtain: u(i/.3) are contained in Eq. 3 (1.3) Ci.
2.e.3 Comparing the polarization modes with those of a 2dimensional oscillator. one obtains (cf. formula 8. u(i>.4) IniV! ~ JVlniViV + O(0).2) requires that e ex is. 1. . 940. p. M. (1. S. I. e = his. The Stirling formula or approximation will appear frequently in later chapters.g. with the quanta represented schematically by dots as indicated in Fig. with riydu — 2 x —w—dv. Gradshteyn and I. U{T) being the average energy emitted by one oscillator.4 CHAPTER 1. W. e. Agreement with Eq. Ryzhik [122]. Oberhettinger [181]. by multiplying U with the number nvdv of modes or oscillators per unit volume with frequency v in the interval v. We now obtain the energy density of the radiation. as in most other Tables. and the second law of thermodynamics ((dS/dU)v Example 1. Magnus and F. p. N * oo. there not called Stirling's formula. . We visualize the iV oscillators as boxes separated by N — 1 walls. i. as for small values of e). = 1/T).1)!P! (1.T)dv. i. v + dv. 1. Introduction temperature T.7) *See e.3.1) u = vmrri (L5) as the mean energy emitted or absorbed by an oscillator (corresponding to the classical expression of 2 x kT/2. h = const.343(2).g. (1. Then W is given by w = (N With the help of Stirling's formula* {N + piy.6) Fig.e. (1.
(1.2.965 and xmin = 0. L for i = 1. 2 2 2 r2 L K — 7T n .1). The number with frequency v in the interval v..UJ2/C2 + K? = 0. (1. The number of possible modes (states) is equal to the volume of the spherical octant (where n^ > 0) in the space of n^. We observe that u(v. We obtain the expression (1.3 (as for ideal conductors).1.. nvdv per unit volume.V 2 ) E = 0.T)dX = ^ehc/*kT_idX. .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.. as indicated in Fig.1. (1. dj\l dM .8) This is Planck's formula (1.3. so that .e.3... We obtain therefore u^T) = Unv = 2^fJ^—i.2. . where^ 2 [2KUY .2.UJ = 2KV. as in electrodynamics. 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. d ™4*±\IL> — —dv = nvdv dv dv _8 3 \ c / 14 8 2 4TTV2 = 3 dv 83 ^ ^ = ^ ^ ' as claimed in Eq.7) for instance. v + dv. is given by . rii = 1.. Then L^j = nrii. '''From the equation I \ JW . so that the derivative of u implies (x as in Eq. so that (lvL\A 0 I I = rr.i = 1.T) has a maximum which follows from du/dX = 0 (with c = vX). KT = I J .1)) The solutions of this equation are ^max = 4.7).3. In terms of A we have u(X. i.
1. but one can expect an analogy.e. Lorentz and Planck that Eq. 2 . n = 0. the behaviour of the system at zero absolute temperature. One might suppose now. Since temperature originates through contact with other oscillators. Later it was realized by H. (1. . which can not be eliminated without a different approach.2 (1. This is Wien's displacement law. Thus we have a rather complicated system here. One expects.l. which can assume the discrete energies en — nhv. then (with x = hv/kT) its mean energy is En=0e = nX dx ^0 — /ii/— In = huf r%e dx 1 — e_x (1 — e~x)z hv . which had also been known before Planck's discovery. . . If an oscillator with thermal weight or occupation probability exp(—nx) can assume only discrete energies en = nhu. that it is easier to consider first the case of T = 0. ra = 0. 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=—.965K = Const. of course.10). However. Introduction he 4. A. . x — oo) the mean energy vanishes (0 < U < * > oo). . we then have at T = 0 independent oscillators.10) Thus here the socalled zero point energy appears. that we arrive at quantum mechanics simply by discretizing the energy and thus by postulating — following Planck — for the harmonic oscillator the expression (1. such a procedure leads to contradictions. We therefore examine such contradictions next.—v <L9) We observe that for T — 0 (i.8) could be derived much more easily in the context of statistical mechanics. that of an oscillation system at absolute temperature T ^ 0. i.e. We are not dealing with the linear harmonic oscillator familiar from mechanics here. which did not arise in Planck's consideration of 1900.6 The first value yields AmaxT = CHAPTER 1. and from which the constant h can be determined from the known value of k.
k dU 1+U\ f ln(l T + U\  U U k ( e . kT/2. {NW •• 1)! (TV1)!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) . 1.2 Contradicting Discretization: Uncertainties 7 Example 1.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.1.ln(TV . 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.i n .. d . i.1)! . 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". for 2 degrees of freedom.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)! . so that 1 f ds\ T — \dUjy .= . In the following we attempt to incorporate the above discretizations into classical considerations* and consider for this reason socalled thought experiments (from German "Gedankenexperimente"). we obtain T V S =fc[ln(TV+ P .+ 1 e \U e u = exp(e/fcT) which for e/kT — 0 becomes > .2 Contradicting Discretization: Uncertainties The farreaching consequences of Planck's quantization hypothesis were recognized only later.In .1 ' u~ e . around 1926.1 Solution: Inserting W into Boltzmann's formula and using In TV! ~ A In T — TV.e.kT This means U is then the classical expression resulting from the mean kinetic energy per degree of freedom. with Heisenberg's discovery of the uncertainty relation. .
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 "zeropoint 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,
(113)
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 selfevident 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 semiaxes 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 particlewave duality.^
1.3
ParticleWave Dualism
The wave nature of light can be deduced from the phenomenon of interference, as in a doubleslit experiment, as illustrated in Fig. 1.6.
Fig. 1.6 Schematic arrangement of the doubleslit experiment. Light of wavelength 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 reinforced 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 ParticleWave Dualism
13
the number of such photoelectrons 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. particlelike, 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 pointlike 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 particlelike 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) =Aeikriut,
where r is the position vector, and to and k are given by E — hio, p = /ik. Thus particles also possess a wavelike 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
ParticleWave 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 particlelike nature in another. We cannot observe both types simultaneously, i.e. the wavelike nature together with the particlelike 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 doubleslit 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 SingleSlit Experiment
Discuss the uncertainties of the canonical variables in relation to the diffraction fringes observed in a singleslit experiment. Solution: Let light of wavelength 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 singleslit 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 ParticleWave 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 wavefront
WW.
We consider rays deflected by an angle 9 with wavefront WW' and bundled by a lense L and focussed at a point Q on the screen S2. Since WW1 is a wavefront, 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 — rexp(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 singleslit 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
(125)
From Eq. (1.24b) we obtain for the angle 9 in the direction of the first minimum Ax sin 6 = A,
1.4 ParticleWave 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 xcomponent is 1h Apx = 2psin0 = — sine A (1.26b)
(prior to scattering the xcomponents 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 = rp' = 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 socalled 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 doubleslit (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 wavelike 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 timedependent 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' < —yEo, 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, ie. [[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 + wgeo, ^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 < 5Bo.
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 lmw2x2\ (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 socalled adiabatic theorem of Ehrenfest, which formulates this in a general formJ
Example 1.4: Angular spread of a beam
A dishlike aerial of radius R is to be designed which can send a microwave beam of wavelength 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)
~ AKR'
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 reexpress 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 cnumber 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
/
^2PiQiH(qi,Pi) dt = 0,
where now (different from the derivation of the EulerLagrange equations) the momenta pi and coordinates qi are treated as independent variables. As
23
^As H. S »(«. x^fdu. which can be observed directly at time t. dH In this Hamilton formalism it is wrong to consider the momentum pi as mqi. . (2.1). (2.P.p% are therefore again observables. (2. A system consisting of several mass points is therefore described by a number of such variables. Compared with an arbitrary function f(qi. Hamiltonian Mechanics is wellknown.1) as d . in analogy with Eqs. All functions u(qi.4) contains as special cases the Hamilton Eqs. (2. It suggests itself therefore to consider more closely the properties of the symbols (2.Pi) is contained implicitly in the canonical variables q^ and pi. One now defines as (nonrelativistic) Poisson bracket the expression^ With this definition we can rewrite Eq. this expression is simply a functional determinant. M. (2.4) as the generalization of Eqs. 6t>0 5t Real quantities which are directly observable are called observables.24 CHAPTER 2.qi(t) qi = hm f . whereas the velocity requires observations of space coordinates at different times. since .1). Goldstein [114] remarks at the end of his chapter VIII. The total time derivative of u can therefore be rewritten with the help of Eqs.1). the standard reference for the application of Poisson brackets is the book of P.g. Dirac [75]. One can verify readily that Eq. the entire timedependence of observables u(qi. . (2.^ ^ j . chapter VII.t). the equation of motion of the observable u.) = £ [wm + WiK) = £ [WiWi . Goldstein [114]. The following properties can be verified: *See e. as mass times velocity.Pi. (2. Rather. (2.2) If we have only one degree of freedom (i = 1). Pi has to be considered as an independent quantity.. which all together describe the state of the system. du \ x^/dudH du dH\ .pi) of qi. It was only with the development of quantum mechanics by Heisenberg and Dirac that Poisson brackets gained widespread interest in modern physics. chapter VIII. i. qi(t + 6t) .2) as This equation is. A. H.n n.e.3). We can therefore consider Eq. one obtains the Hamilton equations* • OH .
B}* (4) product formation: {A.B} = {B. (2) linearity: {A. {qi. the Poisson bracket {A. (2.5d) is useful in calculations. that the very general Eq.6) We can now show. B2}. {pi. then the ordering is taken as in (2. but could be multiplied by a complex number): {A. As an example we consider a case we shall encounter again and again. where A and B are arbitrary observables. {A. a i S i + a2B2} = ax{A. for example.5d) = {A*. {B.A}. If we evaluate the Poisson brackets for qi. it is irrelevant whether we write {A. A}} + {C.Qk} = 0.5c) The first three properties are readily seen to hold. (2.B*}. As long as we are concerned with commuting quantities.B}C + B{A. B}} = 0. Property (2.2. Bx} + a2{A. Later we shall consider noncommuting quantities. B}. in other words without any reference to the original definition (2. we wish to evaluate {A. can be solved solely with the help of the properties of Poisson brackets and the fundamental Poisson brackets (2.5b) (3) complex conjugation (note: observables are real. we expand A and B in powers of qi and pi and apply the above rules until only the fundamental brackets remain. i.BC} (5) Jacobi identity: {A. If.5e) = {A. Since Eqs. {C. (2.6) give the values of these. 25 (2. (2.Pk} = 5ik. which combines the Hamilton equations.C}. These are {Qi.5d) above. like here. we obtain the fundamental Poisson brackets.Pi.e.B}C or C{A.6). B} is completely evaluated. C}} + {B. The original definition of the Poisson bracket will not . that of the linear harmonic oscillator.2 The Hamilton Formalism (1) Antisymmetry: {A.4). (2.3) of the Poisson bracket.B}. (2.Pk} = 0.5a) (2.
8b) We insert (2.8b) p=q.12) .8a) and use the properties of the Poisson bracket and Eqs. Since constants are also irrelevant in this context.7) into (2. According to Eq. (2.p}p + p{q.p\) P.7) q = [q.11a) 'q' = q. from which we infer that q(t) — qocost + posint. (2.11b) (2.H}. and so q = q. Hamiltonian Mechanics be used at all.t).P2} + {q.yPot3 + •••• (2. or q(t) = qo+ Pot .p.\(p2 + q2)} = l({q.t). and Pi are ordinary real number variables and that H(q. (2. and P={p. (2.8a) (2.6).p) is an ordinary function is also irrelevant. q + q = o. (2.p) = ±(p2 + q2). These are transformations qi—>Qi = Qi(q.. (2. Pi—> Pi = Pi(q. (2. In the evaluation one should also note that the fact that g.10) In classical mechanics one studies also canonical transformations.. Then we obtain: (2. (2.qot2 .9) Similarly we obtain from Eq. "4' = q. From Eqs.26 CHAPTER 2.p. we consider as Hamiltonian the function H(q.p. q = {q.4) we have for u = q.H}.9) and (2...10) we deduce q = p = q.q2}) = = 2i{q.
P is canonical in the sense defined above.t).3 deals with the relativistic extension.15a).qk} = 0.P. {qi.B}q.* Hence in Example 2. which transforms qi. i.e.Pi back to their constant initial values. i. and Example 2.e. {Qi. which means that a Hamilton function K(Q. _ax p__dK_ (2.e. . {PhPk} = 0. that (dropping the subscripts therefore) {Pi. In classical mechanics we learned yet another important aspect of the Hamilton formalism: We can inquire about that particular canonical transformation.1 we verify only Eq. this transformation is described precisely by the equations of motion but we shall not consider this in more detail here. . Of course. B} of two observables A and B in terms of either set of canonical variables.15a) °raS {AiB}Q. (2. *H.e. The proof requires the invariance of the fundamental Poisson brackets under canonical transformations. i. P) exists.Pk} = 0. chapter VIII. (2.e.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. Example 2.pk} = 5ik. those at a time t — 0.1: Canonical invariance of Poisson bracket Assuming the invariance of the fundamental Poisson brackets under canonical transformations Qj = Qj(Q'P)>Pj = Pj{Q>P).P. Eq.B}Q>P.13) We write the reversal of the transformation (2.2.14) With the help of the definition (2.2 below contains a further illustration of the use of Poisson brackets. for which Hamilton's equations hold. verify that the Poisson bracket of two observables A and B is invariant.P provided the transformation q.t). i. i.B}q.P One can then show that {A.Qk} = 0. (2.12) Qi—>qi = qi{Q.p <> Q. Pi^Pi=Pi(Q. as {A.p = {A. {QhPk} = Sik.15a). Example 2. {qi.3) we can now express the Poisson bracket {^4.2 The Hamilton Formalism 27 for which the new coordinates are also canonical. Goldstein [114]. (2. (2.
15b) dA {Qk.Pj}q. .15a) r„ ™ v .q/c) and (E. 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 „ „•.F}r ~di d~E ~ d~E~8t Consider F = H(q.p = J2 d~Fk Replacing in the above A by P and B by A.s F 5 7 r ={A.pc).28 CHAPTER 2.3: Relativistic Poisson brackets By extending qi. {A. we obtain analogously dA dQk' Inserting both of these results into the first equation.A}q.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. we obtain as claimed by Eq. Qk}q. we obtain dA {Qk. V V °Qk dPk dPk dQk J E x a m p l e 2.p) . OQk Pk}q. dA {Qk. we have —Et.p^ "Pk. (2.E(t).A}q. Solution: Relativistically we have to treat space and time on an equal footing.pt to fourvectors (in a (1 + 3)dimensional space) define relativistic Poisson brackets. p. Mittelstaedt [197].p^— + {A. their product Et — qp being relativistically invariant.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). Thus we extend q and p to spacetime vectors (t. 236.p = dB dA dB \ rA 1 {A'B^ E b r s r .( 9A = {Pk.P^pr. Replacing here A by Q and B by A. (2.p^v dP.+ dQ. 1 „ H = p + m0gq and solve the canonical equations with Poisson brackets for initial conditions q(0) = qo.B}q. Thus whenever q and p are multiplied.B}QIP.Qj}q.p(0) = poSolution: The solution can be looked up in the literature. § E x a m p l e 2. § See P. The relativistic Poisson bracket (subscript r) therefore becomes du dF dq dp du dF dp dq du &F _ du dF ' {u.
since H is expressed as a function of q and p.p) around some point qo. 2TTE UJ and hence g oc — . and at time t > to in a domain G\(q. .po) = Gi(q.3 Liouville Equation.)•'=(*)**£.po. But now we consider a system whose phase space coordinates are not known precisely. it is the equations of motion which lead from Go(q. of course.3 2. g. numerically zero.t).p) .2. in the case of the linear oscillator with energy E given by Eq. Of course.14) and area A of the phase space ellipse given by Eq. . p= p(qo. We consider first the a priori weighting or a priori probability.1 Liouville Equation. and the space spanned by the entire set of canonical coordinates is described as its phase space.e. Instead we assume a case in which we know only that the system is located in some particular domain of phase space.t0.t).Po.3. and E as a function of t). where dr is the difference of proper time given by dudH 1 du dE(t) du — = 1 du . Since Hamilton's equations give a continuous map of one domain onto another.p) to G\{q. We distinguish in the following between two kinds of probabilities.p).Po. i. H(q.. g oc AqAp. du q\ P = —. also boundary points of one domain are mapped into boundary domains of the other. so that Go(qo.p).17a). dE . Let us assume that at some initial time to the system may be found in a domain Go(q. Then .po is a point on Go.Pi are the coordinates of some mass point m.y. dudH s l u . rrr . (1. Probabilities 29 (This is. if qo.e. we have A = (p apaq — J . 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. Probabilities Single particle consideration We continue to consider classical mechanics in which the canonical coordinates qi. (*. (2.E(t)}r = Hence at Relativistically we really should have clu/dr.. This probability is evidently proportional to AqAp.._. du . but partial derivatives of F do not vanish. I .to.dt y/l  qZ/c2' 2. to .' ±=j1V c \dtj' u^. c dt 2 2 w^ du du dr . one obtains Gi with q = q((lo..p). (1..16) For example. i.
— dp A A — In(AqAp)s = — + — = Example 2.— = — Aq.5 thereafter provides an illustration of the a priori weighting expressed in terms of energy E. J0=0 Hence g oc 8n2IdE. dt dt Aq d ( A p ) dt ' Ap Here d(Aq)/dt is the rate at which the qwalls of the phase space element move away from the centre of the element./. dq Aq dq 2 dq Aq dq 2 qH to the right and q to the left. dpdq and similarly dt = — Ap.4.1): d . Hamiltonian Mechanics If g depended on time t it would be dynamical and would involve known information about the particle. (1. and hence g oc %ir2IdE. Solution: Integrating over the angles we have =2TT fe=7r 2frEsin. dt dq and with with Hamilton's equations (2. in view of this independence it can be expressed in terms of the conserved energy E. as is demonstrated by Liouville's theorem in Example 2.30 CHAPTER 2. Je=0 . Example 2. Example 2. this has the same value at a time to. which means.4: Liouville's theorem Show that A q A p is independent of time i. Pi 2 / \. Thus g must be independent of t. dt dq dq dp dp d2H dqdp d2H = 0.15) — an ellipse of area § dpgdp. the (pg.)curve for constant E and $ is — as may be seen by comparison with Eqs. = 2TrIEsm6.eded<t> = 8TT2IE.p^.14) t o (1..5: A priori weighting of a molecule If the rotational energy of a diatomic molecule with moment of inertia / is 1 / 2 . ? « + •sin 2 6>/' 21E 2IEsin2e' in spherical polar coordinates. 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. Hence from the difference: — .
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. (2. leaving the basis of classical mechanics! Hence we set / «w)^ =.t)dqdp = N.e.17) F ^GT^^ 0 Fig. dn — p(q.p. i.p. Since W has the dimension of a reciprocal action.t).20) .Thus we assume a large number of identical sytems. 2. however.p + dp.3. W=p(q. Probabilities 2.t)dqdp= 1. i.1 The system moving from domain Go to domain G\.p.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.e.Po. 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.*). (2.p.2. d d 1 P = JJdqidpi.3 Liouville Equation. by jr = P(9»P. The total number of systems N is obtained by integrating over the whole of phase space. (2.18) With a suitable normalization we can write this / W(q.19) Thus W is the probability to find the system at time t at q. (2. it is suggestive to introduce a factor 2KK with every pair dpdq without.q + dq.
we take into account. 2. t)dqdp p(q.p.p / . The number of points at time t + dt in domain G.p.a t dt + q+dq. how the system moves about in phase space. that in our consideration no additional points are created or destroyed.e.3 and establish an equation for the change of the number of points or systems in G in the time interval dt. In order to derive this equation. p + dp G P O q qndq Fig. 2.P dt p(q + dq. i. t + dt)dqdp — p(q. i.e. The equation of motion for n or W is the socalled Liouville equation. * q Fig.t)dp{ jt Q. we consider the domain G in Fig. p(q.p) denote the velocities in directions q and p — p{q.p) and vp(q.3 The region G. — if vq(q.p. In doing this.p.2 The ensemble in phase space.32 CHAPTER 2. 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.e. Hamiltonian Mechanics We can consider 2irh as a unit of area in (here the (1 + l)dimensional) phase space.t)dp( — I . 2. We are now interested in how n ov W changes in time. t + dt)dqdp.p. i.
p) dq p(q.t)vp(q.t)vp(q.p(q.p + dp) dp = or  However. (2. dH Mq.19).P.t)vp(q. p. vp(q. Comparison of Eq. so that dt ~~dq\P Hence dp) + dp\P ~dq)~ ~~dq~~dp~ + dp~~dq~ ~ { .2. With Eqs. t)vq(q + dq.p.p. 33 Dividing both sides by dqdpdt this becomes p(q.21) with Eq. .4) shows that p and u satisfy very similar equations.p.p.t)vq(q.p)dtdp +p(q.P)]K(Q.p. (2.p(q + dq. p. the probable motion of the system under consideration.t)vp(q.P)].t)vq(q + dq.p.p.P *' % = iH>P} ( 2 . (2.t)^ = 1.W(q. (2.p)dtdq .p) .21 ) This is the Liouville equation which describes the motion of the ensemble or.p.22) The generalization to n degrees of freedom is evident: The volume element of phase space is .p(q.p) .21) we can also write dW(q.p + dp.p.t)} with JW(q.p)dtdp . Probabilities and thus p(q.p + dp)dtdq.t) dt p(q.t + dt)dqdp — p(q.t) dt {H(q.p)=p = 3H —. = K(9.p) = q = g^. .p(q + dq.p(q. t)dqdp = p(q.3 Liouville Equation. (2.p. put differently. .t)vq{q.p + dp.20) and (2. .t + dt) .p).
since no systems are created or destroyed. for i<«>!/«<*p>"w>(^)".t)(^J. Example 2.4).p.p)W(q. p+dp. We now inquire about the time variation of the expectation value (it) of u.that the density or probability W(q. (2.p) the following expression: (u)=Ju(q. i.e. (2.24) since the total derivative is made up of precisely the partial derivatives contained in Eq. p.26) we described the time variation of the observable u(q.24) implies that p is a constant in time. We shall see that we have two possibilities for this.24).e.p) be an observable. (2. We deduce from the Liouville equation the important consequence that ^ M = 0. 2. Hamiltonian Mechanics where is the probability for the system to be at time t in the volume q.4).p. 2. and this means — since these systems are contained in a finite part V of phase space — that dt We have in particular.p). With Eq. (2.34 CHAPTER 2. We define as expectation value of u(q. Equation (2. t) depends explicitly on time t (if determined at a fixed . The first and most immediate possibility is — as indicated .p (cf.po are the initial values of q. i. (.4 Expectation Values of Observables Let u = u(q.1 the area Go is equal to the area G\. q+dq. that dtj\y v if qo. and hence that equal phase space volumes contain the same number of systems. Thus in Fig.
t).0) = u0{qo. . we have Qo = g(q.t). and hence u(q. Sec.33) po = f(q.e. Goldstein [114].Po.p. (2.31) i. 1 (2.24): W(q.0) =u(g(qo.32) In this expression the time t is contained explicitly in the observable u(q.22).Po. we can also employ a more complicated consideration. so that Q = g(qo. We verify this claim as follows.i).34) S e e also H. 8.p) assumes certain values) that depends explicitly on time. (2.po.po.t) = = W(g(q0. Reversing Eq.t). (2.p.28) = Ju(q.^ Solving the equations of motion for q.p0. We expect.t) W(q0. since W oc p is constant in time according to Eq.0) = W0(q0.t)W0(q0.p). ie.8.p. (2.p0)(^^J.t).po.f(qo.t). Thus we can write.t).Po. Then Eq.Po.W(q.po.f(q0. However.t)}(^y.t). (2. (2. With these expressions we obtain for the expectation value (U)Q: (u)o = Ju0(qo. (2.t). that (u) = (u)0.p) = u(q. of course. W is the density in the neighbourhood of a given point in phase space and has an implicit dependence on time t.2.4 Expectation Values.Po) at time t = 0.27) becomes !<»>  /«(*P>!"W>(^)" (2.29).t).Po. (2. (2. where we used Eq. at t = 0.p.p.t).p.p){H(q.29) The distribution of the canonical variables is given by W(q.30) p = f(qo.p) = uo{qo.p. and the time variation d(u)/dt is attributed to the fact that it is this probability (that u(q.po. we can express these in terms of their initial values qo. Observables 35 point in phase space).
(2.t) dp M du\ ° q 'P=f(qo. (2. Hamiltonian Mechanics so that on the other hand with Eq. i.f(q. (2. (2. Inserting Eqs.t) M f 9u\ V °P / P=f(q0.p.35).po)(~ .t).P.) f / u (q0.p.p).M 27Th i .(dq0dp0\n We deal with the partial derivative with the help of Eq.t) d t _ (du\ \ d<i dq / P=f(qo.36 CHAPTER 2. d dt {U)0 = dt.P.P0.(2.) Wo(qo. / ^fdq0dpc .f({H(q. Taking now the total time derivative of (it)o. (2.p.0) u(q.30): duo(qQ. p = f(<j(q. (2.p)})q=g(QO.u{q.t) dB__ (du\ °P \°PS P=f(qo.35) Inserting these expressions into no we obtain uo(qo.p. (cf. Eq.Po.p.0)=u(q.t).f(q.p. we obtain an expression which is different from that in Eq.po. we obtain («)o = / u{q.36) = (2.t).f(g(q.37) as had to be shown.=9(.t) (2.\„.p)W{q.p).t).po.32).39) Substituting this into Eq.28). (2.37) and (2. (2.t).Po.t).po. (240) .p0) = W(q.32).36) and (2.p.f(q.Po.t).u(q.31)) W0(q0.Po.t). (2.P0.P)}).t).t)W "(q0. (2.t)\—^\ = (u).38) we obtain £<«)„ = . .t)J(q.POit)i p=f(qo. [ du0{q0.f(q.p).t) = (2 30) (2.t) .t).p.e.t).30) u0(g(q.t). (2.t) 0H_ 0 q =  {{H(q." V " " ™0' > • • » « » 0 .p.PO.t) v{g(<j(q.25) into Eq.29) q = g(g(q.o.p.30) Moreover.t). Eq.Po)(^^)n.p.t).
p).41) and (2.W} in Eq.p. lim ——uW — 0.42) the relation d . However. (2.t)} . The phasespace integral of a Poisson bracket like I (2.31).p)) is attributed to the probability W(q.p).28). Observables 37 Here we perform the transformation (2..p. from the properties of the Poisson bracket. / d q d p \ n 2irh J (2. uW} = {H..p). we obtain {H. W}. described as "Schrodinger picture".W(q.p)}W(q.34) and use (2.p.28).41) This expression contains {H.u(q. x r „. (2.t)}.. .42) wxtf)" ~dHd(uW) dqi dpi vanishes under certain conditions. Consider {H.44) iji^ioo dpi (which is reasonable since the density vanishes at infinity). u}W instead of u{H.t) dt {H(q.45) in agreement with Eq. This timedependence is described by the Liouville equation dW(q. .p.t) of u assuming certain values q and p.p){H(q. ^ „.t) dqdp\n 2irhJ (2.25) and (2.p). Alternatively we could deduce from Eqs. so that ~d~t <«>0 = J{H(q.W(q.(2. and the time variation of (u(q. dt <«)o v f u(q. we obtain zero after partial integration of I and hence from Eqs.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 (2.28) and (2. The considerations we just performed demonstrate that we have two ways of treating the timedependence: The explicit timedependence can either be contained in the probability W or in the (transformed) observables.4 Expectation Values. In the first case.p. u}W + u{H. (2. (2.45) the Liouville equation.43) ——uW = 0.2. the observable u is treated as a function u(q.
(c) A multiplication of the quantities among themselves is defined.H(q. " This means: Only in the Heisenberg picture dW/dt = 0. (b) As usual a muliplication by a complex number is defined. The time dependence of the expectation values can be dealt with in two different ways. Eq.t) dt du(q.4) of an observable on the other.p0. This formulation deals with observables u representing physical quantities. does not have to be commutative. the probability of the initial values Wo(qo. as we observed. " T h e author learned this approach from lectures of H.po) is assumed. which are described as "Schrodinger picture" and "Heisenberg picture"— all this on a purely classical level but with the use of canonical transformations. i. the reason being that — since qo.** 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.46) We thus also recognize the connection between the Liouville equation.F(«.t). (cf. but does satisfy the usual associative and distributive laws.po. 2.e.rt. as the equation of motion of an ensemble or of a probability distribution on the one hand.p)}.p) It = {u(q.p).po are constant initial values — we have du0(q0. and the equation of motion (2. (2." and the explicit timedependence is transferred into the correspondingly transformed observables Uo(qo.t) dt __ ~ (2J30) ( 4) du0(qo.38 CHAPTER 2. which describe the state of a system. and probabilities W..po. called u Heisenberg picture".5 Extension beyond Classical Mechanics With the above considerations we achieved a general formulation of classical mechanics. Hamiltonian Mechanics In the other case. (2. .p)}. which.4)) ^ = {«(. The equation of motion is then that of an observable. 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. for which the axioms of a commutative group apply. Koppe [152] at the university of Munich around 1964.
B] := AB — BA in the following way: 'i {q. also called "density matrix".p} = 0 • — ^ ~[q.5 Extension beyond Classical Mechanics 39 Quantities satisfying these properties define a linear algebra. therefore. (2. That this is the case. The quantity corresponding to W in quantum mechanics is the socalled "statistical operator'''. and to be able to correlate these with the above classical considerations. (c) TV (u*u) > 0.6) a socalled "commutator''' [A. if u ^ 0. Moreover. so that we consider this next. by Traceu:=Ju(q. (d) Tr(ut.q} = 0 {p. for a phasespace integral we define the word or symbol "trace". also written "TV". it is helpful to introduce already at this stage some additional terminology.26)) (u) = Tr(wW). y:\p.p]=0(247) One verifies readily that the commutator relations are satisfied by the differential operator representations qx > x.49) With these considerations we have reviewed aspects of classical particle mechanics in as close an approach to quantum mechanics as seems possible. We shall then interpret as "canonical quantization" the procedure which allocates to each of the fundamental Poisson brackets (2. (2. ~[q. that its use here implies the essential properties it has in matrix theory. px > in—. which all apply to (2. (2.48) In matrix theory the symbol "trace" has a welldefined meaning. In Chapter 9 we attempt a corresponding approach for classical systems with .) = TV (vu). Eq. (b) TV(cm + j3u) = oTV u + /3Tr v.p)(^y.p] = l.2. Thus we can write the expectation value of an observable u (cf. can be deduced from the following characteristic properties of a trace.48): (a) TV u* = (Tr «)* = TV^.p} = l {q.q}=0. ox In view of our later considerations. Introducing it here assumes.
since gauge fixing (i. However. Hamiltonian Mechanics a wavelike nature. i. .40 CHAPTER 2. Thus we can now proceed to prepare the ground for the extension of classical mechanics into an operator formulation. and. excepting the Poisson brackets.e. a constraint) has to be taken into account. the Poisson brackets require modification to Dirac brackets. electrodynamics. These aspects will be considered in Chapter 27. obtain corresponding results — as one would envisage in view of the expected particlewave duality in quantum mechanics.e. it will be shown that in electrodynamics.
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.Chapter 3 Mathematical Foundations of Q u a n t u m Mechanics 3. 3. if the elements ipi of M satisfy the usual axioms of addition and K 41 . We found that the Poisson algebra permits extensions to noncnumber formulations. of measurable quantities. i.e. = {ipi} is called a linear vector space on the set of numbers I 6 {C}. which turn out to be those of the theory today known as quantum mechanics. Building upon this. 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 cnumber type as classical mechanics. In this chapter we therefore introduce important basic mathematical concepts of this noncnumber mechanics. we can define the Hilbert space as the space of states of a physical system. quantum mechanics: The Hilbert space as the space of state vectors representing the states of a physical system. i. with the canonical commutation relations or Heisenberg algebra defining the basic product relations.1 Introductory Remarks In Chapter 2 we investigated the algebraic structure of classical Hamiltonian mechanics.e.2 Hilbert Spaces We first recapitulate some fundamental concepts of linear algebra and begin with the axioms defining a linear vector space. A set M. and selfadjoint operators in this space as representatives of observables.
2 . In each case n linearly independent vectors are said to form a basis. if every vector ip E M can be associated with numbers Q .2) Vectors ip\. f > 0 if V T ^ O . 2 . (0 : null element and with complex numbers a and j3: a(tpi + tpj) = a{Wi) lipi = = < M). i.~ip2) '• M. (ipi.i = 1.M. fa.42 CHAPTER 3. are linearly dependent.ip2. and n is the smallest such number. E (3. ip2 of this space can be associated with a complex number (V'l.ip2). ^ i = 0 if ^ = 0. .M is said to be a metric vector space or a preHilbert space. such that n ^ = YJCi^i.  ./?eK).1) aipi + aipj. Mathematical Foundations of Quantum Mechanics multiplication by complex numbers. (ipi + ipj) + ijjk = ipi + (ipj + ipk). n is called the dimension of . ( n (3.4) The vector space . A + tpj = tpj + i'i. .^1) = (01. —»• IK. .ipn are said to be linearly independent. if numbers eti. . .^) called inner product.5a) (3. (V>. . . . If n + 1 elements ^ € M. . (3.^i) . (a.3) If all on = 0..ipn are said to be linearly dependent. if any two elements ip\.aiV>i + a 2 ^ 2 ) = ai(ip. so that n 5 > ^ = 0. (3 5c) + a2{ip. n exist (not all zero). • • • . (aptyi.02)* (hermiticity). i=l (3. n. the vectors il)i.5b) .e.tp2. with the properties (a* G IK): (^2. (3. . x M. i = 1.
^ 2 ) =   ^   2 + 2 ^ i . ^ ) + A 2 (V2. if Wi.9a) we start from if) = ipi + \if>2 6 M.is defined by d(ipi. antilinearity in the first component (also described as sesquilinearity. ^ i ) + A ( ^ i .7) In addition the following relations hold in a metric space M.3.^1) 2 h 11 0 <%«e + M .^) = ai*(V'i. The norm of the vector ip (preHilbert space norm) is defined as H:=(^)1/2.^2)1 < H^ill • 11^211 (Schwarz inequality).e.9e) 11^11=1 We restrict ourselves here to some remarks on the verification of these wellknown relations.V') +"2*(V.*h)=\\AH\(38) (3. linearity in the second component. (310) (^2.2.i>2\\2 = 2'i/'i2 + 2'i/'22 (parallelogram equation).rh) (3. \\1p1 + ^211 < HV'ill + IIV^II (triangle inequality)...V>i + A</>2) > 0 . WA+Ml2 = ll^il 2 + ll^2 2 . meaning oneandahalffold linearity). ^ i ) + A * ( ^ . The distance between two vectors if>i. (3. for ip\.2 Hilbert Spaces 43 where the asterix * means complex conjugation. The first two properties imply ( a i ^ i + a2ip2. (3. »2 2 . A V 2 ) + A2V22 2 \m\\ llwl For if)2 7^ 0 we set A so that (^2^l)2 2 IWI + ll^i. ^ 2 )  . for arbitrary A and ip2 ^ 0: (V>i + AV>2. In order to verify Eq. tp2 £ M: 1(^1.V. which we can write 0 < ( ^ i .9d)   V i   = sup  ( ^ i .9b) = 0 (Pythagoras theorem).9c) HV'i + tp2\\2 + \\1p1 . (3.tp2 £ M.)> (36) i.9a) (3. (3.
\\ih\\. ^ e M are said to be orthogonal if (</>!. Examples of metric vector spaces: (1) Let M. Mathematical Foundations of Quantum Mechanics 1(^2. so that   ^ l + ^ 2   <   ^ l   + ^2.9d). (3. We also omit the verification of the following properties of the norm of a vector tpi € M with ^ € X .44 or CHAPTER 3.11). o ^ V i = o. In verifying the triangle inequality we use this result (3.9c).ii) thus verifying Eq. beginning with A = 1 in the second line of Eq.^1)1 Vi 2 + ^2 2 + 2   ^   .12) We omit here the verification of the remaining relations (3. ll^ill + llVdl. a £ K : ll^ill HVill 11^1 +^211 M If for a vector tp e M: > = < = 0. be the set of all column vectors v\ V = (Vi) = I 2 V with complex numbers Vi.* Two vectors i f i . . HIHI (3. (3.V2) IIV'l2 + V'22 + 21(^2.^)1 <IIV>il.9e).13) the vector is said to be normalized. (3. i=l * Not all the wave functions we consider in the following and in later chapters are automatically normalized to 1. for which CO 2 := V^l^l 2 < 00. (3. (3.  ^ l l = (HlMI +   « 2 . V 2 ) = 0 .10): ll^+^H2 = < <   ^  2 + V22 + 2K(Vl. (3.9a). hence verification in each case is necessary.
all squareintegrable functions / which are "almost everywhere equal".e. */ x } = / /o for x = 0.[$]):= / JM /(x)* 5 (x)d 3 *. which differ solely on a set of measure zero.5c)) / ( x ) = 0 = > ( / . In order to avoid this difficulty. / ) = 0. for which the scalar product.5c). (2) Let M. . Then L2 is the space of all these equivalence classes. and one defines addition and multiplication by complex numbers with respect to these classes. Elements of the classes are then called representatives of these classes. i.g)= [ /(x)^(x)d3x.5b). which satisfies relations (3.e. are combined to an equivalence class [/] (with space L2). (3. = C? be the set of all complexvalued integrable functions / ( x ) on 5 C IR3 (in the sense of Lebesgue) for which / VSCR 3 /(x) 2 d 3 x < oo. is defined by ([/]. etc.5a).w) :=J2v*Wi. 2 and so on. with inner product oo 45 (v. (3. i. But this applies also in the case of any function which is nonzero only on a set of measure zero.3. Js the space C? is not yet a metric vector space although for (cf.2 Hilbert Spaces Then we define v + w := (v{) + (wi) := (vi + Wi). With the scalar product (f. The Schwarz inequality is then oo oo \M\<* ^2\Vi\ ^2\Wj\2. Eq. •'^ \ 0 otherwise. (3.
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 finitedimensional 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
(3i?)
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^aV'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*+iwm)!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 squareintegrable 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 —(x2ir, 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
x2px
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)ex2'2 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 nth 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*,jeL*>.
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, aiBi + 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) :=arj0(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)\f00J^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
(339)
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 socalled test function spaced Our aim is here, to provide a frame in which the formal Dirac bra and ketformalism 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. (341) 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 socalled "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) (350) (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)eikx dxf(x)eikx
=
F~lg{k), (3.58)
Joo
roo
 =
= Ff(x).
\Z2TT JOO
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 BraFormalism
4.1 Introductory Remarks
In this chapter we introduce the (initially position or momentum space representationindependent) 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 onedimensional case —
1
/"OO
S(x x') or 6(k k')
= — /
27T
i
dkeikxe~ikx'
(4.1a)
J_00
/oo
= — /
27T J_00
dxeikxeik'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 BraFormalism
1 fn  / dxum(x)un(x)
^
= 6mn,
(4.2a)
JK
i r
 /
^ J7T
dxum(a;)i;ri(x) = Smn,
(4.2b) (4.2c)
i
 \
^
r
dxum(x)vn(x) = 0.
JK
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 Fouriertransformed 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 ketvectors as elements of a linear vector space and bravectors 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 representationindependent formulation we assign to every state of the system under consideration a vector, called ketvector 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
(43)
is also an element of V. As mentioned, the space can be finitely or infinitely dimensional. In the vector space V of ketvectors 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 ketvectors we can associate a dual space V of bravectors, which are written (x. The bravector (% is defined by the linear function <X{«» of ketvectors \u). For a certain ketvector \u) the quantity x has the value (xit), 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 ketvector \u) G V yields again a ketvector \u') G V. On the other hand, a linear functional x i s a n operation, which assigns every ketvector \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 ketvectors \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 bravector (x.§ Then (x\u) is the number that results by allowing the linear functional (x G V to act on the ketvector \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 ketvector is assigned an appropriate bravector. The reverse is not true in general. See e.g. C. CohenTannoudji, B. Diu and F. Laloa [53], p. 112.
4c) (44d) (v\ = J\*(0(m One defines as the scalar product or inner product of \u) E V and \v) E V the cnumber (v\u). Also: (xi = (X2I) if (xi\u) = (X2\u) for all \u) E V. which we write (u\.4a) \v) = J\(m<% and (4. As in our earlier considerations we demand for (v\u) the following properties: (1) (u\v) = (v\u)*. i.5c)). Hermitian Operators A linear operator A in the space of ketvectors V acts such that A\u) = \v) EV . \v) E V are said to be orthogonal.9a): \{u\v)\2 < (u\u)(v\v). V \u) (in the case of £) with infinite norm!).62 CHAPTER 4. Two ketvectors \u). the norm squared. (2) (u\u) > 0. if (u\v) — 0. The dimension of V is the same as the dimension of V.(t.4b) (4. (4.5) 4. (3) (u\u) = 0. It follows that we can now write the Schwarzian inequality (3.e. if \u) = 0 (see (3. Dirac's Ket.3 Linear Operators. 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. if x\u) = 0 V kets \u) E V. that the vector \u) E V is associated with a vector in V.\u) = (u\0*. or (4. This expression is linear with respect to \u) and antilinear with respect to \v).and BraFormalism Furthermore we have: ( x  = 0. Because of the antilinearity the conjugate bravector associated with the ketvector u) = Ail> + A22> is (v\ = Xl(l\ + X*2(2\.
with n)(1) = Aiw)( 1 )+A 2 k) ( 1 ) . Furthermore. if \v) — A\u) i. the space Vi <S> V2 has the dimension . Then one says: (771 results from the action of A on (x. as we can see as follows. Two operators A and B are said to be inverse to each other. Furthermore linearity applies.e.or Kronecker.. if AB = 1.. BA = 1. If Vi has the dimension A/i. Also A = B it for every vector \u) £ V : (u\A\u) = (u\B\u). The operator A. B..4. AB^BA. The inverse of A is written A . and one writes: (V\ = (X\A.or direct product of vector spaces. Let the following bravector be given: (x\ £ V. on bravectors are similar to those on ketvectors. The product of such vectors is written: \u^u^) = \u)^\u)^. denned as a linear operator on V. Finally we define the tensor. Then (x(^4it)) is a linear functional of \u) £ V (since A is linear). also acts on the dual space V. Then the following rules apply: A + B = B + A.\u) = ((X\A)\u) = (X\(A\u)) = (x\A\u). It follows that (r.3 Linear Operators. we have \u^u^) = X1\v^u^) + \2\w^u^) and so on. ltt) = \u). let (771 6 V be the bravector defined by this functional. This does not always exist. The inverse of a product is (AB)'1 =B1A~1. Let \u)(l> be vectors of the space Vi and similarly ?j)(2) the vectors of the space V2. i. the space Vi <S> V2. i. In this way the operations of operators A.e. Multiplication by unity. The product vectors l?^ 1 )^ 2 )) span a new vector space. This product is commutative. defines the unit or identity operator. Hermitian Operators 63 with A = 0 if A\u) — 0 for every vector \u) £ V. 1 and \u) = B\v).e.
\t) G V. Examples of operator relations: (1) (AB\u)(v\Cy = C^\v)(u\B^Al (4. Let A^> be an operator in the space Vi and A^ space V2.7C) («)(u)t = \v){u\. Dirac's Ket. AW\U)W = \V)W. AW\uW)\uW) = \vWUW).g.7b) (4. by replacing in the above the bravector (t by <tflt) (AB)* = B*A\ (cA)* = c*A\ (A + 5) t = A* + B\ (4. also Sec. Assuming now that \v) G V and (u\A G V are conjugate vectors as explained above. then \v) is a linear function of \u). i.64 CHAPTER 4. we obtain the conjugate relation <tAt«> = (u\A\t)* (4. i. Every such operator is then also associated with an operator.7a) (4. in the product space.7d) .A^] = 0. 3.3).6) for all \u). (v\t) = (u\A\t). since AWAW\UW)\UW) = \vWuW) = AWAW\UW)\UW). here designated with the same symbol. If (v\ = (u\A. Next we construct the following scalar products: (t\v) = (t\Ai\u). i.and BraFormalism an operator in the Mi A/2.e. [A^. and (v\ = (u\A. we write \v) = A*\u).e.e. As a consequence we arrive at the following properties (e. the operator A^ is called hermitian conjugate or adjoint operator of A (cf.e. Every operator A^> commutes with every operator A(2\ i. Since we demanded the scalar products to have the property (t\v) = (v\ty.
4. The operator H is said to be positive definite. 65 The linear operator H is called hermitian. i. . if H — H' and antihermitian. The commutator of two hermitian operators is antihermitian: [H. Verification: Since A — A^ and A\u) = a\u).K\.3 Linear Operators. for we had (\u){v\)i = \v){u\.e. if H is hermitian and (u\H\u) > 0 for all ketvectors \u). It follows that a is real. so that (a)(a)t = a)(a. K] = 0. H^K] The separation of HK into hermitian and antihermitian parts is therefore HK=±{HK + KH)+l[H. or (u\A\u)* — {u\A^\u) = (u\A\u). An hermitian operator A has the properties (a) that all its eigenvalues are real.K}. Hermitian Operators (2) The conjugate bravector of AB\u)(v\C\w) is {w\C^\v){u\B]A]. and (b) that the eigenvalues with respect to \u) are equal to those with respect to (u\ and vice versa. as is also (u\u). i. hermitian The product of two hermitian operators is not necessarily hermitian. but only if these commutators commute.e. so that (n^4n) is real. it follows that (u\A\u) = a(u\u). if H = H' and K = K\ then (HK)* = HK only if tfrf = KH. [H. Obviously the quantity \a)(a\ is an hermitian operator. Every operator A can be written as the sum of two such parts: A=\(A + A*) + \(Atf) anti—hermitian . Analogous considerations apply in the case of bravectors (u. An important theorem is the following.K]^ = = [HKKH]* KHHK = = K^H^ [H.
The entire set of ketvectors spans the extended Hilbert space. Ps is idempotent.and BraFormalism Moreover.e. Then every vector \u) £ "K can be written \u) — its) + \us*) with \us) e S. then the hermitian operator defined on this space is an observable. Assuming A\u)=a\u). Ps2 = Ps(4. \u8*) E S*.9a) . This follows from observing that (u\Ps\v) = (u\vs) = (us\v3) = (us\v). which includes vectors of infinite norm.e. as we can see as follows. since a ^ b.8) The projection operator has the following important properties: Ps is hermitian. it follows from A\u) — a\u) that (u\A = a(u\ or the other way round. Let S be a subspace of the Hilbert space !K. Orthogonality: Two eigenvectors of some hermitian operator A belonging to different eigenvalues are orthogonal. (v\A = b(v\. i. i. The operator Ps which projects onto S is defined by the properties: Ps\u) = \us) = Ps\us).9b) (4. but the continuum vectors are normalized to a delta function. i.66 CHAPTER 4. {us\us*) = 0. Ps\ua.e.) = 0. If these form a complete system.u'). so that (v\u) = 0. i. (u\Ps = {us\.Au) = a(v\u) — b(v\u) — (a — b)(v\u). The above theorem remains unchanged.e. Dirac's Ket. (n\n')=5nnl. so that (u\Ps = (us\. A continuous spectrum can immediately be included by passing to the extended Hilbert space. we have 0 = (uArt) — (u. Next we introduce the very useful concept of projection operators. ( n  z / ) = 0 . (4. and let S* be its complement. (u\u') = 5{v . since a is real. a ^ b.
P\p) =p\p). p 2 . Example: The vector \a) normalized to 1 with (a\a) — 1 spans a 1dimensional subspace. (1 — Ps) is projector onto S*. = (p2p)\p). Then 0 = (P2P)\p) and hence p = 0. i..p = 0.. i.9d) Thus the vectors P\u). if the set of vectors 1). The projection \ua) of an arbitrary vector \u) onto this subspace is ita) = a)(aw).4. Then (a\u) = (au a ) + (a\ua*) = (a\ua) and so {a\u) — {a\ua) and hence {a\u} = c(a\a) = c. as can be seen as follows. Ps is projector onto S.e. so that \ua) = (a\u)\a) = \a)(a\u). i. i.3 Linear Operators. (4. . Let p be an eigenvalue of the projection operator P. c = (a\u). iV) is a set of orthonormalized states. where £ is a continuous index. . \u) arbitrary. Set \u) = \ua) + \ua*) with (a\ua*) = 0. Let a continuum state be written £). 2).9c) This property can be seen as follows. Hermitian Operators This property follows from the observation that Ps2\u) = Ps\us) = \us) = Ps\u). Obviously N PN = y]ln)(re n=l is the projection operator onto the A^dimensional subspace SN.e. (1 — P)\u) are orthogonal. Ps2 = PsThe only eigenvalues of Ps are : 0 and 1.e. \ua) = c\a)..1. 67 (4. The quantity \a)(a\ is called elementary projector. If for these (by construction) the projector P2 onto a subspace S2 is P2= J*I f2\odm. so that (Ps2 ~ Ps)\u) = 0.e. Very similarly we can construct operators which project onto the subspace spanned by the continuum states.
See e. which we introduced earlier. 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.g. in this case the spectrum also has a continuous part.r)(ui. Examples 8.e.68 CHAPTER 4. If there are r such vectors. i.114b). are linearly independent). Let \u\).. .114c). Observables are representatives of measurable quantities. ' The projector onto the subspace with eigenvalue Aj can then be written Pi = ^2\ui.6. In the case of the infinitedimensional Hilbert space with a countable number of orthonormalized basis vectors. see Eq. In general it is possible. Degeneracy will be discussed at various points in this text. Dime's Ket. (11. 4. u2).4 Observables Operators which play a particular role in quantum mechanics are those called observables. P = £n)(n.6. The eigenvalues Aj then form a discrete sequence with associated eigenvectors \ui) € !K. Coulomb potential.1 and 11. one says the degree of degeneracy is r . (11.. that one and the same eigenvalue Aj is associated with several eigenfunctions. which are orthogonal to each other (i. the operator P is correspondingly P2\u)= f2\0m\u). 8.r\ "An example is provided by the case of the hydrogen atom.and BraFormalism and hence '6 Numerous properties of the projection operators Pi are selfevident.e. However.. and Eq. Sec. 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.
i.r'} = Sii>6rr>.r\u) i. Applied to an arbitrary vector \u) G "K Eq. which we do not enter into here.r>(ui. the projection of this operator is the entire space. i.10) follows from the fact that i Applying the operator A to the projector (4. The operators Pi are linearly independent.r \ui. 69 i If A is an observable and if the spectrum is purely discrete.r =^ i. or also as subdivision of unity or of the unit operator. is always implied).10) gives the linear combination it) = ^ i. and the eigenvectors \ui.r)(ui.r \(ui. and APi = XiPi. the relation (4.r (4.r) (the convergence. Together with the orthogonality condition.e.e.r)(ui.10) expresses the completeness of the orthonormal system.\i)Pi = 0. The uniqueness of the expression (4.10) This expression is known as completeness relation or closure relation.4. (4. then PiPi' = 0.r\u). i.e. the operator A is completely determined by specification of the eigenvalues A.r\ui/.r.e. we obtain i i i i i. If Aj ^ Aj'.4 Observables The dimension of this subspace is that of the degree of degeneracy. i i (4. (4. PA = y£pi i = Yt\ui. ^ A i P i = A = ^Aiui.r\ = l.r\u)\2.r)(ui. It follows that the norm squared is given by (u\u) = {u\PA\u) = ^2(u\ui.10).11) i. in the case of degeneracy with (ui. Let us set {A .12) .
the eigenvector \ui): f(A)\ui) = f{\i)\ui). Then we denote by \uv) the eigenvector of A whose eigenvalue is \{v). In a similar way we can handle functions f(A) of an observable A. 1/2)) PA = 22 \ui)(u*\ + / dv\uv){uv\ = 1. written in Dirac's notation. now however. (u„zv) = 8{v . One defines as action of f(A) on. i.v'). for instance exponential functions. (4.13) 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).17) previously. we have the corresponding generalizations. as explained earlier. i. (note: {uv\A\uvi) is the matrix representation of A) A\uv) = \(v)\uv).70 CHAPTER 4. If the spectrum of an observable A consists of discrete as well as continuous parts. > rV2 + / dis\(uu\u)f \uj)(ui\Xi + / \(v)\uv){uv\dv. (4. (3.e. 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. for instance.e. Let v be the continuous parameter which characterizes the continuum. Then f(A) = f(A)PA =^/(AOk)^! + / f(\{v))\uv)(uv\dv. The ketvectors \uv) are orthonormalized to a delta function.and BraFormalism This is the expression called Parseval equation which we encountered with Eq.14) . Dime's Ket.
First of all we can rewrite the integral representation of the delta function.B. We shall return to this theorem.C. In the onedimensional case we write {x\x') = S(xx'). For many practical purposes the use of the Fourier transform is unavoidable. where the symbol ip is already indicative of the Schrodinger wave function ij. i. Therefore we want to reexpress the Fourier transform (3. (3. as a formal orthonormality condition. their commutator vanishes.16) Correspondingly we also have a complete set of basis vectors {\k)} of an associated vector space F^. \k)€Fk.15) and (4. form a complete set of commuting observables.. Finaly we recapitulate the following theorem from linear algebra: Two observables A and B commute.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. Extending this we can say: A sequence of observables A. Since both expressions (4. for which the completeness relation is f dk\k)(k\=l..and bravectors. and (b) if they possess a uniquely determined system of basis vectors. In the differential operator representation the problem to establish such a relation is known as the SturmLiouville problem. x) € Fx. if and only if they possess at least one common system of basis vectors. here presented without proof. B]. again later. Eq. i. with energy E.e. 4.15) where the vectors {\x)} are to be a complete set of basis vectors of a linear vector space in its position space representation Fx.59). (4.17) The Fourier transform provides the transition from one representation and basis to the other. (4. (4.5 Representation Spaces and Basis Vectors We began by considering ketvectors \u) in which u is a parameter of the (energy) spectrum. which is discrete in the case of discrete energies and continuous in the case of scattering states (with continuous energies).e.. In the following we consider more generally ketvectors \ip) as representatives of the physical states.58) in terms of ket.4. [A.. if (a) they all commute pairwise. / dx\x){x\ = 1.16) represent subdi . which all commute with one another.
we can rewrite Eq.Fk.59) or Eq. i.22) All of these expressions can readily be transcribed into cases with a higher dimensional position space.15): 5(x . (3.19) Comparison with the orthonormalized system of trigonometric functions (4. this expression has to be identified with — / dkeikxeikx\ i.21) provides the ketvector \ij)) in the A. The Fourier representation ij){x) = )= V 27T J Ieikx^{k)dk (4. 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). i.72 CHAPTER 4.L e " * * ' = {x'\k)\ V27T (4.e. (4. . (4. Dime's Ket.and BraFormalism visions of the unit operator.1a). called wave function. <a#) = [ dk{x\k)(k\ip). (x\k) = ^=eikx.space representation. which are representatives of the states of our physical system. il>{x) := (x\i/}) : "K * C. ${k) := (k\if>). Obviously we obtain this by inserting a complete system of basis vectors of Fk.18) According to Eq.20) The representation of the corresponding bravector (^1 G IK in the position space Fx is correspondingly written (il>\x) = {x\ipy.x') = (x\x') = (x\t\x') = (x\ f dk\k)(k\x').e. Rather \x) and \k) serve as basis vectors in the representation spaces Fx. (4.e. (4. shows that these expressions are the corresponding continuum functions (the continuous parameter k replaces the discrete index n). V 27T (k\x') = . The vectors \x) and \k) G F are not to be confused with the vectors u) or \ijj) G !K.2a) etc. (4.
Chapter 5 Schrodinger Equation and Liouville Equation 5. 73 . if we do not take into account the complementary part of the universe. Schrodinger [244]. Thus we distinguish between socalled pure states and mixed states of a system and thereby introduce the concept of density matrix and that of the statistical operator. with the complementary system or *E. 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). 5.p.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. which is the quantum mechanical analogue of the classical probability density p(q.2 The Density Matrix We shall establish a matrix p.t). We mentioned earlier (in Sec.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. the calculation of which for specific cases is the subject of Chapter 7. This will then also clarify the role of p as an operator in the quantum mechanical analogy with the Liouville equation. 1.e. We postulate the (timedependent) Schrodinger equation* (with the Hamiltonian as the timedevelopment operator) and obtain the quantum mechanical analogue of the Liouville equation. i.
74 CHAPTER 5.3a) and N J  a ) ( o  = l in the complementary subspace.b. where T means transpose: p = CTC*. 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). i. \a). The states \a) £"K'on the other hand are the corresponding states of the complementary part of the universe. \i).j = where J2 (J\A\i)5baCatCbj* = X>'l4*>/0« a. called the density matrix. The states \i) £ "K. T .e. p" ( C C * ) t = CTC* = p.e.. Schrodinger Equation and Liouville Equation the measuring instruments. The difficulty we have to face is the impossibility to specify precisely the stationary state of our limited system. a (5.3b) The quantum mechanical expectation value of an observable A. i.. \a). (52) (5. i eH ®^' (5J) We assume that the states are orthonormal and in their respective subspaces also complete... of an operator that acts only on the states of the system we are interested in. we have (a\b) = 5ab. are called "pure states".^ Since p is hermitian.b. but also of the states of the complementary set of the universe.. The actual and real state of a system is then a superposition of two types of states. is then given by (A) := {MA\^) = Yl U\{b\A\a)\i)CaiCbj* a. we have to deal with a socalled "mixed state".i.. as well as 2~] «)(* = 1 in the subspaceof pure states i (i\j) = 5ij.i. which in Dirac's notation we can write as = J]C a »i) a.. that is.j ij (54) Plj = (i\p\j)^YCiCJ*a (55) Here p is an hermitian matrix. the matrix can be diagonalized by a transition to a new set of states ^The hermiticity can be demonstrated as follows. where i represents collectively all quantum numbers of the state of the system under consideration. These are the quantum mechanical states which are also referred to as microstates. (a\i)=0 = {j\b).
e.7) p = ^2\ii)(i. The operator Pi = \i)(i\ projects a state \<p) G "K onto \i).\p\i")(i"\ = Y. then (cui/ '^Zi^i)Z is the number of ensemble elements in the pure state i). or (5.^'\i'^i'\> (58) since then (f\p\j") = ^Ui'ij'li^ii'lj") v = ^Ui'Si'fSi'j" v = UjiSj/j. P ^ ) = i)(#) = (#)i) (5. seen macroscopically of Z copies of the system).* We can specify the state of our system. . i.e. but occupy in general different microscopic. in the position state representation or xrepresentation. The ensemble can be represented as a set of points in phase space.2 The Density Matrix 75 \i') 6 "K.. In this sense we have p(x'. Thus the real system is in the pure state \i) with probability uii/ J2i Ui.3a). i. in a specific representation. where x represents collectively the entire set of position coordinates of the (particle) system. Gibbs introduced the concept of ensemble. This transition to a new set of basis vectors in which the matrix becomes diagonal.3 with Pi^YtCaiCaj*. which are all subjected to the same macroscopic boundary and subsidiary conditions. (5.4) we had the expression (A) = Y/Pij(j\A\i) i. which today is a fundamental term in statistical mechanics. can be achieved with the help of the completeness relation of the vectors \i): \i') = J2\i)(i\i') i. a complete orthonormal system with properties (5. (see below) eM.6) (i'\p\i") = Wi'Si'vi = real. quantum states. a projection operator. a (512) *In his reformulation of the statistical mechanics developed by Boltzmann.ji(x / )i*(x).9) is diagonal.e.5.) = YJc t .. If the ensemble corresponding to the system under consideration consists of Z elements (i. This ensemble is interpreted as a large set of identical systems.g. In the following we write simply \i) instead of \i'). as we saw earlier. e. This projection operator represents a quantum mechanical probability as compared with the numbers ui{ which represent classical probabilities. (5. (5. as we discussed previously.10) and is therefore.x) := (x'\p\x) = yju. i i (511) In Eq.j(x'i)(ia.e. Thus the pure states \i) form in the subspace of the space of states which is of interest to us.2) and (5..
we have Tr(p) = l. (A. 3 (5. On the other hand. as claimed by Eq.4).18).3 = pvv = uv. i t (5. In particular for A = 1.) := (mm = (#') W> = l(#')2 > 0. according to Eq. (5.16) (5.14) which we can write (A) = Tr(pA). Hence Wj/ > 0.17) Hence the sum of the classical probabilities is 1. We recall that the elements of p originate from the coefficients Cai in w = Y^cai\a)\i) = Yl ( E ^ M V ^519) .13) {A) = ^(j\A\i)(i\p\j) ij = 52{j\Ap\j).18) Substituting this into Eq. We can also show that Ui > 0. (5. (5.5) above) (i\p\j)=Pij. (5.14) we obtain (Ai') = ^2(j\i')(i'\i)(i\p\j) ij = ^Sji'Su'Pij i. (5. Eq.15) 5^0'l5^w«l*X*b') j i = 1 » or J^wi(it> = ^ W i = l. Thus Tr(j>ii)<i)=l. To this end we set A —>• Ai' := \i')(i'\ (no summation over i'). (5.4)) (5.76 CHAPTER 5. Schrodinger Equation and Liouville Equation We now define the operator p by the relation (cf. Eq. (5. Then the expectation value of the observable A is (cf.
We observed above. ^2ui\i){i\.22) In the following (x\i) is always to be understood as (x\i(t)}. (Question: Is the state of the universe a pure state?) 5. one says. If all ui but one are nought. i. The effect of the interaction of the system under consideration with the surroundings. We see this more clearly by going to the position space representation. and in the other states with probability zero. defines the statistical operator. that these are actually not sufficiently general. (5. since they do not take into account the (interaction with the) rest of the universe. Thus we have a system in a pure state. i. the only remaining one is 1.t).t) we now write tp(x.x'). t) p= i The expression (5.e.t) = {x\i).8). . represents a quantum mechanical probability.e. i. not in a mixture of states.e. in all other cases the state of the system is a mixed state.t) and instead of (x\Pi\x) we now write p(x. In the following we are mostly concerned with considerations of systems in pure states. in which the numbers Wj represent classical probabilities and the operator \i)(i\ quantum mechanical probabilities or weights.3 The Probability Density p(x. thus enters these states. but we see here.23) Instead of i(x.§ {\x)} is a complete system of basis vectors in the position representation space Fx with dx\x){x\=l. (5.t) = \^{x. Equation (5.24) s (x\x') = 5{x . (5. so that p{x. that — different from u>i — P. the system is in a pure state. Then {x\Pi\x) = (x\i)(i\x) = \{x\i)\2.21) p=\i){i\=Pi. by representing the vector \i) by the wave function i(x. which is contained in the coefficients Cai.5.3 The Probability Density 77 We see therefore that p defines the mixed states.t)\2.20) expresses that the system can be found with probability 1 in the state \i). We now consider the latter expression. For the case n we have TJ=J (520) (5.
and hence the integral over all space dxp(x.e. (5. (5. it is not "derived"). i. t). (5. (5. we obtain ih % = ^ E ^ w o i = ^E^ij)oi+^Ewi(^iJ))oi dt +ihYJ^\J){jt{J\) (529) "More precisely this equation should be called Schrodinger equation of motion (since H is fixed.78 CHAPTER 5.27) is iHJ\ jt= m . The generalization to a threedimensional position space is selfevident: JK 1 / p(x. the Schrodinger equation is always postulated in some way.28) Differentiating Eq. Hence we postulate: The time development of a state \j) E "K is given by the equation ih^\j)=H\j).27) is known as the Schrodinger equation. can be found with probability 1 in the space R1.3.8) and multiplying by ih. and the timeindependent equation should correspondingly be called Schrodinger wave equation (i. (5.t)\2. the expression (5. but the linear operator states are moving — even if the system is observably stationary). We now want to obtain the quantum mechanical analogue. We therefore postulate first an equation for the time dependence of a state vector. or the wave function tp(x. We shall convince ourselves later that the postulated equation is sensible (in fact.^ The equation which is the conjugate of Eq.t) = \rp(x.26) 5.e. with respect to time. Equation (5. To this end we have to differentiate the density matrix p. t h e equation of motion with states represented in a system for which the q's or here x's are diagonal).27) where H is the Hamilton operator of the system.t) = / dx{i\x)(x\i) = (i\i) = 1. . (5. which is chosen in such a way that the desired analogy is obtained.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.8).22) can also be written (xi) 2 = (i\x)(x\i). This differentiation requires the time derivative of a state vector \i) E "K. Schrodinger Equation and Liouville Equation The relation (5.25) J Thus the particle whose state is described by \i). 2.
such a formulation is described as Schrodinger picture. (5. M\/n\ (530) where in statistical equilibrium dt £^">01=03 (5. (3. Eq.21).4 Schrodinger Equation and Liouville Equation 79 Inserting here Eqs. (5. (5.24)). (5. both uji and i)(i). We thus obtain the operator form of the Liouville equation.28) we obtain (see also Eq.31) below) 3 j 3 = y £u>J{HPjpjH)+thyE^m\ dt 3 = [H.27) and (5. The correspondence to the classical case is obtained with the substitution h J< > <5'33> and dp/dt = 0 (cf.p}. With Eqs. (cf.32) with its classical counterpart (2.32) Comparing Eq.e. (2. dp/dt ^ 0. not the partial derivative. we observe that here on the left hand side we have the total time derivative.28) the states are considered as timedependent. the relation ih^ = [H.34) .27) in the position space representation: ih(x\\j) = (x\H\j) = {x\H(pi.5.p} + ihJ2^\J)(J\.e. p also contains "quantum mechanical probabilities" (i. (5.31) 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). We can now consider the Schrodinger equation (5. The Schrodinger picture and the alternative Heisenberg picture will later be considered separately.Xi)\j). i.e. The reason for this is that in addition to the "classical probabilities".27).37)) (5. Eq. i. (5.
Feynman [94].36) in in The initial wave function or timeindependent wave function ^(x.Xi){x\En) = En(x\En). n (537) This equation is described as the timeindependent Schrodinger equation.80 CHAPTER 5.and temperaturedependent Green's functions. a (5.4.1 Evaluation of t h e d e n s i t y m a t r i x As a sideremark with regard to statistical mechanics we can make an interesting observation at this point.38) (x\j)t=o = Yt(x\E^(En\J)t=o.8) for p.35) can be written'! ip(x. namely that the density matrix satisfies an equation analogous to the timedependent Schrodinger equation — not. but with respect to the parameter /3 = 1/kT. with respect to time t. which we shall need later.e. i.e. (5.4. (5.. **R. however.t) = exp ip(x. „ / ith—. in Sec. 0) or \il))t = \il>)t=o exp (5. 10. appearing in the Boltzmann distribution. we write V>(x. in p = ^2i^i\i}(i\. With the help of Eq.36) however.0): where H(ih—. operator" the exponentiated oper . In expression (5. For a mixture of states \i) of the system under consideration caused by the rest of the universe we have Ui ^ 1. to) = exp[—iH(t — to)/H\. P.t): > ih—</>(x.t) = Hld . 5. we can replace these states by corresponding timeindependent states.. t ) .35) Here we can look at the Hamilton operator as a "time generator3''. Schrodinger Equation and Liouville Equation or with (x\j) — ip(x. we shall describe as "time development ator given by U(t. i. it is plausible to refer to this equation at this stage. 0) can be expanded in terms of a complete set of eigenvectors \En) of the Hamilton operator H. (5. this is an application of the BakerCampbellHausdorff formula which we deal with in "Later. or H\En) = En\En).** In view of the close connection between time. .. the state \i) is still timedependent. since the timedependence cancels out (the exponential functions involve the same operator H but with signs reversed. since the solution of Eq. T meaning temperature.x 1 ^ ( x .
i.x')^p(x.9) with En > nhu) L0icxePE\ so that Yliui becomes = f3 = l/kT. i ( 5 . we can rewrite p.4 Schrodinger Equation and Liouville Equation 81 Example 5.^ n tne position space representation Eq. x') = ^ i LUiMxWix').39) Without proof we recall here that in the socalled canonical distribution the weight factors uji (similar to those of the Boltzmann distribution) are such that (cf.45) since on the one hand •PEi i i i and on the other hand . Hence we have (with (Ei\Ej) = 5^) in what may be called the energy representation p = YJ"i\Ei)(Ei\. (5.x'(3) = Si e ~^W*V). (5. the expression _Zie^\Ei)(E>\ also as (see below) e0H (5.39) therefore (x\p\x') = J2"i(x\^)(Ei\x'). Eq.40). we obtain p(x.44) r (TrePH)' (5.e. or ^ = ^ e 0Ei _m.42) Inserting here Eq.5. (5. (543) Since H(f>i(x) = Ei<f>i(x).40) 1. (1. (5.41 ) or with 4>i(x) := (x\Ei) this is p(x.1). i (5.
(5.45).x'. [B.L [B. the following relation holds: exp(A) exp(S) = exp ( A + B + i [A. (5.35). . Tr(pN) = / dxpN(x.The quantity pjv is obtained from the solution of the Schrodingerlike equation (5. (5.50) Differentiating this equation with respect to j3. (3).(3).x'.49) In the position or configuration space representation Eq. (5.X'.82 CHAPTER 5.x'. (7. [A. (5. (5.tt tt For a more extensive discussion see R. PN(0) := (EilpNiP^Ej) = (Ei\ePH\Ej) =e"^^ (5. we obtain = HxPN{x. Equation (5.1: Baker—Campbell—Hausdorff formula Verify that if A and B are operators. Schrodinger Equation and Liouville Equation so that p is given by Eq.45) and (5.4> = ^ g ^ .46) In the energy representation this expression is pNij(P) with PNij(P) = 6ij. (5.51) where the subscript x indicates that H acts on x of PN(X. M.47) with respect to (3.45) without normalization as PN(J3) := e~pH.55)).48) = 1. (5. We now rewrite the factor exp(—(3H) in Eq. This solution is the Green's function or kernel K(x.46) is pN(x. Example 5. With Eqs.(3) := (x\pN(P)\x') = (x\e~0H\x'). (5. (5. we obtain 9pN gpP) = SijEie^ = EiPNij(J3).46) we can write the expectation value of an observable <^Tr<„. Differentiating Eq. A}} + • Solution: The relation can be verified by expansion of the left hand side.51). whose calculation for specific cases is a topic of Chapter 7 (see Eq./?). B}} .x. B] + L [A./3).52) Hence this expectation value can now be evaluated with a knowledge of pN. Wilcox [284].47) (5.51) is seen to be very similar to the Schrodinger equation (5.
can — in fact — be quantized in quite a few different ways. Kim and M.51) to (8.. its associated space of state vectors also provides the best illustration of a properly separable Hilbert space.53)). like the Coulomb potential. (8. This is a fundamental topic since the harmonic oscillator. Noz [149]. However.the present form of quantum mechanics is largely a physics of harmonic oscillators.Chapter 6 Q u a n t u m Mechanics of t h e Harmonic Oscillator 6. * Comment of Y. S. E. Since the spectrum of the harmonic oscillator is entirely discrete.1 Introductory Remarks In the following we consider in detail the quantization of the linear harmonic oscillator. 83 . 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. 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. that the oscillator. its quantization here in terms of quasiparticle creation and annihilation operators also points the direction to the quantization of field theories with creation and annihilation of particles. we shall see on the way." Quite apart from this basic significance of the harmonic oscillator.. The importance of the harmonic oscillator can also be seen from comments in the literature such as:* ". 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).
MvT« 7=r)^(VTvir>' + («> 'Note that we assume it is obvious from the context whether q a n d p are operators or cnumbers. Also observe that H does not contain terms like pq or qp. linear harmonic oscillator the Hamilton operator is given by the expression XJ 1 2 . Quantum Mechanics of the Harmonic Oscillator 6. This can be done in a number of ways.1) for the hermitian operators q. the boundary conditions on the solutions ip(x) would be square integrability over the entire domain of x from —oo to oo.2) which for mo = 1/2. we can also proceed in a different way. hence we do not use a distinguishing notation. This equation would then have to be solved as a second order differential equation.^ The first problem is to determine the eigenvalues and eigenfunctions of H. which is called Weyl ordering. When such a term arises one has to resort to some definition.84 CHAPTER 6. in the present case + [EmooJ2x2)ip = 0. of which the only nontrivial one here is [q. p^ih—. h = l.to = 1 reduces to the parabolic cylinder equation. ox and then to the timeindependent Schrodinger equation Hip(x) = Eip(x). so that there is no ambiguity due to commutation. like taking half and half.2 The OneDimensional Linear Oscillator In the case of the onedimensional. . 1 2 2 The quantization of this oscillator is achieved with the three postulated canonical quantization conditions.' ^mgLOh by setting and *~. For instance we can go to the special configuration or position space representation given by q^x. 2mo dx \ 2 2 (6. However.p. To this end we introduce first of all the dimensionless quantities rriQUi fl q. which is — in the first place — representationindependent.p]=ih (6.
A^] = 1.12) in order to obtain the following expressions: [tfA.6. C] = A[B. (6.9) We observe first that if \a) is a normalized eigenvector of N with eigenvalue a.13) ( ^ A ) i t = A\A^A From Eq.8) The eigenstates of H are therefore essentially those of N := ^ f A (6.14a) we deduce for an eigenvector \a) of A^A. A*} = A^A.1 ) .7) (6. (6. for ^ A  a ) = a\a). (6. (6. (6. i\/2 Inserting these expressions for p and q into H we obtain H = hiu{A^A + AA*) = hi* (A^A + ^ .34d): [A. i.B}C + B[A.1) we obtain immediately the commutation relation [A. (6.14a) (6. Prom these relations we obtain ( ^ = ^ . q are hermitian).e. + 1).5) Reexpressing q and p in terms of A. C}B. A] = [A\ A]A = A. i.14b) [A* A. C] + [A.10) then a = (a\A^A\a) = \\A\a)\\2>0.2 The OneDimensional Linear Oscillator 85 (p. (6. .BC] = [A. if A*A\a) = a\a).C]. and [AB. We now use the relation (3.11) Thus the eigenvalues are real and nonnegative. (6. With the help of Eq.e. A^] = A\ (6. A\ we obtain h A + A^ rriQLU y^ (6.6) and p = y/motoh AAi =^.
or \\A\a)\\ = Va.2A)\a) 2 (a .A)\a) A{AAU . or Ata) = v ^ T T . (6. A 2 a) is eigenvector of A^A with eigenvalue (a — 2).17) (6. (6.11) this is not possible.14b) : (AiA)A*\a) = A\A^A + l)a) = A\a + 1) a) = (a + l)A^\a). since this equation implies that the eigenvalues cannot be negative.16) This means. If we continue like this and consider the vectors An\a) ^ 0 for all n.e. (6.  a) is eigenvector of A^A with eigenvalue (a + 1).19) i. unless A 2 a) = 0. we find that A n a) is eigenvector of A^A with eigenvalue (ct — n). Similarly \\A^\a)\\2 = (aAAta) = (al + A^A\a) = (a + l)(aa) = a + 1. (6. (6. we must have (a) Let \an):= Ana)^0. Quantum Mechanics of the Harmonic Oscillator (A*A)A\a) = A{A*A . but (b) An\n>\ llA An+1a)=0. Similarly we obtain from Eq. However.l)\a) = (a . Thus for a certain value of n > 0.l)A\a). This would mean that for sufficiently large values of n the eigenvalue would be negative.2)A a).21) . (6.. unless A^\a) = 0.l)A\a) = A(A^AA . The norm of A la) is Aa) 2 = {a\A^A\a) = a ( a  a ) = a. in view of Eq.86 the relation CHAPTER 6. Next we consider the vector A 2 a).15) Thus A\a) is eigenvector of A^A with eigenvalue (a — 1). (6.20) .2A)\a) = A(Aa .18) = '= a ) A(A^A . unless A\a) = 0. In this case we have (A*A)A2\a) (6 (6.l)\a) = A(a .
17) a by (a — n).. (6. From relation (6. For a — n = 0 we deduce from (b) of (6. (6.5) to shift operators A to the right which then acting on 0) give zero.n ) = " A»a>=» ' = A«a)F = 1 .2 The OneDimensional Linear Oscillator be a normalized eigenvector of A^A with eigenvalue (a — n).25) (6..e. v.20) we obtain from this that the right hand side vanishes.27) According to (b) of (6.22) With relation (b) of Eq (6.18) we obtain for a = n: p t  n )   2 = n + l. (6. Moreover. i. In the following manipulations we always use the commutator (6.28) . But A 2 A t 0) = A(AA^)\0) = A(A^A + l)0) = 0. \ 9 \ I / II 4 T ) .' " /. so that A^\n)^0 In particular we have A^\0) ^ 0 and II^IO))!2 = (OIAA+IO) = (0\l + AU\0) = (00) = 1.n  a . \ I 2 2 x 87 Replacing in Eq.I n U .20) that A0) = 0. that an = 0. (6.6.23) Hence the eigenvalues of the operator iV := A^A are nonnegative integers. I „ . The state 0) is called ground state or vacuum state.26) <0A4 t AA t + A4 f 0) = (Q\2(A]A + 1)0) = 2. (6.20) we have for a = n = 1: A2\1) = 0.24) This is a very important relation which we can use as definition of the state vector 0). or a = n > 0. (6. we obtain an = \\A\an)\\2(6=) An+l\a) \\An\a) (6. so that (a(A n )tA n \a) __ \\An\a)\\2 ( a . HAU+IO)!!2 = = (O\AAAWI\O) = {O\A{A^A + I)A^\O) for all n.
26) the equality l) = A t 0).30) (arbitrary phase factors which are not excluded by the normalization have been put equal to 1). (6. (At)2]A+ + (A*)2[A.Al] = 2A^A^ + (A^)2 x 1 = 3(A^)2.29).e. (6. (A*)2} = [A. In general we have \n) = ^(A^n\0) (6.11) we have [A. But using Eqs. (6. A^A* + A*[A. as we can see as follows. (6. The states \n) thus defined are orthonormal. 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) 20) ^ 0. Similarly we find 2)oc^Ut0). (A^)3] = [A. According to Eq.32) we have (nm) = . and in general [A. i. (6.34) (6.5) and again (6.88 and CHAPTER 6.29) Hence we obtain and in view of Eq. l>ocAt0).31) (6. and in view of Eq.(6.^ ( O l A ^ n o ) .(A*)n] = n(Ai)n\ (6.11). we have 2> = ^=dAi\0). A^} = 2A\ and [A.32) (6.33) . (0^ 2 (At) 2 0) = 2.
6. (6. . whose number is given by the integer eigenvalue of the number operator N = A^A. 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 .1 0).40) (6. ldnm = n\5nm. . .41) The contribution hu/2 is called zero point energy.33).1 (A + ) m . . The terminology is. (6.34) and (6. however.35) Inserting this result into Eq. i. n = 0 . l . we obtain (n\m) = 5nm We also deduce from Eq. in quantum mechanics.1 ) . •(» (6.L ( A t ) n + 1  0 ) = VnTT\n Vn! and with Eq. . .8) we obtain therefore H\n) = hu (A^A + ]. chosen in close analogy to the postulated "second quantization relations" of field theory. (6.J \n) = hw (n + i J \n). we do not have creation or annihilation of any real particles as in field theory (hence the word "quasiparticle"). (6. Here.38) (6. .35) the operators A^ and A are called respectively raising and lowering operators or also shift operators. With Eq. In view of the same properties A is also called annihilation operator and A^ creation operator of socalled u quasiparticlesv.1) = n\n). the eigenvalues of the Hamiltonian H are En = hw\n+\+ i ) .'n\ and = v n ![ ( A t ) M i= + (6.e.1) (6.37) n^)711}^) = v n  n .32): rf\n) = . In view of the properties (6.34) A\n) = ±=A{Al)n\0) . (6. 2 .39) A^A\n) = y/nA^\n .36) + 1) (6.2 The OneDimensional Linear Oscillator so that {0\An(A^)m\0) = = {0\An1A^mA\0) + (0\An1m{A^)ml\0} 89 0 + m(0A n .
e..39)). ( 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 / (6.3 The Energy Representation of the Oscillator The energy representation. also called Heisenberg representation. other elements zero. (6..43) = = y/n(n'\n — 1) = y/n5n^ni...90 CHAPTER 6. (6.'\n) i'\A*\n) (n + l\A'\n) and similarly (n'An) (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 ..46) v^ . Vn + 1. Quantum Mechanics of the Harmonic Oscillator 6. (5. The representation of the Hamiltonian in this energy representation is {n\H\n') = En5nn. (6.7) the energy representation of the operators q and p.6) and (6.39): {n\A. n = 0. Eq.45) Correspondingly we obtain with Eqs.g. A^ from the relations (6.38) and (6.2. other elements zero. 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. and — in fact — we have used this already previously (see e. is defined by projection of operators on eigenfunctions of energy. (6.1.42) We can deduce the energy representation of the operators A. i.44) yfn. En = hu(n + ).
Eq.4 The Position Space Representation 91 It is an instructive excercise to check by direct calculation that Eqs.48) Applying from the left the bravector (x\ and remembering that (x\p\<f>) = we obtain ih—(x\(j)}. (6. Hence ( * Recall J™ dxe™ * ? 2 2 2 1/4 W \ i/4 \ em^x*/2h_ (g 5 Q ) = yfln/w . (6.e.1) and (6.5) are also satisfied as matrix equations. The ground state wave function (/>o(x) is defined by Eq.6. (6. Correspondingly the position space representation is given by the wave function <t>n(x) := (x\n). A\Q) = 0. 6. This is a simple differential equation of the first order with solution (x0) = The normalization constant C is determined by the condition* OO /C 1 = (00) = / / dx(0\x)(x\0) = \C\2 I dx(0\x)(x\Q) = \C\2 / OO J —< emou}x2/hdx = \C\2 mow' so that We choose the arbitrary phase 6 to be zero.3)) ^ (q 2n \ + (6. mow / (6.47) LP 0) = 0 . 2 . i.32). by (cf. (6. ™ou(x 2h \ + J _ d ] m = 0 (649) mQUJ dx CemouJx2l2h.4 The Configuration Space Representation We saw that the eigenstates are given by Eq.24).
51) \l 2h {X q \ m0uP)\l 2h \X m0u. (6.L ( x  ( A t ) »  0 ) . ) e"^2. so that <t>n(x) = ( . Quantum Mechanics of the Harmonic Oscillator This is therefore the ground state wave function of the onedimensional harmonic oscillator. Vn! Now {XlA (6. i. In order to obtain the wave functions of higher states.92 CHAPTER 6.0 = ( . it suffices according to Eq. they are called Hermite polynomials.56) < ^ = We have as an operator identity the relation .2 n n! The functions Hn(£) are obviously polynomials of the nth degree. (6.54) (6 55) i/4 rzri7r 1 / 4 vo.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 .32) to apply the appropriate number of creation operators A^ to the vacuum state 0).l ) > n ( x ) .l ) n f f „ ( 0 . These polynomials are real for real arguments and have the symmetry property # n ( .e. c/>n(x) := (x\n) = . (6.
l ) n e ^ In particular we have #o(0 #i(0 #2(0 #3(0 tf4(0 = = = = = 1.2)m = (±t)m for some function VKO. d£" (6. (6. Magnus and F.12(20 .58b) by W. 93 . 2£.6. (6. gain as an operator relation we have de = een± = een da ee/2±_^ee/2 d£ 1 u ^ d? dt.e.It follows that 0en(Pe/2^Lez 2 n\P?i2 — c2 d d£ ? = 2f. e. (202"2.4 The Position Space Representation i.58a) ' T h i s definition — apart from the usual one — is also cited e. (2036(20. ( 2 0 4 . . p. — = n+ d ~^ i 2 + d 2 de) ' 1+e ~ (6.12.en_ dt. Oberhettinger [181].g.57b) • Generalizing this result and inserting it into Eq.54) we find that (observe that there is no factor 2 in the arguments of the exponentials!)§ Hn{i) = ( . 81.
He'n{£).59c) The recurrence relations satisfied by these Hermite polynomials (the second will be derived later in Example 6. £ 4 .94 CHAPTER 6. (6. Quantum Mechanics of the Harmonic Oscillator The differential equation of the Hermite polynomials Hn(£) defined in this way is ( ^ 2 ^ + 2n )ff"(O=0.3) are (as given in the cited literature and with a prime meaning derivative) Hen+1(0 Hen(0 He'n(0 = ZHenifi) . = ffen+itO+ntfeniCO.59d) The normalization of Hermite polynomials is given by the relations: oo / d£ Hn(OHm(Oe~e oo = 2 n n\^5nm. = nHe^iO.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. Mathematicians often define Hermite polynomials Hen(^) as^ Hen(0 Then He0(O Hex{0 He2(0 He3(H) fle4(£) = = = = = 1. £33£. Magnus and F. (6. £2l.^ + n) He n(0 = 0 (6. (6.(e^/2) or Hn(0 = 2n/2Hen(V2ti). . Oberhettinger [181]. 80.59b) = (l)^2/2^.60a) and I d£ Hen{Z)Hem{Z)e?l2 = n! V & „ m  (6. (6.59a) The differential equation obeyed by the Hermite polynomials Hen(£) defined in this way is (^2 . pp.6 £ 2 + 3.60b) J—< "See for instance W. 81. (6. £.
62) Joo The following function is generally described as generating function of Hermite polynomials Hn{£): F(S. Oberhettinger [181].64) n\ See e. p.. Magnus and F.s)=F(£.^ = e^/42~n/2Hn(aV2). Their normalization is therefore given by OO /"OO / OO J —OO z (6.61) e?l*2nl2Hn{Z).. 2(£  s)£2& s=0 s=0 ~ ds = m sf2]e?= •««) : S=0 2(2e2l) = (2e)22 = ^2(0. W.l ) ^ e^HeniO 2 / 4  . s) as a Taylor series. Thus the generating function can be written F(Z.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£) = ( .6. . (6. F(S.s) = J2 n=0 (6.63) The meaning is that if we expand F(£.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. 93. and so on.s) = eM?(Zs)2}(6.g.
Quantum Mechanics of the Harmonic Oscillator Fig.1. We also see from the eigenvalue t h a t the energy of the lowest (i. W i t h Apx = moAvx we have = moLoAx Ax and thus h/2 Apx h/2 mQuAx Ax I h/2 . 6. In classical mechanics the energy of the oscillator can be zero. But this would mean t h a t b o t h x{t) and p{t) would assume simultaneously the "sharp" values zero. in order to estimate the lowest energy of the harmonic oscillator. for x(t) = 0 and p(t) = mox(t) — 0. 6. In fact we can use the uncertainty relation AxApx ~ H/2. ground) state is fko/2. We see t h a t the lowest state is symmetric.e. i.96 CHAPTER 6. i.1 The comparative behaviour of the lowest three wave functions of the harmonic oscillator.e. Furthermore we observe t h a t the wave function of the nth state has n zeros in finite domains of the variable x.e. mow' Apx = ^m0Luh/2. and we see t h a t even and odd states alternate. E0 ~ —{Apx)2 + IrriQ l mQuj2{Ax)2 moLoh/2 2TUQ 1 moU!2h/2 2 TTIQU = Ku). T h e comparative behaviour of t h e wave functions of the three lowest states of the harmonic oscillator is shown in Fig. 2 . has parity + 1 .
The analogous Example 6. to 0.4 The Position Space Representation 97 The zero point energy plays an important role in atomic oscillators. Sec.: 27 T mov2a2.1. classical orbit i.2 The original path and the transformed path at infinity.1 we demonstrate how the eigenvalues may be obtained by the method of "poles at infinity". But it has a pole at infinity.e. According to "old quantum theory" with integer n. in corrected version instead with n* = n + 1/2. 27rmoi/ <f> V'a 2 — q2dq = n*h.. original path q . E . which is the complete orbit. Thus. setting q = 1/z. and is directly observable. Example 6.4) to obtain by contour integration the eigenvalues of the harmonic oscillator.plane Imz z . /classical orbit Here q goes from 0 to a. and with potential V(q) = 2K2mov2q2 and total energy E = p2/2mo + V(q).2 contains only the answers since the evaluation of the integrals proceeds as in Example 6. like twoatomic molecules. The kinetic energy is always positive or zero. Solution: Consider the simple harmonic oscillator of natural frequency of vibration u. In the plane of complex q the integral has no poles for finite values of q. pdq. back to 0.6.plane pole at z=0 ^ . we obtain the integral . to —a. This observability can be taken as direct proof of the uncertainty relation.1/a O l/a Rez all at infinity Fig.1: Eigenvalues obtained by contour integration Use the corrected BohrSommerfeldWilson quantization condition (cf. 14. n*h. In Example 6. 6. 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)). mass mo. Finally we point out that the Hamiltonian of the harmonic oscillator can be diagonalized in many different ways.
z = 0 Vi 2Kmov(Ka2) = n*h = E/v. for instance — without deriving all properties of the solutions. (11. i.98 CHAPTER 6. but here we leave this . 6.a) m = 2 .65) f ^Va2z2 .5 The Harmonic Oscillator Equation In the above we considered mainly the energy representation of the harmonic oscillator and encountered the Hermite polynomials. 1 .It follows that 1 = s2 i ^V»a*ai ~2f z=0 2wi r m '„2 = 7ria 2 La*Va 2 2 2 ..2: Contour integration along a classical orbit Setting q = 1/z. and obtain with this all properties of the solution.1)! (6.1. 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). With a2 = E. / " ^ .e.) hv. verify that II = d> dq yi92 aq + ft 2TTI 9 /y/F^I 9z V a + bz z=o . harmonic oscillator and Coulomb interaction.c2 = (I + 1/2) 2 (natural units. in q at infinity). . . Quantum Mechanics of the Harmonic Oscillator This integral has a triple pole at z = 0 (i. are the most basic and exactly solvable cases in quantum mechanics. we start with a Laplace transform ansatz for the remaining part of the solution.e. In the following we investigate in more detail the important differential equation of the harmonic oscillator.V62 a2). « qdq + I2 = f dq . £ = n*/w = ( n + .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.3. since these two cases. 2 .2TTI a ^ = . In Chapter 11 we perform similar calculations for the radial equation of the Coulomb potential — as in Example 11. 6.e. E = Z2e4/4(l + n+ l ) 2 (cf.114a)). We can evaluate the integral with the help of the Cauchy formula (the superscript meaning derivative) Jc Hence we obtain (z . hence we consider the following differential equation in which n = 0 . i. a J \d Solution: The solution proceeds as in Example 6.5 .5.1. Example 6. (m .2b = Ze2. Eq. mo = 1/2) the integral I2 can be checked to give the eigenvalues for the Coulomb potential Ze2/q (2 turning points). .( 6 . After removal of a Gaussian or W K B exponential.
(6. Hence we set (with the chosen sign in the exponential) m=y{t)et2l* with *0. . i. since this can give rise to exponentially increasing behaviour (at most a factor t can be tolerated). This has to be true for all values of t.g.67) we obtain \ps(p)e pt] + I P2s(p) + —(ps(p)) + ns{p) e~ptdp = 0. 80. We demand that C(p) := — \ps(p)e~pt] = 0. (this determines the limits). Then (with partial integration) *dt = ~*/Ps(p)e. .69) **See e. fp{ps{p)) = {v2 + n)s{p).** We make the Laplace transform ansatz^ (with limits to be decided later) y(t) = [ s(p)e~ptdp.g. (6. Substituting these expressions into Eq. (11.66) First we have to remove the £2term. p.67) the latter being the differential equation of Hermite polynomials Hen{t) for integral values of n. (6. E.p t dp = \ps(p)ePt} . (6.i ^  (ps(p)) P n P Integrating this expression.J ±(ps(p))e*dp.e. Magnus and F. so that the integrand of the remaining integral vanishes.6. Eq. / dpf(p)e~px the Laplace transform of f(p).t*ij& + ny{t) = O. (cf. Thus — ignoring details — / dpf(p)erpx is the Fourier transform of / ( p ) .5 The Harmonic Oscillator Equation latter property open for determination later: 99 ~dP + 1 1 2 V> = 0 . Oberhettinger [181].111)) the Mellin transform of e~p is n!. W. we obtain P2/2 ln(ps(p)) = — p — nlnp or s(p) <x Tl+l P (6.68) § = J p2s(p)e*dp. and / dpf(p)pn the Mellin transform of / ( p ) .
This exponential assumes a maximum value when pt \—p2 = minimal.3 The path when n is not an integer. i. (3) We can choose a contour once around p = 0 in the complex pplane..70) = 0 P. The quantum mechanical wave function is then (cf. n must then be an integer. Eq. Thus the only solution left which does not involve large values of \p\ is that with § Q for n an integer. so that y(t) ~ exp(i 2 /2). we can choose a path from zero to infinity (the condition will vanish at p = 0). Hence a possible path of integration > is from — oo to oo.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.n+l between the two limits of integration. then for large t we have y(t) ~ exp(i 2 /2). y(t) oc 2vri J pn x dp for n integral and ept p 2 = coefficient of pn in ~^ (6. of course). Therefore we take ptp2/2 y(t) .e. (1) This will be satisfied if \p\ — ±oo. so that p cannot go to infinity. as indicated in Fig. consider the exponential in the integrand. The value of the exponential at this point is ~ exp(£ 2 /2). (2) If n < 0.100 and hence CHAPTER 6. To see this. since otherwise (with p = \p\ exp(i8)) we will get part of the solution from 0 to infinity.71) .3. We can show that if \p\ is allowed to be large. 6. For singlevaluedness. This can be attained if p is allowed to be large (t can be of either sign). 6. when p = —t (which is possible. Quantum Mechanics of the Harmonic Oscillator ptp2/2 » ( « ) ptp2/2 dp with C{p) = (6. = / • Im p Rep n not an integer Fig. (6.
l ) n n ! x coefficient of pn in = = (~l)nn\ e~ ~ x coefficient of pn in V l z ^ e .0) p p»+i v ' Hen(t) = (l)^ e * 2 /2 / 2vri /(9=t) dg " ^(gi)"+i e ?2/2 = e i 2 / 2 ()ne"i2/2. this becomes jfj. (676) . . the term of highest degree in t in the exponential corresponds to taking all p's from exp(— pt). In y(t) above..l ) n n ! x coefficient of pn~v in V LJ_L e p 2 /2 *^ v\ u=0.3. The Hermite polynomial is therefore given by H ^{t){^—f—^r~dp. n—v even ^ 2 ' Hence we obtain the polynomial of degree n Hen(t) = nl £ see below ( _ 2 ) ( n .„y2 v l { ' ^ r (^) \ 2 ' where v = 0. We can also obtain the derivative form of this Hermite polynomial.P 2 / 2 00 C—tY ( . 4 . . . The Hermite polynomial Hen (t) is now defined such that the coefficient of tn is unity.if n is odd.5. We have 27T! Setting q — p + t.72) We now obtain the polynomial form with Cauchy's residue theorem as Hen(t) = ( . if n is even and v = 1. 2 . .6.5 The Harmonic Oscillator Equation 101 with Cauchy's residue theorem.. The exponential function here is described as the generating function of the Hermite polynomials Hen(t). Therefore the coefficient of (pt)n in exp(—pt — p2/2) is (—l) n /n\. pt p2/2 (6..
and elsewhere. We leave the consideration of the recurrence relation of Hermite polynomials (of paramount importance in the perturbation method of Sec. and we obtain then: ptp2/2 —^Ti ptp 2 /2 d : P r epte~p2/2 •(n + l) <t —5 f dp  eptep 2 /2 Here the bracketed expression vanishes in view of our condition (6. 8. (6. (6. n — l)Heni(t). one can construct the tower of polynomials Example 6. Hen(t). Solution: We have to evaluate °° 2 . Starting from Heo(t) = 1.it) _ (n + l)Hen+1{t) (l)""1^1)! (—l)"+ 1 (n + 1)! ' tHe„(t)=nHen1{t) + Hen+1(t). for the evaluation of expectation values. we obtain tHe„{t) (l)"n! ~ i.n + l ) f f e n ^ i ( £ ) + (n. in radiation problems. Hen(t)Hem(t)et ™ 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+<. Example 6. Hei(t) Henj.)} 2 /2dt = ePde(P 2 W)/2^^} J —CO .59a).102 CHAPTER 6. 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). (n. Example 6.75) as Hermite polynomials.3 and 6.7 that we use throughout this text) and their orthogonality and normalization to Examples 6.4.g.1 '2dt using Hen(t) = (l)n27 r . (6.3: Recurrence relation of Hermite polynomials Show that tHen(t) = (n. n + l) = l. Pptp 2 / 00 /2 dp.e.77) = t. as in the normalization of asymptotic expansions of Mathieu functions).4: Orthonormality of Hermite polynomials Establish the orthogonality and normalization of Hermite polynomials. Reinterpreting the remaining integrals with Eq.70).5 is a further application of the method for the evaluation of integrals that occur frequently in practice (e.n— 1) = n. ( n . Quantum Mechanics of the Harmonic Oscillator which we recognize as agreeing with Eq.
5 The Harmonic and hence Oscillator Equation 103 / = (!)" But (2iri)2 { J Jp=0 J q0„ 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 pnm+1 =2ni if Ti=?7i.66) is \Zn\V2w Example 6.5: Generalized power integrals with Hermite polynomials Establish the following general formula for Hermite polynomials: / Jo t2sHen(t)Hen+2r{t)et ir n\(2sy. g) is n! Next. and 0 otherwise \/2n(—l)n+mn! if n = TO.q).5 nm .n)\ i s an+2rij. zero otherwise (6.82) ^ .jQ) n n!I ^—' ^Q fi\(n . (6.78) \/27rn!.(n + 2r)\ ' 2 ~ s r '2dt r 2* ^ i / ! ( 2 r + i/)!(Tii/)!(sri/)!' Solution: From the above we know that He„(t) is the coefficient of p " in (—l) n n! exp(—pi—p 2 /2).q) = = e . Hence the normalized wave function of Eq. 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 eatHen(t)Hen+2r(t)et l2dt = n\(n + 2r)\ X coefficient of pnqn+2r in f(p. the coefficient of qn+2rn the coefficient of p"qn+2r m exp(./(n __ 2r — /i)!.6. g) is the product V^Tre" ^oM!(n/i)!(n + 2rM)! (6.^ + 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 .80) f(p. (6. so that /2 Q2(n+rrf in / ( p .
n + 2)(n + 2. i. the relation (n.e. by n!(7r/2) 1 / 2 . n + 2)]ife„ + 2 (t) + ••• . i.l ) ( n . A « 2 ( r '+ I/ ) 2 coefficient of (2s)! in (n + 2r)\e ' ^ . n + l ) ( n + 1. Prom Example 6.A o?(. so that now with integration limits from 0 to oo K = coefficient of a2s (2s)! in J a = 2. K is the coefficient of a / ( 2 s ) ! in the expression J .i / ) ! ( 2 r + i/)! > . n + l ) ( n + 1. n + 2) + (n. n + 4)_ffe n +4(t) + [(ra.e.ra + 3)(n + 3. we obtain the result (6.Y^ i[ ^0vKnv)\C2r + ») cc2s A (n + 2r)! a 2 ^ . n + l ) ( n + 1.A i ) ! (6. 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) + . /i = (n — v) _ / r o o e .v)\ ^ Multiplying both sides by the value of the normalization integral. n — 1) are given by the first index.e '' ^ o^M ! ( n . n + 2) + ( n . n.r .v)\(s .' '2dt= / Jo t4 Dn(t)Dn+2(t)dt = V2^(2n + 3)(n + 2)!.— — ^ . 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. Quantum Mechanics of the Harmonic Oscillator It now follows t h a t — in later equations with v := (n — /i).77).t • 2l/2dt _ J : Q2 » a2(n+r^) ~ f ° ° f f „ t2/2 „ !ZoHen(t)e2 Wdt . i. .v)\(2r + v)\ > Q (2s)!(n + 2r)! AT 2~ s r 2V v\(2r + u)\{n .r .83) = (n + 2r)!e a / 2 V „ „.l . The result then follows with half the normalization and orthogonality integrals (6.° i i f e T 1 ( t ) i f e „ + 2 r ( t ) e .' . n + l)(n + l .( n + Jr).(t)e«2/2dt Now. 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. n + l ) ( n + 1.^ ) ! ( n + 2 r ." ) a2^) r coefficient of (2s)! in f^ (s . n + 3)(n + 3. .3 we know that upgoing coefficients are 1 and downgoing coefficients (n.78).n + 2) + (n. n + 2)(n + 2. the coefficient of T 2s »{sr u)\ As we observed.79).e..n+ l)(n + 1. n)(n.2s _oo ( a 2 / 2 ) i . n + l)(n + 1.^ / 2 * . i / ! ( n .u)\ 2s<—" v\(n . n + 2){n + 2.104 CHAPTER 6. n ) ( n . As an example we obtain for the integral from 0 to oo (also reexpressing the integral in terms of parabolic cylinder functions Dn (t)): / Jo t 4 f l e „ ( t ) H e n + 2 ( t ) e . n .r+v) a2s coefficient of in (n + 2r)\ V (2s)! h .
and is thus an important point to be noted here. This example also offers a convenient context to introduce the inverted oscillator potential. and those which are timeindependent. In particular we derive the timedependent Green's functions for the case of a free particle and for that of a particle in an harmonic potential. We 105 .Chapter 7 Green's Functions 7. which is a point of unstable equilibrium for a classical particle. in Chapter 21 (with a different calculation). 7.1) be the timeindependent Schrodinger equation which has to be solved. which we write G{x. We consider first timeindependent Green's functions. The other example of the oscillator enables us to evaluate corresponding expectation values of observables in the canonical distribution introduced in Chapter 5.2 Timedependent and Timeindependent Cases In the case of Green's functions we distinguish between those which are timedependent.t). which we write here K(x. With reference to this case.1 Introductory Remarks In this chapter we consider Green's functions of Schrodinger equations in both timedependent and timeindependent forms. These are the Green's functions of the timeindependent Schrodinger equation or. we consider the sojourn time of a quantum mechanical particle at a point. more generally.x'.x'). Let H^ = E^> (7. in Feynman's path integral as the kernel or free particle propagator. of a differential equation of second order. which is normally not discussed. The first of these cases will reappear later.
x') by recalling the completeness relation. x') has to satisfy correspond to those of * (to insure e.x') can be written t G (7. Green's Functions H = H0 + Hi.x'). (7.3) i which has the specific representation ^2{x\i)(i\x') i = 5(x . appropriately decreasing behaviour at infinity).4) where the functions ipi(x) are solutions of HQij)i{x) = E^iix).g. (7. i.g. of the kinetic part p2/2mo and some part of the potential. 'The case of £(°) equal to some Ei is considered later. cf.106 assume a Hamiltonian of the form CHAPTER 7. square integrability.x') or ^2^(x)^*(x') i = S(x . although the explicit calculations can be much more involved.2) where Ho consists. since the generalization to higher dimensions is selfevident. and Hi is some other contribution. Eq.5) M = Z*W=EW i l for Ei0) E * (7 6)  This holds since *For simplicity we consider the onedimensional case. e.e.x').EW)xG{x.x') is defined by the equation* (Ho .g. The timeindependent Green's function G(x. (7. aq2. (7. We can readily see that G(x. the relation X') = 5{x . We can obtain an expression for G(x. i.13). . where £ ( 0 ) = lim E. a perturbation part like (3q4. The boundary conditions which G(x. for instance.e. e.
which consists of the perturbation part Hi of the Hamiltonian and the corresponding correction E .11) = = 3) = fdx'(H0 . (7.EM)X*(X) ( .) = f dx'G(x. We see that the initial condition (7.7.8) is satisfied as a consequence of the completeness relation of the wave functions ij)i{x).2 Green's Functions: Timedependent and Timeindependent 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. We rewrite the complete timeindependent Schrodinger equation H^ = E^ as (#0£(0))* = (££(0) #/)*.0) = S(xx').x'.t) (K for "kernel") is defined as solution of the equation ih—K(x. .x'.x')(E .x'. We can see the significance of the timeindependent Green's function for the complete problem as follows.£(0) . x'\ t) (7. (7.t) = ^Eie^ihM^i(x') i = Y^eE^ihUx)^(x').Hi{x'))^(x').t) since ih^K(x.E® (EE{0) Hi{x))m{x).(*'))*fr') Hi{x'))^(x') f dx'5{x . (JL x'\ t) = H0K(x.7) with the initial condition K(x. . The timedependent Green's function K(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. x'){E .x'. (7.t).x.9) = i H0YieE^i%(x)^(x') = H0K(x. i (7.E^ We can check this by considering (H0 .#. For the solution V we can immediately write down the homogeneous integral equation (again to be verified below) *(a.E^ of the eigenvalue.E<®)xG(x.x'){E ..10) The contribution on the right hand side is the socalled perturbation contribution.
11) an inhomogeneous term which is a solution of the nonperturbed part of the Schrodinger equation? If we keep in mind the anharmonic oscillator with square integrable wave functions ^(x). Sec.e.12) Equations of the form of Eq. integral equations are more difficult to solve than the corresponding differential equations. 17.11) with a system of linear equations of the form Vi = Mijyj. Merzbacher [194]. the function ipi(x) would not be a solution of and it is not possible to add an inhomogeneous contribution. Instead of Eq. But if we assume that E^ ^ Ei for all i. Green's Functions Can we add on the right hand side of Eq. if we assume that E^ = Ej.x'.6) is not defined. however. Maharana. In this case the perturbation is restricted to that subspace of the Hilbert space which is orthogonal to ipj. see E. (7.15) ''This problem and its circumvention can be formulated as a theorem. All such methods of solution are based on the analogy of Eq.Ej ffj(x'))*(a:') is a vector in TL which is orthogonal to ipj..2 (not contained in the first edition!). In cases where an inhomogeneous contribution is given.6) we would have (7. In general. The theorem is also known as Fredholm alternative. that (E .108 CHAPTER 7.t) for the complete problem as follows. see B. (7. (7. W.14) We can now see the significance of the timedependent Green's function K(x. i. .11) are called (homogeneous or inhomogeneous) Fredholm integral equations. MiillerKirsten and A. it is possible to solve the integral equation by an iterative perturbation procedure (see later: Born approximation). But then the Green's function (7. Bleecker [33]. Wiedemann [183]. This difficulty* can be circumvented by demanding from the beginning that 0= f dx'tf(x'){EEjHi(x?))V{x'). (7. then at best such a contribution would be something like cipi(x). D. 2nd ed. (7. Examples in various contexts have been investigated by L. J. Booss and D. This will be different. H. The timedependent Schrodinger equation ih\><l>)t = H\il>)t (7.
g. (7. 109 (7.e. for the oscillator potential.17) as initial condition is obviously W)t = Y.21) dx'K(x.' which obviously satisfies the initial condition K(x.22) According to Eq.17) is the initial condition of \4>)t. Eq.The solution of Eq. Merzbacher [194]. in different formulations.7) with Eq. (7. p. (7.7. (7. . We can write this expression also as§ ip(x. (7.g. <a#) t = ]T n J [dx'(x\En)(En\x'){x'\^)t=0eE^\ (7. H\En) = En\En). x'.18) or.20) is the timedependent Green's function.Q) = 6(xx'). 2nd ed. this relation provides the probability density \ip(x. (7.5.t) describes the evolution of the wave function from its initial value ^(a^O).x'.16) As usual.51) shows that we obtain a very analogous expression for the density matrix PN(P) a s for the Green's function K(x.t) is known. 1 See e. 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. Comparison of Eq. E.t) = ^ e £ " f / « V n W < ( ^ ) .0). e.0)eEnt/in.t) = J2 fdx/^n(x)^*n(x')^(x'. (6. (5.t) = where if(x^'. x'.2 Green's Functions: Timedependent and Timeindependent permits stationary states \En) defined by \1>)t = \En)eE^ih. \En)(En\ip)t=oeE^\ n (7. (7.x'.t)il>(x'.2 for the computation of the sojourn time. t > 0.20) the Green's function K(x. t)\2.19) i/>(x. *>0.158.x'.15) which contains (7. We use this in Sec. 7. i.. t) with the Note that when K(x.42)) Wi=0 = E l ^ ) ^ l ^ = o n (717) Thus at time t = 0 the state \ip)t is a linear superposition of the vectors \En) with coefficients (En\i/})t=o.
23) along the contour C in the plane of complex E as shown in Fig. i.1 The contour of integration.x'. x'.«^«2 W . (7. Inserting for GE+ie the expression above.1.E .e. 24 )  ReE Fig. Green's Functions difference that (5 = 1/kT plays the role of it.25) in agreement with the timedependent Green's function .x') = GE(x. 7. Jc 27T „ hn . we obtain / W : = .x'. 7.t). x')6(t) (7.x') and K(x. With Cauchy's residue theorem we obtain I(t) := J2eEnt/ihMxWn(x'Mt) n = K(x.ie (7 .i /f e «/. we obtain then also the corresponding density matrix. between (with E^0' = E) G(x.t) K(x. As a consequence of the above considerations one wants to know the connection between the timedependent and the timeindependent Green's functions.110 CHAPTER 7. We therefore consider the following integral with e > 0: I(t) := i J ^eEt/*hGE^e(x. In the following we shall derive the Green's function for the case of the harmonic oscillator. t) = J2 e £ " f M i ( x ) < ( x ' ) . t > 0. = J2 ^„(s)C(*') En — E We see that G = GE depends on E.
v ' 2m 0 v h i.27) An equation of this type — called of the type of a diffusion equation — can be solved with an ansatz. and identifying coefficients of the same powers of t on both sides.0) = 5(xx'). we obtain *I£I = h2 —(2AB). A and B being constants.29) and (7.x'. i. / dxK(x.t) = ^eB^X In this case we have dK ~dt and dK dx 2 8K dx2 A A 2t3/2 A tV2 '^'K (7.t).x>.30) Inserting Eqs.29) tJJ 2 2B(x .7.3 The Green's Function of a Free Particle The timedependent 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. .7). B m0 n _ 2ih 2iK (7.X>.x1) t B(xx')2/t B(xx')2/t 2AB 4AB2(xx')21 ~¥l 2+ W 2 (7. We therefore consider first the simplest case with Hn = P 2mo h2 a2 2mo dx2 (7. (7.0) = 1. It is clear that it is nontrivial to solve an equation like Eq.t) t h2 = .— d2 —2K(x. (7. Thus we try the ansatz.e.x'.26) This is the case of a free particle with mass mo. (7. which is moving in one space dimension. (7.28) B(xx')2 t2 DB(xx') 2 /t (7. In this case the Green's function is the solution of the equation d hK(x.27).x'. 2mQK h h2 ih(AB) = —{4B2A).30) into Eq. K(x.e.3 The Green's Function of a Free Particle 111 7.31) The constant A has to be chosen such that K(x.
from K{x.20) and can then obtain ip(x.H We can insert the expression for K(x.t)= f^em2t/2m0iheik(x~X')_ J 27T (735b) We set a = i—. (7. t) = /jJo_ e mo(««') 2 /««.(7.33) V Lirvnt Can we demonstrate that this expression can also be obtained from Eq. j3 = i(xx'). we > have to make the replacements V ^ fdk. (7.36) ^ W Q ^ J emo(ii') 2 = —eP2lia 27T dke~^k^2^2 J„00 /2«tS . 1st ed. (7. 158.33).0)oce"Qx2+ifcox. we have This is 1 provided A=JB= ^ (7 . Merzbacher [194].38) . 32) It follows that x'. Green's Functions For parameter values such that the following integral exists.e.t) into Eq.t) = — / dkeak2+Pk 2vr J."737) \27rii/i in agreement with Eq. *>0. (7.34) K(x.x'.112 CHAPTER 7. 2m0 (7.ihMxWn{x').t) — for instance for a wave packet given at time t = 0 of the form ^(x.36) 2mo Then — provided that the parameters assume values such that the intergral exists — K{x.zEnt. i. (7.oo 27T (7.21). "See the excercise in E..x'. n For a free particle moving in the onedimensional domain \x\ < L — oo. v27r En^^. p. (7. Mx)^M^) „ n so that J =^ .35a) K(xy.x't) = Y.
t) We now set h2 d2 = .7) is the solution of d ih—K(x.t) = h2 2 d2 TW . . (21.39) In this case the timedependent Green's function K of Eq. / ) = ~^K(x.X .7.e. We rewrite this initial condition in terms of £ and use for a = const. 5{ax) = ± J eikaxdk = ArS(x). . (7. the relation* 6(x) = o<y(os). —KK(X. K(x...0) = 6(xx') at / = 0.e. (7.40) Then Eq. 7.t) + ~x K(x. —ih—K(x.x'. x'./ ) + fK(x.25) — in the context of Feynman's path integral method. . / ) (7. §jK(x. i.x'. (7. ..x'.——K(x.33). (7. .3 The Green's Function of the Harmonic Oscillator 113 The result (7.t).e.40) becomes 2 . z .d r.t) 1 + m0u2x2K(x. HQ = ^—p2 ZrriQ + \m0uj2q2. i.42) .x usn ot i.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. . mQLu . (7. 1 2mo u>n ox n with the initial condition (7. We consider the onedimensional harmonic oscillator with Hamilton operator HQ.e.t).8).x'. in this domain *Recall that S(x) = ± f eikxdk. I (7.x'. x'. i.37) will later be obtained by a different method — see Eq. 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.
4 a 2 .48) ocexp[{a(/)C 2 + 6(/)e + c(/)}] (7.b2. (/ = 2a. . and we expect K(x.b2. (7.e. p.49) a=icoth2(//J0 ) w 2 "' 2tanh2(//0) * To ensure that the expression (7.45) in accordance with (7.4a&£ + 2a .45) in this sense. (7. 50. W ) . as the limiting case / — 0. that is. f i. Identifying coefficients on both sides.e. (7. (7.4a 2 )£ 2 . we obttain a' = l . c(0)^. b' = Aab. (7.51) See R.. 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.t) become similar to expression (7.50c) (7.x'.50a) we obtain (7. so that for / — 0: > a^—. (7. we must have /o = 0.f) with a(0)»y. Green's Functions to the particle behaves almost like a free particle. Feynman [94].c o t h 2 / . i. a = .47) becomes (7.42) and obtain the equation (with a1 = da/df etc.50b) (7.46) If we interpret Eq.48).50a) (7.^ .Z'. it • is suggestive to attempt for K the following ansatz^ m.47) We insert this ansatz for K into Eq.) a'£2 + b'i + c' = (1 .114 CHAPTER 7. Integrating Eq. P.33).
In order to satisfy Eq. 174. x') of the density matrix p^ (with respect to the canonical distribution with (5 = 1/kT): pN(x. c(0) = £' / 4 / . 7. (7. 115 To ensure that in accordance with Eq. (7.47) we obtain ^ c o t h 2 / + ^ + ^ c ° t h 2 / (7.x'.c(f) K = 5 =exp Vsinh 2 / into Eq.t) to obtain this element (x.40) of the timedependent Green's function with Eq.50c) yields for c(/).5.t) = .51) for the density matrix PN(X. if we return to x. 2TT^ Inserting a(f).3 The Green's Function of the Harmonic Oscillator Correspondingly we obtain from integration of Eq. (7. t. (7.50b) Kf) = —r^77) smh 2 / A independent of / . as one can verify.x'.e. p.48). as we shall see in the following.2. . we must have (besides A = —£') = /m 0 u.e.55) 'For an alternative derivation and further discussion see also B. B independent of / . (7.48) 6(0) = — £'/2f. Comparing the Eqs. c(/) = i ln(sinh 2/) + ^ 2 c o t h 2 / . with A. in particular for the derivation of the sojourn time in Sec. x . Finally Eq. we must have A = ('.33) for the Green's > function of a free particle.0). (7. (7.£) _  {(x2 + x /z ) cos tot — 2xx'} 2msm(o.* With this result we have another important quantity at our disposal. K(x.i) (7. (5.In B.7. i.X'.b(f).27). (7.54) For t —• 0 this expression goes over into the expression (7. and to ensure that we obtain the prefactor of Eq. we can use K(x. x'. i./3) = J IJIQUI 2irhsmh(hLu/kT) 2hSinh(hw/kT) \{X +X j C x exp ° S n kT lXX J (7. Felsager [91].53) or.45). / —————rexp y 27rmsin(u.
(5.40)) 1 epEi 1 Ui = y. P. For instance we have with Eq. we skip the algebra here. (5. eP E i = i + y.55)..56a) {q2) = = Y. Eq. = jdxp(x. [ [ 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.uij>0 >• 0.eft >E*) E (7 57) ' of the number of systems of the ensemble occupies the quantum mechanical state i. ^—• Irpiv J pN(x.e. p.x. 52. Thus the system is in a mixed state and the expectation value "Cf.x.» l.116 CHAPTER 7.52)) the expectation value of an observable A in the canonical distribution (i.e. at temperature T): {A) = Tl{pA) = ^ > .56a) the expression (7. . <2 (7. )x Inserting into Eq. we obtain § < Z> = ^ c o t h ^ ^ ° — . R. Eq. 11 For T ~+ 0: u>0 .x.P) 2.p)dx W ) = —^r = —p . Feynman [94]. Green's Functions With this expression we can evaluate (cf. 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"}. (5. (7.56b) w v ' 2mQu} 2kT 2m 0 u ' What is the meaning of this expression? At temperature T the fraction (cf.52): Tr (PNq2) Jx2pN(x..
setting a = mooj/h.J (w — ie)< e (n+l/2)Et e i(n+l/2)u.QUJ (7. 1 mo^ exp lmow / 2 2 ft X (7. t) of Eq. of K(x.54) is K(x. (7. and we replace to by LO — ie.50) we obtain. We return to the Green's function (7. x'. <*2>o dxx a •K j e IT I da r dxeax *\1/2 da a J _ 1 _ _ h 2a 2mou j —( .x'. 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. 2 + x'2)le^2xx' to—>w—it t—>co 2 . x'. 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 (7.59a) E 0 = ^fiw.t "With the normalized ground state wave function of the harmonic oscillator given by Eq.e. If we consider the system in the pure state \i). which means in the oscillator state \i) with eigenenergy fku(i + 1/2).t) w—>UJ—ie to—n t—>oo moco/irh ^ ' piuit g—iwt exp m ^\(x 0 ^ . We assume t > 0 and t —> oo.21).3 The Green's Function of the Harmonic Oscillator 117 (7.7. In this case the Green's function K(x. (6. e > 0. This is the first and hence dominant term of the expression (7. i.59b) n+ l \ ftwt here 2hf exp _ i ( n + . /2N 2~7T: = eE°/iht(t>Q(x)(t)0(x') for £ > 0.54).56b) is that with respect to this mixed state (whose cause is the finite temperature T).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.
5.x'. Green's Functions For t large (i.1 W a v e packets The simplest type of wave is the socalled plane wave or monochromatic wave of frequency UJ represented by the expression exp[i(krwt)]. i. Considered classically. the velocity of planes of equal phase. 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. as in Eq. and with this we estimate the sojourn time T. lie on surfaces.118 CHAPTER 7. 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. However quantum mechanically in view of the uncertainties in position and momentum.59a). 7. i.61) The word "plane" implies that the points of constant phase <p := k • r — uit at t = const. (7. The wavelength A is given by The phase velocity v<p. (7. is defined as vv = . the particle will stay there only for a finite length of time T.1 we estimate T semiclassically. In Example 7. which are planes. (7. a particle placed at the maximum of the inverted oscillator potential (which is classically a position of unstable equilibrium) will stay there indefinitely.5 The Inverted Harmonic Oscillator We encountered the inverted harmonic oscillator already in some examples.e. In a very analogous manner we obtain the solution of the equation for the density matrix.(3) = J2^EnMx)K(x') n „ ^ ° e^MxWo^') (760) 7.62) . k = k.e. to infinity) the contribution with n — 0 dominates.. of dpN as pN(x.e.
65) Let f(k') = \f(k')\eia and <p := k'x .)¥'(*:")) +ism(<p(k') . ' t) dk'. e. (7. xt— + — = 0. One defines as centre of mass of the wave packet that value of x for which d(p — = 0. i ) = /*/(k')e i(k '" r '' .e. A wave packet is defined as a superposition of plane waves with almost equal wave vectors k. t) describe the motion of a classical particle? For reasons of simplicity we restrict ourselves here to the onedimensional case. / oo J—oo . oo (7. If we assume for / ( k ' ) a Gauss distribution. the fundamental postulate on matter waves in Chapter 2): E = hu.7.)ll/(fc")l[cos(¥'(fe.e. i.63) The relation u = w(k) is known as dispersion or dispersion law. then ^(r.Jt + a. a > 0 .k ') 2 . (7. e . We now ask: How and under what conditions does the time variation of the function ^(r. to the function ip given by oo / fik'y^'x^dk'. The wave packet describes a wave of limited extent. u = w(k).64) where / ( k ' ) differs substantially from zero only near k' = k.e.<p(k"))}dk'dk". ^ ( r .Q ( k . /OO / / / oo /*oo \f{k')\\f(k")\ei^W^k"»dk!dk" l/(fe. t) is the spatial Fourier transform of this Gauss distribution — as we know. for which \ip\ assumes its largest value: OO . i. duo' da .4 The Inverted Harmonic Oscillator 119 Every frequency u belongs to a definite (particle) energy E (cf. or dco' da The centre of mass determines the particular phase. essentially again a Gauss curve. i.
i. (7.67) The centre of mass moves with uniform velocity vg called group velocity of the waves exp[i(kx — cut)]. The threedimensional generalization is evidently du . as claimed for =0. we can use a simpler method. however. dE v 9 = . .1: Fourier transform of a Gauss function Calculate the Fourier transform of the spatial Gauss function e . p = hk the group velocity is equal to the particle velocity v. (7. a > 0. In the present case.120 CHAPTER 7.fc w> vv = ~a^ = S r a d P E «i7 Sd u . we obtain the deBroglie relation p — hk. Thus the expression g{k) is solution of the following first order differential equation g'(k) + Aff(fc) = 0. Solution: The function to be calculated is the integral dxeax OO ezkx.e. For E = hio. = — (768) It is instructive to consider at this point the following examples. Green's Functions This expression assumes its maximal real value when <p(k') = ip(k") = const.69) In general one uses the theory of functions for the evaluation of this integral. Differentiation of g(k) yields g'{k) = / dxe~ax 2a y_oo 7 f°° 2 {2ax)eikx = 7 2a / f°° Jx dx With partial integration we obtain from this 7 f°° 2 k g'{k) = — / ikeikxea* dx = ..— g(k). Example 7. 9 cku dk dhw dhk dE dp dE dp' We can also argue the other way round and say: By identifying vg = v.^ g=r a rad. .
(7. (7.74) Example 7.2: Representations of the delta distribution Use Eqs.0 0 7T 7 0 7T e From Eq. with partial fraction decomposition. \a — b\ (7...75) ' E . Appendix A.1 verify the uncertainty relation Aa.. (7.7 °) With the help of this example we can obtain some useful representations of the delta function or distribution as in the next example.. (7.a) + 6{x . (7. —±=ek*/ia2 / 4 " = lim 1 — — lim ^e* = The second important example can be verified by immediate integration.71) (7.72) Solution: From Eqs.4 The Inverted Harmonic Oscillator Simple integration yields g(k) = c e " f c 2 / 4 a . Since /•c 121 9(0) = J—00 dxe J_ we obtain S(fc) = ^ e " f c 2 / 4 a ( 7 .70) we obtain 6(h) = lim — f°° dxeax2eikx = .3: The uncertainty relation for Gaussian wave packets For the specific Gauss wave packet of Example 7.73) 27T 7 . XZ (e > 0).73) we obtain the requested representation of the delta distribution with 1 f°° 1 € Six) = lim — / dke~e^eikx = lim .&)] for a + b. (7.70) to verify the following representations of the delta distribution: X2/€2 5(x) = lim ?—=e ^ o eV7r and 5{x) = \\m^r^. g .Afc = 81n2..69) and (7. where c = 9(0) is a constant. (7.69) and (7. W. . or see H.r „• With the help of Eq. J. MiillerKirsten [215]. We have — 1 / * ° ° dke~eWeikx =  I / " 0 0 dke~tk cos kx =  1 .43) one can verify the following important relation* 5[{x a)(xb)] = T—!— [S(x . Example 7. 2 f + .7.
Solution: In Fig. at time t = 0).77a) small values of a. Thus a sharp maximum of the function f{x).2 The Gaussian curve. (7.2 we sketch the behaviour of the Gaussian function p—x 11a ~—ikox (7. According to Eq. requires according to Eq.r)1/2a Re f(x) Ax Fig. Green's Functions 1/(27i)1/2a — 1/2(2.CHAPTER 7. The wave function corresponds to the function f(x) in the above considerations (e.* o ) 2 / 2 .78) It follows that the product of the uncertainties is AxAk = 8 In 2.t) is a measure of the probability to find the particle at time t at the position x. 7. is Afe = 2 V 2 1 n 2  (7. (7.e.g. when Ax is very small.76) 2ira The uncertainty Ax is defined to be the width of the curve at half the height of the maximum. i. where g(k) = max<?(fc)/2.77a) g(k) = e .) leads to a very broad maximum of the Fourier transform \g(k)\. 7. the width Ax is a measure of the uncertainty of the probability.77b) (7.70) the Fourier transform of f(x) is (7. Physically . Thus a slim maximum of the curve of /(a. where  / ( x )  = m a x  / ( x )  / 2 .« 2 ( * . The breadth A/c of the curve g(k) around k = kg. i. A simple calculation yields A i = 2v/2~In2a. In quantum mechanics the square of the modulus of the wave function ip(x.e. and hence implies large values of Ak.
2 ) and U)t. This means the particle is localized at x = 0. many supposedly elementary problems request the calculation of the sojourn time of a quantum system near a classically unstable equilibrium configuration. (7.. (7.76) ex 2 /2a2 ^2na a>0 um 1/0*01 = lim a^O . which means all values of the momentum are equally probable. In the limit a — 0 we obtain from Eqs.20) the wave packet at time t > 0 is obtained from that at time t = 0 with the relation «. 7. nr = S(x).u> = l.54) with the substitution ui — iu.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. The function g(k) is therefore described as momentum space representation of the wave function f(x). For reasons > of simplicity in the following we set in addition t mo = l.". at time t = 0.7.* We obtain the Green's function K{0 for the inverted harmonic oscillator from Eq.5. The result (7. so that Ki0(x. Barton mentions at the beginning of his paper. .h = 1. Barton [15]. Barton recalls that W. Thus G.81) *G.e. 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.«> = / ^ C ' W . Estimate the maximum length of time compatible with quantum limitations before the pencil falls over. O ) . E. As G. In the same limit the momentum uncertainty Afc grows beyond all bounds.x'.71) > and (7. ^In Eq. (7.75) implies therefore: The more precise the coordinate of a microscopic particle is determined. (7.54) these parameters appear in the combinations moui/h (dimension: l e n g t h . Drastic idealizations are then required in order to reformulate the classical situation into that of a quantum system.79) According to Eq. <7*» We assume that initially. 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=ex2l2b\ (7. (7.. In the reverse case a sharp localization of the particle in momentum space implies a correspondingly large uncertainty of its spatial coordinate. the less precise is the determination of its associated momentum.4 The Inverted Harmonic Oscillator 123 p = hk is the canonical momentum associated with x. i.t) = V 27Ti s i n h t exp 2 sinh t {(x2 + x'2)cosht2xx'} .
124 CHAPTER 7.85) = —r= o2 7 2 sinh 2i.t) = J oo dx' Setting A2:=±we can rewrite ip as ex b2 and AB sinh t' (7. = .83) sinh 2 t + (i/2b2) sinh 2i 2 sinh £ 2A 2 sinh 2 1 (7.79) into (7.. 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 .84) dx' oo tvirl/2b With some algebra one can show that (B2 2y and A2 sinh 2 t f7 s ^ sinn / + icotht) = .cosh t sinh t' Wisll2b ^ (7. + cosh 2 1.80) we obtain „/2 oo ex P 2 s i n h t { ( : C + x')^ht2xx'} \/27usmh .87) <9 2 „9 2 sinh21 i? = A + B f / ^ d x e " ^ / 2 " 2 = v^F<*.(i/2) sinh 2t 2 2 2C 2 (i) + *c^.88) .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. Green's Functions Inserting (7. (7.(Ax' + Bxf \/27risinh + (B2 + ^cothi)a..t) = J oo P .2 (7. ' 2\Aswht % cosh t (7.84) we obtain* ip(x. A .83) ip(x. 9 (7.82) i/f(x. (7.81) and (7.86) Evaluating the Gaussian integral of Eq. .
4 The Inverted Harmonic Oscillator The real part of Eq. C2(t) (7.89) = exp[x2/C2(t)} 7rV2C(t) (7.. In taking the modulus of ip(x. (7.86) has C2(t) := b2 cosh 2 t + 72 sinh 2 1.90) with the expected limiting behaviour \im \ilj(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.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). •&(t)=//c(t) = / or.7.t)\2 Ji == 71 / V " Jo dO ? 6' (7.t)= where Co(t) / dx\^(x.t)\' 2 exp[x2/b2] rrVaft In the following we choose 6 = 1 .92) The sojourn time T is now defined as the mean time T given by T:= r*° ioo 2 (M)s r*(f4 d^e.91) C(t)' (7. since dy Jo the derivative dT_ ~dT For b = 1 we have f°° dt= / (7.t)\' 2 125 (7.£ 2 (7. _2 1 exp C 2 (i) = l + 2sinh 2 i. t) the phase ip drops out and one has (x.95) . We have therefore .94) / dxf(x)=g'(y)f(g(y)).
97) sinh 2 1 = .97) and obtain dT ~dl or with £ = ?y/7: 2 2 d^e. C(t) = .99) .= \ / l + 2sinh 2 t. B. p. 136.=o V(i v W + v2) oV "dT J L 1 f A x 2 __H_I r'=s/21 Adx'. Dwight [81].126 We set I C(t)' so that dT ~d~l Now.1 ) 2 \n J ^"•""ff^^V^K?.96) °° l Jr.(4 2 . V and (cosh t = y 1 + sinh2 t) 1 r CHAPTER 7 Green's Functions dr} dt C'lt) ' C(t)' (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.=0 V^ C'(t) I Jt=0 V^ (7.2)(p_7?2)(7.. (7.98) We insert this into Eq. £ 1 dr\ 2Z %_ rl 2 2 yfx Jr. + 0(e'2) (7. we obtain dT _ 1 ~dJ~l 'See for instance H..From this we obtain C(t) C'(t) Z2 ^(r?2_.
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.3 How stable is the particle on top of the eggshell in quantum mechanics? E x a m p l e 7. U) (7.4 The Inverted Harmonic Oscillator 127 for I sufficiently large. 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. Zwiebach [294]. 7. It follows t h a t T~lnZ.100) Reintroducing the dimensional parameters we had set equal to 1. Fig. p. (7. T h e following Example 7. See e.101) A more detailed evaluation of the constant can be found in the paper of Barton cited above. this is T ~ ]n(ly/mou>/h). B. 238. is instructive in revealing the basic quantum mechanics involved in the finiteness of the sojourn time. 2 2 2mo mow x The classical equation of motion is therefore' it — uj x = 0 (7. Such a field arises in the spectrum of states in string field theories and is there called a "tachyon". this equation is the equation of motion of this scalar field with negative masssquared.g.4: The sojourn time calculated semiclassically A tiny ball of mass mo is placed at the apex of an upright egg.4.102) 'As a matter of interest we add that in the theory of a free scalar field x. also motivated by this paper. The equation there describes the classical "rolling down" of this tachyon. Explain the parameters entering the calculation.7. .
(7. the smallest macroscopic length.e. pifi) have to be replaced by the positions of the spatial and momentum maxima of a wave packet. i.102)) p(0) = ^/(p 2 >. Ap= v /(p 2 )<p)2. Eq.p(0) cannot be determined with arbitrary precision. We ask: At what time T > 0 does the ball reach the point with horizontal coordinate x = I. i. 2 x(0) ^ where A i .g. and solution x = A cosh Lot + B sinh cot.128 with x2 = LO2X2 + const. A p are defined by the mean square deviations given by (see Eq. so that AxAp= In our semiclassical consideration we set therefore x(0) = y/{x*). (x 2 ) h 2mou 2mou> and so l^J^ULl). (7. . / 7 ^ 2 =2J(x }. ^(x2)(p2). v A x A p = \ (x2)(p2) v = m0u(x2). 2V ' Let A and Bio be the values of x and x at time t = 0.5)) Ax = <J(x2){x)2.e. and for simplicity we take (cf. so that e. Green's Functions x = Aco sinh cot + Bco cosh cot. 2 10 x(0) = B u . Thus we set I = A cosh LOT + B sinh coT ~ (A + B)e"T. For h —> 0 the time T —• oo in agreement with our classical expectation. CHAPTER 7. (p2):=mW(x2). but instead are subject to an uncertainty relation of the form AxAp > h. For a symmetric state like the usual ground state of the harmonic oscillator we have (since the wave function is an even function. Thus quantum mechanically x(0). so that raoto _ We assume now that at time t = 0 the ball is placed at the point x(fi) with momentum p(0). x(0) = A. the expectation value of x is the integral with an odd integrand) <x) = 0 = <p). (10. Quantum mechanically x(0).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. 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 .
It follows that in general one depends on some approximation procedure which is usually described as a perturbation method.(i)+/^(2) + ". EW+(3EW+(32EW +  ((3 can also be thought of as a kind of "bookkeeping" parameter in retaining corresponding powers of some kind of expansion). i. but this is no longer the case for an anharmonic oscillator potential like ax2+(3x4.e.Chapter 8 TimeIndependent Perturbation Theory 8. one would set tt E = = ^(°)+/ty. In the case of the harmonic oscillator potential ax2 the Schrodinger equation can be solved exactly with ease. For mathematical purists the question of convergence of the series is an even bigger challenge. The perturbation method generally described in textbooks — and frequently called RayleighSchrodinger 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.2h2 cos 2 a # = 0.e. In general perturbation series do not converge. in closed form) only for very few potentials. Frequently even the calculation of the next to leading contributions ip^.E^ is already a bigg problem. An example permitting convergent perturbation series is provided by the trigonometric potential cos 2x with onedimensional Schrodinger equation given by ip" + [E. 129 .1 Introductory Remarks The Schrodinger equation can be solved exactly (i.
. the expansions in ascending powers of the parameter h2 can indeed be shown to have a definite radius of convergence. the expansions in descending powers of h2. Schafke [193]..* i.g. TimeIndependent Perturbation Theory In the case of this equation. explore the latter in somewhat more detail and finally consider methods for deriving perturbation solutions of the Schrodinger equation. T.g.g. This convergence can be shown with D'Alembert's ratio test since''' lim . Meixner and F . p. Actually expansions of this type which are ubiquitous in physics are socalled asymptotic series which were originally also described as semiconvergent series in view of the decreasing behaviour of their first few terms. at the end of Chapter 26) — and by comparing the results of WKB. N. Watson [283].2 However. the expansions i> = v (0) + /*V (1) + /*V 2 ) + • • •. E = E^ + h2E^+hAE^ 71 72 7 + . see e.= 0 < 1.130 CHAPTER 8..g. (8.2 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. h hl are in the strict mathematical sense divergent.2) n—»oo n It is interesting to compare the behaviour of the series (8. W.3) *See e. E. In the following we explain the difference between convergent and asymptotic series. e. 581. 8. 0. (8.1) and its terms with the behaviour of the following series and its terms: f(x) = 1 . Cll E = a_ 2 n + a_i/i + a0 + — + —r \ . The series n=0 of the exponential function is well known to converge absolutely for all real and complex values of x. J. It will be seen later (e. concerning also the uniform convergence of the exponential series. ^For further details. Whittaker and G.• + (ir~ X X* 1' 2' r?l Xn + Rn(x).+ .. perturbation theory and path integral methods — that the lightminded way in which perturbation theory is sometimes discarded is not justified..e. which is known as the Mathieu equation. .
. Whittaker and G.3) with the convergent series (8. for larger values of x.0. p. T.22222 . The theory of asymptotic expansions claims that if the expansion is truncated at the least term.29629. .. e.. +0. Comparing the asymptotic series (8. E... 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.1). i.. / ( I ) = 1 — 1! + 2! — 3! H • However.. From the Weierstrass product which defines this function* one obtains the series _ M*l)! = 7(*l) + E [ ^ .e. . Watson [283].g.. . We can see this as follows.000 002 00 . It is evident that the larger the value of \x\.0..376 . + 0. However. As a second example we consider two series expansions of the gamma function or factorial T(z) — {z — 1)!. + •••. .5) Cf. . we have ^ ' ~ = 3 + 9 ~ 2 7 + 8 1 ~ 2 4 3 + 7 2 9 ~ 2187 + " ' ' 1 .22222 . the series (8.0.0.3) can still be used to obtain almost correct values of f(x) for x sufficiently large.g. after reaching 6/27 = 0. 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.8. + 0. and we see that the moduli of successive terms first decrease and then.222.3) diverges for every value of x.4) for every value of x. and in fact increase indefinitely.2.. Thus e. Normally a divergent series is characterized by an ever increasing behaviour of its terms as in the case of x = 1.001 + 0.4938.l n ( l + ^ ) n=l L ^ ' (8.. We observe in the first place that since lim n—>oo co (8. .9876. for x = 3. 235.000 000 06 + • • • . N.. the better the approximation obtained. begin to increase again. /(1000) = 1 . in spite of this the series (8.33333 .0.2 Comparison of Asymptotic Series with Convergent Series 131 where Rn(x) is the remainder sum. This type of behaviour is characteristic of the terms of an asymptotic expansion as we shall see in more detail in Chapter 20.
5) are practically unknown. 236. 16. for ln(z — 1)! one can also obtain the Stirling series ln(zl)! = ln(2vr)5 z+ (z. Thus one can say that the vast majority of expansions of this type in physics is asymptotic and not convergent. at a later point in his treatise Messiah admits that the expansions are mostly § Cf. in fact. Vol. the dominant approximation is.1 between Eqs. Whittaker and G.41) and (16. T. TimeIndependent Perturbation Theory where 7 = 0. Schiff [243].5). 371. p. whereas applications of the convergent series (8. N. It is interesting to observe the comments of various authors on this point. § However. ip) can be represented by rapidly converging power series in e. although this has not been investigated except for a few simple problem^. (16. Sec. In fact. Watson [283]. p. **A.1. that E and \ijj) (i. . 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.J In2. It seems that Schiff tries to cling to the idea that a proper series has to be analytic. In many respects Messiah aims at more rigour in his arguments. meaning convergent." These "rapidly converging power series" in physical contexts are most likely very rare cases. Messiah [195]. ^E.e. This latter observation hints already at the importance of asymptotic series in applications. it is sensible to make the assumption. I. 152. On the other hand Schiff" says:" We assume that these two series (for \I/ and E) are analytic for e between zero and one. 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.132 CHAPTER 8.5772157— Here the series on the right is an absolutely and uniformly convergent series of the analytic function." This is a very clear statement which does not try to pretend that a perturbation expansion would have to be convergent. E. Merzbacher [194]. It is therefore not surprising that he** says: "7/ the perturbation eV is sufficiently small. p. This is not appreciated by mathematical purists. 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. 11 L. the leading term of an asymptotic expansion.
7) as / e'* dte* dt = 1 .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. Messiah [195]. 2 (8. (8. (8. *R. Dingle [70].ft somewhat afraid t o say so himself a n d therefore with reference t o investigations of T . It is therefore suggestive to write iy poo <t>(x) = 1 . p.8. ft Jx du = 2tdt. (8." Of these three authors t h e first. 198 (German edition). which have been studied in great detail.11) A . (8..2. II. possess asymptotic expansions which have for a long time been important a n d accepted standard results of mathematics.2 Comparison of Asymptotic Series with Convergent Series 133 asymptotic.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.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. We know from books on Special Functions t h a t all of these functions. In fact. Merzbacher.. He says. solutions of second order differential equations. we can rewrite Eq.* and we follow some of t h e considerations given there. . Kato: "Indeed the perturbation expansion is in most cases an asymptotic expansion . Replacing t h e exponential in Eq.7) This function has been considered in detail in t h e book of Dingle.^=e~x2 / e^^dt. 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..= / e~t2dt. 8. there is no need for such bias. B.10) Vn Changing t h e variable of integration to u = t2x2. Vol. (8.
e. (8. (8.14) sin7T2.' we have 2 nH 7T 7T (8. du.12) VK Jo n V oo 7 n=o x) 2 2*Qn\(±n)\\x2) (8.e.134 we obtain CHAPTER 8.16) . Using the reflection formula (*!)!(*)! = for z = n + i . we obtain 1+ i. 2 9 f00 / Jo du e" 2(u + x2)^ u <£(*) = 1 T h e binomial expansion .i)[(_!)«' Then.15) < 1 we can insert this expansion into Eq. Then e (n —\)\( n\ 2du u \n Xy/n o du VK n = 0 e 2 + XypK ~"(1+ ^ °° /•oo —u (ni)!/ n! (nn\ 2)! u du+ Jx : e u du 1 ^ ( 1 + n=0 oo /_u_ V x2 ^ ) K ^ n=0 (8. since (—4)! = y/n. TimeIndependent Perturbation Theory <P(x) = 1 . « \~2 =E n=0 l 1 V5F(ni)!(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.^ X e U ~ ( 1 + * \ 2 .13) presupposes t h a t < 1.^e~x \Ar i.12). 7T (ni)!sin{7r(n+^)} ( „ .
8. The expansion (8. (8. As in the case of the gamma function.19) implies either that we ignore the remainder or the correction term given by the integral in Eq. the asymptotic expansion of the error function is.18) or that we insert the binomial expansion (8.15) into (8. i. (8.17) 0(X) = i. Whichever way we look at the result. Frequently "=" is written instead of "~". and in fact so slowly. whereas in a convergent expansion the terms first increase in magnitude. that a large number of terms has to be summed in order to obtain a reasonable approximation of the quantity concerned.r<7r. In the foregoing discussion of the error function we assumed that x was real and positive.l)! = / Jo Then 2 cHnxdt.e. It is fairly clear that the above considerations remain unaffected for argx < 7T.e. ' arg.12) ignoring the fact that the latter is valid (i. (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. We now want to relax this condition and allow for phases argx ^ 0. We now see how the asymptotic expansion (8. i.—yS!L^LLZ e x OO / _ 1\ X2 /CO dueu L \ / V n = Q 1 and we write 0(X)~1 V 7 n=0 v 2^. In an asymptotic expansion the first few terms successively decrease in magnitude. (8. in general. We can actually understand the deeper reason for this in practice. convergent) only in the restricted domain u < x2. much more useful than the convergent expansion. argx = 0.e. /•oo T(n) = (n . 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.19) approximates <f){x) the better the larger x is. .19) originates.19) where "~" means asymptotically equal to which in turn means that the right hand side of Eq.
2 y V . VK Jo V^ Jo (8. 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). Proceeding along lines similar to those above we write <f>{iy) = ^=ey2 I* e^^ds. (8. (8. however. V7T Jo Changing the variable of integration to v (8. (8. (8.^ f V e ^ .20) where we set t = is.26) or else return .^ = .15). we set x = iy in Eq. The situation becomes critical. We can therefore rewrite the integral as (j)(iy) i 2 fy2 y —=e / e~v 1 . when argx = 7r/2. thus with \v/y2\ < 1. (8. =dv. TimeIndependent Perturbation Theory since for this range of phases the decreasing nature of the exponential is maintained.e. We therefore consider this case in detail. we can try to proceed with Eq.23) Throughout the range of integration 0 < v < y2 the integrand is real.21) = y2s2. i. (8.22) we obtain 2 4>{iy) = .136 CHAPTER 8.vds.7) and obtain 4>(iy) = —= j e~l dt = = / es2ds.24) We can read off the binomial expansion of the factor in the integrand from Eq. dv = 2sds = .
26) we can write* 137 v* Jv2 ie « „2 OO / 1 n=0 M ~k n! . Returning to Eqs.27) Then . We know that for v < y2 the integrand is real.30) yvn f Jo V y J Cf.2 Comparison of Asymptotic Series with Convergent Series to Eq.8. 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.e. (8. and the larger y. i. However.29) where 5t means "real parf. for  arg x\ < 7r/2 and for arg x = IT/2 the error function 4>(x) possesses different asymptotic expansions.^ ) * dv JO V y /»00 /»' yV^ 1 i Vsfv 2 ey n oo J0 Jy' 1%) 'dv. (8. i (8.28) n=0 We observe that this expansion differs significantly from (8.24). i.17): J£° e~vvndv = n\. If we proceed with Eq.e. .28) ignores the integral contribution of Eq.28) because then the correction term oc f°£ becomes smaller and smaller.. ey is much larger than something of order 1/y. (8. the better the approximation expressed by Eq. (8. which will be studied in more detail below.19). (8.2 yir (8. and so the integral is real. for large values of y. We have therefore <j>(iy) = ^e^TL fV e~v ( l .27). „2 \v J v^(ni)! y(nj)l ^—' y2n ie"2 /00 ^ ieir /•" W7r /„2 y ^—' _v^{n\)\(v n! .^ ^ p . e ^ ( i y ) ^ ^ . (8. Suppose now we consider Eq.^] * dv. For v > y2 the second integral is seen to be f imaginary and thus drops out in taking the real part.24). — (8. (8.27) and (8.28) we observe that the expansion (8. y y 2 ( n _i)! y arg(* = iy) = i * . Eq. we obtain 4>(iy) = ^ey2TZ f°° e~v (1 . (8.
31).33)) would have • a maximally increasing exponential e+x .e.25) and integrate from 0 to oo.26).. since (8. B. y.28).31) and (8. We have therefore found that XTT n=0 ^ ' and .* We observe that this Stokes discontinuity is a property of the asymptotic expansion but not of the function itself. *Cf. we use the expansion outside its circle of convergence and hence obtain a divergent expansion. (8. (8.30) can be written </>(iy) = ^ey2 fV ev(l^) yVn Jo V y J "dv = ^!rV"E^(4)V (8.8)) 4>(x) = <f>(x) we see that _x2 OO / ] \  n=0 for 7r/2 < I arg x\ < 3TT/2. and • higher order terms of the associated series all have the same sign. R. Eq. (8. This result is identical with that obtained previously. is obtained for <f>(x) with arg x = —ir/2. Dingle [70]. For the sake of completeness of the above example we continue the phase to  arg x\ > IT/2. Chapter 1. With a similar type of reasoning one can show that the same expansion. Looking at Eqs. (8. e xZ ^ (n — £)! „ ^ ) .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. .e. (8.3!) which is (8. i." — E ^ ^ forarg. TimeIndependent Perturbation Theory If we insert here the expansion (8. 7r = . (8. . Since (cf.138 CHAPTER 8.33) we see that the Stokes discontinuities occur at those phases for which these expansions (i.2 OO rz=0 V . The corresponding phase lines are called Stokes rays.
rgx<n/2 argx>7r/2 _ X1T •<—' v — ( n=0 £ (ni)! X2)n ' oo (8. e. we have e z2/2 _ 2 e(x Rxj+2ixRxI)/2_ The exponential is maximally increasing for XR = 0. though. Equation (8. p.34) the contributions 1. 1/1 ipv{x) v1 vl (±ix). 9 . and 4>(x) argx=7r/2 = 2 r2 ^ + <f>(x) a.g. Thus if we set x = XR + ixj. i. 5. Magnus and F. (6.36) has an exponentially decreasing solution and an exponentially increasing solution for zero phase of x. and so may not be universally applicable. R.8.36) 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.35) 8. See also Tables of Special Functions.2). pp.32) and (8. Eq. Chapter 1. Oberhettinger [181]. Later we shall make extensive use of parabolic cylinder functions Du(x) which are solutions of the equation dx2 + v+ :X y =o . . B. —1 cancel out. in the sum of the right hand sides of Eqs.e. W. '"Cf. i. These rules.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.2.e.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. which can be written (for v not restricted to integral values) D^(x) DM(x) f xu(f>u(x). 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. for arg x = 7r/2.13. Dingle [70]. apply predominantly to asymptotic series of the solutions of second order differential equations. 91.37) Cf. (8.
Eq. x~l/~1ipu(x). = 7r/2. i. ti asymptotic series (8. (14). 9) and XXI.ei<v+VxvHv{x\ argx = £ .36) is invariant under these replacements.39) (as in the example of the error function).38)) has a Stokes ray at Dl '(X) the arg x = 7r/2.40) § Cf. TimeIndependent Perturbation Theory i. We also observe that the late (large i) terms of the series in (8. as can most easily be seen in the case of the Mathieu potential.e. R. > > The equation (8.38) 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. We consider Dv (x). D^(x) = = xv4>v{x)Jr\^l'1^'Kl'1\A~v~X^^) xv4>v{x) + \a. 0 < argz <  (8. with a real proportionality factor. Chapter 1.37). Chapter 17.x 2 oo «*)  — E^fi=0 v .140 where^ CHAPTER 8. . Thus D{J\x) (i. v — — f — 1. B.7. (8.39) 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. 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. ' S u c h symmetries are extensively exploited in the perturbation method of Sec. (838) We observe that one solution follows from the other by making the replacements^ x — ±ix. Dingle [70]. Chapters I (in particular p. 8. (8. 1 1 Cf. Dingle [70].e. i. B. 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. Hence Dil){x) = xv(j>v(x). R.
8. = ir. (8. (8.e.41) xuct)v{x) + a{x)~u~lipu{x). for xj = 0) and therefore possesses a Stokes discontinuity there. Applying Dingle's rule (1) (to the Stokes discontinuity of ^v{x)) we obtain on the ray D<P(x) = x^M^ + aUxy^M^ + l^e^e^lxrMx)] (8. sm(iri/) = —a/3/2 or a/? = 2sm(irv). Thus for arg x > ir/2 we have D$\x) = xp(t>u{x) + = 2]iaei<v+^x1/l^v{x) ^ < arga. (8.e. e +fx2 _ 2 2 e±(x Rx I+2ixRxI) is maximally exponentially increasing (i. On reaching ir the part containing ^ ( x ) . the extra phase factor u ei7r/2 a n c j t n e p h a s e ew7r/2 0 f x <j>u(x) (so that as the rule requires the added contribution is 7r/2 out of phase with the first). Applying rule (2) we obtain D$\x) = U™ + ^iap) (xyMx) + a{x)vlMx) (843) for IT < arg a. < 7T. We therefore proceed to continue (8. < 37r/2. The value of Du (x) on the other side of the Stokes ray. another half of the discontinuity appears on leaving the Stokes ray on the other side. (8.44) + ^iap) = 0 .e. i. as yet an unknown real constant.41) to arg a.41) the Stokes multiplier a is. In (8. = ir. with \x\ = (—x) for arg a.42) T ^(eiwu i. i. 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. we must have at arg x = 7 in the dominant factor on the right of Eq. Since Dv (x) is real when x is real.40).2 Comparison of Asymptotic Series with Convergent Series 141 for argz = 7r/2.e. where in the first line on the right hand side the second term contains the real proportionality constant a/2. is given by • Dingle's rule (2): One half of the discontinuity appears on reaching the Stokes ray.42) el™ + ±iafi\ {x)v<t>v{x) + a{x)vl^u{x). beyond arg x — 7r/2.
i. (8.—^ ON • — ^ > + ^ ^ L . (z)\(ziy. (8. (8.e.45) Thus the right hand side of this equation remains unchanged if v is replaced by —v — 1.e.46) For a given function a(y) this equation determines /3(u).e.14).48) We compare this with the reflection formula (8. a{y)P{y) = a(u .48) we can set <8 49) ' ^ ) = {_V\)V PM = ^f> or ( 8 .51) . i. a(u)P(v) = 2sin(7r^) = 2sin7r(i/ . Hence the left hand side must have the same property. 8.1) (8.e.42) that a multiplies ipv(x).42)) for argx = ir (in the second line of Eq. argx = vr. (8.45) is therefore given by a{v)a(v . TimeIndependent Perturbation Theory Now a and (3 can still be functions of v.1) = ±a(y).1). (8. Eq. . Dl \X) is (cf. i.1)! TT.50a ) (aMb) »W = f "M=(=^)i In order to decide which case is relevant we observe from Eq.1) = 2sin7ri/.47) Equation (8. (3{v .v . We can see that the equation is satisfied by (3{u) = ±a(u .45) that af3 = 2sin(7r^)) D^\x) = (8. or sin7r(2: + 1) sin irz' W ( = 7 ^ ) ! = 2sil"r2Comparing this with Eq.38) = cos7ru(x)u(f)1/(x) + (—x)ve~*x COS7T^— —r a(u)(x)l'1^u(x) (2i . (8.1) (8.l)0(v . (8.42) we insert from Eq. V i=0 \U0 v 2w .142 CHAPTER 8. i.
e^ . which vanish at x = ±oo.. solutions of Eq. (2inl)! (n1)! n! {n2i)\ So far we have been considering the asymptotic series expansion of Di. (8.53) We have therefore obtained in a natural way the quantization of the harmonic oscillator.8.e.A / ± sin^Cx)"M* 2 V ^ .e.36).. We observe that for v — n = 0.50a) because this enables us to define normalizable parabolic cylinder functions..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.1 ) ( .36)) in the domain 0 < arg x < IT.1 ) ! 4~L i\{2x2Y i=0 V2^(x). In a similar way we can examine Dv (x) and its Stokes discontinuities (at argx = 0.7). ' (X) (cf.52) (argx = 7r). It should be noted that in Eq. (2x2)1 = COSTTU(X) v . the error . Chapter I.. (8. (8.*"ei°> f ( ^ _  _ L _ = (!)««(. u 1 y> {2i + v)\ 1 n 0. (8.2^1 v y 7T i=0 ' (2i + v)\ z!(2x ) v (8. For details we refer again to the book of Dingle.2. Then (in the second step using (—u — l)\v\ = — ir/sm(Trv)) DW(x) ~ c o s ^ ^ * f ^ .7r). Eq. the second contribution vanishes and we obtain £>«(*) ..e.) (8. i.53) the factor (—n — 1)! is to be understood in association with (2i — n — 1)! in the numerator. (8.)(.35)..I / .„ _ix2^(2ivl)l e 2 > —. i. 8.1. i. However.3 Derivation of Asymptotic Series from Differential Equations 143 We choose (8. —f. Proceeding along similar lines we can continue the expansion into the domain (2) ir < arg x < 2ir. ' ^ (i/l)!z! ^ .
. x. x—>±oo (8. cf. We write Eq. x°. and demonstrate.38) can be obtained.54) we consider again the important equation of the harmonic oscillator. Eq.54) (8. (8.56) ~l^(8 57) Q{x) =  We then set y = es^ so that the equation becomes (8. (8.58) Since x2 is the highest power of x appearing in x2Q(x) we set S(x) = ]S0x2 + SlX + S2 In a.36)) 0 where + x2Q(x)y = 0. (8. % = {. Instead of dealing with Eq (8.e. i. • • • can be determined.144 CHAPTER 8. Thus 2S 0 Si 5o + 25 0 5 2 + ^ + ^ and so on. Solving these equations we obtain = = 0.37).59) and equating to zero the coefficients of powers of x2. \ + U (8.36) in the form (coefficient of — x2 here chosen to be ~.^ + 1 ) .60) into (8. ^ = 0 dxA dx with the boundary conditions lim 4>(x) = ± 1 . S2. previously we had ^. + — + ^ + • • • . we obtain equations from which the coefficients So. how series of the type (8.55) The large x asymptotic expansion can also be obtained from the differential equation. TimeIndependent Perturbation Theory function can also be obtained as the solution of a second order differential equation. S\.e.. (8. i..60) 2 x x l Inserting (8. ^ + 2 . (8. Eq. 5„=4 Sl=o.36). 0.
37). 2 X 2Wi.36) have correspondingly e±x / 2 ) . p.91. In particular we shall need the asymptotic expansions of parabolic cylinder functions in various domains. i. 2 ) + V2(l)! lt1! 2 1 — v 3x\ 1 2 2' 2 where iF\(a.61) (the dominant terms of (8. and it is clear t h a t higher order terms follow accordingly. In the following we are mostly concerned with asymptotic series in some parameter. a z a a ( + 1) z2 which has the unit circle as its circle of convergence.62) 4 \ 2 XiFi 2'2 .38).b. The derivation of the asymptotic series from integral representations (cf.8.+ X M W = Q. etc. function.z) is a confluent hypergeometric T. 4~l~2 ' i i _i. (8. do not enter here into their derivation and simply quote the result: W i t h \x\ .35) and (8. therefore. (8. T h e parabolic cylinder function normally written Dv(x) is frequently defined via the Whittaker function WK^{x) which is a solution of Whittaker's equation d2W dx2 Thus* Du(x) = 24 + 1 K 1 + T+ . Magnus and F. . Oberhettinger [181]. Nonetheless we shall need asymptotic expansions like (8.3 Derivation Then of Asymptotic Series from Differential Equations 145 eS(x) = 2 e±^ xS2 1+ 0 (8. / i N .37) with (8.19). Whittaker and Watson [283]) is now somewhat elaborate (although in principle the same as before).38) in a variable for the solutions of approximated differential equations. » 1. are asymptotic series in a variable x. \x\ ^> \v\ and (a) for  a r g x  < D„{x) 3/4TT: e 4 x x i/(i/l) 2x2  v(y\){V2){yS) 2Ax2  .e. n 22e"T JziL (8. We. T h e expansions (8.63) **W.
TimeIndependent Perturbation Theory (b) for 5/47T > arga? > n/4: Dv{x) ~ e 4 v x v{y\) 2x 2 V{y\){y1){yS) 2. 349.4x 2 27F (*/!)!  e^e^x""1 1 + (v + l)(i/ + 2) 2x 2  (i/ + !)(»/ + 2)(i/ + 3)(»/ + 4) 2. top of p. 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 3.64) (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. 8. N. . T. Watson [283]. E. If Sn(x) is the sum of the first n + 1 terms of the divergent series expansion .146 CHAPTER 8.0.1 a2 an a0 + — + ^ + . 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.4 Formal Definition of Asymptotic Expansions Finally we introduce the formal definition of an asymptotic expansion. Whittaker and G.66) 7 =e(^) 2 i(3x)2ei(^)(?Jx) decreases exponentially.67) Cf. Thus there is no contradiction^ between (a) and (b).+ — + ••• of a function f(x) for a given range of a r g x . whereas e~x '4 increases exponentially.4x2 (8. 4 (8.65) 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.
68) The definition(8. (8. . It may be noted that Whittaker and Watson's internationally aclaimed text on "Modern Analysis". A critical discussion of the definition can be found in the book by Dingle [70]. in particular with regard to possible exponentially small contributions to the expansion for which e.5 RayleighSchrodinger Perturbation although Theory 147 lim \xn [f(x) . we set E^En * > ^ n = En°1+eEnV = + e2EnV + . In fact.72) For the eigenvalue E near En and the eigenfunction \l/ near ipn we assume power expansions in terms of the parameter e. which is assumed to be small. possesses a whole chapter on asymptotic expansions.69) and the consequent nonunique definition of the expansion. i. 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. The equation to be solved is the equation H<S> = EI/J with H = H0 + eV.68) is (effectively) simply a statement of the observations made at the beginning. (8.Sn{x))  = oo n—>oo (8. The definition is due to Poincare (1866).g.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 (871) / dx%l)*m(x)lpn(x). lim xneW x—>oo = 0 (8. 8.5 RayleighSchrodinger Perturbation Theory RayleighSchrodinger perturbation theory is the usual perturbation theory which one can describe as textbook perturbation theory as distinguished from other less common methods.e. thereby (unknowingly) discarding the vast majority of expansions which have been used in applications to physics. (8.73) Vn + q/#> + e2T/i2) + • • • . first published in 1902. Chapter I.
4 0 ) M = / ^C(4 1} . °° i J or 0 = EWJdx{rnVil>n). We obtain then the following set of equations: Hrj. (879) 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. (8. so that we choose / • dxrpM^O.n = 40)V>n. 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.76) The first equation is that of the unperturbed problem. (8. we write in the position space representation ^ ^ E ^  (878) A contribution to ipn can be combined with the unperturbed part of the wave function \P. (8.77) The states \tpn) of the unperturbed problem span the Hilbert space"K. . i. ( ^ . i. Ho^+V^n = EW^+E^n.V)i. TimeIndependent Perturbation Theory We insert these expansions into Eq.e. (8. J f°° dxrn E(^ 0) .w » .70) and equate on both sides the coefficients of the same powers of e.i4 0) )V4 1} = {EM .n.78) into Eq.As a consequence the state vl/) is a vector in this space.e.75) HoW + vW = E^+E^+E^.77) and obtain {Ho ~ 4 0 ) ) E "i^i = (Ein] ~ W n . With the help of the orthonormality condition (8. (8.e. ^ ) = 0. We insert the ansatz (8. i.74) (8. The second equation can be rewritten as {Ho .148 CHAPTER 8.
the procedure forbids degeneracy and we can consider here only nondegenerate eigenstates. multplying Eq. with En .4 0 ) M 2 ) = ( 4 1 } .82) *> = ^ + E . In order to appreciate the structure of these expansions it is instructive to go one step further and to derive the next order contribution. i. We multiply the equation by ipm. (8.8. (il)i. i. We therefore consider Eq. the coefficients a\ . (8. (8. (8. and hence It follows that to the first order in the parameter e: En = 4 ° ) + e ( ^ .84) (885) ^2) = £ a S 2 ) ^ (excluding i = n for reasons discussed above).V)^ Setting + 42tyn.76) which we rewrite as (H0 .5 RayleighSchrodinger Perturbation and so Theory 149 E& = {n\V\n) = Wn.^(EW ~ V ) ^ + E& / J—oo dx^miln . It is clear that we can proceed similarly in the calculation of the higher order contributions to the expansions of En and \I/n.79). (8.Vil>n) (880) In order to obtain ipn . (883) Thus the procedure forbids the equality of any E\ '.Viprt e (8.a\^(E:S0) „• 4 0 ) )^ = / J—oo dxrl?m(E$V)1>n.84) from the left by ip^ and integrating over the entire domain of x we obtain coo J —( oo />oo / oo dxi.m ^ n and integrate: oo /•oo / oo dxrmY.e. ^ n ) + 0(e 2 ). S ' ti^+Q^)• We observe that Eq. we return to Eq.e.83) makes sense only if Ef] ^ EnV for all i + n.i ^ 0.
Inserting the expression (8.V'>pTl (E^Eny V'fc (8. (8.^ ) + $& w £ (^fc.2 (V> m .4 0) ^+ e 2 E (lpk. V ^ i ) X (Q) (Vi.79) •/—oo oo / Using (8.Vlpn) E^ .^(8. (8.86) E^ iy^n JJn J J i ' i^n For m ^ n w e obtain from Eq.(1) (i&0) . ~Ej E E.£i°») 2 Hence +^£o (£«? .90) + £ ( 4 0) 4 0) )(4 0) ^ 0) ) ^ +' .Vlpn)(lpn. (8.e.^j)(V .B i 0 ) ) ( £ f . n) £f) (8. ^ n ) j&iift + ^n . FVn) + e 2 ^ ( V > n . TimeIndependent Perturbation Theory Setting m = n we obtain with (^„.V^n) . En0) (8.V^j)(^j.V4>n){lpn.86) into the expression for En we obtain En = 4 ° ) + e ( ^ n . ipn ) = 0 i.Vlpn] M _ P (0K 2 ^r . j.VV .89) • • • where we guessed the term of 0(e 3 ).85) together with (8.81) we obtain (2) _ _ {lt>m.88) (Q) Inserting Eq.^w .Vlpn)E.88) into xjjT we obtain similarly * r i + f ^ i^n (lpi.87) (i>k.150 CHAPTER 8.
91) The coefficients Cj follow from the recurrence relations.. We then obtain (lpk. These recurrence relations allow one to reexpress expressions like Xm^n as linear contributions of ipi with constant coefficients. .lpn+i) = ^ Cj5k. sometimes expressed in terms of projection operators which one can construct from the states \ipm) (see e.89). in the form i Adopting this procedure for the perturbing potential V(x) we can set V(x)i. We can therefore rewrite tpn ': The perturbation theory described in Sec.90) can be found in just about any book on quantum mechanics. 8.n(x) = J2crtn+ii (8.g.7 is based on this procedure.90) is the term with only one summation J2 m the contribution of 0(e 2 ).(0 4>k The sum J2j^n includes a term with j = k which is exactly cancelled by the other contribution. i. These functions generally obey one or more recurrence relations which are effectively equivalent to the differential equation in the sense of difference equations. An ugly feature of the expansion (8.e. Merzbacher[194]). (8. + l^t /„(0) V~^ c kjCjn &(Efi»Ei )(EfE&»)\ 0) „(0)w.n+i = Ckn(2) We can now rewrite ipn ' as / (2) _ V"^ c ^ n knc0 ~ 2j ~ TJfi)0 ) k^n (4 ^) 2 n(0). The functions ipn(x) are eigenfunctions of some secondorder differential equation.5 RayleighSchrodinger Perturbation Theory 151 The expressions (8.8.Vlpn) = y^Ci(lPk.. One would prefer to lump it together with the other contribution.
in a central force problem all have the same energy unless the system is placed in a magnetic field. Em can now be linear combinations of ipni'ipm. is not uncommon. + c22^m). Em = 4 ° ) +eE$ + e2E$ + .ipm) = 0 = (^m. (8.96) (Ho .97a) (8. i.4 0) )</4 1} = (En1] ~ V)(cu^n (Ho . i.. (8. (8.ipn). .152 CHAPTER 8. I — 1 .94a) into Eq.96) are effectively the known equations of ipnitpm and so yield no new information.e.70). For example the magnetic states enumerated by the magnetic quantum number m.4 0 ) ) ^ } = (Em] . (8. so that En = 4 ° ) + eE™ + e2EP + ••• .Hence we write * n = CllV'n + CisV'm + € ^ + d 2 ^ 2 +••• . Equations (8.93b) The lowest order eigenfunctions belonging to En. are allowed to be equal since otherwise the expansion becomes undefined.97b) with ipn y£ ipm. Proceeding as before we obtain Ho(cniln + Ci21pm) = En0) (cU1pn + C121pm) = EnVn. degeneracy. .e.95a) (8. In order to deal with the problem of degeneracy we consider the simplest case.94a) * m = c 2 i ^ n + c22tpm + e ^ + e ^ + ••• (8.. . (8. (H0 + eV)^n and (HQ + eV)Vm = Emym. Equality of eigenvalues. all n.93a) (8.6 Degenerate Perturbation Theory It is clear from the expressions obtained above for En and tyn that no two eigenvalues En '.97a) and (8.We now insert Eqs. that of double degeneracy En = Em . —I.97b) are .94a).V)(c2l^n + ci 2 ^ TO ). i. .93b). (8. TimeIndependent Perturbation Theory 8. however. . Equations (8.93a).94b) with (ipn. (8. (8.95b) H0(c2l^n and + c22ipm) = En°\c21^n + c22^m) (8. m = I.e. (8.
Vil>m)] V^m).ipm ) the equations (iPn.y>m.£ < « = ag)^) . ^ ) = ( V .EJ®(lPn.6 Degenerate Perturbation Theory 153 inhomogeneous equations.m). c2iV^n + c22V7pm).ci 2 (V„.ipn n) = 0 = (ipm. Wm./>£>) = C2l[E£> . t f ) = a k 1 } 4 0 ) " Eg>aU = 0.98b) + Culpm) + cuV^m). ^ o ^ ) ) = 0. (^n.H0^) .^ ( ^ . ^ ) ) = CU[EW .C 21 (^ m .m. Equations (8. We now multiply these equations from the left by ip* and integrate. Similarly Wn.^ ) ( ^ m .V^n). V^n)\ . (8. (^m.cnVipn and tyi. But (8.(^04°))41)) = EM^CuTpn (ipl. . we obtain with (ipn.J#»a&> = 0j since ii4i = En (degeneracy). V^n). £ aW^°Vi) .H0il>W) = cn[EW . cn(ipm.99) and ( ^ ^ O ^ ) " ^ ^ . Vi/>n)] . H0lpV) . and ( ^ m . (8.C22(Vn.H0xl>$) . H0lP$) = C22[E$ ~ (Vm. Vlf>m)] . Vl/.99) can now be rewritten as (frvvA \1"'Vr\ )(cn)=w(cn) (siooa) . Then (^.(lPn.8.(Vn.98a) Setting I = n.
1.100b) determine C2i. splits the problem into two independent onedimensional systems.C22. in particular in parabolic coordinates (see Sec.5). like that providing the Stark effect (proportional to rcosO).3). (8. 9} ip.Vrl>n) . At x = ±00 the constant is zero. A onedimensional system.ci2. \nip = ln(j> + c o n s t .Similarly Eqs.154 and V (^m. if)' (/>' — = —.E™ (Vn. If one wants to perform a perturbation calculation in such a case. (8. It follows that ip and <j> are linearly dependent. It is then possible that a special perturbation. the secular determinant.6. d>" E=^.^Vm) 0. as we saw.E)tp = 0. . i> 4> Hence there is no degeneracy. Example 11.^ ' V = — (•>!>'<t><t>'i>). . i.Vlpn) CHAPTER 8.. An example which emphasizes the significance of the spatial dimensionality in this context is treated in Example 8.This completes the derivation of the first order perturbation corrections if an unperturbed eigenvalue is doubly degenerate. (V>V</>V>) = const. Let ip and < be two eigenfunctions belonging to the same eigenvalue E. In this particular case therefore nondegenerate perturbation theory can be used (cf.Vl/>m) J \ C22 J m \ C22 J V ' These matrix equations have nontrivial solutions if and only if (lf>n. Then / > i>" 2— = V ip Hence 0 = V ' V . i. dx i. cannot be degenerate because there is only one quantum number corresponding to one definite energy and hence cannot be degenerate.100a) determine the coefficients Cn. the degenerate perturbation theory must be used. However.e.e. Nonetheless we shall see that the Schrodinger equation of the hydrogenlike problem can also be separated in some other orthogonal coordinates.1: Do discrete spectra permit degeneracy? Show that in the case of a onedimensioanl Schrodinger equation with discrete point spectrum there is no degeneracy. 4.101) This secular equation determines the first order correction En • The equations (8. there is degeneracy in the case of the hydrogen atom with Coulomb potential (Chapter 11) which is based on threedimensional spherical polar coordinates r. ip = c<p. 11. Example 8. Solution: We consider the Schrodinger equation with potential V in the simplified form i>" + (V . just as there is no degeneracy in the case of the onedimensional harmonic oscillator (Chapter 6).e. TimeIndependent Perturbation Theory (l/>m.
V0)<f> = 0 (8. See also R. Dingle [72] and H. the method has to be applied in each domain separately. one quickly arrives at clumsy expressions which do not reveal much about the general structure of a higher order perturbation term.7 DingleMuller Perturbation Method 155 8. We consider an equation of the form Dcj) + (E .e. W. i. in particular in applications to periodic potentials. cf.103). and such that one can even obtain a recurrence relation for the coefficients of the expansion. J.7 DingleMuller Perturbation Method We have seen above that although the procedure of calculating higher order contributions of perturbation expansions is in principle straightforward.102) where D is a second order differential operator and E a parameter. with some restrictions on the function V. also possible in areas not of immediate relevance here.102) can be written D<f> + (EQ . B. H. The subdivisions (8.104) are chosen such that the equation Dcf) + (E0 .8. In the following we describe such a procedure as first applied to the strong coupling case of the cosine potential or Mathieu equation. Muller [210]. and the solutions then have to be matched in their regions of overlap. Muller [73].105) (8.V)(p = 0. of course. Applications of the method are. We assume that V can be expressed as the sum of two terms. J. . (8. T For instance spheroidal functions can be studied with this method.103) such that if E is written correspondingly E = E0 + e. W.t Since in general a solution is valid only in a limited region of the variable. W. (8. (8.e)<t>.106) (8. the considerations given below would be valid even if D contained only the derivative of the first order. B.V0)<t> = (v.104) "This method was developed in R. Muller [216]. Thus the method is a systematic method of matched asymptotic expansions. It is immaterial here whether D does or does not contain a first order derivative — in fact. V = V0 + v. anharmonic oscillator potentials and screened Coulomb potentials. (8. J.* In various versions this method is used throughout this book. Dingle and H. Eq. a potential. It is therefore natural to inquire about a procedure which permits one to generate successive orders of a perturbation expansion in a systematic way.
We write the coefficients "(m. v = {D+EmVo)<t>m. m + s) which are determined by the recursion formulae. s an integer: (ve)(pm = 'Y^{fn.Em+s)(/)m+s. and so v (8. m + s)0m+s.108) with coefficients (m.V0)<f>^ .+s ' (8.V0)<f>m+a = 0. the equation Dcf>m+S + (Em+S . (8.VQ) ^ s c s4>m+s = 22 s C s(Em ~ Em+s)<fim+s = ( u . Consider (D + Em= (D + E m V0)(4>W + ^(1)) = (D + Em .V0)4>m = 0. the function 4>(0) = <t>m (8. Then if ]e < \E\ and \v . TimeIndependent Perturbation Theory is exactly solvable and has the eigenvalues Em and eigenfunctions (f>m.e\ < \V0 .E0\ (the latter in a restricted domain).Vo)(j)m+s = (Em .m + .113) . The next step is to use the recursion formulae of the functions 4>m in order to reexpress (v — e)4>m as a linear combination of functions <fim+s.107) represents a first approximation to the solution (f> of Eq. Considering (frm+s we have.109) D(j)m+s + (Em . where m is an additional integral or nearintegral parameter (depending on boundary conditions).) This result is obtained as follows.m + s)" in order to relate them to "steps" from m to m + s. since D<f>m + (Em .110) If we write the next contribution <f>W to <p(°> in the form m+st (8.m + s)4>m+s s (8.102) with E ~ Em.111) s then (see below) 0(1 = E (m.e ) ^ = 2 ( m .156 CHAPTER 8. (8.
m + r) m ^ (8 118) r^O ' ~~ Em+s)\Em ~ Em+r) together with n _ / x y^(m.m + s) 157 (B\IA\ s^O: 1 cs = J m _ ^m+s . We observe that the coeffcients of expansions (8.m + s)(m + s. and the second expansion is an equation from which e and hence the eigenvalue E is determined (i. Thus (m.m)4>m in (8. (8.8.m + s)(m + s.102) to order (1) of the perturbation (v — e). V ^ V ^ {rn. .m) 1^1^ IW .m + s)(m + s.T? S The first expansion determines the solution 4> (apart from an overall normalization constant).m) + > ^ . (8.^ (Em . Eq.7 DingleMiiller Perturbation Method and so (m.113) must vanish. the coefficient of the term (m. .m) E J^ ( m ~ Em+S) ' • fa. (8.e.Q (m..115) Repeating this procedure with cf)^1' instead of 4>^ we obtain the next contribution 0(2) =YY together with (m.m + s)(m + s.118). (8. /01_. The factors (Em — Em+S) in the denominators result from the right hand side of Eq.m) — —. (8.n\ ^119) + . (8.m) — 0 (first approximation).119) follow a definite pattern and can therefore be constructed in a systematic way.m + r)(rn + r. £ ^ T^o (^m ~ Em+s){Em — Em+r) ( m ' m + s ) ( m + g ' m + r ) $m+r (8.116) m ^ v (m. We then find altogether ^  l^m  Em+S) + y^y^ SJ.„N This procedure can be repeated indefinitely.110) and are therefore related to the inverse of a . . .Em+S) = 0 (second approximation).TP \(T? .114) To insure that 0 = ^(°) + f^ ) H is a solution of Eq. (8.119) is effectively the secular equation). .117) „ .
[q(llq2 + 1) + 2(33g 2 . Imposing normalization later one obtains the normalization constant as an asymptotic expansion. A further conspicuous difference between this method and the usual RayleighSchrodinger perturbation method is that the first approximation of the expansion.[ 4 ( 8 5 g 4 + 2q2 . as already mentioned after Eq. . i. S. The eigenvalue expansion obtained is an asymptotic expansion. One should note that in this perturbation theory every order in the perturbation parameter also contains contributions of higher orders. § H.I2q + 64) + 256Z 3 (41 g . This can in fact be done and will be dealt with later.e.7 1 9 2 + 32g + 2976) 3•215g2 +32Z 2 (252g 2 . this is — so to speak — the price paid for obtaining a clear systematics which is essential. in which q = in + 3. The systematics of the coefficients suggests that one can construct the coefficients of an arbitrary order of this perturbation theory. Other applications can be found in the literature.2). .423) + Z(2720 9 3 . An explicit application of the method (not including matching) seems suitable at this point.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. does not occur in any of the later contributions. as will be seen later.2 therefore illustrates the use of this perturbation method in the derivation of the eigenvalues of the Gauss potential. since it will be used extensively in later chapters. for the investigation of the behaviour of the late terms of the perturbation series. This means the solution is not normalized. i. (8.158 CHAPTER 8. . as will be seen in later chapters. (f>m in the above.e. TimeIndependent Perturbation Theory Green's function. and is typical of a large number of cases. for instance. ^The potential A r 2 / ( 1 + gr2). Miiller [211]. The result is the following expansion.* Example 8. . W. An additional aspect of the method is the full exploitation of the symmetries of the differential equation.6q + 1)1 + 24(5? .38). J. (8. 2 k2 + g2 = ga(2l + q)^[3(q2 „3 + l) + 4(3q~l)l + 8l2] 32ag r .120) n4 . n = 0 . has been treated by R. Kaushal [146].1)( 2 + 64J3] — . for instance. The following Example 8. 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. 1 . 2 .9) + 4096J4] + 0 ( l / g 3 ) .
_ „!2 ^ ( " « 2 r r 2 > . <«»> In the limit g —• oo we can neglect the right hand side and write the solution ijj —* I/IQ.2a2 A.l)ip(a . (8. \2 ..2) where 1/ . We now insert this equation into Eq. (. „2 J„2JL.1). Therefore in the general case we may set k2 + g2 = ga(2l + q) . multiply the equation by onehalf and set h = —a/g.123) the contribution fl<°> Ah + . Then a?ip dz2 fk2+g2 2ga l(l + l) 1 z*)i> ±(T'(^*and S=z2.b.• 0 ( ° ) = ipq(z) = zl+1ez2/4<S>(a. i/> with Dq = d2 . The solution i. . S) is a confluent hypergeometric function. 11. .122) where A is of order 1/g. 1 1(1 + 1) 1 . (fc2 + <?2) 4ga and b = I \—. . 2 .a + l)if>(a + 1) + (a. We then set V>o(2)=2i+1ez2/4X0(2) Now the function XQ{Z) satisfies the confluent hypergeometric equation (cf. 2 The solution XQ(S) = $ ( a .^z2) = i>(a) leaves unaccounted for on the right hand side of Eq. Setting q = in + 3 implies k2 + g2 ga(2l + q). . Sees. 1 .V ! 4 I z S *! V2 ipq(z). 3\ 2 V ' 2. . 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. (8.2 '(l + l) .b. a . .7 DingleMuller Perturbation Method Solution: We expand the exponential and rewrite the Schrodinger equation in the form rfV (ir 2 .0(z) = zl+1ez2^fa. a)ip(a) + (a.121). (8.8.6 and 16. Then the equation is Dqi> = Ah+y 2 4 ^ 1 1 » fc i\ . b.z2 is a normalizable function if a = —n for n = 0 .123) The first approximation (note the last expression as convenient notation) ^ .
S m (a. fa.1) = a .120). In fact. „ . .e. a + r) with the boundary conditions Sn(a.i ^(i)=£V+i i=0 J2 j = (i+2).a]2+L ^ J1[a + 2 . the latter determining to that order the eigenvalue.a]i It is clear that the various contributions can be obtained from a consideration of "allowed steps". a) = 1.b = (g + 3) . a ] i + L ' ^ J 1 [ a + l . So (a.2a = I + q.Z. of course. a ] x + L _ 2 J 1 [ « .3). o + r) + S m _ i ( a . a]i + h fa.^ 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. (a. . a + r) satisfy the following recurrence relation: Smi(a. ( 5 m ( a .2. a + r) = 0 for r > m. fa. a + r — l)(a + r — l . (8. 0 = h[a. a + j)i/>(a + m).160 where (a. TimeIndependent Perturbation Theory (g . 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. This equation shows that a term fiip(a + j) on the right hand side of Eq. The coefficients S m ( a . (a.a + 2]i P .a—lli. one can write down recursion relations for the coeffcients of powers of hl as will be shown later in a simpler context. a + i) = 0 for i ^ 0.a — 211 . As a check on the usefulness of the perturbation method employed here.j^0 ' + jU+1 J ^(a + j) with h[a. a + r + l)(a + r + l . a ] i + 0(/i3) [a. _ J [a . when j = 0.120). a + 111 r [a. i. a + r) + S m _ i ( a . where [a. a) = b . a + i)(a + r.a + j}i+1^(a + j). Now we make the following important observation that Dqip(a + j) = jip{a + j). Hence the next order contribution to tp^ becomes oo i+2 r [a a . a .a]i = .123) and so in R(°> can be taken care of by adding to the previous approximation the contribution fitp{a + j)/j except. one could now use the RayleighSchrodinger method and rederive the result (8.a]1=0. a + r) = i 2 \ m ?n z J 0(a) = J^ j= — m S m ( a . If we now evaluate the various coefficients in the last expansion we obtain the result (8. 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.l . a + 1) = a = In general CHAPTER 8.
except for the canonical algebra which has to be formulated with Dirac brackets replacing Poisson brackets.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.Chapter 9 The Density Matrix and Polarization Phenomena 9.e. We therefore begin with the appropriate classical considerations with a view to generalization. i. the generalization to many particles. although not exactly straightforward.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. 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 oneparticle systems. light. and we shall then see that these lead to analogous results as in Chapter 2. 161 . and this means electrodynamics. proceeds along similar lines. 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 wavelike nature. as we shall see in Chapter 27. 9. Regrettably this has to be left out here in view of lack of space.
2) (9. In the latter case one has the possibility of {A0R x A 0 /) • k > 0.5) which is described as right polarization as compared with left polarization for which (A0R x A 0 /) • k < 0. The Density Matrix and Polarization Phenomena potential A(r. 9. (9.oei(krwt> which satisfies the transversality or Lorentz condition k • A = 0. In general the constant Ao is complex. t) = A(XR cos(k • r — cut) — Ao/ sin(k • r — cot) and at r = 0: A(0. This is the equation of an ellipse as indicated in Fig.i). t) = A 0J R cos(cji) + A 0 / sin(u. t) given by A(r.1) Fig. It follows therefore that A(r. When Ao/? is perpendicular to Ao/ and Ao# = Ao/ the ellipse becomes a circle and the field is said to be circularly polarized. When AQR is parallel to Ao/ the ellipse becomes a straight line and the wave is said to be linearly polarized.1 The vector potential ellipse.1.e. (9.t) = 3ftA. A 0 = A0R + iA0I.4) (9. (9. i.6) . 9.162 CHAPTER 9.3) (9.
2 Reconsideration of Electrodynamics 163 As a consequence of the Lorentz condition (9.A i = 0 = k .2 *i*4 0) (98) with coefficients F^. In classical electrodynamics.e. We set therefore A 0 = Ai + A 2 . p.8).A2 is the sum of the moduli.A«). (9. (9. the intensity of a wave. the connection must be linear.e.11) 'This means we consider the polarization. as e.A(°)). the direction of polarization.y)) components A i . Rae [234]. the vector potential A. Here we are interested in observable properties of light waves — like.2) the vector potential A is completely determined by its two (say (x.A 2 . for instance. 17. which leads to an outgoing wave which we give the label "1". A2. / = 4 0 ) * 4 0 ) + 4 0 ) * 4 0 ) = (A(°\ A<°>) = A(°)2. a calcite crystal or a film of nitrocellulose. (9. 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.AW). (AW. we recall. i. i. (AW. however. not. The intensity 7 of a light wave with amplitude coefficients A\' . (9. determined with the help of a polarizer. Observable quantities can only be those which are invariant under translations and rotations. .k (910) We give the initial wave the label "0" and subject this to an experiment. the fields E and B are observables. (A(°). Then their connection is given by Eq.9) The intensity is characterized by the fact that it can be looked at as the "expectation value" (1) of the unit matrix 12x2. represented by a filter "F".A« we can construct only the following scalar products with this invariance: (A(°\A(°)). See for instance A. I=<l2x2>:=E40)*l*40)i. i.7) We consider now measurements with polarized light.g.e. with k . From the vectors A(°).t in which k and ui remain unchanged.9. AW = £ fc=l. ie.
15) This matrix is hermitian since (Pife)t = (AiAtf = (A*Akf = (AkA*f = {pkif = (Pik). Thus.k l f c 4 ° \ (9.17) .164 CHAPTER 9.k i (9. (9. To this end we observe that as a consequence of Eq.15) in order to rewrite Eq.13): j2^FikAk = Y^AiF*kAi = Y.12) i. (F)=Tr(Fp). (9.A*kF£A* = E and hence F = Fl A p i tkAk. it must be real and hence (F) = <F>*.13) This consideration of translation and rotation invariant quantities is somewhat outside the framework of our earlier arguments.16) i.e. i.e.e.A( 1 )) = ^ ^ F i. the connection between the initial polarization and the final polarization. (A(°). We write the observable quantity as (F):=Y.k (9. Thus (F) = J2FikPki = ^(Fp)a. (9.A*FikAk. Then according to Eq. (9. pki = AkA*. (9. But we can convince ourselves that we achieve exactly the same as with our earlier consideration of observables. We use definition (9. if (F) is an observable quantity.13). i. i. so that (F) can be looked at as the expectation value of an observable F.14) We now introduce a matrix p — called density matrix — which is defined by the relation pik := AiA%. (9. The Density Matrix and Polarization Phenomena Only the second last of these combinations describes the experiment.13) the quantity F must be an hermitian matrix and thus representative of an observable.
1 The classical polarization vector or wave vector A corresponds in the quantum mechanical case to the wave function ip. like that of a bulb. This admixture is described by introducing matrices p.21b) This state is still a pure state.15) with A\ = a.e. For a 45° polarized wave we have 1 . But light. 165 (9. mutually orthogonal polarization directions x and y described by the matrices (see also below) * .18) The vector A describes a particular ray of light. it is a statistical admixture of light rays whose polarization vectors are uniformly distributed over all directions.9. is in general unpolarized.21a) follows from Eq. i.2 Reconsideration of Electrodynamics In particular we have / = Tr(p). b = 1 we obtain the expressions (9. we have (9 19) ' In this case there is no preferential direction.22) For a — 1.* For a lightwave travelling in the direction of z. is given by a single polarized wave.(AAt) = ( £ £ ) .e. i. .( J o ) . The socalled pure case. (9.e. In the present case of a polarized beam of light we have two possible.d "»=(o ! ) " In the case of an admixture with equal portions of 50%. i. that of a pure state. 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. (9.19). The density matrix for the pure state (9. A2 = b as </*) . "incoherently. <»^> (9. b = 0 and o = 0.
(9. Then according to Eq. as a consequence of absorption). Equation (9.™ V 0 n kmAm / ^^jPim^mk' .8) describes the relation between the initial and final polarization vectors A^°) and A^1) respectively in a measurement of the observable F. l l\ " i °\ ) ' ^925^ 9. 1 = .166 and CHAPTER 9. b = —=. Pi35° = \ \ 2 ? )• (9. (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 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 .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. We set in analogy to Eq. (9.8) A(1)=E^40)> k R = R(<*).g. We now inquire about the the way p changes in the process of the transmission of the lightwave through an apparatus. in which the intensity I = Tr(p) is not conserved (e. (926) 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".Z^ 3.15) Pik — A i A k . The Density Matrix and Polarization Phenomena P45° = I 1 For a 135° polarized wave we have a = and l M .
The corresponding condition here is that no absorption of the wave takes place. In the case of Eq.32) (9. (9.31) (9.30) Ti{FRp{®I$) = Tr(F. the observable F remaining unchanged.q. and p(0' remains unchanged. here this contains the dependence on a. that Io = Tr( / 0 (°))=Tr(pW) = / 1 . the dependence on a is contained in F' = R^(a)FR{a).30) the dependence on a is contained in pW(a) = R{a)p^{a0)RJ'(a) (9.27) The measurement of an observable F in the new state with superscript " 1 " then implies according to Eq.9.33) i. (9. the observable is transformed. in other words without creation or annihilation of systems. in the state functional) in analogy to the time dependence of W(p. (9. (9.31). We can interpret the result either as (F) ( 1 ) = Tr(F{Rp(0)R}}) or as (F){1)=Tr({RtFR}p<0)).e.34) .t) in Chapter 2 in what we described as the Schrodinger picture there. The Liouville equation describes the motion of an equal number of systems in phase space elements of the same size. The two cases are completely equivalent and can therefore be described as Schrodinger picture and Heisenberg picture representations.17) (F) (1 ) = Ti{FpM) = = Tr(tiiFRpW) F' = RtFR. (9. 9. i. i.29) where (i. that the initial intensity Jo is equal to the final state intensity I\. (9.4 The Liouville Equation It is natural to go one step further and to derive the equation analogous to the Liouville equation.e.28) (9.e.e.4 The Liouville Equation 167 pW = RpMRl (9.pW). on the other hand. In the case of Eq.
B]. j£ = i[H.21) and (2.35) Thus the matrix R has to be unitary.F}. the matrix M is antiHermitian: Mf = .35) we have 1 = R*R = (1 + M W ) ( 1 + Mda) ~ 1 + (M* + M)da. Eq. = Tr(R^Rp^). For p ( 1 ) ( a ) = p ( ° ) ( a 0 + da) follows from Eq. (2.B}:=i[A.M .168 i. We set therefore M^:=iH. (9. In view of Eq. .33)) we obtain correspondingly in the Heisenberg picture the equation ^ = +i[H.2) has to be taken into account. Comparison of these equations with Eqs. For an infinitesimal variation of the state of polarization we have R{a) = 1 + Mda.36) (9. In Chapter 27 we shall see that the Poisson brackets here are actually Dirac brackets because the gauge fixing condition (9. (9.37) (9. i.39) If the dependence on a is not contained in p^1' but in F' (cf. (9.40) Equations (9. The Density Matrix and Polarization Phenomena Tr(p(°)) = Tr(p^) so that = Tr(Rp^R^) RlR = 1.40) can be considered as quasiequations of motion.p].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. (9. (9. (9.e.39) and (9.46) suggests to define Poisson brackets of matrices by the correspondence {A. (9.32) in the Schrodinger picture that p^iao + da) = = = or ( w i t h p(°> —> p) (9.38) H = Hl R(a)p(°\a0)R\a) (liHda)p(0)(a0)(± + iHda) pM(ao)i[H. CHAPTER 9.p(°\ao)]da.e.
10. we establish their Heisenberg equations of motion.2 States and Observables Every state of a physical system is represented by a certain ketvector \u) which is an element of the corresponding Hilbert space "K. Heisenberg and interaction pictures. (10. To insure that the 169 . the Schrodinger.1 Introductory Remarks In this chapter we are concerned with the formulation of quantum mechanics in general. Thus in particular we establish the uncertainty relation for observables in general.1) This expectation value remains unchanged if \u) is replaced by e%a\u) with a real.Chapter 10 Quantum Theory: The General Formalism 10. 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). Thus the expectation value with respect to the vector describing the state in the Hilbert space does not depend on this phase factor. and consider timedependent perturbation theory.
g. B We first establish the following theorem: If A and B are observables and hence hermitian operators obeying the commutation relation [A. Pi • Pi. The first step in the investigation of a quantum mechanical system is to specify its dynamical variables A. Quantum Theory: The General Formalism expectation value of the unit vector or identity operator is 1.170 CHAPTER 10.3) These fundamental commutators determine the commutators of arbitrary observables A = A(p.Qj] = 0. (10. (10. Such a correspondence with classical mechanics is not always possible. 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. 10. we must have (u\u) = 1. In order to avoid multivaluedness one always starts from Cartesian coordinates in position or configuration space.Pi are postulated to obey the following fundamental commutation relations (again leaving out "hats"): [Qi. For these the following rules of canonical quantization are postulated: Postulate: The operators q~i.q). \pi.2.2. we assume this now. Here. y. and assume that A = A(p. q). however.y.Pj] = 0. The phase space coordinates qi.Pj] = ihSlj. Naturally one can also write (10. z and for definiteness here — which will not be maintained throughout — the "hat" denotes that the quantity is an operator.4) *Even in those cases where this is possible the operator correspondence may not be unique.* a conspicuous exception being for instance a spin system.1 U n c e r t a i n t y relation for observables A.Pi are called fundamental observables. where i = 1. E.B] = ih.2) (F(A)) = {u\u) <"lf. classically qiPi = PiQi. . [qi. which clearly does not hold in the case of operators. such as for instance of the angular momentum operators Li.3 represent the three Cartesian coordinate directions of x.
7b) = = [A.10) . Thus AA = AA = ( i 2 ) V 2 . (10.B.0)1/2 = AA.(A)2 = (AA)2.7a) (B) = 0. Ai = = = ((A2) .5) are subject to the following inequality called uncertainty relation: AAAB In proving the relation we first set A:=A(A). so that (A) = (A) . (10.13) AB = AB = (B2)1/2. Then [A.2A(A) + (A)2.9a). (10..7b).2(A)2 .2 States and Observables then their uncertainties AA.9) ((A2) + (A)2 .12) the Schwarz inequality (cf.B\u) G IK as vectors) we obtain (for an application see Example 10. (10. (10. we have A2 = A2 .8) and.6) B:=B(B). Eqs. C = A..5)) (for A\u).11) (A2) = (A2) . Applying to the expression (10.. (10. 111 (10. = \\A\u)\\2\\B\u)\\2 (10.{A)2)1'2 {{A2) .(A)2)1'2 (1 °= a) ((A2 + (A)2 ..(A) = 0. AB defined by the mean square deviation AC=((C2)(C)2)1/2. (10. Hence (AA)2{AB)2 = (A2)(B2) = (u\A2\u)(u\B2\u). using Eq.1) (AA)2{AB)2 = > (u\A2\u)(u\B2\u} \{u\AB\u)\2. (4. (3. B}= ih. Since A = A— (A).10.B] = ABBA > ^h.2A(A)}  {A)2)1'2 (10. (A{A))(B{B)){B{B))(A(A)) (10.12) where \u) 6 "K is a ketvector representing the state of the system.
Zi (10. and with Eq. since A and B are hermitian. and (b) when (AB + BA) = 0. From this we obtain (uiJBu) 2 = i22 + / 2 . A\u)=cB\u).4). Quantum Theory: The General Formalism Next we separate the product AB into its hermitian and antihermitian components^ and use Eq.3.16) I = h. (AB + BA) This means (AB + BA) is real.19) (u\AB + BA\u) = (u\AB\u) + (u\BA\u) c*(u\BB\u) + c(u\BB\u) = (c* + c)(u\B2\u).15) as (u\AB\u) = R + iI. 4. We can therefore rewrite Eq. Then AB = \(AB + BA) + \(ABZ Z BA) (1 = 8) \{AB + BA) + \ih. (10. for a complex constant c. (10.e.18) >R2 + I2. and hence for condition (b) 0 = = tSee Sec.172 CHAPTER 10. R= (AB + BA).17) as had to be shown. Zi Zi (10. (10. (10. Zi + BA)\ + ^ih. (10.14) Hence (u\AB\u) = /^(AB where. The equaltosign 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)*.13) (AA)2{AB)2 i. (10. AAAB>I=h. . The condition (a) is satisfied when. (10.
we obtain {ALi)2{ALi)2 > \{u\LiLj\u)\2.13) to operators Li — Li. i.e. q) = ih. For states \u). i. (Ly)2=0.Lj] = (LiLj + LjLi) + ih£ijkLk. i. we have LiLj = (LiLj and ^ = y(u\LiLj so that \(u\LiLj\u)\2 Hence {ALx)2{ALy)2 > l + LjLi) + [Li. the following uncertainty h2{Lk)2. + LjLi\u). (Lz)2^0. 9 = h{u\eiikLk\u).e. ALX 10.e. (Ly)=0. Separating the product into symmetric and antisymmetric parts. (A{A))\u)=i(Zc)(B(B))\u). for which one component of L is "sharp"? Solution: Applying Eq. (ALZ)2(ALX)2 > \h2 = (Ly)2. Example 10.10.3 OneDimensional Systems 173 Thus c = — c*. in view of Eq. . implying that the vector it) has to satisfy the following equation: A\u) = i($Sc)B\u) (10. = 0. implying {Lx)=0. What does the relation imply for states u) = \lm).z. Thus in this case Lx.7a). For these we have in the onedimensional case the relation [p. m\lm)). ALy = (u\L±\u) ^ 0. 9ftc = 0. {u\LiLj\u) = R + iQ. (Lx±iLy) ^ 0. such that Lz is "sharp".1: Angular momentum and uncertainties Show that in the case of angular momentum operators Li.3 OneDimensional Systems We now consider onedimensional systems with a classical analogue as described earlier. i. ALZ = 0 (these are states \lm) with Lz\lm) it follows that (Lx)2 = 0.y.) 2 > = x. (10.20) or. (10. = 5R2 + G 2 > 9 2 = hi2(Lk)2. h2(Lz)2. (ALy)2{ALz)2 > ~h2(Lx)2.e. The observables like angular momentum are functions of p and q. Ly are not "sharp".i relation holds: (ALi) 2 (AL.
Then (see below) we can derive for A the following commutator equations [q.B]C + B[A. This is why the operator p requires a representation in the space of eigenvectors of q.pn]=ih^(pn). Eq. Next we have a closer look at the operator q.23) = ihg(q2) (10.q). as remarked earlier. Quantum Theory: The General Formalism Since this says that p and q do not commute (as operators). (10. We obtain these equations as follows.174 CHAPTER 10.e.p2] = [q. . i. We have (cf.p] = ^ihp ~ Proceeding in this way we obtain [q. they do not have a common system of eigenvectors.i h ^ ^ .p) can be expanded as a power series in q and p.22) d ih—(p2). Using [q.qn] = ih^(qn).p)]=ih^^and \p. for q\x) = x\x) we have p\X)=\x)(th^y Now let A be an observable. q]q + # > v] = ~2ihq and \p. i. (3. Similarly we obtain \p. A — A(p. and (c) the spectrum is not degenerate.21b) (10. Q2] = [p.BC] = [A.e.C\. We demonstrate that (a) its eigenvectors have infinite norm (in the sense of a delta function). (b) its spectrum is necessarily continuous.p)] = .21b) now follow with the assumption that A(q.p] = ih we obtain therefore [q.21a) Equations (10.34d)) [A.A(q.21a) and (10.A(q. (10.p}p + p[q.
25) It is shown in Fig. For the cases (b) and (c) we consider the operator U(a — pipa/h (10. .e. an operator. Replacing in Eq. Since p = p\ follows t h a t U\a) so t h a t U(a)U\a) = 1.10. 6K(xx') = — fK dkeik^x'^ 2TT JK sin K(X K{X — x') — x') (10. we obtain [q. = 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).24) lim SK(x — x') — S(x — x').26) it Here p is an observable. U] = ih^= aU.3 OneDimensional Systems 6K(x) 175 Fig.e. > We begin with (a). U is unitary.28) i. We have (x\x') = = where f^ J—oo dk(x\k)(k\x'} (4 = 9 ) ±.21a) A by U. (10. 10. and a a cnumber. i.f^ ^ J—oo dke ikx —ikx' (10. (10.27) (10.1 The delta function for K — oo.1 how the delta function arises in the limit K —> oo. 10.
176 i.q)U^U\ilj). 10. 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. Evidently every eigenvalue has only one eigenvector. q any other observable F(p.29). WU = 1.00).e. cf. b a cnumber). The eigenvectors have the same norm as \x): (x\U*U\x') = (x\x') = S(xx').e. In an analogous way by defining the unitary operator U(b) = eiqblh (q operator. CHAPTER 10. Eq. (10.1 T h e translation operator U(a) We saw above.30) (10.31) . Thus all these values belong to the spectrum of the operator q. and that the eigenvectors do not have a finite norm and are normalized to a delta function according to (jPx\Px) = S (Pxp'x) With the help of the observables p. q) can be constructed representing a dynamical variable. Quantum Theory: The General Formalism qU = Uq + aU = U(q + a). that with U(a) = e~ipa/h = U.00). which means that the spectrum of this operator is a continuum.29) This means: U\x) is eigenvector of q with eigenvalue x + a. (10.q)\i>) = (i>\U^UA(p. i. Since this holds for any arbitrary value of a in (—00.3. we can see that also p possesses only a continuous spectrum (from —00 to 00). or qU(a)\x') = (x' + a)U(a)\x'). the spectrum is not degenerate. ^Observe that {iji\A{p. we obtain qU\x) = U(q + a)\x) = {x + a)U\x). or qU\x) = U{q + a)\x) = (x + a)U\x). (10.
3 OneDimensional Systems 177 Moreover.x")c(a. we have q\x' + a) = (x' + a)\x' + a). \c{a. (x'\p\x"). U(a)\x') = c\x' + a). since q\x') — x'\x').e. (10.x') = = i.a). for instance. The operator is therefore called translation operator. (10. (10. It also follows that (x'\U(a)\x") = (x'\x" + a) = S(x' .32) Since U is unitary.x')5(x" = (x" + a\c*c\x' + a) x').37) With the help of the latter expression we can calculate. In the expression of Eq.10. it follows that 5{x" .37) we set (with e infinitesimal) U(e) ~ 1 Then (x'\U(e)\x") = 6(x' . It follows that U(a)\x') = = U(a)U(x')\0) = U(a + x')\0) \x' + a).x" .hm or „'i„i™" s> i ™' ~/ (x'\p\x")\ = 8'{x' x"). the operator U(a) acts to shift the value x' by the amount a.33) (10. i. Comparing Eqs. (10. offdiagonal elelemts of the {x}representation of the momentum operator.x") U{x'\p\x" % pz.e.e) (x \p\x ) = .e) ~ 5{x' . i.e. (10.31) and (10. \x') = U(x')\0).39) .5(x> . i./ (10.35) (10.x')\ = l.34) {x"\U\a)U{a)\x') c*(a. (10.e.e.x" .x") .x" . We choose the phase such that U(a)\0) = \a).32) we see that \x' + a)ocU(a)\x').36) i. (10.38) h 6(x' .
i. so that U{t. We introduce another postulate: Postulate: The time dependence of the vector \uE(t)) is given by the relation \uE{t)) = eiH^t0^h\uB{tQ)).to) is called the time development operator.44) (10. i. since 5(x) = . (io.tQ) = eiH(tt<Mh. See Eqs. 5. Differentiating U(t.43) (10.L fdpeipx.f dpipeipx.178 CHAPTER 10.§ It follows that (x"\p\x') = d'(x' This expresses that p is hermitian.° ° d(p) (ip) e"« = S'(x).45) (10.to)m0)). . a conservative system. (5.t0). The operator U(t.to) exists which is such that^ m)) = u(t. H\uE(to))=E\uE(to)).40) 10. to) with respect to t we obtain ihjtU(t. .x") = (x'\p\x")*.4i) We assume first of all. it follows that 5'(x) is an odd function of x.4 Equations of Motion Let \tp(to)} be the state vector at time to of a completely isolated system (i. Quantum Theory: The General Formalism We mention in passing that.4. Let \UE) be an eigenvector of the Hamilton operator H with eigenvalue E.e.27) and (5.t0) = HU(t. 6'{x) = !. (10.42) ^Observe that 6'(x) = £ f ^ dpipe** = ± / . 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).35). This is in conformity with our earlier postulate in Sec. In this case the classical Hamilton function does not depend explicitly on t. (10.e.e. 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.
(10.48): [/(Mo) = 1 . (10.t) = 1. (10. In the immediate considerations it is not necessary to restrict ourselves to systems with classical analogues. (10.to) can be expressed as an integral as the solution of Eq. But obviously (cf.e.45) has very general validity.49) . Ufa. (10. t0) = H(t)U{t.t) = l^eH{t). Hence U(t.t)l]U(t. (10.48) with U(to.t)U(t.47) e.tiM*i.t0) [U(t + e.10. t0) (10.41) we obtain ^(t 2 )> = = J7(t2.t).to). i. Eq.t 0 )^(*0)> U(t2.*l)^(*l.44)) we have U(t.*l)^(*l)> = I7(*2. From Eq.t0) or ih—U{t.t0)\^(t0)). = U(t0.t0)]1 Let us set for small values of e: U(t + e.t0) = U(t + we have U(t t) = lim ^ + Mo)^(Mo) = n (1046) = U(t.48)). Evidently U(t. The operator H is d efined by this relation. (10.t) = l. t0) = E/(t 2 . Then with U(t + e.to) (differentiating we obtain immediately Eq.4 Equations of Motion 179 We now demonstrate that Eq.t0)U(t0.to) = 1.t) and therefore [U(t.z / dt'H(t')U(t'.
Now let \tp(t)) be the state of the sytem at time t. Then \Mt)) = u(t.51) Since </. (5.(£ + dt)) = U(t + dt. t)\t/>(t)) =U~ and since H is hermitian.180 CHAPTER 10. >(*)> = i.e. i. (io. to) can therefore be interpreted as a product of a sequence of infinitesimal unitary transformations. (10. Since the only measurable quantities are moduli of scalar products. (10.t0)\ii>s{to)) (10.t0) m0)).41). 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. The operator U(t. and these remain unchanged under unitary transformations. The eigenvectors of these observables are also constant in time.53) . whereas the physical quantities are represented by observables in "K which do not depend explicitly on t. the predictions of different descriptions are identical. it follows that U(t + dt. the Schrodinger equation ju(t. the instant at which a measurement is performed on the system. requires the identification H(t) = H= Hamilton operator. Comparison with the equation postulated earlier. In the following the subscripts S and H indicate Schrodinger and Heisenberg picture quantities.52) is the operator of an infinitesimal unitary transformation (recall that U = 1 + ieF satisfies the unitarity condition UU^ = 1 provided F — F^). Eq. Another such description is the Heisenberg picture. 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.5o) ihjtm)) = Hm)). 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.30).e.t) = l % ^Hdt) \il>(t)) Hdt (10.
we obtain ih^AH at = = U^HAsU + ihU^^U at + U^AsHU (10.57) and rf[As. even if As is not timedependent. (10. so that a matrix element (i. i.55) and using Eq.(tf HU)(rfASU) (10.59) This equation is called Heisenberg equation of motion. t0).HH]. 181 (10. a scalar product) remains unchanged: AH(t) = U\t. . As(q. (10.56) U^[As.With explicit time dependence.e.H]U + iW]^U. if ^ = 0.55) Thus AH is explicitly timedependent.H]U = = = U]ASHU AHHH — U^HASU . Differentiating Eq.54) Observables transform correspondingly with a similarity transformation.t0)\ips(t)) = \ips(to)) = constant in time. {AH. as a rule. Here H is the Hamilton operator in the Schrodinger picture.e. It does not always have to be the case that As does not contain an explicit time dependence.48). However the associated observables are timedependent. For As(q. (10. In the Heisenberg picture it is HH = U]HU. but here.58) HHAH (U*ASU)(U^HU) [AH. i. = so that ihjtAH or ihjtAH = = [AH. (10. we have to take into account the contribution dAs/dt. (10. t0)AsU(t.4 Equations of Motion and \^H) = U\t.e.i). we consider only cases without explicit time dependence of As. subject to a continuous change.HH].HH]+ihU^U. 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. Thus. in order to obtain the Heisenberg picture (description).p. for instance in cases of nonequilibrium or if part of a system disappears through absorption.10.p) in the Schrodinger picture dAs/dt = 0.
Pi): and where the expressions on the right follow from Eqs.HH}=0. An observable CH is called a conserved quantity. i. For an extensive discussion see B.21a) and (10.21b).182 CHAPTER 10.13) to derive the uncertainty relation for energy and time.e. A. In particular. We also observe that the momentum pi is conserved when \pi. since in the transition from one to the other the commutator remains unchanged. Finally we add a comment concerning conserved quantities. we used in essence unitary transformations of matrices. Quantum Theory: The General Formalism Wave mechanics is now seen to be the formulation of quantum mechanics in the Schrodinger picture. We end with a word of caution. as in the transition from the configuration or position space representation to the momentum space representation with the help of Fourier transforms. Orfanopoulos [224]. We see that formally one obtains Hamilton's equations of classical mechanics. if ffCH = 0. provided the Poisson brackets in the latter are replaced by commutators. .H] = 0. In many cases the Schrodinger picture is more amenable for explicit calculations. in the case of the fundamental dynamical variables qi. (10. This holds also in the case of the Schrodinger picture.61) Thus conserved quantities commute with the Hamilton operator. [CH.2: Energy and time uncertainty relation" Use the Heisenberg equation of motion and the general relation (10. (10. 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. In our treatment of representations.Pi we have in the Heisenberg picture (replacing in the above AH by qi. In the treatment of the pictures we used unitary transformations of operators (and vectors). Solution: We have for an observable A (A) = (i>H\AH\TpH). 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. Example 10.
183 dV ' \ dt >"•"»» + ( £ ) • (10.10.7a).62) The Hamiltonian H does not depend on time.e. > (10.p) we then have 0=^M) = U[A.15): (AAf(AH)2 Using Eq.66) This means.62) and dAs/dt > \WAH\i. i. (10. if we set AH = AE we obtain TAAE (10.\HAM\2) \(nA.H]) = 0. We construct the following vectors using Eq. We defined earlier stationary states as states with a definite energy E and wave function * = V( r ) e x P ( — i E t / h ) .e.2 > h.64) = 0. dt in. (10. a position or momentum observation is independent of t. i. Let \ip) = \"4>H) represent the state of the system. since \IPH) is timeindependent. (10. If r is the smallest such characteristic interval of time. This means that for a physical variable A(q. H\tl>) = {H(H))\f). H] = 0.4 Equations of Motion Then.67) 1 d(A)/dt is oo and AE = 0.13) and (10.e. we can replace ([AH.e. in a different form: The eigenvalues of an observable are called "good" if [A. of A\<t>) = a<£)) are then called "good quantum numbers". (10.Such a characteristic time interval can be defined for any dynamical variable. (10. . is displaced by A A in the interval A T = TA. and TA O C (10. i.68) To put it The eigenvalues a of A (i. this is practically the same as at time t for times t with \t — t'\ < T. then if a measurement is made at time t'. (10.H}\f)\2.HH]) Hence we obtain by ih(A). 2 d{A)/dt " and.)\2 ( = \{i. The probability density P ( r ) =  ^ ( r )  2 = (^r>(i#> is independent of t. (A). the "centre of mass" of the statistical distribution of A. AW = {A(A»V>. at ^±AH>h.63) so that with Eqs.65) and rA AA d{A)/dt' .
as is observed. (10.71) From Eq. Quantum Theory: The General Formalism 10. the wave function ipEo(x. Energy and decay length of the radiated aparticles are characteristic properties of naturally radioactive nuclei. Recall for instance radioactive decay.Eo(x)eiEot/he'^2h9(t). the wave function ^ ( x . (10.H ) ( £ . (x)/ E o (E')5(E . i. The wave function of such a state with energy EQ and lifetime r = h/'y > 0 could.5 States of Finite Lifetime When E is real and P(x) is independent of t and J dxP(x) = 1. then fE{E') = S(EE') and integration with respect to E' yields again t()E(x. i ) = J dE'^El^)eiEltlhfEo{E').70) by From Eqs.t) — ipE(x. for instance.t) of such a state must be the superposition of states with different energies about EQ and a corresponding weight function / so that ^ ( x ..E1) 2irhfEo{E)4>E(x).£ ^ + n/2)^ o ( x ) e{^1/2+i(EE )}t/h 0 o =0 . /oo (10.70) we obtain: oo / oo TPEO (x.e. On the basis of the uncertainty relation AEAt > h these must have an uncertainty AE in their energy spectrum.70) we can determine the function fEo{E) integration with respect to t. This means that in the case of decaying particles.69) and (10. t) = ^(x) exp(—iEt/h) describes states called bound states of unlimited lifetime. have around EQ the form ipEofct) = i. (10. t)e iEt h ' dt = V>£0 (x) / JO <fte{7/2+i(i5£b)}t/ft (10. states with sharp energy (AE = 0). (10. (10. This means.69) (Observe that if ip consists of only one state with energy E in the domain of integration of E'. But the number of radioactive (thus excited) nuclei decreases exponentially.72) = V'EoW 7/2h + 0i(EE )/h  oo .(x)fEo(E')e«EE'Wh 2irh f dE'^E.184 CHAPTER 10.)exp(iEt/h)).tyEt/hdt = = = / J—oo dt J dEJrl>E. First we obtain from Eq. States with finite lifetime (At = r ^ oo) are those whose probability density falls off after a certain length of time called "lifetime".69): oo poo p / oo *PEo(x. the particle density diminishes in the course of time (t > 0).
whose real part EQ specifies the energy of the state and and whose imaginary part 7/2 specifies the lifetime r = h/j. 10.g. £/ = £/W(Mo)^'(Mo).e. ^(*o))> then \xp(t)) for t > t0 follows from Eq.e. where U is the solution of Eq.11). We saw previously that if the state of a system is known at time to. (10. They are called "resonance states". We obtained the result (10. Let U^°\t. (10. (10.41). the formula says that JEQ{E) possesses a simple pole at E = Eo — 27/2. (10. which means that a nucleous with discrete energy EQ does not possesses some other admissable level close to EQ. i. and is called "interaction picture" or "Dime picture". to) be an approximate but unitary solution of this equation.to) = HU(t. i. (C/(0) unitary).75) (10. i. We see therefore that the states with lifetime r < 00 and TAE > h belong to the continuum of the spectrum.74) Thus the solution of this equation is the main problem. 185 i. (14.toM{to)).10.t0). that** EQ.e.73) precisely by assuming with fEo{E)^8{EEQ) a "smearing" of states around the energy EQ (with uncertainty AE).6 The Interaction Picture Prom the last two equations we deduce for E close to ipE0(x). This description contains effectively parts of the other two pictures and is particularly useful when Eq. A state is discrete if its immediate neighbourhood (in energy) does not contain some other state. of course. i. W)) = U(t.45). ihU(t.45) can be solved only approximately (in actual fact.6 The Interaction Picture One more motion picture of quantum mechanical systems is in use. (10. Eqs. Considered as a function of E. .105) and (20.76) **For specific applications see e. 7 > 0.e. The state with lifetime r — h/'y is not a discrete state. every unitary transformation of states and operators defines a possible "picture").e. VE(X) ^ ^ (£ > = dbfift+T/* (la73) This result is known as the BreitWigner formula.
t0) 0 = l. with Eq. (10.81) But [/(°) is unitary. £/(°)£/(°)t = 1.186 Then CHAPTER 10. (10. = HU<®U'.e. i.e. dt so that #(0)t ( t ) It follows that H{0\t) = dt = H<®{t). (10.78) (10. .82) (10.83) ih^U^ ishermitian. with Hamiltonian H = H^ + H'. [/' has only a weak ^dependence (i.e. For a "good" approximation U « [ / ' ' we have HU^ihjtU^^0. i.e. H(°Ht) = ih^uW. i. be defined by the relation H(°)rj(0) _ i h ^ l = 0 (exact)> n (0) = eiH(o)t/aj (1080) i. varies only slowly with t).78) at with the initial condition c/(0)t f ^ ( 0 ) _ ^ ^ 1 V \ at J U\t0.79) ^'~°i.e. In addition we have (10. Quantum Theory: The General Formalism ih^(uWu') i.e.77) C/(o)t itiuWfLu^HUMu'ih^U' at or i. Now let H^(t).h^U> = at ( io.e.
lfe(*)}.6 The Interaction Picture 187 We use now the subdivision of H into unperturbed part and interaction. and obtain from Eq. (10.78). Then ihjt\Mt)) = ih~u^\Mt)) = u(0)ihi^W and H'^it)) = U^H'U^U^\^s(t)) + ih(^) \Mt)) (10. (10.e. (10. (10. (10.85) = where we used Eq.88) and as interaction picture version of an observable A the quantity Ar(t) = tf(°>t AgU^.80).U^HU^U' H'fi'.87) = We define correspondingly as interaction picture state iM*)> = t/ (0). U<®\Mt)) = \Mt)). H = H{®{t) + H'. (10.78) at (10 85) (10. We now multiply this equation from the right by U' and insert the first expression on the right (of Eq.92) . We also set H^U^H'UW.89) where ^s(i)) and As are state and observable in the Schrodinger picture. (10.90) = U^H'\^s(t)). 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^.85) multiplied by U') on the right of Eq.84) U^ where H' is also hermitian.91) Since the Schrodinger equation is given by ihjt\^s{t)) = H\1>s{t)) = ( # ( 0 ) + H'Ms(t)). U^H'U^U' (10.86) at = U^HU^U'. (10.10. (10. (10. i.
94) Then we have (normally with dAs/dt dt (io=8o) — 0. (with ihjtAI(t) = [AI. i.90) and replacing the left hand side of this equation by the right hand side of Eq.e. so that with Eq.60a) or (10.60b) with iff0) instead of H.e.95) Thus in the interaction picture the physical quantities A are represented by timedependent observables which satisfy a type of Heisenberg equation (10. (10. (10. (10. H'j = UWH'UW. i.80) (and multiplying by t/(°)t) ih~\Mt)) = u<MH'uM\Mt)). a perturbation so that H ~ H(°)). (10.e.188 CHAPTER 10. ot i.89). In order to obtain the equation of motion of an interaction picture observable Ai. we have i. we differentiate with respect to time the operator defined by Eq. (10. near zero and \tpi(t)) is almost constant in time. (10. Since H'j is assumed to be small (i. ^ (0) l^(*)>. (10. Quantum Theory: The General Formalism we obtain by starting from the right hand side of Eq. the vector varies with time (different from the Heisenberg picture).e.92) ^°)^V/(t)) + ^ ( ^ )  V ' / W > = ^ (0) ^ (0) l^(*)> + ^ .89)^0.94) _Hf)Al 1 + lhdAl+AlHf).e. (10.e. cf.o)AU(o) at +uWAsHWuW UWHWU^UWASUW + iww^uw dt +U^ASU^U^H^U^ (10.H\0)}.55)) at + imm<Ms_u(o) at dt _umH(. comments after Eq.93) Thus the vector ipi(t)) satisfies an equation like the Schrodinger equation. dAs/dt=0) . i. (note the difference between H' and H'j) ihjt\Mt)) = H'AMt)). A^t) = U^AsUi0).
7 TimeDependent Perturbation Theory 189 10. We obtain the operator U^ from Eq. (10. {ipn\<Pm) = Snm.80) as E/(°)(t.e. ih±\Mt))=Hi{t)\Mt)). (10.7 TimeDependent Perturbation Theory We consider timedependent perturbation theory as an application of the interaction picture. Integration with respect to t yields \Mt)) = \Mto)) ~ \ \ dt'Hi{t')\^j{t')).101) Iteration of this inhomogeneous integral equation yields the socalled Neumann series. We assume again that the spectrum and the eigenvectors of Ho are known. (10.97) The equation to solve is Eq. H' = H'(t). Since .93).10. (10.e. (10. \Mt)) = +' \Mto))^fdt'Hi(t')\Mto)) ft J dt' f dt"Hi{t')Hi{t")\^I{tQ)) + (10. the system is in the Schrodinger eigenstate \ips ) of iJ 0 with energy Em. (10.i0) = e<Ho(tto)/ft.100) (10.102) We assume that before the perturbation H' is switched on at time to.e.94) Hi = uW(t. i.99) (1098) Instead of the original Schrodinger equation we now solve Eq.87) and (10.96) in which the perturbation part H'(t) depends explicitly on time and HQ replaces H^0' in the previous discussion.t0). ^ \<Pn){Vn\ = 1n (10. i. i. H0\ipn) = En\(fn).t0)H'uW(t. We have the Hamiltonian H = H0 + H'. where according to (10.98).
(10.105) rt n Jto h ' Jt00 t rt  x  l h ' Jto t0 (10. Quantum Theory: The General Formalism HQ is independent of time.107) Jto With this we obtain from Eq. the time dependence of \ips } can be separated as for stationary states.103) The actual state \ipg) of the system is solution of ih\^s) = H\^s). (10.e.103) for m — n and use Eq.104 ) Here we insert Eq. assuming that this supplies a sufficiently good approximation: IMQ) = \fm) lz\ n dt'HW)]^). (10.88) and obtain > n ^ V s ) = {<Pn\eiHot/h\ll>s) = {<Pn\eiHot/hUM\Mt)) = <¥>*(*)>• (10. We express this state \tps) as a superposition of the states \ips ) n of #o with coefficients which as a result of the timedependent perturbation depend on time.^ V ™ ) . (10. This probability is the modulus squared of the projection of the Schrodinger state \ips) o n t o \ips )„.102) and truncate the series after the first perturbation contribution.108) . Hence we set ^> = £°«(*)l4 0 ) >n. we can write l 4 0 ) ) m = elHot/hWm) =e . i.105) In the interaction picture the iJostate m at time t is \Mt)) = eiHot/hWP)m = \<pm).190 CHAPTER 10.106) We insert this expression into Eq. n We wish to know the probability for the system to be in a state \ips ) n at time t after the switchingon 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 . (10. (10. (10.
(10.8 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^. Thus we set H'(t) = H'0(t).10.Jr(En—Em)t _ ^ (ipn\H'\(pm) {En — E„ sm{±(En . (10.8 Transitions into the Continuum In many practical applications one is interested in transitions into a continuum. (10. We assume that the perturbation is switched on at some time to.111) Different from Eq. » — < (10.110) With this we obtain from Eq. if in the scattering of a particle off some target its momentum p changes to p'. (10.Em)t] _ sin2 at aH {±{EnEm)}H Consider the function St{a) :1 sin2 at ir a2t t 7T for a —• 0. t o = 0. A further example is provided by adecay. < KO?t .Em)t' (<Pn\H'(t')\pm) = Wnm(t). or if by emission of a photon with continuously variable momentum an atom passes from one state into another. One reason why we begin with these simplifying assumptions is also to obtain reasonably simple expressions.112) for a^0. for instance.109) the transition probability Wmn(t) = jp Jo .Em)t} \{tfn\H'\ipv h 2h(En . This is the case.25) we here have the positive quantity (note t in the denominator) s i n 2 { ^ ( £ n .En (10.109) 10. where the momenta of the aparticles form a continuum. but that otherwise the perturbation is timeindependent.
E7m)t} = (10. so t h a t Wnm would be zero.Em)t} r i 7 TTT. Otherwise.Em). (10.2. t—>oo Hence also (shifting t to the other side) sin2{^(ffn.2 The delta function for £ — oo.114) It follows t h a t for large but finite times t Wmn(t) 2vr ~ — t5(En Em)\(vn\H'\tpm)\2. we would have En — Em ^ 0. . though sufficiently small.— = o (En 2h En 2h5{En . 10.ii5) This expression has a well defined value unequal to zero only if the energies En belong to the continuum. Quantum Theory: The General Formalism T h e behaviour of this function or distribution is illustrated in Fig. Hence lim St(a) = 5(a). We assume t h a t the matrix elements for transitions into this infinitesimal element can be taken . if the energies belonged to the discrete spectrum.113) {±JEnEm)¥ or n m irtSt En — ^ 2h 1 s i n 2 { ^ ( £ n . 10.io.192 CHAPTER 10. and no other state would be nearby. > 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. Fig. In the case of transitions into the continuum we have to consider the transition probability into the interval dEn at En.
Schwabl [246].e.117) in many applications. i.t The result makes sense as long as the uncertainty AE in the energy satisfies for finite but large times t the relation .115) and (10.3: Application of test functions to formula (10.116) The transition rate V is the transition probability per unit time. Pauli [227]. Fermi [92]. 266.* In view of the usefulness of Eq.113) this is* hm . then Eq.113). 75. r = J E ^ W =p(£m)y(</>n#W2n (10117) The formulas (10.4). (10. f See E. (10. Then.117) provides the outgoing particle current (see Example 10.10. in the article of W. Fermi gave this formula the name "golden rule" . We evaluate the integral 0(0) = 1. _ 2irh AE > > Se.117) were derived by Pauli in 1928. so that initial and final states belong to the continuum. '"Recall also the integral J ^ ° dec sin 2 x/x2 = TT.r Z x 1 ^ ) = 4>{Q). Solution: For test functions <p{x) € S(R) the delta distribution is defined by the functional I &(x)4>(x)dx = <j>{0). t We choose a test function which falls off at infinity faster than any inverse power of x: <Kx) = e .8 Transitions into the Continuum 193 as equal./ t>oc TZ J_00 dx . for instance from a potential. F Jo . If we are concerned with pure scattering.113) Use the method of test functions to verify in the sense of distribution theory the result (10. sin 2 (tx) ax i—( * According to F. p. pp. . Let p(En)dEn be the number of states in the interval dEn. the transition probability into this set of states is J2Wmn(t) = fdEnP(En)Wmn(t) n •* = p(Em)2lTl{iPnlfliPm)l2t.W . Example 10. In the case of the functional (10. (10.118) where 8e is the separation of states around En which in the case of the continuum goes to zero. (10. if p(En) varies only weakly with En. 142.
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. 77^ + . formula 3. .ln(l + 4t 2 ) + — t a n _ 1 ( 2 t ) = — tan 4t7T tTV T J< (oo) = 1. Jin § m = See I. 1 .b2 4 p 2 + (a + b)2 2 b _j 2pa V 1 fc Urn . E x a m p l e 10./ cfc a n p 2 +fc 2 . Hence the result is 0(0) as expected.a 2 ' i that for a = b = t . .( 1 / 4 ) ln(l + 4t 2 ) + t tan _ 1 (2<).. . . Solution: We set ¥>n(x) : so that {<Pn\H'\<pm) = ^3 / d x e i K ' x V ( x ) . with n .194 CHAPTER 10.4: Differential cross section from the "golden rule" Obtain from the "golden rule" (10..947.k. . iknx . K = k 0 . 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. E = h2k2/2fi. V +(ab)2 a _x 2p6 7 m —^ '. The ingoing particle flux is therefore v / L 3 .117) for H' = V(~x) the differential cross section da _ / n V dH ~~ \2Trh) f dxV(x)e where K = k m — k n . . M. ^o n . Then p(Ek)dEk Since Ek — h2k2/2[i. Then ^ . S.^ V v 4 ) = ^ T 3 e * k ° *^ko)e*° * = — v . Gradshteyn and I. L3/2 ¥>m(x) : L 3 /2 The periodic boundary conditions ipn(xi) = fn(xi ± L/2) imply kiL/2 = ±nj7r. . with v the velocity of the particles. L \3k2dkdUn It follows that T.„I. x . . .? 2TT / L \3k2dkdUn 1 /V(x). § We have r Jo \p\x dx sin ax sin bx  V.( ^ V V • < ih 2 _ i k n . = ( — ) dk= 2ir (—  2irJ k2dkdfl. * ^ = . . 2 . 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). . dEk = h2kdk//j.tan p 2 + a 2 .e " 1 * 1 = 2 lim — .<p). p. p = 1 we obtain I = . Quantum Theory: The General Formalism with the help of Tables of Integrals. and is obtained as ih. and hence . 2 T / ^ M.. Ryzhik. 491.. = 0 .
9 General TimeDependent Perturbation The differential cross section is defined as da •• Theory 195 outgoing particle current x r2dfl ingoing particle current da / /i V (hko)/(lJL3) J dxV(x)eil where for purely elastic scattering k = ko. .t) = ^On(t)^(x. specifically in the case of the Coulomb potential. For a possible application of this result.10. see the discussion after Eq.t). We expand the wave function V( x . Thus we assume that the eigenfunctions (fn = (x</?n) of Ho are known.69)./continuum = <San(£)</4°)(x. 193.9 General TimeDependent Method For a better understanding of the "golden rule" derived above.^ We start from the Hamiltonian H = HQ + H'. (10. by which we mean without leaving the Schrodinger picture. (10. The problem is to solve the timedependent Schrodinger equation ih^\i(>) = H\r/>).121) .t)+ / n a{E.^°>(E.t)^°\E^t)dE (10.122b) S e e also L. we present another derivation using the usual method of timedependent perturbation theory. (10.120) In the case of Eq. (21. Thus we set ^(x. Schiff [243]. 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. pp.x)^(E'.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.x) = S(EE').119) where H' = H'(t) is a timedependent perturbation term and Ho\ipn) = En\ipn).122a) (10. (10. 1 rfx^°)*(x)^)(x) = 8mn. 148. t) = (xl^) therefore in terms of the eigenfunctions of Ho with timedependent coefficients.
Next we use the orthogonality relations . into dip (10.120). f 6kn in case of discrete tp's.124) and obtain ihSdn(t)ipn^)eiEnt/h + San{t)En<pn(x)eiEnt'h = San(t)(HQ + H')vn{*)eiEntlh = San(t){En + H')vn(x)eiEnt/h..x. = e^EE'^h5(E Thus with Eq.126) + H')<pn^)eiE^h}. ^0)(E. Quantum Theory: The General Formalism The discrete and continuous eigenfunctions of HQ are Vi 0) (x.196 CHAPTER 10. Observe that these conditions do not contradict Eqs.„„. . (10..125) where an(t) — dan(t)/dt. Then J d^*k(X)[ihSdn(t)cpn(X)eiEt/h = J d^*k{x)[San{t){En + San(t)En<pn{*)eiEntlh] (10. V>(x.. .122b).127) we obtain from Eq. (10. We multiply this equation from the left by </?£(x) and integrate over all space.E') = 5{E . (10. since f rJ>(°>(E.x. (10. (10. (10.t)rJ><®(E'.128) . t) = San(t)<pn(x)eiE^h.E').. (10.„.123) We now insert these expressions into Eq.126) ihak{t)eiEktlh where = San{t)eiEntlhH'kw (10. . i.x.129) . dnptMrnW = { ^ _ En) (10.x). / *. .i) = e.e.t) = eiEtlhy(E...t)dx.iJ5 »*/Vn(x).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). The second term on the left of this equation and the first term on the right cancel out.122a) and (10.
10. We observe that the amplitude ak of a definite eigenfunction tpk has effectively replaced the wave function ip. (10. (10. (10.130) along with the parameter A in front of H'kn.e.^ [a^(t) E h + } + \a£\t) + A2a£>(t) + • • • ] .134) . i.133) The numbers ak = const.9 General TimeDependent Perturbation We can rewrite the equation as Theory 197 iMk{t) = SH'knan(t)e^EkE^h. and we set ak°}=5km or ak0) = 8(EkEm). SH'kn(t)e^~E^ha^(t). Now we set ak(t) = ak0){t) + \a£\t) + X2a[2\t) + ••• (10. We obtain ih{ak°\t) + \ak1\t) + \2ak2\t) = SXH'kn(t)e^.130) we develop a perturbation theory. Here we assume (our assumption above): At time t = 0 all coefficients ak are zero except one and only one. SH'kn(ty^E^l^\t) (10. We restrict ourselves here to the two lowest order equations. (10. (10.130) H'kn by \H'kn and at the end we allow A to tend to 1.e. In the first place we replace in Eq. they are fixed by the initial conditions. let's say am / 0. Comparing coefficients of the same powers of A on both sides. since it is equivalent to the latter. to perturbation theory in the lowest order.131) and insert this into Eq. (10. Thus first we have ak0) = const. i. 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 order to solve Eq.120).132) These equations can be integrated successively.130) This result should be compared with the Schrodinger equation (10. we obtain the equations: iha[°\t) iha£\t) iha[s+1\t) = = = ••• . 0.
135) and (10. We shall see that in this approximation the scattering amplitude is effectively the Fourier transform of the potential.ml2sin2{^frn (En . (10. H'^y^"^^/hd£.e. (10. ) ' X V(x')dx'. (10.198 CHAPTER 10. whereas ak is a Fourier transform in time (a distinction which is easy to remember!). The amplitude / B o r n is a Fourier transform in spatial coordinates. ifm{kl) = f J —oo (10. i.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.e. ip) in the context of timeindependent perturbation theory. so that ihakl\t = oo) = 0. (10. Later we shall discuss scattering off a Coulomb potential in terms of the socalled Born approximation of the scattering amplitude which we there write f(9.136) Comparing Eqs. Quantum Theory: The General Formalism depending on whether Em belongs to the discrete part of the spectrum of Ho or to its continuum. i.k .Emy in agreement with Eq.135) We put the additional constant of integration equal to zero. In this case we obtain from Eq. an expression of the following type AS?) = " 2 ^ 2 / e j ( k .136) we see that both "amplitudes" are Fourier transforms and look similar. What is the physical significance of Eq. Equation (10. ipm) into the state k.132). (10.137) . Integrating the second of the equations (10. (10.e.135) / Jo L Hfrfc ~ Em)/h Hence the probability to find the system at time t in the state n ^ mis (with Al) 22 i rjv I1 r i r IW a(l)m2 K ( j l = i^kl\t) = H'km e^E^dt> = H'km 6 \nnm\ ei(EnEm)t/h i(EnEm)t/h _ j [EnEmy 4g.111). we obtain it*™ = f J—oo SH'kn{t')e^E^'lh5mndt'.135) assumes a particularly simple form if — as we considered previously — the perturbation H' is taken to be timeindependent except that it is switched on at a particular time t = 0 and is again switched off at some later time t > 0.
meaning the inclusion of the creation and annihilation of field quanta). Angular Momentum In considering the Coulomb potential in a realistic context we have to consider three space dimensions.2.2 Separation of Variables. the position coordinate and the mass of particle i with i — 1.y. such as the electromagnetic field (the quantization in these contexts is often called "second quantization".m.Chapter 11 The Coulomb Interaction 11. ( 1L1 ) We let pi. Z<i = Z) y(r) = _ M ^ . 199 (1L2) . The electrostatic interaction of the particles is the Coulomb potential (later we take Z\ = 1. r=  r .1 Introductory Remarks The quasiparticle quantization of the harmonic oscillator is of fundamental significance in view of its role as the prototype for the quantization of fields. The Hamilton operator is then given by w= jt + ji_^M. 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. The Coulomb interaction on the other hand is of specific significance for its relevance in atomic and nuclear physics. Let r : = (x. 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. Both cases therefore also serve in many applications as the basic unperturbed problem of a perturbation theory.Ti.i be respectively the momentum.
2Pip2 m\m2 pip2 mi + m 2 . 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 ' . P = pi + p 2 . But now P2 2mo = "ip/pi 2 \mi m p2\2 m2J  = m0/pi2 2 \m\ p22 m2. mi + m 2 va\ (11. Solving Eqs. Y. r} representation they have the differential operator forms Setting R = (X._d_ _ _d_ dX dx2 ~ dXl d dx2 nifi^i [ ' etc.3) and centre of mass momentum P and the relative velocity p/mo by setting r» . y.5) The momenta P . Z) and r = (x. /'Pi p = m0 P2 A v. so that as expected we obtain the expression for p in Eq. p are such that as operators in the position space representation or {R. Similarly d_ _dx^_d_ dx dx dx\ dx2__d_ _m^_d_ dx dx2 mi dx\ mo_9_ m 2 dx2 etc. (11.2/ mi+m2 The mass mo is usually described as the reduced mass. where m 0 = mim 2 . (11.4) P = p i + p 2 . .200 CHAPTER 11. z) one verifies that j9_ = dxi_ _d_ 8X ~~ OX dXl dx2. mi + m 2 iTi'2 r2 = R ^ —r. (11. r = rir2 mi + m 2 (11.U7&. The Coulomb Interaction We introduce centre of mass and relative coordinates R and r by setting R = m i r i + m2r2 • . (H4) \mi m.3) for the individual position coordinates we obtain ri=R+ •^ —r.4). so that as expected according to Eq.
r = —.z). Pj (which in the following we write again ri. Pj over to operators ?i. Angular so that Momentum 201 ^ ^ .y.2 It follows that p2 2 P2 n2 n2 n2 (H. Ri.Ri. 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). with M := mi + vn^R = ^ P. (11.12) (11. P on the one hand and those in r. r) is a canonical transformation. (i. (1114) m0 We observe that the equations of motion in R.8) and so W =777 v + 7 T . The motion of the centre of mass is that of a particle of mass M = m\ + 1712 in uniform motion along a straight line. p = W.j = x. we mean n = rx s x.r2) to (R.e. Canonical quantization of the classical theory implies that we pass from the Cartesian variables rj.2 Separation of Variables. 2(mi + m<i) 2mo 2m\ 2m.= ^. .Ri. etc. (11. ' d and 'AW OH'ATJ P =  ^ dR <9# <9r ' (ii. p on the other are completely separated.+ ^2.Pj). (11. [Ri.10) One can convince oneself that the transformation from (ri.+ ^(r) 2(mi + m 2 ) 2m 0 (H. (11.pj]=iMij. (11. pj. for which we postulate the commutator relations* [ri.Pj} = ih6ij.pj.16) Although we write ri.15) In accordance with Eq.+ P.11.13) * and = M ' 1V1 P = 0. Hence Hamilton's equations apply and we obtain.10) we set H = Hcm + H.9) ft = ftcm + ^ = centre of mass energy + energy of relative motion.pj.il) ""Op' i. The canonical quantization must always be performed in Cartesian coordinates.
18) In the position space representation the time independent Schrodinger equation is therefore 2M' A fl ) + h2 2mo A r + V(r) * ( R . it is reasonable to go to spherical coordinates r.202 CHAPTER 11. (11.Q 2M' + V(r).19) For stationary states with energy E the equation possesses a complete system of eigensolutions * which can be written tt(R. 0 < r < oo. df\ 39 J + 1 d2i/j r2 sin2 9 dip2" (11.21b) + E. = rcos9.r) = *(R)V(r). (11. 9. —n < tp < ir.17) (11. 0 < 9 < ir. Thus the twoparticle problem is practically reduced to an effective oneparticle problem.20) r2MAR_ $(R) = ER*(R). r ) = J E t o t a i*(R. (11. In these coordinates we have /\rip 1 d r2 dr \ tdil> dr + r2 sin 9 09 \ S m 1 d_ . where Hcm$(R) = and Hip(r) = with ^total — ER (11. V(r) = V(r).24) . as in the case of the Coulomb potential.22) If.23) y — r sin 9 sin tp. The Coulomb Interaction nc H = 2m. 2mo (11. (11.21a) L*+v{r)] ip(r) = Ei/j(r) (11. i. ip: x z = r sin 9 cos ip. to that of the Schrodinger wave equation h2 Ar + V(r) tp(r) = Eip(r).r).e.
A fc2! d ( d^ ^ = _tfU2dA r\dr and (compare with Eq.11. n ( .31) .3.3.2.pr]=ih. = (p2 + ^ V (r + 0). that the above expression follows from l:=rxp=ift(rxV).k a clockwise permutation of 1.33) i. i—>0 (1L25) (11.)In Example 11.32) (11. the operator —ihd/dr is not hermitian! The operator pr obviously satisfies the relation [r. We have I2 = (r x p) • (r x p).3. (11. where (r x p)j = ^eijkrjpk. .i ' .1 it is shown that pr is hermitian provided lim n/>(r) = 0. (l + . k an anticlockwise permutation of 1. i. k = 1. _ h2 \ . = —1. (11. 2 1 9 / 1 9 . (11.27) We can now write (see verification below) p V = h2Ar^ where 2.29) . otherwise. i( i._0di})\ x We now demonstrate that 1.26) However.12 — I2.2. Angular Momentum 203 We can rewrite this in a different form by defining the operator * : =«. (11. with tijk — + 1 .j.28) . if i.e.24)) 2 + M\ d dr2 J = tfL^(r^A\ r2 dr\ dr J d2ip (11. j . is the operator of the relative angular momentum. (11.5r = .2 Separation of Variables. = 0.j.2.
43) r2(p2p2).25)) r • p = —ihr • V = rpr + ih and hence 8 = (r • p)(r • p .36) It follows t h a t (note the part defined as —5) l2 = = ( r 2 p 2 .42) (11.34) 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.k ~ rjpkrkPj) (11. The Coulomb Interaction = &jj'&kk' — Sjk'Skj'.204 It follows t h a t 2_^£ijktij'k' i CHAPTER 11. ih6jkrjpk (11.ih5jk)pk (11. * ' S Since r =xz we have (cf. (11.ihr • p r2p2(rp)2 < + ih(rp).15) we have fjPkTjPk = fj(rjPk and TjPkTkPj = = rjPkipjrk rjPj(rkPk + ihSjk) = TjPkPjTk + ~ ihSkk) + ihrjPj.k Using Eq. (11. (11. (11.j. r— = r • — or or (11. (11.37) .(r • p ) 2 + 3ih(r • p) . Eq.38) + y* + z .40) (11. (11.39) + ih).44) .ihr • p ) .ih)pr and x 2 2 9 9 2 2 Q O (11.27) we obtain r{prr)pr = r(rpr .41) o = r pr. This means that l2 = so that 9 9 1 (11.35) j.ih) — rpr(rpr With Eq.
¥ p)r 2 drdn s = (p r 0.6d0dip. h f l d —(r<l>*(r.2.46) (11.e. (pr4>.6l. r := r : lim rip{r) = o i .— (riP(r. S o l u t i o n : Set d£2s = sin.48) . <p)*Pri>(r.e lim rip(r) = 0..7r].<P) = Etl>(r. 6.e. 11.[ 4>(r. <p)* . 9 € [0.<p))ril>(r. <p)*rip(r} 0.6.1: Proof that pr is hermitian Show that if Vv lipe H = C2(R3).e.30) I2 is to be identified with the square of the angular momentum operator.2TT]. (11.[ i J With partial integration with respect to r this becomes (<l>. <p) d °° .<p)*?(ri.2 Separation of Variables. reR3. 6. pr = . Pri/Oi V0. prip) = = I <j>{r.e'iP^} .(r. i. ip 6 X>Pr C 7i.<p G [0.Q ' 2mor2 with the condition (11.<p))drdns.l2} = 0.ip) = (<£. (11.<p) (11.! /'). <p)r2dfls = .26).e. H 1 2mo pi + K) +V(r).30) yields l2 = I2 = sin sine— BinO^r 09 + d^2 (11. 9. Then {<j>.11. (11. /'(r.Pril>) J / r= dQs rcj>(r. Angular Momentum 205 Comparison of Eqs.47) Example 11.e. r—>0 il>(r. or J .<p)drdna " r r=0 J \ll§r~(r4'(r. The Schrodinger equation of the relative motion can therefore be written V1 V 2m.45) oe Thus in Eq.d_ r dr the operator pr is hermitian. i.28) and (11. 8.. 0.1 Separation of variables We deduce from the expressions of H and l 2 that [H. <p))r2drdns % J r or r<t>(r.d.
Since 1 = r x p . The Coulomb Interaction i. We can now proceed as in the case of the harmonic oscillator and determine their eigenfunctions.e.ypx) = ihlz. (11. i.50) Thus the components of 1 do not commute pairwise. g i = x.ly]=ihlz. 11. zpx] + [zpy. i.lz (or l . 2 2 2 = 0.49) By cyclic permutation of x. [ly. (11.pz]x = ih(xpy .53) Either of these operators is the hermitian conjugate of the other. we have [h.xpz] = [ypz.lz]=ihlx.206 CHAPTER 11. z we have altogether [lx. ly\ — —tfi(lylx + Ixlyji [hi h\ = 0. y.ly) are simultaneously "sharp" determinable variables and hence have a common system of basis functions. ly] = [yPz ~ zpy.51) (11. Example 11. for any two of the components there is no common complete system of eigenfunctions. 2 . xpz] = y\pz. [lyA = 0.2: Verify that [h. ?3 = z). (11. We therefore first search for a complete system of eigenfunctions of l 2 .l2] = 0. In order to determine the eigenvectors of the operators l 2 and lz one defines l± = lx±ily. 3 .52) This means that l .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.e.lx] = ihly.qj]=iheijkqk S o l u t i o n : This can be verified in analogy to the preceding equations (i = l .e. However. (11. so that [lz. /z> *xJ = l <i\}ylx + tx'j/Jj [h. zpx . and this implies in the terminology we used earlier that their determination cannot be "sharp" simultaneously.l2} and correspondingly [lx. These operators here play a role analogous to that of the quasiparticle operators . [lz.q2 = V. z]px + py[z. they are incompatible variables.lx or l . H and l 2 possess a common system of basis eigenvectors.
* Thus we have (1 being a bounded operator) lz\l) = l0h\l) Then from Eq. i + ] = [l2. assuming a finite dimensional representation space.60) Similarly for l+ with the help of Eq. (11.lh)\l) = {l0 . With the help of Eq.l)hl\l). Hence we must have Z+I0=0.i] = m. Therefore (lo + l)h cannot be an eigenvalue of lz. . (11. (11.49) we obtain the following relations: [iz. (11.i+] = m+.51). (11.3 Representation of Rotation Group 207 A.).59) = 9) h(l0 r)\lr).58) {l0 : max. Similarly we have lg{l)2\l) and more generally lz(l. (11. since lz is hermitian.54): izi+\i) = (i+iz + i+h)\i) = (lo + i)ta+\i). [ig.52): 0 = [ l 2 .54): lzl\l) = {llz . (11.62) 'At this stage it is not yet decided whether IQ is an integer or not! But we know that IQ is real. and with Eq.r) (11 = {l02)h{l)2\l) (11. (11.h~lz(1156) (11.11. (11.e.54) l 2 = \{l+l~ + 11+) + ll = l+l.Y\l) = (lQr)h(lY\l). l\l) is eigenvector of lz with eigenvalue (IQ — l)h.+ ll. [i+M = 2%iz.55) One now defines certain ketvectors as eigenvectors of lz which span the space "K\ of the appropriate states. We write or define: / .57) i.61) But lo is by definition the largest eigenvalue of lz.r) := (Z_)rZ) so that lz\l . y .Z_] = [ l 2 . A^ in the case of the linear harmonic oscillator.54) On the other hand we obtain from Eqs. in particular one defines \l) € CKj as eigenvector of lz with the largest eigenvalue lo. (n. In the present context the operators are called shift operators for reasons which will be seen below. (11. (11.
65) for r = n: l2\l n) = l2(l)n\l) = lo(lo + l)h2\l . (11.66) It then follows from the second of relations (11. IQ(IQ 1 0 (11.64) we obtain on multiplication by Z_: lVl\l) and more generally l2(f)PJ> (11.. i. (11.68) .h)\ln) {(lQnf(lQ.l2\l) = l0(l0 + l)h2l\l).n). l\l) is also eigenvector of l 2 with eigenvalue Zo^o+l)^ 2 .\ll)..+ i2z (11+ + 2izh) + 1 \ \i) = (izh + ti)\i) i0(i0 + i)h2\i). [12. i. (cf..e.e.n) (1 = 60) {l)n+1\l) = 0. Eq.\lr) terminates at some number (let us say) r = n (n a positive integer).64) i. (11.+ 11+) + l2z\l) = i+i. i. \l). Since l 2 commutes Hence Z) is eigenvector of l 2 with eigenvalue with l_.n)}h2\ln). (11.56) CHAPTER 11.63) + l ) ^ 2 .65) We assumed that "Ki is finite dimensional so that the sequence of eigenvectors of lz.54) (n = 57) (11. (11. l\l . The Coulomb Interaction (l+l.e.208 Thus for l 2 we obtain: V\l) ' = 11..67) i (11 60) = But according to Eqs. (11.e.Z_] = 0.56) that l2/n) = (11 6) (l+l+l2lzh)\ln) (lll.60) and (11. (11.y\i) l)h2\lr).55)) we have 12Z_Z> = l.60) = Vli\l) = l0(lQ + l)h2li\l) i0(io + l0(l0 + i)h2(i. (11.From Eq.
.70) they satisfy the eigenvalue equations (with m = l0Jo l. \l — n) are eigenvectors of lz with eigenvalues loh..n){l0 . lo = l0n.69) Prom this it follows that IQ must be either half integral or integral. § See e. Messiah [195].m) = = mh..m).m±l\l±\l. Brink and G.. According to Eqs.65) these are eigenvectors of l with eigenvalues Zo(^o + l)h2. (H71) These expressions should be compared with the corresponding ones in the case of the harmonic oscillator. The basis vectors \l). y/(l0±m + l)(l0^m)h.n) 2 .l)h \/(lo + m + l)(l0 .1)..m .1). (11. Satchler [38]. The dimension of the representation space is therefore n + 1 = 21Q + 1.l. lz\l.m + 1)(Z0 + m)\l. The phases are chosen such that§ l+\l.m / + l)h ..m) — mh\l. .m\lz\l.  U o .m) l\l.2. The operators l+. II. y (lo .71) can be established in analogy to the method used in the case of the harmonic oscillator.. *D. i.l_. R. We therefore write the 2/o + 1 orthonormal vectors: \l. (11.3 Representation of Rotation Group so that with Eq.m).m) = IQ(IO + l)h2\l.. (11.Z 0 ).g.l0): l2Z. 17. m + l)h. lQ = .m y/lo(lo + 1) — (m — l)m\l. —loh. p. p..m) with — lo<m<lo2 (11. lz) are then: (l.m) = mh\l.67) we obtain (observe IQ(IQ 209 + 1) = —lo(—lo — 1)): *o(*o + 1) = Wo .e.. We write these* for integral values of IQ: \l. In the following we consider primarily the case of integral values of IQ.m)\l. (IQ — l)h. we skip this here.n)} = (l0 .70) According to Eq.11.m).m) (l. (11. As in the case of the harmonic oscillator we introduce normalized basis vectors. Vol. A.lz transform the vectors \l). The nonvanishing matrix elements of 1 = (l+.n ...{lo .. The normalization factor in Eq. \l — n) into one other. 24.m) = = = = y/lo(lo + 1) — m(m +l)\l.l)h. (11. M...63) and (11.m) : Mo).. where lz\l..
ip. for the simple "rotor". mhY£(9. We now perform the analogous procedure for the angular momentum or.. and m = —IQ. (11. (11.73) 1 d d sm0 06 \Sm9dd) + (11.2.With the ansatz Y(9. . The Coulomb Interaction Z+Mo> = 0..210 Then. . ip) = e(9)$((p).l0)=0.1.77) $((p) = m2<b(y). . 11. .. $ oc eim*.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.<p) = i)n2Y^{e^). In these coordinates we obtain: h l± so that h2 = xpv .75) the variables contained in the expression (11. (11. l_\l.yp.74) can be separated: 1 d sin 9 36 2 1 sm6— 89 j — sin 0(9) = h(l0 + 1)0(9). as this is sometimes described. 9. We . 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.. as above.e. d2 sin2 9 dp2 _' (11. i. as remarked earlier.<p) = i0{io + lzY£(0. with i2Y£{e.72) In view of the spherical symmetry it is reasonable to use spherical polar coordinates r. /o. ih d_ dip' lx ± ily = ±he^ d I ^ ± zcotfl^ . for which the eigenfunctions \ip) are given by l2 l0(l0 + i)h2\ip). CHAPTER 11.<p).76) d2 (11. where IQ = 0.74) We choose the eigenfunctions of l 2 and lz as basis vectors of the ("irreducible") representation. —lo + 1 .
<p) = (iy 47r(Z + m)! In particular for m = 0: 1/2 PI11 (cos 6)eimi?.76) and (11. orthonormalized system of eigenfunctions. m > 0. The functions Y™ satisfy the orthonormality condition dfilimT' = / Jo < W sm8d9Yr(9. so that im(ip±2ir) The functions Ylm(8. (11.=o m=l sin 6 The connection of these hyperspherical functions with the associated Legendre functions P™.\ = W . (11. is given by (2Z + l ) ( Z .80) y? Furthermore Y°l = V^fcosO).78) and the completeness relation oo I ' YrVMYTVrf) = °^~' ± J2 Y7n*(f}.79) . and that their phases are such that Yt0(0. (11. y2° = \ (3cos 2 01) 2 V V 4vr V 16?r 3 W—. Uniqueness implies immediately that m has to be integral.r o ) ! YT{e.(p) are called spherical harmonics.i». We mention in passing that in spherical coordinates d lz = xpy .77) and o these are determinable by the requirement of uniqueness of the original wave function and its square integrability.n>).11.<p) Jo = 6mm'5w. / 3 / 5 J — casO.t»'\Y7n((l'.4 Angular Representation of Angular Momentum 211 observe that Z and m define the eigenvalues of Eqs.V>)Yir'(8. This is the case if one demands that l 2 and lz possess a common complete. (11.82) .(5cos 03cos0).^ = *(n .0) is real and positive.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. (11.* ? ^ . These functions are defined in such a way that they are normalized to unity on the unit sphere.ypx = tfrK(11.
i..88) we obtain for every value of x ^2 ai(K + 0( K + * . is also in use.m = 0. For the solution we use the ansatz of a power series. The Coulomb Interaction The number I is referred to as the "orbital quantum number" and m as "magnetic quantum numbed. f. Consider the equation for Pi(x).g. (11.e. The associated Legendre functions or spherical functions PJn(x) satisfy the differential equation (1 . (11.2(m + l ) x i f + [1(1 + 1) .2.1. (11. (11. of Ytm.Y^O»[(K + i)(/e + i ~ 1) + 2(re + i) . (1 .p. exp(imip)..x2)Pf .87) we obtain the equation (1 .2xy' + 1(1 + 1) 1x The number m must be an integer already for reasons of uniqueness of the wave function. i.83) Their connection with the Legendre functions Pi (x) is given by the relation pr{x) with the differential equation =£^ Pi{x) (iL84) (1 .x2)P[' ..90) . Instead of the values I = 0.l)xK+i~2 i .2xy' + 1(1 + \)y = 0.88) y = ^aiXi+K. oo (11.m(m + l ) ] i f = 0..212 CHAPTER 11.2xP[ + 1(1 + l)Pi = 0.86) y = 0.85) (11. i. It is instructive to pursue the appropriate arguments. Setting y = (lx2)m>2Pr(x) m . (11. i=0 (re lowest power > 0).x2)y" . 22 (11. If we did not know yet that I = IQ = an integer > 0.. respectively..1(1 + l)]xK+i = 0.d.e.89) Inserting this expansion into Eq. in the replacement ip —> if + 2K. the designation s.e.h. we would now have to search for those values of I for which the solutions are square integrable.x2)y" .e. i.
ai(re + !)« = 0. For the solution in accordance with (11.46)) p.93) where Xl(r) is the solution of the radial equation (cf. This is precisely the case. 93. For \x\ < 1 we can obtain from this the coefficients of a convergent series. However. Eq. Schmidt). Murphy [190]. (11. Comparing the coefficients of xK+3 we obtain aj+2(K + j + 2){K + j + 1) = OJ[(K + J)(K + j + 1) .29)) 2 fe2l 92 Pr = & r 7Tororz We set (11. We see that this happens precisely when I is a positive or negative integer or zero. but also those for a different series for a. l 2 and lz are the solutions of the Schrodinger equation of the form (11. when the series terminates after a finite number of terms.78) to be square integrable. (11. (11.1 we obtain by comparing coefficients QOK{K 213 — 1) = 0. these are of little interest to us here. . Margenau and G. l(l + l)h2 _2mo + 2m. (11.92) This is a recurrence relation.*" 11. (11. > 1.1.26)) yi(r = 0) = 0. (11. p. M.95) Vl = rxiThe hermiticity condition for pr then requires that (cf.91) from which we deduce that K = 0. Eq.5 R a d i a l E q u a t i o n for Hydrogenlike A t o m s The common set of eigenfunctions of the operators H.96) + V(r) . . We have therefore (multiplying Eq.1(1 + 1)].94) ^For further details see H.5 The Radial Schrodinger Equation for Hydrogenlike Atoms For i = 0.Qr2 with (cf. they have to be polynomials (recall the orthogonalization procedure of E.11.E Xl(r) =0 (11.2 + 2mo dr 2mor V{T)E Viir) = 0 (11.94) from the left by r) h2 d2 h2 2 + H! + \)IL. first line of Eq.
(11.e. nor the condition of normalizability for l ^ 0. We assume that V(r) has at r = 0 at most a singularity of the form of 1/r. In these cases the behaviour of yi near the origin is dominated by the centrifugal term and thus Eq. The wave function of the relative motion is the solution ^(r) of the equation A 2mo Here •4>{r)=Ylm{eM and Vl(r) (11.214 with CHAPTER 11. The equation also has an irregular solution.96).99) e ip(r) = Eip(r). Jo Jo We thus have a problem similar to that of a particle of mass m in the domain (0. He + . since it satisfies neither the condition yi(0) = 0. The Coulomb Interaction (11. i.97) / r2dr\Xi(r)\2 = / \yi(r)\2dr. In order to verify this behaviour we set <VflV>r> yi(r) = rs(l + air + a2r2 \ ). In the first place we investigate the behaviour of yi in the neighbourhood of the origin. (11. an analogous consideration applies to hydrogenlike atoms like. oo) of a onedimensional space. and others.102) .98) 2mo e2 _ 1(1 + 1) yi = o e+ 2 y'l + h r r2 2m0E e = (11. (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.101) 2 r ^ We set  6 = ( _ e ) 1 / a = ( = (11. Obviously the irregular solution has to be rejected. we obtain the socalled "indicial equation!'' s(s .96) can be shown to possess a regular solution there which behaves like rl+1{\ + 0(r)) in approaching r = 0. We now consider the hydrogen atom with V(r) — —e2/r (i.1)). and substitute this into Eq. for instance.100) with (11. Z\ = Z2 = 1 in Eq. r m0 = meMp me + Mv (11.1. s = 1 + 1. which behaves like 1/r1 in approaching r = 0.e. Equating the coefficient of 1/r2 to zero.
108) This equation has the form of a confluent x$" + (6 .+if>x/2 vi(x).e.v) vi(x) = 0. for E < 0 they > decrease exponentially.100) as^ d2 dx2 1(1 + 1) x v___\ 4 2/1 = 0. see e.11. (11. (11. p. i. T This equation is known as Whittaker's equation.105) 11. (11.529 x 1 0 " 1 0 meters) 1 V/2 e2f hc\ m0c^1/2 2E 2moE J (11.g.6 11. Magnus and F. hypergeometric (11. (11. *The factor 47reo appears in SI units.a * = 0.e.88. d2 x—j 1 + (2l + 2x) ax d j dx (/ + 1 . behaves there like xl+1) extended to an exponentially decreasing branch at infinity.105) we obtain the equation for V[(x). It remains to determine the function vi(x) for a complete determination of the solution. Inserting the ansatz into Eq. .e. (11.107) equation. (11.105) t h a t for x —• oo there is a solution behaving like exp(—x/2). We set x — We set as Bohr radius* a = and 1 v = rea rriQe ft V (47Fe 2KT.6 The Discrete Spectrum of the Coulomb Potential 215 For E > 0 the solutions oscillate in the region r — oo.x ) $ ' .6. We set therefore Vi X J. i.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. Oberhettinger [181]. We can see from Eq. W.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.104) W i t h these substitutions we can rewrite Eq.103) °y moe2 ( = 0. (11.
(11. i.111) Important properties of the function T(z) are: (1) zF(z) = T(z + l). The Coulomb Interaction The regular solution of this equation is the confluent hypergeometric series (also called Kummer series) 1 . The function T(z) is defined by the integral /co T(z) := / eHzldt. 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. (5)r(n + l ) = n ! . since the exponential function (generated by the infinite series) would destroy the asymptotic behaviour of the wave function which we separated off with Eq. (11.112) The gamma function is a nonvanishing analytic.! ) ! = imlFI' c a l led inversion or reflection formula.1/2)!.(0_i)!(6 + n l)!n!' (11.z(zl)\ = z\.106).J. However. particularly pure mathematicians. + ' ' ~ 6 1 ! + 6 ( 6 + l ) 2! + Z. . 02zl / i\ (3) r(2z) = — ^ r ( z ) r ( z + . the factorial for an arbitrary argument.216 CHAPTER 11.z) = ^f^. called dupZication formula. ^ ( 2 ^ ) ! = 2 2z z!(z . 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. (11. it is convenient to use only the factorial notation. more precisely. are very strict in reserving the factorial exclusively for positive integers and the gamma function for all other cases. with many of these around in some calculation. meromorphic function.109) In our case we can write the series The expression* r(x + l) = x\ is the gamma function.e. (z)!(z . (2) r ( z ) r ( l . and for both cases. (4) r ( i ) = V5F. and hence the *Some writers.
I) r(vl p)sm{n(i/ I . One defines as the principal quantum number the number n. if the series terminates for certain values of u. Eqs. i. (11..114c) (Recall that for an arithmetic series a + (a + d) + • • • + (a + (n — l)d) = na + n{n — l)<i/2.104)) 1 n =v = K.101) to (11.e. for z. With (cf. For every value En. . . every n. which are also described as its nodes. (11. = n = Z + l + n'. i. 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 . and so to the number of different values of m. which is thus seen to be a consequence of demanding the square integrability of the wave function.p ) } n (l)Pr(^Q n'\ K T{vlp) ' (n'p)!' Note that 11 + 1 is the number of possible orientations of the angular momentum vector 1 with respect to a preferred direction. 2 . Thus the infinite series must break off somewhere.2.V))T(v . The quantum number n' is called radial quantum number. .11.v) T(l + lv) _ ~ K sin{7rQ/ . (11.1) + n = n(n .) 'Observe that with property (2) of the gamma function we can reexpress the ratio of gamma functions in Eq. in our case: a = 0. n' = 0 . • • • • <»•»*> It should be noted that the magnetic quantum number m does not appear in this expression.» • » • » . This is achieved. This number is equal to the number of finite zeros of the radial part of the wave function (excepting the origin). must be a polynomial.1) + n = n2. the orbital or azimuthal quantum number I can assume the values 0. 1=0 (11.110) as follows: T(l + 1 + p .113) This is a quantization condition.e.. i..6 The Discrete Spectrum of the Coulomb Potential 217 latter would not be square integrable..a " \ 2moE we obtain the energy spectrum of the discrete states with angular momentum I ie t m0e2 / 1 N1/2 * =  ( s ) ' ^ . .1. .e. 1 . d = 1.
l = 2 to n = 2.1 Hydrogen atom energy levels. .d. where the superscript 21 + 1 at the top left of I gives the degree of degeneracy.1 = 0 (called principal series...\ I = 0 or 1. and hence the fine structure of the energy levels is excluded.§ One also writes n2l+1l. 2 . nf .... ' O n e describes as Lyman series the final state with n = 1. nd. np. hence s).4.. hence p). . The following spectroscopic description of states is customary: nl —> ns. from an initial state n > 2. The above treatment of the hydrogen atom has obvious shortcomings: (a) The hydrogen atom was treated nonrelativistically. I = 0.'.1. hence d). I = 0 to n = 2.1 = 0 (called sharp series. ne.'. .1. from an initial state n > 2. stand for I = where n denotes the principal quantum number and s. 0. The spectrum of the hydrogen atom thus has the form shown schematically in Fig.1 = 1 (called diffuse series. One describes as Balmer series the final state n = 2. from n. 11. respectively.1 = 1 to n = 2. and from n. . / = 1.. relativistic corrections are of the order of v2/c2 ~ 0(En/moc2) as we illustrate in Example 11. 11.218 CHAPTER 11. (c) The nucleus was treated as pointlike and not as something with structure.. . The Coulomb Interaction E=0 degeneracy lEl 3 31s 21s 33p 23p 3 5 d 9fold IE I 4fold IE I 1 1 1s none Fig.p. (b) The spin of the electron was not taken into account.
Magnus and F. 844. I. With the help of Eq.2. . Each can be recognized by reference to the generating function.11. Thus Messiah [195] defines the original or stem Laguerre polynomial as : >Lr(z) = *L°r(z) = e'^(e'zr). We mention here three different definitions used in the literature. (u.z) = (r + k)\ Lkr(z). Vol. Appendix B.115) we can reexpress this function as Lr{z)  r\k\ p=0 \* p\ (rp)!(fe+p)! (U 118b) ^This definition of *LJ? is used by A. Apart from factors.2 (11. in fact. for instance. Messiah [195].110) is usually reexpressed in terms of orthogonal polynomials known as associated Laguerre polynomials.r f . (11. M. 11.1.p)\(2l + 1 + p)\ The spectrum (11.2l { + 2. k + l.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.r = 0 .{z) is the associated Laguerre polynomial as normally defined and mostly written Lr (z).6.414(3).118a) where Lk. (11. Unfortunately. several different definitions are in use.117) The function *Lk(z) is a polynomial of degree r. We distinguish between different definitions in use. 2 . . it is given by^ *Lkr(z) = ^ ^ f 1 *ir. . Gradshteyn and I. p. and I. . "1'f„+11" P] „. . • ( .= t . 84. • (11.115) ' =0 (ri . For clarity and later use we denote the function used there by its normal form L^. l . The mass of the hydrogen atom is given by massif = Mp + m e + E < Mp + me. (11. however. the polynomial contained in y\ of Eq. 3 . formula 7. Unfortunately the definition mostly used in mathematical literature differs. Oberhettinger [181]. a + 2:*) = t v .6 The Discrete Spectrum of the Coulomb Potential The associated radial wave function is yi(x) 219 = xl+1ex/2<f>{n'. so that one always has to check the definition. p. see. • fc.x). Ryzhik [122].114b) yields the binding energy of the states. W.115) is effectively the associated Laguerre polynomial normally written Lk(z). S.
is defined by^ dk (—'\\k(r^\2 (11. Appendix B.220 CHAPTER 11.1. pp. Ll(z) = 33z + ±z2. The reader is there warned that '•'•care must be taken in reading the literature to ensure that the particular convention being followed is understood'.121) In Example 11.120) ' r=0 It will be seen that the orthonormality condition is" z ezJc*Tk.. 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. It is said that this is the definition preferred in applied mathematics.119) dz dz In particular one has (we cite these for later comparison) LQ(Z) = 1. Example 11.. N.(z)*L TJz)dz 3 = KP + kV}Opqit Jo (11. L2(z) = l2z + z2. 162 . .„\*Kk. and Ll(z) = l. which we write ^Lk.164.3): e2t/(lt) (1 .2.z . Sneddon [255]. r r Z r=0 v tr k k (11.„M„ z"*L. Vol. I. Messiah [195].t) + k l r)! E E (zY (i +(fc +h)\(ri)\ i\ r=0 L OO p OO i=0 E (k + r)\ *L (z) = J2L ( y. . (11. ^ F o r this definition and the following equations we refer to I. Sneddon [255]. L1(z) = l . N.3 we show in a typical calculation how this normalization condition can be obtained with the help of the generating function. **I. 163. p.** The associated Laguerre polynomial here. L\(z) = 2z. For our purposes here it is convenient to refer to yet another definition of the associated Laguerre polynomials.122) "See A.d z—^z + {k + l~z)—+r Lkr{z) = 0. . 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 generating function of the associated Laguerre polynomials so defined is (observe (—t)k as compared with (11. Solution: We consider the following equation of the form of Eq.124a) (i*) pi The orthonormality condition is (cf.120)) it)k& zt/{lt) ~*P p=k fc (11. (11. and obtain with this the normalization condition (11. Example 11.4 + 2z. (11. i L1(z) = l . (11.119) derive the generating function of associated Laguerre polynomials (11. (11.6 The Discrete Spectrum of the Coulomb Potential The differential equation of these associated Laguerre polynomials is** z 221 ~r^z + (k + lz)— + dz dz d2 n . . '"''Comparison with Eq.11.119): ty"(t) + (a + 1 . Then.1 8 + 18z 3z2..4z + . N d rk 0.121)..3: Generating function of Laguerre polynomials With a Laplace transform ansatz for the solution of a second order differential equation like Eq.120).z .123) For the purpose of clarity and comparison with the polynomials of the other definition it may help to see that here in particular % ( « ) = !. Jfpd>2s{p))eptdp.. ^L2(z) = 2 . and partial integration. with y'{i) = — / ps(p)e~ptdp ty'(t) = t f ps(p)eptdpp2s(p)ept s(p)e Pt J ^(ps(p))eptdp. (11.124b) Comparing Eqs.% ' ( * ) + by{t) = 0 with ansatz y{t) = I s(p)e~ptdp.122) we see that tLkr+k(z) = (l)k*Lkr(z) = (l)k(k + r)\Lkr(z). ^L\(z) = . The limits will be determined later.118a) and (11. (11. *L\(z) = . and *L\(z) = .1 ..124c) Example 11.107) implies k = 11 + 1 and r = n + I.8): (11.
the only possibility is the last. Case (1) with p —• oo implies y(t) ~ / dppa~1e~pt ~ t~a..119) with a and b as there. We demand (to be considered later for integration contours) the following condition. (11. p° Possible paths satisfying this condition are: (1) If b < 0 from 0 to oo. Then dp = dq/(lq)2.] from the above equation: C(p) := p(p + l)s(p)ept = 0.a = 21 + 2.li I u {p(p+l)s(p)} : p(p + l)s(p) dp a + 1 b p+ 1 p(p + 1) a + b+l p+ 1 b p Integrating the equation we obtain s(p) and hence y(t): a+b+l^b ln[p(p + l)s(p)] = In ip+ir+o+'p (p+l)a+b f (p+l)a+b The condition C(p) = 0 now implies for possible contours of integration in the plane of complex p: C(p) = p(p + l)s(p)e'pt („ _i_ na+b+i = ^ ' e~pt = 0. (11. since for n > b : (b — n)\ —• oo. Hence the expression in square brackets vanishes. (2) from —1 to oo." b + 1 e~ptdp = coefficient of pb in g(p) := (p + \)a+be~pt. case (3) similarly. Case (2) implies the exponentially increasing solution ~ e + t .) 2 J J p 6 + i ( l . e ^ t '?/( 1 '?) = coefficient of qb in /"OO / = = / Jo Ll{t)Lac{t)tae~tdt taet{P/(lp)+q/(lq)+l)dL l t I * * [" (2TT.124d) In order to obtain the orthogonality and normalization of the associated Laguerre polynomials./ ( l — <?) —— = G(t). Then the preceding two equations (inserted into the original differential equation) leave us with I eptdp d {p(p + l)s(p)} .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 + 1 = 1/(1 .222 CHAPTER 11. The Coulomb Interaction since d(p2s)/dp = ps + pd(ps)/dp. s bK> (lq)°+i (11. t > 0.^ . 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. It follows that 1 d i I _i. j p = (P 0 "V?. 2TT_ Jq=0qb+l(lq)a+l 1 .p ) « + i q c + i ( i . agreeing with Eq. which removes [.. this is the rejected irregular solution.p ) o + i 7 t = 0 .( a + l ) p s ( p ) + 6s(p) dp 0. (3) around p = — 1 if a + b is a negative integer. n! (6 — n)!(a + n)! An alternative integral representation is obtained by setting p = q/(l — q) or q = p/(p + 1).q) and P~ t ( . we insert this contour integral representation of the polynomial in the following integral and obtain __?(_) = P da <* . and (4) around p = 0 if b > 0 and integral. Thus in the case of Eq. . with requirement of a finite solution. p and one has (as polynomial of degree 6.
. is introduced for dimensional reasons.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. The wave function of the state (n.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^ exx^k^Lkn{x)^Lkm{x)dx. f a p)(i 1) r + 1 l (1P9) J The coefficient here required is obtained from the expansion as . Jo (11. m) is 3 2 &nlm = a~ / NnlFnl (—^Y^O. and obtain r taet[P/(iP)+q/(ig)+i]dt Jt=0 With this the integral I becomes = . Eq.(x). so that I = (c + a)! 1 f . (11.?).—: <f h_nJ_[ dp = (a + c)\ T^ if c = o.111).3 T h e eigenfunctions According to the previous discussion we obtain n2 orthogonal eigenfunctions in association with the eigenvalue En. (11.11.112).c (al)\ d(col)! c(c + a)\ c!a! with use of the inversion formula (11.—i. /•OO / Lg(t)£?(t)taefctt: (a + c ) ! .125a) and the factor a~ 3 ' 2 with a = Bohr radius. 11. where Fnl(x) = yi(x) (11. Clearly only the second form is sensible.6. I.127) . and 0 if cj=b. 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) = xlex^LZ.
2.v (6n2  6nk + A:2 + 6n .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 . and for the explicit derivation of the second case below Example 11.(l) n+l.7. This behaviour is indicated in Fig. For the ground state (n = 1.3(2Z + 1) + 2} 2 {2(n + 0 .130) We see that the mean value of r is the larger the larger n is. 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 + /„.8):* OD  <n!>3 (nk)\ (nfc) (n\)3 i£ n (3) = _ . 2 (11. + ..2Z .1 + 1) (11.128) These expressions permit us to obtain the mean or expectation values (1/r) and (r).6(n + Z)[(2Z + ! ) .3fc + 2). 11.n+l na l2 ° ^ 2 na n+L+lV) (2n + 2/ . = 0) we have ) r /u =.! ] + (21 + l ) 2 . The Coulomb Interaction One finds in particular (see also Example 11. a (r)is = ~a. (11.224 CHAPTER 11. thus on the average the electron is the farther away from the proton or nucleus the larger n is. (11.131) .(2Z + 1) + 1} = ^[Zn21(1 + 1)].
In this case (r) = = For (r 2 ) one obtains < r 2 ) = n 2 ( n + ^ ) ( n + l)a 2 .6 The Discrete Spectrum of the Coulomb Potential r.j .20. i.2 Hydrogen atom wave functions.e.(r) = x W 2 n + 1 =.133) for large r. / ! J — . we have and tp becomes particularly simple.^ We obtain therefore for the quadratic deviation (r) .134) *These are taken from I. Sneddon [255]. with n +1 replaced by n and 21 + 1 by k. \f{r ) ~ n a.11.J n 1s 2s 3s *r Fig.( n .l ) n ] =  [ 2 n 2 + n] an[ n + (11. 11. In the case in which I assumes its maximal value n — 1.2 1 A r = V (^2) .132) (11. = l a nl n N'nll rFnl.2 I 3/2 Kl r 225 \2 1 R„.. This corresponds to E ~ — e2/2n2a like classically with circular radius n 2 o. or (r ) ~ n a 2 2 ^ [ 3 n 2 . v ^ V2n + 1 (11. In the Kepler problem of a particle of mass mo in Classical Mechanics with the Newton . Problem 5. S. pp. 173/174.
pJ = —— and j> Pidqi = nth. in which periods are obtained from the BohrSommerfeldWilson quantization condition. = ngh. i. The above treatment of the hydrogen atom has obvious shortcomings: (a) The hydrogen atom was treated nonrelativistically. identifying yj (r2) with aKeplen w e see that T K ep l e r = ^(r2)^ = ^(n2af/2 = *^a3'*n*.e.226 CHAPTER 11. and separating this in cylindrical coordinates r. relativistic corrections are of the order of v2/c2 ~ 0(En/moc2).135) This expression becomes very small for large values of n. (b) The spin of the electron was not taken into account and hence the spin fine structure of the energy levels is excluded. The Coulomb Interaction or TT = ^ = (11. Since 6 is a cyclic coordinate.e.— > where ampler is the length of the semimajor axis of the elliptic orbit. „ _ 'Kepler — the Kepler period is given by o Z7ra 3/2 Kepler .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). the states Ytl are predominantly concentrated in the xyplane). for these values of I it is not distributed with uniform probability over the spherical surface. determine the energy E by evaluating the BohrSommerfeldWilson integral of "old quantum theory". . we consider in Example 11. Thus. Solution: We have H = yjm%c4 + pIn cylindrical coordinates.4 the relativistic equation in cylindrical coordinates. In order to obtain an impression of relativistic corrections in a simple context. \/v vT 7 v^ See also Example 14. i = 0. but rather like classically in a plane (i.r.3. This means. Example 11. we have pg = const.6. (c) The nucleus was treated as pointlike and not as something with structure. p2 = p2 + — pg • eV. the electron remains practically localized within a spherical surface of radius (r). 6. However. Thus we have to evaluate the following gravitational potential written V(r) = —mofi/r. r.
11.4 Hydrogenlike a t o m s in parabolic c o o r d i n a t e s The Schrodinger equation of a hydrogenlike atom can also be separated in parabolic coordinates £. 86. Schiff [243]. .9. The deeper reason for this are symmetry properties of the problem.2. but depends very slightly on ng alone which gives rise to the "fine structure" of hydrogen lines. ( v c + i ) . where the constants are identified by comparison. as also indicated in Fig.136b) £ = r ( l + cos#) = r + z. p. 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. 11. it is possible to separate the Schrodinger equation in other coordinates. (and subtract moc2 to We then obtain with a = e2/hc ~ 1/137. y — 0: V ^ c o s tp. <p = <p.* See L. y=y/%qsanp. ip are the spherical coordinates. (11.6 The Discrete Spectrum of the Coulomb Potential 227 integral which we achieve with the "method of poles at infinity" explained in Example 16. r\ = r ( l — cos#) = r — z. where ip = 0.6. which correspond to a new choice of polar axis.T).V . It follows that (with h = 2nK) Z2e c h?ni 2 2ixnrfr = — 27r nehtjl  Ze2E >^m2c2E2/c constant'. Thus E2 2EeV(r) e2V2(r) nrh = a> prdqr = ilc2 . where r = ^/x1 + y2 + z2.. Whenever there is degeneracy. (p).(p .136a) (11. 11. 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^la Z /r. which are related to Cartesian coordinates in the following way. z=^rj). 2 2 Therefore the energy is no longer a function of n = nr + ng.3.d r Kmlc2 + T U^\±AnlK2^Ur 2 2 \ c / r * + * ! + $*—».
In terms of the coordinates (11. 2 . 1 .137) One now sets the total wave function ip = £(£)!N"(r7)$(</?) and divides the equation by ip.22) becomes — as we show in Example 11. and r\ = const. We rewrite the first of these equations with the help of the relation M (w ^  m + e) .228 CHAPTER 11. d_ dirt) + TUQE fm0Z±Z2e2 2h2 V ^2 A V m (ri'N) Arj1 2n Since these equations are alike. there is really only one basic equation to solve here. (11.3 Parabolas £ = const. . one obtains the following two equations: £ ) + TJIQE A m2n  « ' m = o. . so t h a t the variables can be separated. The Coulomb Interaction Fig. a separation constant. Multiplying the remaining equation by (£ + r /)/4 and setting the £part equal to —A. Then 1 d2$ izimip $ oc e $cV = —m where the separation constant m2 must be such t h a t m = 0 .3 — h2 ( 4 r d (rd^\ d_f df\~ + 1 d2xl>\ £?7 dip2 2Z1Z2e< Z+V •ip = Eip. for uniqueness of the wave function. with zaxis as axis of rotation. = 0. 11. .136b) the Schrodinger equation of relative motion (11.
To obtain a closer analogy with the separation in the spherical polar case.105)) Hence we write I = (m — l ) / 2 .6 The Discrete Spectrum Then of the Coulomb Potential 229 A 22 (£ 1 / 2 £) + moE 2h2 A m2 — 1 +£  4£ 2 (cT1/2£) = 0. m0Z1Z2e2 ^ 2 = n (say). .125b) (dividing by p 1 / 2 ) £ oc e . E x a m p l e 11. . . i. We leave the calculation of the degree of degeneracy to Example 11.i n +n = n i + n 2 + m + l = we obtain the energy eigenvalues given by E = n i = n' \(m + 1) = 0 . . . 1 . ds2 = dx2 + dy2 + dz2. 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. .' / 2 p m / 2 L 2 Z + i = m ( / 9 ) ) Analogously we have N oc eCT/2<7m/2L™ (a).106) and (11. . Adding now / . In the spherical polar case we had (cf. (11. from Cartesian to parabolic coordinates. . we set (since E is negative for the discrete spectrum we consider) ITIQE =*"• H J?P ^( <9p: 1 / 2 i m0ZiZ2e2 A = n". T h e n by comparison with the spherical polar separation the answer is as in Eqs. as well as the Laplacian A = V and the Schrodinger equation.114b). . " . (11. p = /?£ and cr = /^r?.6. . 2 . . n 2 = n" . Eq. ph h2p2 _ h2 m2Z2Z2eA hn 4 2 _ ~~ m0Z2Z2e4 2ft2 n2 ~ 2m 0 ~ 2m 0 This agrees with our earlier formulas like (11. 1 .o.e.± ( m + 1) = 0 . 2 .11.
2d? dr)}.230 CHAPTER 11. and so dz = (d£ — drj)/2.e. 2 + ^dV . (n _ !) n = n2 §See e.e.tx2 . i. " . We use the method of selector variables.g.e. see Eq. i. The general formula for the Laplacian is§ ^ follows that 1 [9£\ A = ' 9 \ gr. Observing that the highest possible value of m is n — 1 (i. Beware! For m = 0 there is only one mstate.1 in this expansion is seen to be n. > . 712 and m which give n. W. dx = d{ y £77 cos ip) = .2) and summing over m.[df + dr]2 .E E n l „ . Solution: For the wave functions ip = £(£)J\f(r])&(ip). n 2 (1UJ)2 The coefficient of a . Oberhettinger [181].136a). Magnus and F. (11. . Here we have (see preceding text) 711+77. n the double sum ]T]2 is TI — m. .114c).l ) ( n .136a).e. last chapter. Similarly we proceed with dx and dy.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.. for m / . (11. Thus dz2 = . </. 2 _ = (9id02 + (gvdvy + /„ . We then sum all possibilities and select the coefficient of ojn~1 in the result. Since the lines of constant £ are orthogonal to lines of constant 7 (they cut perpendicularly). \— < / — cos ipdr] — \/£r] sin ipdtp.2)d/2. all cross terms cancel in the sum of 7 the squares.2 . Thus we want to obtain the coefficient of wn~1 in £2 . / — cos ipdl. Eq. and attach a factor to to each term introduced in the wave function. for instance. (11. when m = 0 = 77. 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. . = e±im^e("+^'2{par'2L^{p)L^ (a).2)d} = (n. 2 2V? V^ One now performs the square of this and obtains similarly dy2.* 71 — 1 n—1 n— 1 n n+ 2 E ( n ~ m ) = n +2 E n _ 2 E m= n + 2n(n .2 = n — 1. z = (£ — rf)/2. + l 4 {^^)drl f i + v \ .?e 9£ J 4 t d \ g^gv 9 \ dri \ gv dr] J t d \ g^gr.1) ~ . etc.2 = n — m — 1 and therefore the number of the coefficient of u}n'm1 . Thus. We consider first the case m = 0. / J„^2 . we have to find the number of values of the quantum numbers n\. as t h e two signs of t h e exponential indicate.g<p 9 \ . 'Recall that the sum of an arithmetic progression is given by a + (a + d) + (a + 2d) + • • • + {a + (n . 0 there are two.l)a + (n . The Coulomb Interaction Solution: We begin with z of Eq. N2 where g^dS.. and one obtains 2 = _ 1 4{ £ ) * (t + 7 )\ j. i. are elements of length. In the case m ^ O w e have n\ + 77. 9 \ 9<p\ gv 9f\ _ 1 a2 a / a\ a / ay £77 dip2 E x a m p l e 11. {g^Y.
i.1 (cf.{ n i {2ni + TO + 1} + {ni j=± n 2 } 6(nin2)(ni+n2+TO+1) ^±n2} ^7—. i.e. { 6 ni2 + 6 n i ( m + l) + (m + l ) ( m + 2 ) } . and determine the average value of v for a given state. We obtain the volume element dV in parabolic coordinates from the above results as dV = g£gvgvd£dr)dip = — (£ + rj)d£. Apart from the power of the leading factorial (which results from the definition of the associated Laguerre polynomial in the generating function).g. we now have (pa) 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.) = ^eF(p . With the help of the wave functions we derived above.eF / nh2 \ {v)=3(n1n2)^—lr2j Thus in this case nondegenerate perturbation theory could be used as discussed in Example 8.8 and 11.(p)L™ (<7)]2(p + a)dpda rev where JJ = / Jo dpe~ "p"[I.—r~n = 3(mn 2 ). . s 0/ . . Thus e. with Eq.6). Z2 —> e.124c). which will then give the change in energy of that state to the first order (i.128) if one makes there the substitutions n —> n\ (or 712) + m.7: Stark effect in hydrogenlike atoms 231 Consider a perturbation v = eFz.e. L ni\ ny.127) and setting n = r + k. I. where F is the electric field. In fact.r+k(l) = / Jo e~*xk[(k + r)\Lk(x)f = [(/= + r)\flk.11. if we expressed the perturbation in spherical polar coordinates. eF{z). with the connection » (11.•* n (l) = / Jo It follows that . Solution: Using formulas and results of above we have to determine (v) = eF{z) = leF(e.9) one finds that _ = %2 jm+2 {n2+m)\ (ni + rre)! +1 : > in. Ikr+k.128). (11. we can relate the expressions Iq to those of Eq. these expressions agree with those of Eq. (11. fc — m.a).. 8. with Z\ —> Ze. = j (2ni+m+l). with RayleighSchrodinger perturbation theory). . with z = r cos 9. .6 The Discrete Spectrum of the Coulomb Potential Example 11.r. ( ! ^ ± ^ ) l { 6 n 2 + 6 n i ( m + 1 ) + ( m + 1 ) ( ? n + 2)} ny.drid<fi oc (p + o)dpda. (11. {p — a) — = dxe~*xkvLkn (z)] 2 .e. . Sec.™(p)] 2 By contour integration (see Examples 11. 2(ni +712 + TO + 1) Therefore the change in energy of the state to first order is given by. However. we would have to use degenerate perturbation theory and obtain the same result.
g )(l.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 .( 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)!. L™(t) = coefficient of qn in G(t) e tq/(lq) (1 . dt = ds/a.p ) r + 2 [(lp)(lg)]"l+1(lpg). .124d) of the alternative integral representation obtained in Example 11. _ „ .p ) ( l .232 CHAPTER 11. s = at.9) with the help of the generating function G(t) of associated Laguerre polynomials obtained in Example 11.p ) ( l .l)!r! X We now evaluate the integral / as the coefficient of pnqn _ ( m + l)! [ ( l.l ) ! ( m + l)! in the expansion of p " ( n + m + 1)! p " " 1 ^ + m)! 1 _t . tt„:_ (1 .1)! n\ This verifies the expression I™t in Example 11. c positive integers.1 ) ! (n .p n _ 1 ) in (n — 1)! J p"(n + m + l)! n! p n .9 ) ] m +h 1 yJo 0 ' 1 —pP 1 q 119 lpq _ (l9)(lp) = a With (see Eq.pK — — —L .b.9 we consider integrals with an arbitrary power of t.a. the integral / is the coefficient of pnqn in the expression _ / . The Coulomb Interaction E x a m p l e 11. (11..128).n+2 in the expansion of i n c o e f f i d e n t o W (m + 1 ) 1 ( 1 .3: /"OO I := Jo et[L^{t)]2tm+1dt.0 ° e *[l+P/(lP)+9/(l<j)] t m+l di) t  P  9 l [ ( l .x ( n + m)! ( n .^~T^——^ ) = coefficient of (pn . (n + r . In Example 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 . and is a special case of Example 11. the integral arising here becomes with Eq.3).8: Laguerre linear expectation value integral Evaluate the following integral (which is different from the normalization integral of Example 11. E x a m p l e 11.7 and (remembering the definition of the associated Laguerre polynomila there!) the second of relations (11.9: 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. (11.1 ( n + m)!"[ _ (n + m + 1)! (niy J n! (ra + m)! _ (n + m)! (n .9 )« Solution: Using the generating function for both associated Laguerre polynomials.3 by one power of t. ) (l_pg)(m+2) Now the coefficient of qn in (1 — g)(l — p<j) .
i(a) = / Jo dttaet{p/(lp)+q/(lq) + l+a} = Q . (a) j / = 1.(*) from the coefficient of a1: />00 233 1(a) = / Jo dte.™( r > (1 + a)P r £^0y^w Jc+6_2r = .e.q)(l .(i) = i ! ( . . 7£. — (cr)!r!(6r)! Kv. where (a+**£)! c j n Jo = .e.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.a t [Lg(i).6 The Discrete Spectrum of the Coulomb Potential Solution: One considers the following integral and obtains /£. ° ^ + a ( l .f ]. c + b odd). In case (a) we have u \ 7 V^ ir fb + c\ 7(a) = Jo / Tli u.?(t)t a e. so that = (q+fc±£) .p) .a _ 1 = [1 + a ( l . i.111) of the gamma or factorial function.r ) ! a ' .l ) * x coefficient of a* in the expansion of 7 ( a ) .e. = c (q + fe)! W ° (^)!(^)!(^)! ~" (2) In the case of one power of a. (lp)(lq) + a(l q)(l .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 + aapj c\a\ c!a!(l + a)a+c+1' [l + ^ P ] " [1T^P]Q+C+1 y>^ d(a + c + C . .^ ( l .11.e.Pq + Q(1 " 9)(1 " P)1_a_1 = ti f p^ X COeffident f ap)]"01. we have to have y + (c + b — 2r) = 1.124d) of an associated Laguerre polynomial.p ) ] . we can reexpress 1(a) in the form: /(a) = (2^)2 J J p t + l ( l . we have 2r = 6 + c = even.w(r) = and / a ) = r\(P ~ r .a f + c)\(vr)\(l + ayrP „ • ^0^r(cr)\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 «.L. i.p) _lpq It follows that (using Cauchy's residue theorem in integrating around q = 0) 1{a) = (2^ / / ^ ^ [lpq c [ 1 .1 + y w)\(a)y (r — w)\w\(j3 — r — l)!(j/ — w)\ Special cases: (1) In the case a = 0. c + b — 2r = l (i. c + b even) and/or j / = 0. Using the contour integral representation (11. c + 6 — 2r = 0 (i.
234 and (only w = 0 and 1 arise in the sum) CHAPTER 11. 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. b\ V cited in Example 11. + (!±<)«. In view of the slow falloff of the potential at infinity. is the work of J. Taylor [267]. 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. Zie given by h2 _ A _ ZiZ 2 e 2 V>(r) = Ei/>(r). These results are seen to agree with the expressions 1^. .~* 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 . also with regard to earlier literature.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. like the asymptotic condition and definitions of the scattering amplitude and the phase shift. Since such a rigorous treatment is beyond the scope of this text. R. 2mo Here we set (this defines the velocity VQ) j? E ^ 1 (11.1%^ 11.^)(.7. however parallel to the Coulomb case. The Coulomb Interaction b + c\ 2 J (b¥)\(Pb¥w)\{a) ($±£ w)\w\(J3 ^±£1)1(111.)!' f^)(.138) " 2 ^ =2 ° ' 2 m Uo 7 = ZtZ2e2 ^T (1L139) * Screened Coulomb potentials are considered in Chapter 16. the standard principles of nonrelativistic scattering theory. . ^A very readable source to consult. cannot really — and this means in a rigorous sense — be applied in this case. t We have the Schrodinger equation of relative motion of particles with charges Z\e.
e.z). (11. z = rcos<9..M) = l + .7 The Continuous Spectrum of the Coulomb Potential Then Eq.138) becomes 235 ^A + k2 . „ az 1. (11. b. (11.140) The simple Coulomb potential permits particularly simple solutions... Oberhettinger [181]. (11.b. According to Eqs.140) and using (11. .146) See.142) This equation is of the form of a hypergeometric equation. 86 .136b). v = ik(r .. pp.z) + W2(a.+ ^  + .b.^ W ) = 0. where ^. i.108) and (11. writing Vv = Aeik^v)/2f(ri) we obtain the equation d2 x d f(v) = 0. W.144) *(a.e.109) we can therefore write the solution ipr = Aeikz$(i>y. that we encountered previously.87 under Kummer functions.z) = Wi(a.143) and v = ikrj. . A = const. a(a +1) z2 ] (11. with the asymptotic expansions (11. .b. what is the > behaviour of 3>(a. (11. for r — oo) • ipr = Aeikzf(r . In particular the equation possesses a regular solution of the form (also ip ~ exp(ikr). for instance.11. Magnus and F. (11. (11. Our question is now: How does ipr behave for r — oo. i.141) Inserting this expression into Eq. (11. z) for \z\ —> oo? We can find the appropriate formula in books on Special Functions:* $(a.z).145) U ' ' j " T(6a)1 J ^ 71=0 r(a)r(a6+l) n! ' (7T < axg(z) < vr).z)).
a) (z) na)"~ r(la) T(ba) n\ ' (11.236 n\Zzab CHAPTER ^T(n '£ fo 11.l.154) .150) ^(rz)]1"^ 1 + 0 T(«7) i.ik(rz))} =^ + ^d.e.^l*7 l + O (11.i 7 ) (11. We now define a quantity f(9) by t h e asymptotic relation lb O < ^(.147) (—7r < arg(z) < 7r) It follows t h a t for r — oo.151) exp[—ryln fc(r — z)] — exp[—ijIn = kr(I — cos 6)) exp[—ryln 2kr — ry In sin 2 (0/2)]. with 2. Without t h e logarithmic phase factors.(11.ik(rz)) where ^ ~ Aeikz——^—f't^ e ife(rz) + W2(n. A(j)11 A where \ " r(i + \i) 1 j(kz+"flnk('rz)) 7 f ( l + Z7) _ e i(fcr.e.l.151) with Eq. if tbr were „ikr ibr ex el"z + f{6) r (11.149) (11.7 ln*(rz)) k(r . = r cos 0.z) + la)T(n + b. The Coulomb Interaction W2(a.148) . > A = Aeikz[W1(i1.kz+l^ktrz)) C . J\") ci(kr~/\n2kr) (11. a comparison of Eq.153) The asymptotic solution represents t h e stationary scattering state (E = const. i.) of a particle with incident momentum hk in t h e direction of z for large distances away from t h e scattering centre. (11.152) Since z = r c o s # .b.152) determines the quantity f(9) as /(*) = 7 r(i + i7) exp k(lcos6)r(lij) 2sin2(f) ?7 In sin 2 7 2 exp 2 i a r g r ( l + ij) — ij In sin ( ^ 2&sin2(f) . (11.z)T(l . (11.
the Coulomb potential has an influence on the incoming wave even as far as the asymptotic domain.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.1 The Rutherford formula The above considerations permit us to calculate the differential cross section which is a measurable quantity. 2im. also called phase anomaly.4 Variation of the observed differential cross section with 0. In other words. We define as current density the quantity h J := 2imo [^(v^)v(v^n.7. 11. (11. we see that this expression is the exact analogue of a current density in classical considerations. As a consequence of the masslessness of the photon. z — — oo.Q hk mo v0.. The problem of the logarithmic phase. The quantity /(#) 2 would then be a measure for the scattering in the direction 9. This cross section is defined by the ratio of the outgoing particle current (in direction 9) to the current of the ingoing particles.156) .n . For the incoming wave ipi = exp(ikz). that we encounter here is a characteristic of the Coulomb potential. the current density is * h . 11.11. the effective range of the Coulomb potential is infinite. do dQ K 12 K Fig. (11.155) Since the gradient is proportional to the configuration space representation of the momentum.
This implies that scattering takes place for the attractive as well as the repulsive potential. In Eq. The Coulomb Interaction In the case of the Coulomb potential we now ignore the logarithmic part of the phase (see discussion later). In reality this divergence does not occur. We recognize the result (11. We see that a depends only on the modulus of the potential (i. the square 7 2 ). (11.e. 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. which yields the scattering cross section. (11.152) as 1. We make a few more important observations. is called scattering amplitude. since in actual fact the Coulomb potential occurs only in a screened form.153): 4fc2 s i r  The expression /(#). This implies effectively that the Coulomb potential becomes (virtually) a screened potential or Yukawatype potential of the form V . 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. 'screened ZlZ<2e2 cr/r0 ° .152) yields as density of the outgoing current Jr = ^\f(0)\2vo(11.238 CHAPTER 11. i. Correspondingly the other part of (11.159) as the Rutherford formula which can also be derived purely classically. That this formula retains its validity here in quantum mechanics is something of a coincidence.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.154) we chose the prefactor of the incoming wave in (11. for 0 —> 0. thereby screening itself off from the surroundings. in the current this would in any case contribute only something of order 1/r.
11. of the equation r fi.t). i.4. _ x o _ v o ( t _ toh io)_ em0^VoKtt0)^o)ix (1L167) 0 l/> () ) fxXn t)~ gik(xxo) . and which provides the uncertainties arising in quantum mechanics.t) = e .t). which in the case of free motion (zero potential) are simply plane waves. whose maximum moves with the particle velocity. The particle velocity v is v = — . (11. 2mo V + F(x) k (11.*'^ e1 piEk{tto)/h (x x i)= (2^3^ ( k ) roT ^ k o l * . t 0 ).e. 11.i i f (''°>/Vk(x. t 0 ) .e.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. v0 = . (11163) (11. 2 (11162) where Vk(x) = ^ ( x . i. t 0 )^k(x.166) Let the incoming free wave packet at time t > t$. (k). t0) = e. i. the wave packet in the absence of a scattering potential. / dk 7^340(k)4(xxo.2 fr2 . m0 m0 and V.8 Scattering of a Wave Packet We saw earlier that in quantum mechanics a particle is described by a wave packet.x° ' r j . The wave packet is a superposition of waves ip(x. and if H is timeindependent one has the stationary wave function ^(x. 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.165) ^ ( x ) = £ fe Vk(x). be given by ^ f dk AL. (11.161) where Ak0(k) ~ A(k — ko) and ^ k (x.yJ ^ j A k o W 727)3/2 e 0. of continuum wave functions. t) = U(t.11.^ ^ % ( x ) .164) Ek~Eko + {kk0)v0h.k(x) is a solution of the continuum with energy h2k2 E = Ek = — >0.e.
k0) ~ k0 + and the expansion of E^ in Eq.10 we show that the solution V'k(x) of Eq. ip) could be carried along but we ignore it.xo. i. k = kez.t) = J ^ A k o ( k ) e . the stationary scattering wave.ip) is the scattering amplitude with respect to the incoming plane wave. Then with k = ^(ko + k . From Eqs. can be written A (X) = ( 2 ^ + W J ^ ^ ^ ( ^ ( x ' )  eikx 2mn f P»fexx' (11168) We saw earlier (cf.167) by replacing the plane wave exp(zkx) by this scattered wave. i.163) we obtain </>ko(x . (11.i) dk (2TT (11.166).161) and (11.ik .169) and (11.171) The phase of fk(0.154)) that for x — oo.169) The wave packet in the presence of the potential is obtained from the free form (11.e.<p) + Ol Akr l (11. ^k(xxo) ^k(x) ~ = e.170) A ko (k) e . (2TT)3/2 eik* + —fv(P.167) we have ^k0(xx0. (11.e.^ ( * . The Coulomb Interaction In Example 11. (11. With Eqs.240 CHAPTER 11.ko) k0 (11. (11.172) h h k0 {rv0(tt0)} (11.x °Vk(x).iB *(**°)/ R °'tj + 1 pikr e * .t o ) / V k ( x . this solution > has the following asymptotic behaviour in which f(6.165) ko • (k . (11.x + _ / k ( e j ¥ ) ) ikxo (2TT)3/2 zkxo = <>(* r y (2vr)3(27r)3/2e M ^ ) e *o)/^] e ikx 0 (11.173) . Eq.k 0 ) 2 ~ Jk20 + 2k0 • (k .x 0 ).
11. t) f. however. 11. . r ' ) by the equation (A + fc2)G(r. (11. and determine the Green's function G(r. 11. U(T) = ^ V ( r ) .171). Adding a suitably normalized solution of the free equation.x 0 .10: Schrodinger equation as integral equation Demonstrate the conversion of the Schrodinger equation into an integral equation.154).8 Scattering of a Wave Packet so that with Eqs. t) .r')l/(r>fc(r').(«)/ ipko(x .fco(tto)/ft]^0o)(x/xo. (11. *. r ' ) = 47r<5(r .174) r /k0(^. Here ^ (x' — xo.r')V(rW).^e (2^3(27r)3/2e y (27r)3(27r)3/2e (11.e.*0ko (x . '. A particular solution of the Schrodinger equation is — J G(r.r'). (A + fc2)t/.167). E x a m p l e 11. the solution can be written iMr) : (2TT)3/2 ^dr'G(r.z Fig. (11. r 241 (0 LO) f ^ Afc o(k)rikxnri[fcor£fcoato)/ftl i(kko)x' = /k0^.to).(r) 2rrao V(r)V(r) 7/>(r) (2TT)3/2 ^dr'G(r. T h e scattering amplitude is t h e same as t h a t given by Eq.y)eiko'Xoe'[fcor£.5.^y /k0^.5 Scattering of a wave packet with scattering centre O. i. Solution: We define the Green's function G ( r .r ' ) .x 0 .to) has t h e form of t h e wave packet in radial direction as indicated in Fig.r')U(r')ip{r')dr' /4T7 (as we saw earlier) . with the energy of the maximum momentum of t h e wave packet.
which is such that the contribution along the infinite semicircle vanishes./2 dk'd(—cos 6)dip iki. e > 0 small.242 CHAPTER 11. These are the socalled outgoing and ingoing Green's functions. 11. . Hence we write G(r)G+(r).fc ) oo dfc' 1 d nik r 7rr dr / . Imk . we relate it to a properly defined contour integral with the two simple poles at fc' = ±Vfc 2 + ie = ±(fc + ie/2k).O O ™ fc2 (e*fc'r _ e ifc / r 1 fc ) =  / fc'dfc' i(fc' 2 fc 2 ) Since this integral does not exist. displaced slightly away from the real axis as indicated in Fig.fc2 ' 2 ^ i fc72^"fc2 Integrating out the angles we obtain 1 G roo />7r 2n fc' rZn j.6. one obtains another Green's function given by G_(r)= r eikr. The Coulomb Interaction We find a set of Green's functions G(r. Thus we consider the contour integral taken around the contour C+. G+(r) 1 d dk' lim e>0 TTT dr Jc+ fc'2 — (fc2 ~ik'r 1 d 2ni nr dr \k' + k+ ie/2kj ik r J__d_ 7rr dr 2ni N ^ residues k.e \ r dr\ k ikr 1 ikr 0 With the choice of an analogous contour C— in the lower half plane.Rek' kfc/2K Fig. « = 52 / J _ Z100 7rr 7o / A fc'dfc' 2 / 2 • fc2 (fc' *(fc' .6 The contour C+ in the upper half of the k'p\ane. r ' ) = G ( r — r ' ) with the Fourier transforms G(r) = J dk' 3 (k')e ik ' r . 11.=k+ie/2k 1 d f. dk' ik'*r 2TT2 fc'2 . 9{k') = 1 1 S(T) = j G(r) 1 ^ A j dk'e i k ' r .
21 + 2.11. under the stated conditions 2/07 = Z\Zie2 = —Mo. z) = W^a. .145)) F(a.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. 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.179) = = elkr{kr)l+1F(l e (kr) ikr l+1 + l + ij. e. z). Then yW (11.175) yl' + k Here we set (as in the case of the hydrogen atom) yi = eikr{kr)l+lUi).z2 27fc 1(1 + 1)' yi = 0. Then £^<fc + (2l + 2 £ = ~2ikr. b. (11. Eq. 2ikr)] [Wi{l + 1+11. (11.e.9 Scattering Phase and Partial Waves 243 11. (r) (11177) Similar to above we obtain a regular solution y) ' with 0j = F(Z + l + i7.2Z + 2. (11. Eq. b.e. i. If we proceed from the radial Schrodinger equation.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.g.0 + W2(l + 1 + iy. (11. 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. whose continuation to infinity is again given by (cf.2l + 2. i. we define first the scattering phase for the case of the Coulomb problem proceeding from our previous considerations.£). Before we introduce the general form of the socalled partial wave expansion.176) " 0—A ~ (I + 1 + iy)<t>i = 0. where Mo is a parameter in Chapter 16.21 + 2. (11.140) (previously we solved this only for the bound state problem) § . b.178) (11. z) + W2(a.
180) ^ ' (11. (11.i 7 ) I T 2l l—X—i^j ( + 2) 2ikr( ?.ry) = e"^r(Z + 1 + ry).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 . (2kryr{i" V 7 .147) this becomes asymptotically (r) Vi lkr eikr(u„\l+l (kr T(2Z + 2) '—(2ikr) \T{1 + 1 .181) so that eldlT{l + 1 . Then e2i5< = Reie Re~ J2i9 9 = Si.i 7 ) „ikr—ij\n2kr T(2l + 2) e e* jTv(lliy)/2 2l+i —ikr r(z +1 . +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. The Coulomb Interaction Using Eqs.l e*"" + T(l + 1 + H ) T(l + 1 .244 CHAPTER 11.n) • e „—ifer+i7ln2fcr _eMr(i7Jl)/2 r(z + I +17) T(2Z + 2) /2 • ikriy In 2fcri7r(J+l)/2 r(z + 1 .' .«7) „ . Then (r) r—»oo Vi 7T7/2 T(2Z + 2)e*7'*e i8i 2'+1I\Z + l + i 7 ) i5.ii)' (11.146) and (11.j syyll iy il( _ . T(2l + 2) (2fcr)*T« (Alliy ikr .183) r(Z + 1 + 17) := Reie. .
e.e. Squaring this expression we obtain the wellknown formula for the binding energy of bound states (as a comparison with Eq. (11..187) contains almost the entire physical information with regard to the Coulomb potential. considering I as a function of E extended into the complex plane. For instance the poles of S are given by Z + l + z7 = . indicates that — looked at analytically — the scattering amplitude f(6) possesses at E — 0 a branch point of the square root type.u\ respectively with the following asymptotic behaviour: u(±) r^o e±i(fcr_7ln2fcr. (11.11. Analogously one defines as nonregular spherical Coulomb function or Jost solution the outgoing or incoming wave u\ . However. where the continuum of the spectrum starts. . i. a positive integer..9 Scattering Phase and Partial Waves Hence 5j = argr(Z + l + i7). 245 (11.139) n' = 0. (11. i. One defines as regular Coulomb wave Fi(j. kr) the regular solution with the following asymptotic behaviour: r 0 (11. That the energy E appears in Eq. ._W2)_ ( 1 L l g 6 ) The factor e2lSl between the Jost solutions (in the regular solution continued to infinity).188) in the form y/E.2..184) The phase 6i = 8i — lir/2 is called Coulomb scattering phase.n ' . with Eq.e.114a) reveals).e. i. i. In the present quantum mechanical considerations I is.1. g = e«* = rSi + 1+ *V*ri (11. The expression e 2i5j _ 2i5i—iirl is called Smatrix element or scattering matrix element.185) J F} ^° sm(AT7ln2£T + ^ ) (also called regular spherical Coulomb function). considering I —• an(E). as we saw.
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.7 A Regge trajectory of the Coulomb potential. The Coulomb Interaction and plotting these for different values of n.191) . we obtain trajectories which are known as Regge trajectories. 11. For the Coulomb potential a typical such trajectory is indicated in Fig. lman(E) jump from E<0 toE>0 o A LU infinity  E<0 Rea n (E) Fig.246 CHAPTER 11." These Regge trajectories can be shown to determine the asymptotic high energy behaviour of scattering amplitudes and hence cross sections. 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. (11.189) where as before the incoming wave in the direction of z is ikz ipin = e (11.190) and the scattering solution „ikr Vwt = / ( 0 ) — + 0 V. In these cases. (11. we obtain the partial wave expansion of the scattering amplitude as follows. and thus in general. 11.7. We mentioned above that the logarithmic phase does not arise in the case of shortrange potentials.
. 21 .k)=me2iS^k\ in which r\i^\ indicates absorption.192) Thus the incoming and outgoing spherical waves have a definite relative phase. (11. (11.195) with the scattering matrix element S(l.^(cosfl) (1L194) This therefore yields the following partial wave expansion of the scattering amplitude = ^lY. (11.193) where r\x e2l5t determines the change in amplitude and phase of the outgoing spherical wave. The mathematical expression can be found in Tables of Special Functions (there cf. Magnus and F. The incoming plane wave can be reexpressed as a superposition of incoming and outgoing spherical waves with appropriate components of / and with a definite relative phase. pp.11.g.22. W. Oberhettinger [181].9 Scattering Phase and Partial Waves 247 For a spherically symmetric potential the amplitude / depends only on 9. r e. There the expansion is given in terms of Bessel functions Jv which have to be reexpressed 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 .196) "See e.i * r 1 v^ f ne2ibl — 1 \ = k^2l W +1 \^r^)Pl{cosd)—.~ J > Z + l)Lil*e~ikr .i f c r . It follows therefore that pikr ^scatt ^ f{0) = Aegeikz = ~ ^(21 + l)Lilneikr  rHe2iS<eikr\pl(cos9) eikr\pi(coa9) eikr " 2 ^ E ( 2 Z + l){e i ' . The potential disturbs this phase relationship (by delay or absorption or redirection). 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.eikr\P^cos9).^2l + l^S^k) lik (11.r ^ V ^ Wostf).
This page is intentionally left blank .
g. one of the main and now wellknown methods to perform such calculations. which in one form or another. In particular we consider cases which illustrate the occurrence of tunneling and that of resonances.* *See e.e. i. For various general aspects and applications of tunneling we refer to other literature.e. Thus in this chapter we consider traditional "square well" potentials in onedimensional contexts. in texts on quantum mechanics is rare. the computation of finite lifetimes of a quantum mechanical state.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. which a particle in classical mechanics would never be able to intrude. It is not surprising. that the explicit derivation of such quantities in nontrivial contexts. i. In fact.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. 249 . in general. that quantum mechanically a particle may have a small probability of being in a spatial domain. Razavy [236]. M. These newer methods will be presented in detail in later chapters. Similarly difficult is. can be found in any traditional text on quantum mechanics. Such a vital quantum mechanical phenomenon is socalled tunneling. It is very instructive. however. namely the instanton (or more generally pseudoparticle) method. was developed only in the last two to three decades. therefore. states of a finite lifetime. as — for instance — for trigonometric or double well potentials.
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. We assumed this already in the preceding since we interpreted the normalization integral / Jail space dx\ip(x. Correspondingly one interprets as probability current density the expression j x ( < ' >=iH v ^7 v <4 = [W{x. M^ Then dt + Vj 1=0.t)\2 =1 (12. t) was developed by Born in 1926. Quantum Mechanical Tunneling 12.t).e.1) as the probability described by ijj(x.5) L j • dF IFo is the probability that per unit time a particle passes through the surface FQ. i. (12.2) we have = o + ~v[(vv>*)V'V>*(W)] = vj.. . According to Eq.t) = HiP(x.t) that the one particle concerned is somewhere in space at time t.t)\* (i2 2)  With the help of the Schrodinger equation ihiP(x. so that p = T/>2 is the probability density for a spatial measurement at time t. r (12.250 CHAPTER 12.t) (12. ihr(x.2 Continuity Equation and Conditions The statistical interpretation of the Schrodinger wave function ^(x.4) (12.
If there is no absorption. Gosdzinsky and R.33. we can do this also here. From this follows that both p and j have to be finite and continuous everywhere. 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. However. It is therefore sensible to demand that V( x ) a r i d V'0(x) be continuous. It is therefore quantum mechanically possible for a "particle" to pass through a classically forbidden region. 2m E 2 V0 > 0. An important consequence of the wavelike 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. as shown by P. Schiff [243].2a6(x)]il> = 0.12. although in general small. higher dimensional delta potentials do not have properties like that in one dimension. finite and singlevalued at every point x in order to insure that V( x ) i s a unique representative of the state of a system. 12. which is the case we consider here. In much the same way that we define there reflection and transmission coefficients. the equation can be integrated at every point. where „ mpVo V(x) =+V0S(x). the problem is analogous to that of the scattering process from a potential in a onedimensional case. Tarrach [119]. 'Maybe the reader dislikes being confronted again with the onedimensional case. Delta potentials in more than one dimension do not permit bound states and scattering. pp. since this potential implies a discontinuity of the derivative at its singularity (as will be seen below).3 The ShortRange Delta Potential It is very instructive to consider in detail a potential V(x) = VQ6(X) with the "shape" of a onedimensional delta function. Such a region is for instance the domain where V is larger than the total energy E. This case plays a special role. This effect is known as "tunnelincp. Thus it is possible that the probability for the particle to be in such a domain is nonzero. 29 . with regularization their study is very instructive for illustrating basic concepts of modern quantum field theory. ~hr> k = n^ (12 7)  . t For the singular delta function potential a modification of these demands is necessary. 1 0 ~x o a= f See L. (12. even for discontinuous potentials.6) . as we mentioned above. Nonetheless. The Schrodinger equation for this case is* ip" + [A. If ^>(x) and V^>(x) are known at every point x.2 .3 The ShortRange Delta Potential 251 The timeindependent Schrodinger equation is a second order differential equation in x.
252 CHAPTER 12. except in domains of the potential V.2 a J 5{x)4>(x)dx = 0.e.= 2a^(0). (12.ik(C . Consider the case of a purely outgoing wave in the region x > 0 (no reflection back from infinity). i.(ik . (12. The continuity conditions at x = 0 are therefore for V : and according to Eq. 2 (12.e.11a) ikC . (12.e 1 + B = C.: (i2iQ) Ce for x > 0.14) One defines respectively as reflection and transmission coefficients R and T the squares of the respective amplitudes.8) (12. . a 2 — a 2 a — ik a2 + k2 ' . e Q (12.e. _ fQr x < Q^ eikx _ geikx ^=\ ~ Blkx *x . that the derivative of i/> is discontinuous there: W\U + k2 J i>dx .9) States with k > 0 correspond to those of free particles that move with constant velocity v = hk/mo.9) for ij.ikB) = 2aC or 2a(l + B). (12.11b) 2aC. [<//]0+ . a — ik £Qr < { eikx a eikx •t" . i.12a) (12.13) ~A R= for x > °(12.' : It follows that ik .1) .[^']o.ikC = i. Quantum Mechanical Tunneling Integrating the equation over a small interval of x around 0 we see. 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.12b) a=—^— C = a + ik and hence B = C1 Thus for i/> we obtain = ?—. i.
( 1 2 .17) da#(s) 2 = 1.K 2 < 0.a l s l. . i. We write the solution ^ so that ^(0+) = ^ ( 0 .e. possesses simple poles at a ± ik = 0. (12. z<0. Proissart [223] or other books on scattering theory like R.16) The corresponding wave function proportional to etkx for x < 0 associated with the pole for k = — ia is^ rl){x) = V^e~aW.(K)C = . See e. the quantity T. . f Ce~KX c.i 8 ) . K = a. Thus ^ = Ce. poo nc s~t2 / oo dxe dxC2e~2aW = 2a\x\ _ u 2C2 / JO a Hence C = y/a and therefore ^ = VaeQK (12. Newton [219].2 a C . R. when normalized to 1. Omnes and M. as coefficient of the outgoing wave.19) Thus the delta function potential supports exactly one bound state.g. 'Note that the physical pole has negative imaginary part.K C . and C follows from the normalization: OO A.2 o ^ ( 0 ) . oo (12.12. or h2 E = —a2. 2m 0 (12.) and V>'(0+) .e «x = J \ CeKX for for x > 0.3 The ShortRange Delta Potential and T= ^rr a — ik 2 253 =1R.e. at k2 = —a2.15) Furthermore we see that similar to the Smatrix which we encountered in the case of the Coulomb potential. i. = . / oo In the bound state problem the Schrodinger equation is ip" + [k2 + 2a6(x)]ip = 0.2 = .ip'(0) = .
—a].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. p. x€[a. Quantum Mechanical Tunneling 12. p. (12. or in the lectures of E. Fermi [92].1 The square well potential.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. 55. The Schrodinger equation is iP"(x) + 2 ^[EV(x)}iP(x)=0.20) for for \x\ > a. 12.1. . We set 2m0(E + VQ) h2 so that ip" +fc2V>= 0 ip" + K il) = 0 2 (12.22) + De Feikx + Ge~ikx *We follow here largely F.a]. oo]. \x\ < a.254 CHAPTER 12. The latter contains also specific applications in nuclear physics. x G [a. (12. We want to consider the case of stationary states for E > 0. 58. 12. V(x) V=E a v=y Fig. Similar treatments of square well potentials or barriers can be found in most books on quantum mechanics. Schwabl [246].
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.12.24) yields F ^ G. (12.27) v ' Similarly Eq.23) Ce~iKa + DeiKa iKCe~iKa + DiKeiKa = = Fe ifca + Ge"*fca. ( V C n vD ).25) ^\=M{a)lCD where 1 / f l — « \ irea+ifca /i . . ) A l ( pika p—ika pina K ina kC p—ina \ _p—ika J \ Kp—ina J ( c\ U/ (12.ikBeika = inCeiKa + DiKe~iKa. I (12. ). M(a) . (12. Fikeika .ikGe~ika.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 Aeika + Beika = CeiKa + Df.26) K^gifoa—i/ca fc.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'Jgita+Jfca Q _i_ K\p~iKaika \ .iKa ? ikAe~ika and ..f )e" t\ K\0—tKa—ika . fc Af(a) = . 2 l ^i i K\ iKaika O \ (1 + %)e ina—ika (1 . (12. (12. (12.e.
4T(£)e ifci. (12.33) One defines as transmission amplitude the ratio of outgoing to incoming amplitudes.—2ika [cos 2/ta + \e sin 2«a]e~ (12.31) we have A AT(a) = )MFo B ^r)sin2na . x € [a.a .a ] . xe[a.34) into Eq. \R\2 = T(E)^sm2Ka (12. i. (12.a].35) with with with x G [oo.e sin 2na IJlika e B = r)F sm2na. (12.22) we obtain ' Aeikx + AT(E) I sin 2Kae~ikx T/>(X) (12.i^ 2 sin2 2KO 1 + ry?7* sin2 2/ta Inserting Eq.^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.256 CHAPTER 12.37) . Quantum Mechanical Tunneling Hence with the complex quantities IK (12. the quantity T{E) := £ = A lika cos 2/ta — je sin 2/«x (12.36) One defines as reflection coefficient \R\2 the modulus squared of the reflected amplitude divided by the amplitude of the incoming wave.e.34) and as transmission coefficient (note e is complex!) \T(E)\2 = 1 . i.32) [cos2«. .(1 + e 2 ) sin2 2na 1 . = < Ce~lKX + DeiKX .e. (12.oo]. we obtain A = F( cos 2na — .
formula 406.e. Li ( — H— I sin 2na = 0.38) as expected. which is satisfied if either / = g or / = — g'1. 2\k K. they have solutions.39) 2 i. B. in the domain of A imaginary and K real.40) we obtain cot(na) — tan(/«j) — i K IK . (12. 2m0\E\ h? ' (12. However.e.42) These equations have no solution for K.e. (12. (12. (12. i.43) Thus we set k — —i ' H .35) and using Eq.44) . .\ sin2 2KQ •n 4 sin2KaT(E) — \K\ >  IRI2 (12. Dwight [81].41) i. 83. (12. and k real. i. nor when both parameters are purely imaginary. an equation of the type f~1f = g19. p.4 Scattering from a Potential Well On the other hand according to Eq.12.31): 1\T(E)\2 257 1 .12.e. We see in particular from Eq.34) that the amplitude F = AT{E) of the outgoing wave has a simple pole where cos 2KCL e sin 2KCL = 0. for VQ < E < 0. cot(fca) = i— or tan(fta) = K K . (12. (12. where cos 2/ca With the relation§ cot(2«a) = [cot(«a) — tan(/ta)]. (12.
(12.258 and obtain the equations CHAPTER 12. At their positions cos(2/«z)£. T(E)e2ika = — r . from Eq. i.46). (12. (12. ± l . 64 . .l .e. "Note that dtan(2K<z)/'d(2na) evaluated at ER is 1. 2m0\E\ Ktan(«a) = — y——^— . and is itself infinite there.35)) for E > 0 sin2Ka = 0. The vanishing of A implies that because k is imaginary the factor exp(ikx) does not diverge exponentially for x — . 2m0\E\ and Kcot(Ka) = y—^—L. (12. (12.c o .49) Taylor expansion of the denominator in powers of (E — ER) yields . (12. Eq. Quantum Mechanical Tunneling .46) This condition implies the eigenvalues E^ER = n2^^ V0>0.l ) n . F.+ \tan(2Ka) where T is given by" 2 1 (K 2\k 1 (K k\ /n . Also. i.e. t2 2 (12. A = 0. This can be understood. t&n(2Ka)ER = 0.47) These states in the continuum are called resonances. The remaining wave part becomes exponentially > decreasing. = 2/rn (EER). We now return to the scattering states. The amplitude of the transmitted wave assumes its maximum value where (cf. (12. around the values given by Eq. k\ d(2Ka) K) dE \ E R 1v/2^a 2 h 2ER + V0 ^ER~(ER + V0) r 1 (12. i.65.' We observe that T(E) has poles at those values of E which supply binding energies. (12. . „ .45) These are the equations that one obtains from the connection relations for the case of Eq.20). 2Ka = nn. Schwabl [246].50) See e.34).fl = ( . pp. Then according to Eq. ± 2 . .g. . The infinity of the amplitude of the outgoing wave corresponds to a zero of the amplitude of the incoming wave.e.48) We expand T{E) around these energy values. (12. n = 0 .42) which then leads to bound states with associated even and odd wave functions.
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.' ''' 5a 3a \ a ~~~. T(E) has poles at E= ERi: (12. These are precisely the resonance states.2. a 3 a. 12. This result is physically very plausible.^ l / .52) Here ER is the energy of the resonance and 2/F the lifetime of the state. 2a<x<2a.5 Degenerate Potentials and Tunneling We obtain therefore T(E)e2ika = = (^ 259 iT/2 EER + iT/2' (12.51) i. V(x) = 9(x + a)VQ6(ax).12. = 0. The Schrodinger equation to be solved is ^"{x) + ^(EV(x))i. s \ V0 /• ~\ 'X . 12.e. V(x) x=2a x= =2a . We restrict ourselves to the period — la < x < 2a. 12.2 The PenneyKronig potential. . . We consider a potential consisting of a chain of rectangular barriers as depicted in Fig. A potential of this type is known as a PenneyKronig potential. Since we have reflection as well as transmission there are states with repeated reflection and finally transmission.' ' 5a \V / Fig.
2 = W h 2 _„2 = 2mo(E . i. VHV1 Hip(x) = Eil){x): VHV~lV^){x) = EVtl>{x). .e There is the additional boundary condition (for Vo = oo) ^ ( .1. = EVip(x). i. and we choose (12. x + 3a <> ™ since the particle cannot penetrate into the infinitely high wall.2 < a . E^En = ^ ^ t n = 0.. This boundary condition implies the quantization of the energy with ^ ( 2 a ) = n7r. We set . and hence for = Eip(x).53) *w«w^( « j)'*^{i i ^ .. Quantum Mechanical Tunneling (a) We consider first the case of VQ = oo.. 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.e. HV^(x) = H.e. (12. In view of the periodicity we have ^(x) = ^(x + 4a).Vo) n.55) The Hamilton operator of the problem.2. the differential operator is even in x..a ) = 0 = ^(a).. with V as parity operator.56) Thus the problem has a twofold 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. . Hip(x) so that with Hi/)(x) — Eip(x): H[^{x) ± t/>(x)] = E[^{x) ± xl>(x)\. (12.260 CHAPTER 12.
V'(O) = const. 12. We therefore have different boundary conditions. V+(0) = const. — a].2a. We can write the solution as consisting of the following pieces: { Asink(x + 3a) BeKX + Ce~KX with with i £ [ . ±Asink(3a — x) with x€[a. We > obtain even and odd functions ip± by setting ^ = ±1.62) = nBe~Ka T nBeKa = ^n[BeKa q= B e " M ] . At x = ± a we now have to connect ip and ip' of neighbouring domains.12. so that (W meaning Wronskian) W[tl>+. we shall not exploit here.2a]. xe[a. BeKX±Be~KX with xe[a. that these are such that for VQ — oo they approach those of the previous case. = Ka ±Asin(k2a). however.2a].2). The central region of the solution is the classically forbidden domain. This means that i>±(x) = <  = ±1.58) D sin k(3a — x) with xE[a. (12. At x ~ — a the connecting relations are Asin(k2a) kAcos{k2a) and at x = a: BeKa±Be~Ka K.. VV(°) = °. We expect. (12.61) = Be~Ka ± BeKa = ±[BeKa ± Be'Ka]. This wave function is no longer restricted to the interior of a box.a]. (12.il>]?0. (12.Be Ka (12. of course. which.5 Degenerate Potentials and Tunneling 261 We restrict ourselves again to the period — 2a < x < 2a.a].60) The boundary conditions that ip± have to satisfy.59) Asink(x + 3a) with x & [~2a. — a].. (12. are ^_(0) = 0.63) = KBe~ F = . 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. ^kAcos{k2a).
in the present case it is easier to continue with the above equations. + 0 ( e .2 e . we obtain k v ' eKa^eKa  tanh(«a) v ' These are transcendental equations for the determination of the eigenvalues of the even and odd eigenfunctions. It is plausible therefore to use appropriate expansions. and is — as we see — less than that belonging to the lower sign case.nir) + OU2ak . (12.+ • • • 1! cos^(n7r) 2ak .Thus we observe a splitting of the asymptotically degenerate eigenvalues: kesympt.65) and we expand tan(2afc) about nir. i. Taking the ratio of the equations in both cases. say.— (1 ± 2e~2A + O (—). the odd solution with. (12. i.e.67) The value of k belonging to the upper sign.e.66) Inserting these expansions into Eq. » we expect the eigenvalues to approach asymptotically those of the case Vo = oo.nir)2}} = 1 ± 2e~2Ka. for VQ — oo. i. we set tanhx = l . For large values of Vb.68) .nir + 0[{2ak .262 CHAPTER 12.. However. say A. 12. F ^odd (12.e. is the one belonging to the even solution. In general one now rewrites the equations in matrix form.e. Quantum Mechanical Tunneling Fig. i.3 Level degeneracy removed by tunneling.even. the limiting value for K = oo: tan(2a/c) = = tan(n7r) \ (2ak — nir) 1 — ^—. k = fc0dd. (12.mr)2}. (12. = AA.64) we obtain ~[(2ak k . where Akoce'2™. (with reinsertion of nir/2a for k) 2ak ~ nir .4 x ) and cothx = 1 + 2e~2x + 0 ( e " 4 x ) .2 a .
Classically the car is unable to overcome the bump. w(x0) •• = = mogy(x). ^v + ^[E.69) where the potential V(x) is (with g the acceleration due to gravity) V(x) y'(xo) Setting w(x) = 7r**E. Fermi [92]. but quantum mechanically there is a finite probability for this to succeed.3. 12. 57.4.. Example 12. 0.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.4 A car meeting a bump in the road. Solution: The wave functions i/>o>Vv in regions V = 0. . 12. 12. y(x) = sin and y(xo) = 1.5 Degenerate Potentials and Tunneling 263 Thus the effect of tunneling — here an exponentially small contribution — removes the degeneracy of the asymptotically degenerate states. Compute this probability. .V{x)\i. a sinusoidal bump in the road with a height of one meter at its peak as indicated in Fig. . p.** " W ~^7~ Fig. and is a reformulated version of a problem set in lectures of Fermi. b > a. 2m0E „ 2mnE .v = 0. exp r rb rb / •J a 2mo (V(x) . 2mn. The following example is an attempt to transcribe the quantum mechanical effect into a macroscopic situation..e. as indicated in Fig. V ^ 0 are given by the equations ^o + " J .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. 1pV 2ro 0 4>o '. i. exp .2 dxy/V(x) (12. (12.^ 0 = °' Hence in dominant order •00 — exp .70) xo = (a + b) 2' .
Emde [143]. which give /•TT/2 dx sin'M . 369 and 8. Jahnke and F.72) Assuming a mass of 1 ton of the car (i. We have 8m 0 : exp ^(ba)/4 x 2_1 ^^W(s x °' 9191 (12. : 2 ^ B ( ^ . and assuming a length of b — a = 100 m of the base of the bump. p.2 .9191 ~ 1.**.71) The integral can be looked up in Tables of Integrals. We can now evaluate the probability. h = 1. we obtain ( . (15. . S. p.384. Quantum Mechanical Tunneling = y/mog 2(1) — a) /""'' / dw sin1/2 n Jo w.2. 1000 kg) and using the following values of the natural constants: g = 9.1. Ryzhik [122]. M.y) is the beta function (see Eq.264 we have ro / dxyfV(x) Ja CHAPTER 12. 14.055 X 1 0 _ 3 4 J s . (12. n I .807 m s . p.1. Gradshteyn and I. **For instance E. ^ ) = 2 ^ . dx sin ' x •• Looking up the value of the factorial in Tables.1 / 2 ) ! = v^r) TT/2 /.4 x 10 3 9 ]. formulas 3. We find™ (with ( .17)). 950.1 / 4 ) ! = (4/3)(3/4)! = (4/3) x 0.l .2 [fr/2 ~ I)'] (Ml)! 2 ' where B(x.e.621. one evaluates an approximate probability of exp[6.
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.1 Introductory Remarks We distinguish here between three different types of linear potentials. 13. Thus. Then considering the probability distribution determined by the squared modulus of the wave function. in Chapter 15.3. 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. We shall see that in this particular case the wave function can be written as a superposition of de Broglie waves.* In Chapter 14. we consider the linear potential in three space dimensions that forbids the particle to escape. This is the problem of Galilei in quantum mechanics. whereas the first case permits only a continuous spectrum. 144. p. Siissmann [266]. Finally. the last case allows only a discrete spectrum. Example 14. 265 . in the book of G. we consider the corresponding case of a particle above the flat surface of the Earth. This is the quantum mechanical version of the problem of Galilei in one space dimension. each of which propagates with the classical momentum and energy of a Galileian particle. 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.Chapter 13 Linear Potentials 13. for instance.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. * A highly abridged treatment of this problem is given.
x > 0.t) = m0g with potential V(x.t) = c\x\ + V0. In the present case of quantum mechanics we have h2 d2 2mo dx2 m0gx ip(x.2) which is the case with a discrete spectrum. from d_ m x2 0 dx + V(x) 0. (13. (13.t).t). 13. (13. The present case is an exception and a good example of its applicability. We consider here the case of Eq. g > 0.= l V 27T Joo dxelkxip(x. Therefore we use the Fourier transforms il){x. Recall the classical treatment of the freely falling particle. (13.1) where g is the acceleration due to gravity and we can put Vo = 0.t) = mogx + V0.e.3) In general the momentum representation is of little importance.2.1 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 timeindependent homogeneous field of the gravitational force F(x.266 CHAPTER 13.1).= l V 27T J_oo 1 dkeikxijj{k. (13. i. 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.t) = . The equation of motion follows for instance from the derivative of the constant energy.t) = . and yields rriQX = mog. We see " that the case g < 0 can be related to the reflection x — —x.4b) .4a) f°° ${k.t). V(x) = —mogx. (13. c> 0. A different > "linear potential' is the socalled "confinement potentiaF V(x.
6) the latter equation becomes d ~( m0gt ^(^M<=+=?.t) m0gx /2TT m0gxip(x. (13.t) h2k2 d ~J(k.9) .t).t) dketkxiP(k.2 The Freely Falling Particle: Quantization 267 and consider the following partial integration in order to reexpress the last term in Eq.t)im0g^k.6) (and to avoid confusion we do not introduce new functions and variables): k > k + mogt (13.5) m+tm9Wcr{ktt)=2^^ktt)t (13.d m +imo9 .8) With this result and the substitution (13.7) In the wave function on the left hand side of Eq. d \ ~f1 dkr[ k + m0gt ^r't (13.3) yields us now by inserting for ip(x.6) the argument k is then replaced by this.7) applied to every quantity in Eq.4a) d ~ f I dk ikx dk 2vr (im0g) iP(k. To this end we consider the following substitution in Eq. Applying the total derivative d/dt to this function. (13.t). we obtain the differential operator on the left of Eq.< (13. Equation (13. We rewrite this equation in the following form th (13.13. (13.t) the expression (13. (13. • fd ~(' m0gt %h . (13.e.e. The vanishing of the boundary contributions is to be understood in the sense of distribution theory (i.3) as a partial derivative d/dk: d lKx dkeAkx{imQg)—4){k. 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.6) and wish to solve it. multiplication by a test function and subsequent integration).t) dk nikx k=oo '2K (im0g)il)(k. i.6).
10) k by m0gt back : k —> k h The solution of Eq.11) solves the partial differential equation (13. We observe t h a t t h e solution is actually a function of kt — k — mogt/h.6). (13.i) . the third term from the integrand) > / mog \ o exp dk h J h dt y ' 2mo k2i>(k. (13.11) *[k?f.6). (13.t ) = ipQ(k)exp Jo fdt' k+ m0gt' H (13.e.t ) ' U ( f c .t) = i>0[k 1— ) exp 2m0i J0 k+ ^(t't) (13.t In Example 13. Our next step is to reexpress the solution (13.t) ok lm og^— exp dk •ghf dt'fc+^(t't)W(fc. To obtain the solution of the original equation (13. we have to reverse the substitutions and replace in (13.6). i. Thus addition of both indeed leaves the expression on the right hand side of Eq.6) is then $(k.11). (13.12) . E x a m p l e 13.= k rmt. We define the wave number kt as kt. We have thus obtained the solution of the transformed equation. On the other hand the second operator expression implies dtj>o imog^—ip(k.1 we verify t h a t this solution indeed solves Eq.6) to the solution (13. the following result h 2moi tp[k\ ^—.1: Verification of momentum representation solution Verify by direct differentiation that the solution (13.6).10) Note the amplitude function ipo(k) in front. (13. t ) .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. Thus we have first (the second term arising from the upper integration limit.11) in a neater form.gh [ d t ' j f c + ^ ( t ' . Solution: We apply the two operator expressions on the left of Eq.t).
13) rn0_t+2 + mlg2t3 .2 The Freely Falling Particle: Quantization 269 Then the integral in the exponential of the solution (13.17) = J_/ C X ) h2 [It(k) ../„(*)] 2m^ig Inserting the explicit expression for the argument of the exponential.t) = 1 / dkipo ( k mogt 2 h ( 1m\ig \ With . 27T ^ . _ k t  h2 3 Inserting here the explicit expression for fct.gfci + 2 g2m0 2/i 2 ' (13. we have ip(x.= 1 / f°° dkelktx^{kut) dkipol k V —t 1 exp iktx K ) (13.15) we obtain / = .11).13.o o 2 —  exp iktx h mZgH' . (13. m0g (13. m0g ~7T' k Z k _ r ^ h t Y + i n Smog (13.o ut = 2mo —kf 2mo A.19a) . we obtain rp(x.t) = .2„2+2 mfaH 3^3+3 .16) Next we evaluate the Fourier transform of ip(k. (13. we can write the integral / = l h l h (hr + 7 7 (h 3m0g 3 m0g \ 1 H 3m0g With the definition « * ) • ! ( * = * .+^f (13.11) can be written j'A ^. Inserting ip(k.t) defined by Eq.} fA ' _ _tkt 2 k+ t) 2 k .I0(k)]. .— [It(k) . .14) (13.4a).t) into this integral and recalling the property (13.18) h =k mogt p h S' h .
t) =y= dkip0(kt)) exp i< fctx V27T J . 13. (13.t)\2 for t — oo (distant future or past). 7 u>t'dt' (13. ipo(k) is the probability amplitude with wave vector k at time t = 0 and varies in the course of time with the timedependent wave vector kt = k — mogt/h.v = kh/mo. To this end we convert Eq. Linear Potentials f dt'. (13.270 and CHAPTER 13.20) expresses the wave function tp(x. (13.t) as 1 I00 ip(x.gk— + 2h 3 2?7lo 2 2 h mQgk ^ _ mjfo2 2 i + ft2 fcr 2mf)g 03mg £3 .The associated kinetic energy of the particle is mv+ = huJf. Such a decomposition is possible only for homogeneous force fields.2.22) 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 .23) .2 Probability distribution at large times Next we inquire about the behaviour of the probability density \ip(x.21) In other words.18) into a • different form in order to be able to perform the integration with respect to k.t) as a decomposition into de Broglie waves. the relation (13.19b) h3 3 we can rewrite ip(x. The fcdependent terms in the argument of the exponential are T:=ikX^{^fk2+r^k 2rriQig [ n nr (13.20) (13.o o Jo This result can be interpreted as follows. These wave vectors vary in accordance with Eq. fm0gt\ \ 2h ) (13.ut> = Jo t2 g2m0 i 3 k t .19a) exactly as for a falling particle. 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.
2h h 24ift ' mogt^ r m.18) thus becomes ip(x. may be put in front of the integral./m 0 a. /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. ftf fti m0gt\ 2h J' " J' m0x ht mo5* 2ft + 2mo ~ftT f.t) = 1 1 oo 271 27T /. being now independent of £. (13. TOO#A ft and.13.t) exp i\ mogxt mox2 — h 2h 2ht mo<?2£3\] r (vn§x 24ft mogt\ (13. We evaluate the Fresnel integral in Eq. m0xcl ht m0gt 2h gt O v: °° 2hti (13.26) d£e ie £ kt^ m0' 0.24) We now set fit / 2m 0 V" 0 V m0a. Then ip(x. (13. (13.QX 2 ^ 0 7 (m. rnogt m0g2t3 tX^r~.QX ht V ht 2ft For i — oo the definition of £ defines a distance xc\ given by > k i.t) 1 exp r . integrate and then take the limit fi —> 0).28) (for a verification replace a by a + ifi. It follows that ip(x.e.2 The Freely Falling Particle: Quantization The integral (13.25) The amplitude ipoi^t) then becomes for i — oo: > 4>o(h) \t\—»oo 4>o[ k \ .29) .27) xd = vt Thus xc\ is the distance travelled by the particle as determined in classical mechanics./oo dkip0(kt) exp l x ixI exp oo . fi > 0.26) with the formula 1/2 (13.
this expression can be brought into the following form: if *\ 0 ip{x. V(x)=m 0 gx V=E Fig.3 Stationary States In this section we show that the Schrodinger equation with the potential (13. . (13.dA 4>o(h (13. 1/2  ±^(2mlg/h^x^ / 2i (13.33) oo for E <C \mogx\. 13.1) has only a continuous spectrum.e. xc\). (13. For stationary states we set xl){x.1 A turning point at x = —XQ.3) the timeindependent Schrodinger equation i^ + ^ [ E + mogx]<l>(x)=0.31) and obtain from Eq. These are incoming and outgoing waves.30) We see therefore: The large time asymptotic behaviour of the wave function is determined entirely by the classical values of x (i. Linear Potentials Replacing here x by the large time value x c i.t) = eiEtlhct>{x) (13.32) For x — oo the function cf){x) behaves like > 4>{x) ex exp exp dx { 2mo (E + mQgx) \ H2 . kt and ut 13.t) ~ W —exp i V irlt m ktxd J cot. For x exponentially increasing and decreasing solutions are possible.272 CHAPTER 13.
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. t) = Eil)[x. so that ih dtp{k.35) It follows that /W tm0g \ 2m0 i. t) = — = / V27T J .13. However quantum mechanically this is possible with a small. but also d 00 dt Eijj(k. probability.e. We determine the constant from the condition that the continuum solutions are to be orthonormalized to a delta function.34b) It therefore follows from Eq.3 The TimeIndependent 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. 13. (13. i.4b). Thus the spectrum does not contain states which are normalizable over the entire domain x e [—00.6) that also for 4>{k. and we can establish for the resulting continuum solutions the orthonormality and completeness relations. i. E)dk.36) m0g \ 6m0 with c = const.e. which is not possible for real particles. i. is reflected. The Fourier transform is given by the relations (13. at which E — V = 0.1. H2k3 4>{k) — cexp Ek (13.00].t). cf. exponentially decreasing.0 0 dkelkxE^(k. Akx (13. its momentum p would be imaginary there.e.34a) Then d &k 1 f°° ih—il>(x. t).e. and hence is continuous and extends from —00 to +00.t) = dt (13.t). The point — XQ.e. we .t) — (p(k) > E +^og~Mk) = 1 (h2k2 ^m. 1 f01 dketlcx^{k. at which the classical particle bounces back. i.e. Fig. i. J \ „ (13. In view of the simple form of the potential in the present case the equation can be integrated in the momentum or k representation.4a) and (13.
L [ dkeikxMk).41) J—oo Next we insert (13. (13. In a similar way we can verify the validity of the completeness relation.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 . i. (13.38) we have for <j)E{k): OO /"OO / oo (13.36) into (13.a : ' ) .274 have / Using d(x) = — one verifies t h a t with CHAPTER 13.36) (with c replaced by (13. V 2lT J Mk) = ..(k) = / J ~oo dx<t>E{x)<\>EI(x) = S(E  E') Inserting (13. / <>~ikx VZ7T J (13.37) (eikxdk.L dxeikx(f>E(x).38).40)) into (13.e. / dE(j)E(x)(t>E(x') = ^ s . apart from a constant phase which we choose to be zero.M *K Joo Setting dt cos ( ts n Jo st). (13. </>E(x) = . 13.e.43) y=2^x+2^E. Linear Potentials OO dx<ffE(x)<i>E>(x) = oo 8(EEf). so t h a t oo / <fiE(x) J dke ikx s/2nmog exp m0g \ 6m0 Ai(s). h? h2 .39) dk4>*E{k)4>E. the relations oo /*oo dE4>E(k)4>E(k') = 6(kk').£') = <*(#£')• 1 27rm0g' = /o . (1340) i.39) we obtain dkcc*exp m0g It follows t h a t cc = &(# ..
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^^eU*3*3) _^Pi(Sy~tz)ei(s3t3) S(y = j£j**w I ^LJt{yz) 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. In the next section we derive the important asymptotic expansions of Ai(s) for both positive and negative values of s. (13. and in Chapter 15 in the computation of energy eigenvalues for the threedimensional linear potential. 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= ^ . / s M . the function 4>E{X) i and setting s y = — = 1 .J * . and we shall see t h a t one branch has periodic behaviour.^ .. whereas the other is of exponential type.44) is then expressed in terms of the Airy function: .13.45) One can now derive normalization and completeness integrals for the Airy functions. For instance dxAi(y / • + x)Ai*(x + z) = 5(y — z). (13. . For a further solution Bi of Airy functions we refer in Chapter 14 to Tables.46) This follows since with Eq.^x + ^ E (13./ 2m 0 E (13.3 The TimeIndependent 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.
49) (13.r 3 cos 30 J + i ( sr sin 9 . (13. 13. Thus we must have sin30<O..TT < 30 < 0. Linear Potentials 13.4 The Saddle Point or Stationary Phase Method The integrand of Eq.276 CHAPTER 13.\zz.TT < 6 < 0 for r > oo. i. (13. (13. o (13.47) Consider the following integral along some as yet unspecified contour C in the plane of complex z: / := i .r 3 sin 30 ).43) contains the phase (with x — z): > *(*) := sz . or n <6 <TT o for r > oo.51a) and since with sin 30 < 0 also sin(30 + 2ir) < 0: TT < 36 + 27r < 0.50) Im z domain (b) / ' ^ ^ ~C domain (a) 0 ^ ^" ~ ^ \ x Zs=i^S Fig.2 The contour C for s < 0 with ends in the angular domains (a) and (b). .e. We want the integral to exist and so desire that lim e**^ = 0.e. .f dzJ*W 2TT JC with z = reie.48) We have $(z) = sre1 1 A S„3i6 r e sr cos 6 . (13. (13..51b) . i.
13. we choose zs = iy/^s. 0 = &(z8) = s~zt s = za „2„2i6>.57) $"(z s ) = 2% s.51b).45) exists for the contour C. where the derivative of the phase function vanishes.54) i ¥~s)3/2' (13.e.58) . It is reasonable to choose the contour in such a way that only a short piece of it contributes substantially to the integral. (13.56) 2n(s)^Jc But I daefr"?*"^.S )V2 (13. i. provided its ends at infinity satisfy these conditions. and the main contribution to the integral comes from the a region around the saddle point.2. i.54) where a is the new variable with —oo < a < oo.52) = exp(±i7r) Let s be real and s < 0. and we set with a real: z := zs + a (S)V4' (13.55) s < 0. (13. (13.48).( \ / ^ ) 3 2z. We now choose the path of integration C at fixed O z = —iy/^s with a e f [—oo. oo] as indicated in Fig. Then the ends of the contour lie in the domains specified by Eqs.z3s and V^s + . Inserting (13.51a) and (13. Hence _ 2 / s)3/2 /•oo 2TT( s)!/4 doe J—oo CT2 i*3 3(. at which the phase becomes stationary.53) Since we shall have the contour C in the lower half plane. so that exp(±2i9s) — 1 and — s — rl. Then Ze TeC : V^se±i7r/2 = ±i\fzrs. —i ${zs) = szs .55) into (13.4 The Method of Stationary Phase 277 The integral (13. we obtain da 2KJC (*)V4 c 1 e **(Zs) exp i{$>{Zs)+l(zZsf<$>"(Zs) + (13.13. (13. Then 6S = ±7r/2. (13.e. Such a piece can be found in the neighbourhood of the socalled saddle point zs.
59) In Example 13. S.44) and (13. as one can see from the following equation^ ri' + Ffz •2^2. (13. Linear Potentials where the contribution oc er3 is obtained from &"'(zs) whose evaluation we do not reproduce here.278 CHAPTER 13. . According to Eqs. V 2h2 (13.45) 4>E(X) ~ Ai(z). Gradshteyn and I. (13.32) with the solution (13. M. With the help of the substitutions (13. (13. For (—s) — oo the integral converges towards T/TC.61) 2m0E = f 2 \1/3r 2 2 2 fi h \h mog2 Equation (13.2/32u = 0 (13.42) into an equation with a more appealing form. 2 s 3/2 . This is an important aspect in the WKB method to be considered in Chapter 14. (13.32) therefore becomes dz2 + zd) = 0.62) This is the Airy differential equation with one solution 4>{z) = Ai(z). Sec.56) decreases exponentially around a — 0 for real values of a. We observe that the Airy function has a periodic behaviour in one direction but an exponential behaviour in the opposite direction. Ryzhik [122]. > so that Ai(s) 2 v ^ ( ._ 1 .63) The Airy function can be reexpressed in terms of cylinder or Bessel functions Zv(z).s ) V 4 exp \(s) 3/2 (13.2 we show that for s — oo: > Ai(s) 1 .64) tSee I. where — with e = (2/h2mog2)1'3 — we have /2^m3g3y/3 x\ = e[E + mogx}.44) we can convert Eq. cos . V^s / 4 V3 (13. 8. (13..60) The reason for describing the point zs as a saddle point is that the integral in Eq.491. but would increase exponentially for imaginary values of a — thus describing a surface around that point very analogous to that of a saddle.
4 The Method of Stationary Phase with solution u = z1/2Zi/2/j(7A 279 (1365) With expansions (13.^ e ^ / V2 4 [°° Jo dae'2^. kl dpe*' = ~e^'4 V2 f° Joo daeW2 and f°° dpe'^ JO = . (13. the conditions (13. Then._ . We have therefore r /*0 poo zs = ±y/s = z± / Jc dze^M = / Joo dze^^ + JO dze^z\ In these integrals we set respectively P with *(z) = #(z±) + i ( z .51a) and (13. . Hence the saddle points are given by 0 = * ' ( z s ) = s — z2.. Example 13.43)) Ai(s) = — [ dzeiit(z\ *(z): 1 Ai(s).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. P Introducing a new variable a by setting p = ai\a\. cos  . Thus there are two symmetrically placed real saddle points on the real axis. It follows that for s —> oo: Ai(s) ~ .51b) are satisfied.2: Periodic asymptotic behaviour of the Airy function Obtain with the saddle point method the asymptotic larges behaviour of the Airy function Solution: We have (cf. * " ( z ± ) = =F2Vi.13.s 3 / 2  .z ± ) 2 * " ( z ± ) . 27T Ic Jc 3 This time s is real and positive.59) and (13. Eq. changing to the variable p. with *"(z) = 2z. we obtain f° Joo dp=—[a+ i\a\]da. Choosing the contour C to lie along the real axis.
3 The saddle point at z = n + ix. n!= / e~zzndz= / e~z+nlnzdz. ~ 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. 13. z o = n . i.*. since f'(zo) = 0 and (z — ZQ)2 = —x2. Jo Jo \ .Be 2 Fig. Then. Solution: Briefly.e. Linear Potentials E x a m p l e 13. one obtains n! ~ where we used f^° dxexp(—w2x2/2) = V2^nn+1/2en. Thus here / ( z ) = — z + n l n z . . ^/2KJW1.3.111).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.dz = idx (observe that this implies an integration parallel to the axis of imaginary z through the saddle point). 13. the method of steepest descent evaluates the integral J^dz by expanding f(z) about its extremum at ZQ with z — ZQ = ix. and hence integrating as described above through the saddle point indicated in Fig. (11. / " ( z o ) = —n/z2 = —1/n.280 CHAPTER 13.
named after its first promulgators in quantum mechanics: Wentzel. The central issue of WKB solutions is their continuation in the sense of matched asymptotic expansions across a turning point (i. Kramers [153] and L.e. of misunderstandings and other aspects may be found in the monograph of Dingle. Dingle [70]. B.Chapter 14 Classical Limit and W K B Method 14. 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. for misunderstandings pp. A. Jeffreys [144]. i. 292 . ' T h e most prominent earlier expounder is H. A critical overview of the history of the method. ^ The central problem of the method is the topic of "connection formulas". the linkage of solutions above a turning point to those below. and the role played by h in our considerations here.* Descriptions of the method can be found in most books on quantum mechanics and in some monographs. Froman and P. N.317.295. where *G. For the history see pp.g. *R. H. § See e.1 Introductory Remarks One of the most successful methods of solving the Schrodinger equation is the WKB method. Wentzel [282]. It goes without saying that such comparisons are very instructive.e. Froman [99]. Brillouin [37]. The WKB result is therefore frequently described as the semiclassical approximation. 316 . will there be that of a coupling parameter. O.* The method had. been developed previously by others in different contexts.§ 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. 281 . Kramers and Brillouin. however.
(14.2 Classical Limit and Hydrodynamics Analogy In this section we consider the limit h —• 0. We start from the Schrodinger equation ih <9* =HV at with H=£— p2 2mo + V(T).V(r)]^(r) = 0. It is then possible to derive in a general form the relation known as BohrSommerfeldWilson quantization condition.t) = eiEt/h7P(r) and \V(r.7. For stationary states ^/. for which the observation of position and momentum is independent of the time at which the observation takes place. the function S being called phase function. V>(r) = A(r)eiSWn.g. so that _ Aip AA+^V• h AA+CVAn (AVS) + IVS h VS + AAS) iS(r)/h (14.1) (14. 14. In this equation we set. Classical Limit and WKB Method E — V changes sign). A(r) and S(r) being real functions of r. • {VA + %AVS [ h B iS(r)/ft +^{vSVA+^A(VS)2 *See e. It is for these states that one obtains the timeindependent Schrodinger equation A^(r) + ^ [ E . the derivation in Example 18.* We illustrate the use of this condition by application to several examples.t)\2 = \iP(r)\2.2) dip(r = I VA{v) + ^ ( r ) V 5 J eiS^lh.3) . the energy has a definite value E with V(r. We shall see that in this limit the > Schrodinger equation describes a steady stream of noninteracting Newtonian particles.282 CHAPTER 14.
10) with a continuity equation by defining as probability density p:=*** = [A(r.1) and separating real and imaginary parts.Q A 2i 8A 1 (14.t)}2. (14.5) Apart from these we also wish to consider the equations that follow in a similar way from the timedependent Schrodinger equation.7) h2 AA 2mo A 1 (VS)2 2mo = 0 (14. (14. We set in this equation.* VAVS+AAS + V(r)A. (14.10) m 0 ^ . V(r. (14. h h Taking real and imaginary parts of these equations we obtain the equations: + dt and 2rriQ 2m.2 Classical Limit and Analogy with Hydrodynamics 283 Inserting this into Eq. we obtain the following two equations: E .13) .t)eiS{r't)/h.t) = and obtain dA i_dS_A + dt h~dt 2 h r A(VS)2 AA h2 2mo ih A(r.V(Y) + and 2VA • VS + AAS = 0. (14. (14.4) (14.14.11) We define as probability current density the vector quantity J:=K where V* = ijf* V* rnio i_ „ *.12) VA A (14. h h (14. We can identify Eq.6) 2mo dt now with A also a function of t. h2 ih— = H$ A + V(r) * . These equations are equivalent to the Schrodinger equation.+ VAVS + AAS = 0.
(14.J = 0. %(1"1' I<2) and V J = — V • (A2VS) mo Prom Eq. (14.21) . (14.n. Next we investigate the significance of Eq.01.17) With Eqs.9) for the phase function S. ^ + V • J = 0.16) BA 1 moA—. TTIQ (14. . the implications of this equation are also those of the equation of continuity. (14.. (14. (14.+ AVA • VS + A2AS = 0.16) this becomes m 0 1 rr + m o V .19) m0 m0 it is suggestive to define the following vector quantities ± = at v := I p = JVS. m0 But now (14.18) ot ot This is the equation of continuity that we encountered already earlier. + h2 AA V = — ^ .14) A2VS. Classical Limit and WKB Method This means # * — V * = — AVA + — and so J = — A2VS. (14.20) With this we can rewrite Eq.15) and (14.10). ie. we have mo—(A2) + V(A2)VS + A2AS = 0. (14. Since J = — A2VS = PVS.10) we obtain = — [V(A2) • VS + A2AS}. Since we obtained this equation from Eq. mo <™> (14. (14.9) as 9S 1 mo^ 2 + T. ot 2 Multiplying this equation by 2 and recalling that the function A was introduced as being real.284 CHAPTER 14.
(14. d — = (14. we obtain ^ V 5 +  v ( v v ) + V V = 0. where. they are asymptotic expansions. or that of equilibrium in electrodynamics) and dS/dt = —E (from ^ oc exp(—iEt/K)). Such series are typical for expansions in the neighbourhood of an essential singularity. This result is very general.22) (14. the wave function \I/ describes effectively a fluid of particles of mass mo. 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. 'Recall the formula V ( u • v) = (v • V ) u + (u • V ) v + v X rotu + u X rotv. a singularity different from that of a pole). The equation of continuity then describes the stationary flow of a fluid.20). (14.23) (14. In the first • place we require the ansatz (14. This means. This is the analogy of the Schrodinger equation for h2 — 0 with hydrodynamics.25) dt We recognize this equation as the classical equation of motion of the particle of mass rriQ. with the help of Eq. that the quantum theory implies an essential singularity at H = 0 (i. > The motion of the particle or rather its probability is altogether given by the above equation of continuity.1/h. This expression with h in the denominator shows. r o t v = ( l / m o ) r o t g r a d S = 0. However. V • J = 0. and hence''' ^(m 0 v) + m 0 (v.e.24) d 1 dr d •— = d Lv _ . . i. 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. The particles have no interaction with each other. which for very small values of h yield the dominant contributions.2) for the probability amplitude. so that V ( v • v) = 2(v • V ) v + 2v x rotv.14.e.2 Classical Limit and Analogy with In the limit h2 — 0 this becomes > Hydrodynamics 285 ^ + W 2 + F = 0. dt 2 Constructing the gradient of this equation. > In the case of stationary states we have dp/dt = 0 (which is analogous to the condition of stationary currents. An expansion in ascending powers of h therefore starts with contributions of the order of 1/h2. 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.V ) v + V 7 = 0.
similar to our earlier procedure with A and S real functions.3. §We follow here partially A. 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. p. Kramers and Brillouin. 318. B.286 CHAPTER 14.§ y = A(x)eiS^'h where W(x) = S(x) + . z (14.$ 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.28) The calculations we performed at the beginning imply the following equations (compare with Eqs. or h2/2mo(E — V). i. As remarked at the beginning.In A(x). (14. such as "LiouvilleGreen method' and "phase integral method'.4) and (14. i = eiW(x^h.2.5)) and 2 ^ f dx ax + All=0. the method is widely familar under the abbreviations of the names of Wentzel. (14.30) ax *For details see R.3 14. 6. (14. h2 l 2m0 {EVy corresponds to an expansion in decreasing powers of (E — V). Classical Limit and WKB Method 14. Sec. Messiah [195]. There are also other more specific descriptions.e. .[EV(x)}y = 0. We consider the onedimensional timeindependent Schrodinger equation y" + ^. Dingle [70].27) (14. We set. The expansion in powers of h2.26) We observe already here that an expansion in rising powers of k.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. is an asymptotic expansion.
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. (14. (14.2rn0(E V) = h2 lS'A 2S'0 (14. (14. r//\ 4V^ 2 Up to and including contributions of the order of h2 Eq. + (14.33) 3 (S")2 . (14. Since the expression (14. . Eq.32) s" = sz + h2s'( + •••.e. Integrating this second equation we obtain ll[dX so that = \l^dX' A = c(S')" 1 / 2 .31) \a.4 (S>)2 1 S'" 2 S> (14.32) is therefore (S'0)2 + 2H2S'QS[ .V) and (14.V) = h2 In the WKB method one now puts S := S' = S0(x) + h2S1(x) + (h2)2S2(x) S'0 + h2S[ + ••• . (14.34) Equating coefficients of the same power of h we obtain (S'0)2 = 2m0{E . 3 / 5 .30) corresponds to the equation of continuity (i.35) can be integrated.28).35) .3 The WKB Method 287 Equation (14.10)).14.A = const.\ In S'.36) Equation (14. Li This expression can be substituted into Eq.2mQ(E .29) becomes (note that this is still exact!) (S')2 .
27) we obtain now with Eq. Classical Limit and WKB Method has the form of a momentum.39) where a and /3 are constants. and instead is asymptotic which we shall not establish in detail here. B. (1. p = h/X) k{x) :Then S'n +. we expect that in any case So » ft2Si.37) For Eq.19).39) and (14. Dingle [70].31): y = exp = ~ ~iW(x) h exp L nH™ exp ^S . Eq.E)' Then y(x) = V^(a 7 exp dx + 6 exp T(x) (14. (14. (14.I In S' + In c h 2 2 2 exp ^{S0 + h Sx) . i. it is reasonable to set (cf.288 CHAPTER 14. For V(x) > E (this is the classically forbidden domain) we set l(x) = ^==^====.41) We now ask: What are the conditions of validity of Eqs.± ln(S'0 + h S[) + In c (s'0y/i and hence exp :5 0 + 0(h2). (14.and So = ±h / — — A J k(x (14. y/2mo(V(x) . 1 S e e R.40) rx dx v "7 W)\.41)? We note first that the expansion in powers of h2 does not converge. E > V(x).' For the expansion (14. Chapter XIII. The solution is seen to be periodic provided k(x) is real. (14. . y = c'{k)1/2 exp or y(x) = a \ A ( z ) cos I / dx W) dx + 13).e.33) to make sense as an asymptotic series. (14.38) h y/2mo(EV)' (14.
36)) 2s'0s[ = [(s&r 1/2 ra) 1/2 With S£ = ±ft/A. E > V(x).V{x)) ^ This condition is satisfied provided E or E — V{x) is sufficiently large or (14.3/2 s&T + 3 {So Il« we have (cf. we obtain 1/2 1/2 2S[ and hence ±2hS[ = [kL/2]"kL'2 = lk~1/2X = Al/2 [lAl/2A//_lA3/2(A/)2lAl/2 so that It follows that (14. Setting R := A' = \moKV'{x)\ \2m0{EV{x)\W .14. (14. Eq. Since k(xy 1/52\S'0 (14.45a) .43) The condition So 3> ^2<Si therefore implies that /f>^i/^)k = H/y/2m0(E is sufficiently small.42) [(^).36).1 / 2 ]" [§0SS).3 The WKB Method We have 289 S0 = ±h and Si follows from (14.
46) The desired equation has the solutions y(x) = aXll2 exp ±i dx / (14.V{x) ~ Or . Classical Limit and WKB Method we can rewrite the inequality (14.47) Differentiating this twice — we omit the details — we obtain the equation y" + l A2 (A 1 / 2 )" *i/2 y = o. In terms of X the original Schrodinger equation is y" + = y = 0.e.a)V'{a). around the classical turning point. 14. In a similar way we have for E < V(x): l'(x) <C 1. In the neighbourhood of E = V(x).44) as Idxsj2mQ(EV{x))l 1 + ^R2j » ^Rh. V(x)) (14. How do the solutions of Eq.45b) so that the condition So » h2S\ is also satisfied by demanding that i ? C l . (14.48) behave in the neighbourhood of such a point called "turning point'. We can derive the equation whose exact solutions are the WKB approximations of the Schrodinger equation.290 CHAPTER 14.V(x) = V(a) . i. Roughly speaking one therefore requires E to be large in both cases. the WKB approximation is in general invalid (see below). (14. h = 1) .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).48) The (dominant) WKB approximations are exact solutions of this equation. and hence (with UIQ = 1/2.3. (14. at which E — V{x) changes sign? We note first that in this case near x = a: E . *2 A(z) = V / 2m 0 (£ .
1 there are WKB solutions on either side of the turning point and some distance away from it.3. The transition is provided by matching relations. 14. 14.48) possesses a singularity of the form of l/(x — a)2 at every turning point x — a. 14. (14. since this requires different approximations there. which connect one domain with the other. On one side. We assume we have a turning point as indicated in Fig. 14.e.1.14.3 after stating them first here. in the domain E < V the WKB solutions are exponentially increasing or decreasing. 14. i. V=E Fig.3 The WKB Method with the derivative relations 1 1 a)5/4' 4(x 1 5 16 (x. In the immediate neighbourhood of such a point the Schrodinger equation can therefore not be replaced by Eq. and on the other side. We shall not derive these conditions here in detail but rather make them plausible in Sec. in the domain E > V the WKB solutions are oscillatory.1 The regions around the turning point at x = a.1. As indicated in Fig. We define solutions ?/i and ?/2 with branches to the left and to the right as follows (for .48). 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.a ) 9 / 4 ' 291 (v^r 1 (x — a>2 These relations imply that Eq. (14.
P u t differently: We could add a lot of contributions to Ay± without affecting the asymptotic form of Ayi.1) is then: VI exp {~[T) Vxcos fx dx TV Ja ^ _ 4 (14.51) 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.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 / —. in order to make the formulas independent of whether the potential is rising or falling.292 CHAPTER 14. Thus. one could simply replace everywhere /•a / pa • • • b y / Jx J x . The first W K B matching condition in the direction indicated by the arrow (and for the potential decreasing from left to right. (14.49) x 3> a Vxsin and 2 exp V2 VACOS (L X dx A ra dx X for for i « a . and E ^ V for a ^ x the same formulas apply with the same direction of the arrow. as in Fig. has the same asymptotic behaviour in the domain i « a a s Ay\. (i x dx 7T ~~k~ 4 / One should note t h a t the solution ^4?/i + By2 for A ^= 0. a interchanged. This means the asymptotic part suffices only in t h e exponentially decreasing case. 14.H— J =^ v/exp + m (14.50) x S> a.52) For a potential rising from left to right. 14. however with the limits of integration x. Note also the extra factor 2 in yi. (14. Classical Limit and WKB Method definiteness note the + signs): Vtexp yi ~ S + Jx I X for for i « a .
2 The overlap regions around the turning point at x = o. (14.51) — always with the decreasing exponential on the left — is valid in both directions.54) .14.46). whose exact solutions are the dominant WKB approximations. the equation V approximately linear around turning point V=E Airy solution WKB . is singular at a turning point x = a.53) Since at x ~ a.3 Linear approximation and matching How do we arrive at the matching relations given above? We observed above that Eq.Airy overlap* Fig.a).3. the equation becomes approximately y" = (xa)xiy. y" + Toy = o. the matching relations have to be altered accordingly.e. (14. 2 1 X (14. 14. as we saw. i. Therefore we consider now this original equation in the domain around x = a. 1 ~ (x .48). If higher order contributions are to be taken into account. Thus around that point we can not use the WKB solutions and hence have to find others of the original equation (14.3 The WKB Method 293 In that case Eq. 14. (14. One should remember that the matching relations above are those for the dominant terms in the WKB expansion.
i I 772 \3 + > for ia>0.294 CHAPTER 14.5960. (14.56) We have discussed some aspects of solutions of this equation in Chapter 13.60). formulas 10.57b) UV2 3 11 by J dx . Cf. R. With the substitution z = (x . (14. . (13.57b) To the same degree of approximation as the equation y" = (x — a)xiy.a)X\/3. This is given by the following expressions. Stegun [1]. In the following we require the asymptotic behaviour of the solutions Ai(—z). (14. (14. Abramowitz and I. > for a.Bi(—z) for \z\ — oo. B. A. J (X (i4 58)  Replacing in Eqs.^ 4 I 7T x — a <C 0.55) the equation becomes the Airy differential equation d2y dz2 zy. or M. we have r x * f x ' /2(x a)i/2d*=^(x .1 / 4 by \^.59). A second solution is written Bi.57a) and (14. p. Dingle [70]. (13. Classical Limit and WKB Method where xi is a proportionality factor which we choose to be constant. p. (14.57a) COS 2^2/3 _ SZ 1 1 ?(z) /2 3 es^ Bi(z)~< 1 1 for x — a C 0. or to the literature:H 3/2 2^{zy/* Ai(z) ~ < ~~FL~T77 y/TTZ1/* 1 1 s e K*) n for ~. In particular we encountered a solution written Ai. as > may be seen by referring back to Eqs. — a ^> 0.4.a)3/2 ^ /2 =^ 3/2 o Q.291. 448.
2.3 The anharmonic potential well. we enter the domain of validity of a WKB solution.l. We summarize what we have achieved. (14..14. by Ai(z).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 BohrSommerfeldWilson quantization condition given by rb dx 1 fb i / — = . In the direction to the left as indicated in Fig. The latter domain is a small region where the solutions from either direction overlap. 14. e.l v/2m0(EV(x))dx 7r = (2n + l). F T' dx (Z)1^ by ^ harmonic oscillator V=x 2 +x 4 V=E linear approximation Fig. These solutions.59a) . As an illustration of the use of the WKB solutions we apply these in Example 14. 14.g. are only valid in a small interval around the turning point at x = a as indicated in Fig. In the neighbourhood of a turning point the Schrodinger equation becomes an Airy equation (in the leading approximation). As we approach a limit of this interval in going away from the turning point. At and around the turning point the solution of the Schrodinger equation is therefore given by Airy functions. In this way different asymptotic branches of one and the same solution are matched across a turning point.52). Example 14. and hence must be proportional (in view of the uniqueness of the solution there). n = 0.1 to the quartic oscillator. 14.3 The WKB Method and 295 •. In this sense we have now verified these relations.1 the Airy solution becomes proportional to the exponential WKB solution and to the right to the trigonometric WKB solution.2. however.(zf2 by J we obtain the matching relationsX (14.51) and (14.
. . 14. We argued earlier that the WKB method is suitable in the case of large values of E.3. In these domains the dominant WKB approximations are 1 /j/i(x) = cVlexp / fa dx\ I —/ — \ r for i « a. (14. Classical Limit and WKB Method where x = a.3. Consequently only the lowest eigenvalues would be reasonably well approximated by those of the harmonic oscillator. as in the case of the harmonic oscillator itself.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 . The potential now has to be inserted into this condition and the discrete eigenvalues En of the problem are obtained. whereas the approximation of the eigenvalues by comparison with the harmonic oscillator in the domain of the minimum. (14.44) this implies large values of n. (14. Hence we search for solutions which are exponentially decreasing in the far regions. (14.59b) Solution: We consider an anharmonic potential well as depicted in Fig. . The corresponding eigenfunction would be approximations of the proper eigenfunctions around the origin. the continuations of these functions into the central region 2 are: ya(x) = c v A c o s I / J. 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 wavelengths and is therefore valid for large values of n. n = 0 . Here we are interested in the discrete states (the only ones here).b).(x) = c ' v Acos I / — I for a < x < b. Let E = V(x) at x = a and x = b. 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 ( .l. regions 1 and 3 in Fig.61) Evidently these functions have to continue themselves into each other (since the wave function has to be unique). 14. Expressed as an integral over a complete cycle from one turning point back to it the relation is: <b dx^2m0{EV(x)) = (n+jh.y o dx TT T provided 6 _ 4 /. ^ A = (2n + l ) . In the neighbourhood of the minimum at x = 0 we could approximate the potential by the harmonic oscillator. 1 . i.2.51). b are the two turning points. The condition for this is ya{x)=yb{x) for xe(a.^ . for which the WKB approximation surprisingly yields already the exact energy eigenvalue as one can verify by evaluating in its case the BohrSommerfeldWilson rule.296 CHAPTER 14. 2 . I " ' (K rx dx iz\ _ ( (b dx_ 7T _ fx dx _ • f / fb dx\ f* dx\ ( fb dx\ .60) Using the matching relation (14. /n = s u ^ TJcos^ya TJ+cos^a T J .e. extends only over a length of very few wavelengths and is therefore valid for n small. yz{x) = c'vlexp ( [x dx\ I — / — ) for x 3> b. (14. .. Sometimes this distinction becomes imprecise. According to Eq. j/i. . 2 n = 0. Then around these points the potential would be well approximated by a linear potential (as we saw above). .
The corrected form* is. This relation is sometimes wrong. ri2 I Jqi :pdq n + 1 — . 2 .x number of turning points TTH. .1. 2.4 Three cases with different pairs of zeros. Bohr. In this socalled "old quantum theortf (i.63b) *The author learned this in lectures (1956) of R. Consider the Schrodinger equation in the abbreviated form d2iP(q) dq2 f(q)t/>{q) = 0. Sommerfeld and W.14. A. (14. and knows of no published form or script. 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. . (c) V Fig. . equivalently. (14. n = 0.x number of turning points h. .64) (14.e. pdq = n + 1 — . Wilson by supplementing classical mechanics by Planck's discretization. . Dingle. with n — 0.1. B.63a) or. 14.4 BohrSommerfeldWilson Condition 297 14.4 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.
0(g) ex . where /(g) is positive.67) Case (b): Next we consider. Jqi ^ fpdq J =2 pdq = (n + l)h.66) J<1\ Thus in this case the socalled old quantum theory is correct. i. where n —. 14. this is a trigonometrical zero. In quantum mechanics we have dq2 + Ki/j(q) = 0. i.e. There are three cases. 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. this means dqyf—f = (n + l)7r. where /(g) is negative. Before we continue we recall from Eq. 14. i.65) We use the knowledge of these solutions to find the eigenvalues of the equation. Case (a): The case of two trigonometrical zeros. We have the case of trigonometric solutions. (14. (14. —/ = K J h It follows that / pdq = (n + l)Trh= (n + l)h.4(b). i. (14. as illustrated in Fig.51) the linkage between the trigonometrical and the exponential solutions across the position go in space at which /(g) = 0. In this > case the wave function for g > go is .298 CHAPTER 14.e. / • n an integer > 0. Since sine and cosine have zeros spaced at intervals of TT.4(a).e. together with q — — oo. this is an exponential zero. the case of tp(q) — 0 at q = gi. the wave function is to vanish at points q = q\ and qi with the function /(g) remaining negative in between as illustrated in Fig.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. if /(g) is negative.
Hence we have pdq= i n +  .63b) to obtain the eigenenergies of the quantized harmonic oscillator. or Pdq (n + l ) " 2 h. 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 . the sines must be proportional.e. Solution: As in the case of the quartic potential. (14. i. (14.63b).2 and 14.4 BohrSommerfeldWilson 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. E x a m p l e 14. ^~f(q)dq Jqo LJqo + 7T J'% On fq / i 1 V~f(<i)dqjK (n + l)7r.69) pdq = (n + l ) " 4 h.68) (14. and • • for the wave function to be exponentially decreasing to the right of q'0.3.4(c). Hence we obtain the condition I'o Vf(l)dq 90 (n + 1) IT. Case (c): Finally we consider the case of i/j(q) = 0 at q = ± o o exponentially as illustrated in Fig. Thus in this case the socalled old quantum theory is wrong.77T (where we reversed the order of integration to obtain a positive integral and multiplied through by minus 1). 14. (14. As illustrations we consider Examples 14.63a) or (14.67): • the wave function for q > qo must be proportional t o (Z)1/4 sin [ ^W)dq+>x Jqa . For the wave function to be exponentially decreasing to the left of qo and using Eq. this is the case of two turning points. i. ri\ / J do y/f(q)dq (n + 1) 1" (14.70) We can therefore summarize the results in the form of Eq.2: The harmonic oscillator Use the relation (14.14.e. Therefore their arguments differ only by (n + l)ir. Since both cases refer to the same region qo < q < q'0.
Example 14.2 3 J x (n + 3 / 4 ) 2 / 3 . whereas a turning point does. one obtains E ~ 2. in principle. ^Thus a trigonometrical zero is the condition of absolutely no penetrability beyond it. Thus there is one turning point at E = mogq.807 m s . The WKB Method this requires evaluation of the integral with qi = ^' E/2K2mQu2 / 2 m 0 ( B .2 . so that E = hu{n + 1/2). where g is the acceleration due to gravity. Calculate its quantized energy.71) For m 0 = l x 1 0 _ 3 k g .63 x 10~ 3 4 J s .V(q))dq The result is E/u.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.5 A particle above a flat Earth. . permit this although with rapidly diminishing probability. 4 / x CHAPTER 14.5. and h = 6.28 x 1 0 . 14. as illustrated in Fig. 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. :: V=m0gq Fig. Solution: This is the case of one trigonometrical zero of the wave function ip at q = 0 and an exponential zero at q —> oo. g = 9.300 With potential V(q) = 2ix2mov2q2.
= — <*— = h(n+\.73) Jx! rp2a dx I^<E 2 v) = ^ n — = .( 1 — = — = it — = h\ n H — 2TT OJ 2K J p dE V 2 In the case of the harmonic oscillator we have hi n E (14.2. (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 £ = .75) which verifies the formula immediately.74) p hjtl *2 —t\ _ mo f 2rr 2K 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 .14. (14. d dE V)={n+)n.AE(q0 2 = 2n + 1) = — exp 7r dzy/E zn + V(z)/h where q=zo are the left and right barrier turning points. Solution: The proof is contained in Example 18. Chapter 16) one obtains the quantization relation (for Mo = const. 9 .5 Further Examples 301 14.5: Period of oscillation between two turning points Use the BohrSommerfeldWilson 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 ( — = . ra 2/ = 0. h M0 _h 1^3/a ~ 4 0 / 2 J V \ 3 _ 2N3 \M0) ~ ~Ml ' . and u> is the oscillator frequency in either well. In the case of screened Coulomb potentials (cf.7.f V ft h JXl h 2TT (14. 1 . E x a m p l e 14.5 Further Examples E x a m p l e 14.72) \\2dxp=Qdx^{Eh Hence.l. 2TT W 2K J p dE V where p = ^/2mo{E Solution: We have — V). with p = modx/dt.) M0 N = l + n+1 : + • 2/B Hence dE K 2 d dE[ M0 2VB.
It follows that the energy between the minimum and this maximum lies in the range 0 < E < £ m a x = ^aSb2. Fig.77) Determine (a) the turning points. and (b) evaluate the WKB exponential exp[—2/b arr j er ] and the WKB prefactor 2/ w e u. (14. Solution: The potential has a finite minimum at q = 0 and a finite maximum at q = 2a2/3 as indicated in Fig. 14. (11.~ \ (14.y3) = 0.6 The cubic potential. q = y+ a . Example 14.302 and T CHAPTER 14. 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 .e. Eq.y2)(y . The WKB Method ^ 2 JV 3 2TT ~ ~Mjf (14. which determine the imaginary part of the energy.<?2i<?3 for E > 0 given by E — V = 0. i. 14.a .76) in agreement with the result obtained from the classical Kepler period for the Coulomb potential (cf. where 1 2 • 1 2 y = Q.79) . The q term in this equation can be removed by transforming the equation to a cubic in y. where .6: WKB method applied to the cubic potential Consider the cubic potential V(a)=1b\\a*~q).133)).78) (a) We determine first the three turning points at qi = <?i.6. (14.
80 a n d 361. q3 = a2[\ .85) T h e angle 9 = 60° c o r r e s p o n d s t o E = 0. 0<6O°.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.§ ( c o s 0 ± \ / 3 s i n 6 0 a n d § sin 6 sin# (14.g i ) ( g 2 .6. _ \ / 3 — tan(9 2 V92 . we h a v e qx = . M i l n e . i. (14. 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 .78)) c o s 30 •• 4P3 ( ' \ aeb2} . B . (14. 78: sin 120° = v ^ A c o s 120° = .T h o m s o n [196]. e > 0.e.c o s O + cos(6» . (14.93). p . M .ui=K(k). F r i e d m a n [40]. g 2 ~y{3}>0. 37.g i < 1. 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.120°)] = . (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 — 0.2 c o s ( 0 + 1 2 0 ° ) ] . 6 = 0 i m p l i e s E m a x of E q . F . 14. 91 < 93 < 9 < 92(14.a 2 [ l .1 / 2 .2 c o s ( 0 .2 c o s 0 ] . we set 0 = 60° — e.1 2 0 ° ) w i t h 0 < 6 < 60° for 0 < E < £ . p .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 . h = 1) ^barrier := / d( }J^2~(E~v) = ^~= dq^/(q .80) I n s e r t i n g t h e r o o t s yi. a n d t h e a n g l e 0 = 0° t o E = £ m a x . 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 . D w i g h t [81].9 2 [ .gi = 1. D . H e r e cos(<9 ± 120°) = .9 l ) f f / du s n 2 u en2 d n 2 u . tan60°=%/3.k 2 2tan# \/3 + tan! .81) m a 91 .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.120°)] = a 20? 2a2 3 2a 2 • cos 0  91 . V3 + t a n 0 2 . f o r m u l a s 236. V cose' + sin 8 (14.93 2a2 [ . B y r d a n d M . 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. ' P .79) a n d t h e n s e t t i n g m o = 1. W e o b t a i n ^ fbarrier : = 6(92 ~ 9 3 ) 2 ( 9 2 . .= s i n t 3 v 3 (14. p .3/2.' : ± 6 0 ° w h e n E = 0. q3 ~  .q)(q .9 3 — = = 2 " 3 " [ .c o s ( 0 + 12O°) + c o s ( 0 .08. 212. p .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.cos 0 + cos(6> + 120°)] = a2 92 .< 1. 92 = a2[l . (l_fc' 2 a + fe'4)l/4.\tan0^ v/3.{ V 3 e } > 0. 92. ui=sn 1 \l— — = l. § T o h e l p we cite H.04. 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.2/3 i n t o q = y + a 2 / 3 .14.84) 92 . (14. gi ~ y {V3e} < 0.0^6O°.
with K(k = 1) = oo.91 . The WKB Method elliptic integral of the first kind. (14. V?2 . Byrd and M.2(fc2 . . (14. G(fc) =86aS (ITS£ li[fe./ dq~ ^2 b JQl 1 \/(qqi){q2q)(q.86) 15 tan.^ V3 / V V3 y a / c o s e J. Friedman [40]. E(u = 0) = 0.92) ''P.k). D.S . F.304 CHAPTER 14.85).90) hi JlnynEv) V(qqi)(q2 Vz b Jq q) I" bJqi where P fe = 9391 9291 = . Next we evaluate the required integral across the well from 91 to 53 at energy E.00.2)*w+2(fe4+k'2)E{k)] v ' G(k) For one complete round from one turning point back to it. 0 (14. (24.^ + ^ ) (l + fc'2)2 .87) Inserting G(fc) together with the expressions for the prefactors into (14. E(k = 1) = 1) (•u\ = K{k) i Q G{k) = / Jo Next we set s dusn2ucn2dn2u = j[k'2(k2 15fc4 .qz) (14.) .91) a2[cos0^] y/3tanfl ? L . A:') = # ( * ' ) • (14. Thus Jwell : = [13 1 f13 rf 1 / dq J^^EV) 1 y/2 1 V 2 f[13 = — —. we obtain in the limit of E = 0: j d ( ^ ( E „ y ) = 2/barrier = ^ ! .~ y ^ ) a< [ COS 5 + . sinV = q3 qi ~ = 1. 91 < q < q3 < 92 bJc. 72.?)(?3 .2)K(k) + 2(fc4 + k'2)E(k)) k ^i — . 1+72 = and J o s 0 + ^ = (1 .fc'2 + fe'4)"1/4.sm t %/3 £\. E(u = K(k)) = E(k). the integrand of the integral across the well is effectively the inverse of this momentum. (14.91 F(TT/2. g= 2 . V3 cose + ^ = V3 2 1 + 72 v/l + 3 7 4 C (cose) 2 = 1 + 3 7 4 = 4 d .84) we obtain Carrier = 2a 2 sin6> 2 b( . as in (14. Whereas the integrand of the above barrier integral is effectively a momentum. 2 2 . We can evaluate this integral again with the use of Tables of Integrals. Eq. V 93 . . Thus Iwen=lgF(ip. formula 233.= —^ = k a2[cos0+^] v ^ + tanfl V=".32)). E(u) the incomplete elliptic integral of the second kind. p.89) This expression will again be obtained later with the help of configurations called "bounces" (cf.
4 2n + 1 fc' = 4 — ^ — . .2.89) 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 2KE ab tt I. formulas 8. The same expression for E is obtained by applying the "method of poles at infinity" to the BohrSommerfeldWilson condition (14. Thus one obtains from (14.85) (14.114. (14. With q = 1/z the integral (14. we obtain 15 . we cannot expect the ratio exp(—2Jt>arrier)/2J'well oc w exp(—27barrier) evaluated at k = 1. 32 (14.96) Expanding the coefficients of the elliptic integrals in Eq.94) C °S3 2 (lfc2 + fe4)3/2  l+ g f c ( l + fc) + Comparing this equation with Eq. (14.l. 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.113.80). 906. To obtain the latter we have to use the quantum mechanical expression approximated by E = hw(n + 1/2). in fact through logarithmic contributions contained in the argument of the exponential for k ^ 1. Gradshteyn and I. to represent a physical decay rate. Ryzhik [122]. M.3 and 8.5 Further Examples Hence (using K(0) = rr/2) 305 ab ab (14.63b).88) in rising powers of fc' . and hence at E = 0. pp. (14. .14.97) 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. we obtain > . we obtain E a6b2k'4(l + k'2). 2a 5 b K(k) + 2 E(k).93) where w is the harmonic oscillator frequency in the well. 905.95) Comparing this with the harmonic oscillator approximation E — En = ab{n + i ) . so that even in the case of the ground state the zero point energy will contribute to the prefactor..3.2 IE 4 fc' ln 8 (" «V iU^fc' 4 . (14.** However. (14.! With these expansions one obtains ^barrier = 15 ahb 2 + 8A . S. a°b n = 0.99) **For the potential approximated around the origin as V ~ a2b2q2/2 the eigenvalues are E = fat)(n + 1/2) with it) = ab.
15 a . (18.> tf = = 2/wfiii = ±tv8a°b——P.178)) call the Furry factor set equal to one.11)) dq 2TT y/2§?(Ev) It would be interesting to derive the same quantity with the perturbation method and to compare the results.7 . We observe that with \j2je as 1.„. It follows that (for one complete orbit back to the original turning point) .e x p [ . .306 CHAPTER 14.96) for the imaginary part of the energy as in the Breit..Wigner formula (10./^~^l^ab ^aub .100) 'barrier = ~0°b In [ ^ — j . the result agrees with the ground state path integral result (24.. _ exp[2/barrier] >. The WKB Method For k' —• 0 the first and the third terms dominate.73).j . i.102) . so that we obtain 4 . also Eq. . (0) i hw En = E\> ..e.J + 0(VoJb). (14.105) 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..ba / 25a5b\n+2 ±l_l eT5 a f >.2 7 b a r r i e r ] Z Z7T where (cf. /25a5b\™+5 e exp[2/barrier] ^ ( r ) Correspondingly we obtain for the full period 2/ w e ii: 2/well = ? K ( f c ' ) ( l . i. 2n + 1 — / 26a5b \ . this has not yet been done. 7 ~ —. n ~ V27T . _ T T • (14.101) (14. 8 Bfc .— (14.103) 27 w e l l 27rVn+y ' fn With Stirling's formula in the form of what we later (with Eq.fc'2 + fc'4)1/4 * ± t ^ . +3 the result becomes exP[2Jbarrier] = ± i _&g_ (25a5br+J e . oa ba Thus with the W K B method we obtain exp[2/barrier] — barnerj = 8 s.e.( n + l ) e _ JL a * 6 and we obtain for the ground state (ro = 0) . V (14. (20.
led to a spectrum which contains only scattering states. Quigg and J. See particularly E. the Coulomb potential. L. C. L. The study of this potential. Quigg and J. now known as quantum chromodynamics. Thacker.M. T. B. K. the classification of the various states of nucleons and mesons and other particles. This potential became widely popular in the spectroscopy of elementary particles.* 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. There are numerous.1 Introductory Remarks The particular linear potential we considered in Chapter 13. indications from the nonabelian generalization of quantized Maxwell theory. This would mean that the force binding the quarks together would not decrease with increasing separation as in the case of e. 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. [233] and H.g. which permits only bound states. i. after the realization that quarks as their constituents might not exist as free particles and. 307 . Rosner [269]. in fact. Gottfried. that this is indeed the case. [232]. that it may not even be possible to extract individual quarks experimentally with any finite amount of energy. Lane and T. along with inclusion of angular momentum and spin effects and their interactions. at least indirect.e. describing free fall under gravity. In the present chapter we consider briefly the threedimensional potential V(r) oc r = r. Eichten.Chapter 15 Power Potentials 15. C. Kinoshita. D. K. [231]. Yan [82]. Rosner [230].
~ (I + l)rl (15.1.e.m2. (15. 15. (15. These particles therefore cannot be separated by any finite amount of energy. i. For a central potential V(r) we write (15. and so u'(r) .3) with the boundary conditions''' u(0) = 0. h2 2n V 2 * ( r ) + [V(r) . m\m<i M= : . mi + m2 and r is the relative coordinate.2 The force The Power Potential F = . and obtain 2/x 1(1 + l)h2 u"{r) + EV(r) u{r) = 0 2 2/ir (15. We leave consideration of the logarithmic potential to some remarks and Example 15.1) i.4) Instead of considering the linear potential. and u'(0) > [u(r)/r]0. </>)R(r). we return to the Schrodinger equation in three dimensions. Power Potentials general power potentials. In order to have a clear starting point. (15.e.2) tt(r) = Ylm{9.6) = (I + l)u(r)/r = R ~ rl. we consider immediately the more general power potential V(r) = \r\ 'We have u(r) = rR(r).r acting between two particles is constant and directed towards the origin (of the relative coordinate).5) u'(0) u[r) r=0 = R(0). v > 1. This force maintains this value irrespective of how far apart the particles are separated.u ~ r i + 1 (/ + l)R(r). where \x is the reduced mass of the two particles of masses mi.V 7 = const.£ ] * ( r ) = 0.308 CHAPTER 15. A > 0.4>)ru{r) = Ylm(9. R(r) = ^ . f dr{u(r)}'2 1. and the normalization [ dr\y(r)\2 = 1.
63b) for the onedimensional Schrodinger equation.63b): 2 and V(0) = 0 (15. Onedimensional case: We consider first briefly a particle of mass TUQ in a symmetric onedimensional potential with the symmetry V(x) = V(x) without loss of generality. (15. (15.4) in the threedimensional case.10) Here rc is the turning point and n — 1. We therefore proceed as follows. Then k(x) = k(x). We also assume that V'(x) > 0.9) into the threedimensional case in the following form: / Jo C dry/2n(E . This means that we transcribe Eq. .2. or f l H " ^ ) ! .2.V(r)) = (2nl) +1 N.8a) ^ JO = (2Ar + l ) £ . (14. which has been reduced to an effective onedimensional problem in the polar coordinate r..9) Threedimensional case: We transcribe the above considerations of the onedimensional case into that of the threedimensional case by restricting ourselves in the latter case to S waves (no centrifugal potential!).8a) we can write the quantization condition (14. (15."MA. and there identify the Swave states with those states of the onedimensional case whose wave functions vanish at the origin — these are the nonsymmetric ones with N odd — as required by Eq.63a).. 3 . so that with Eq. In order to to obtain the .2 The Power Potential 309 In Chapter 14 we obtained the BohrSommerfeldWilson quantization condition (14.15. and the turning points are at a — — b := — xc.1. N = 0. d5. (15. (15. odd integer ^=(nll)vh. is to be interpreted as the principal quantum number in three dimensions. . We now want to find a quantization condition for the threedimensional problem.7) k(x) = h/y/2m0(EV(x)). (15. .8b) This condition implies that we have only one pair of turning points. x > 0..
E A fv.11) rc = (15. 1W or Xv \E I It follows that (I) / ^""^(l*)1/2.\rvc = 0 .13) we obtain _ X_ „ _ so that A£_ (15.13) dt .y) defined by (xl)\(yl)\ y ^ (x + y .'(1*)1/2= J1F\lt)y1dt. Power Potentials eigenvalues E —• En for the radial potential (15. (1516) (15. (15. Now. _ —vr" *dr.14) Prom Eqs. (15.y) T(x)T(y) _ T(x + y) ("!)/" /•! / d"(1"")/. . c (nj\irh. (15.15) I ' = ^ 1 (( 1) B{x.6) we have to evaluate the > integral / : = [ ° dr^2n{E\r»)= Here the turning point rc is given by E .310 CHAPTER 15.12) i/z Jo We set dr 1 l A v ~ E \ (Et\l/v so that (15. dr = .17) The integral on the right is the integral representation of the beta function B(x.12) and (15. s.
1. t h a t in the limit v —• oo the potential approaches the shape of an infinitely high square well. The case v = 4 is that of the pure anharmonic oscillator also discussed by M. (12.19) or En = _ 1)^(3 + 1)^/^2^+2) r()ir(i)^7^ • v 0 1 Fig. . ^ '"Compare with Eq. T(z + 1) = z\. for instance simply graphically as indicated in Fig. 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 ) .18) \v \E or 3 E2 r(§)r(i)V2M (n ' (15. i r ( i ) = r ( l + i ) = ( i ) ! . Weinstein [281].2 The Power Potential 311 so t h a t by comparison y = 3/2 and x = 1 + (1 — u)/i/. 15. $T(z) = {z. (15.1 Approach to square well with v —> oo. and n — — ]7r/i. One can convince oneself now.15.* In this limit the eigenvalues become:^ E„ (nh^ni) Lr(f)r(i)^7^ rv"ir •n2 2/i 2^2 (15.1)!.55).
In the case v = 1. 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). (6.41) for the .2. = (n>r(§)A 2/3 3TT \h n 2/3 L r(§)v^7^ J (15. For the following we set in Eq.25) "Comparison with Eq. (13.oj = ^/2X/JX. eigenvalues (JV + h)tkj — (2n . 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 = 0.22) 0 which selects from the usual and with the boundary condition 0(0) eigenfunctions the odd ones.61) h2 = I.2) implies A s fj. Power Potentials In the case of v = 2. we obtain En = (n . since these are the ones we are interested in (cf. > and with Eq. (15. the harmonic oscillator. discussion at the beginning of this section). 15. so that the appropriate Schrodinger equation becomes d2(j) + (Edx2 x)<t>(x) = 0 .312 CHAPTER 15. we obtain E„.oj2/2.21) (4nl).24) is discontinuous at x = 0 (this means the derivative there jumps from positive to negative). In the onedimensional consideration these are precisely the odd wave functions as illustrated by an example in Fig.g — —2.23) The eigenvalues for the linear potential can also be obtained from the zeros of the Airy function. Physically it does not make sense for the probability amplitude to have a discontinuity there.mo = \. The potential V(x) = X\x\ (15.1 + \)hx> = {An • l)£fiw.> r ( 2 ) A V 2 in \)*\W \it^T& (15. (6. x > 0. the case of the linear potential.//L This result agrees with that for the onedimensional harmonic oscillator r2 in the form" cf>"(r) + ^(E\r2)cf>(r) =0 (15.
x)\x=0 = 0.15.59) applies. this wave function has the required exponentially decreasing behaviour at infinity. i. In the domain —a < x < + a .29) . dz2 According to Eqs. (13. E > V or s > 0. we must have as a result of the boundary condition <p(x) = 0: 4>(x)\x=0 oc Ai(E . (15.2 The Power Potential 313 V(x) V(x)=x turning points exponential fallofl Fig. i. (13.e. and for s —> oo the trigonometric behaviour of Eq. (14.e. 15. s = E. Ai(s) "v^ G ' cos s 3/2 _ t (15.28) In particular at x — 0.57a) one solution of this equation is the Airy function <l>(z) on Ai(z) = Ai(E .27) i.60) applies.26) For z = s := E — x —> —oo.56) and (14.e.e. W i t h z :— E — x this is d24>{z) + z<f>{z) = 0. \(s?» (15. i.2 Behaviour of an odd wave function at the origin. i. (15. Ai(s) 1 2v^(s)V4 exp x —> +oo.x).e. Eq.
p.31) (an odd function has an odd number of zeros!).. 201.94249 9.. Rosner. n = l. The expression (15. Power Potentials Table 15.82814 i.08181 5.3.1 demonstrates the quality of the WKB approximation by comparison with the exact values. (2/3)E3/2 or 3 / 1 2/3 E — En — . = im — (7r/4).e.7 T Tl * 2 V 4 Actually this expression is only valid for E and hence n large. Quantum Mechanics with Applications to Quarkonium.. to obtain the zeros of the Airy function numerically.00852 11. with permission from Elsevier.29) implies that cos (  ^ . so that n — 1/4 ~ n.00767 11. the eigenvalues En are determined by the zeros of the Airy function. of course. copyright of NorthHolland Publ. Table 2.33811 4.04017 11. .. i.2.93528 12.1: SWave Energy Eigenvalues for V(r) — r (with h = 2/x = 1) from f 3 7 r ^n 2 ( i>j2/3 . [231]. i.314 CHAPTER 15. (15.^ = !(2nl). but for large > values of E Eq. *This Table is reprinted from C. It is interesting to note that in the dominant approximation both methods agree.78445 7. Co.52056 6.02137 10.93602 12. Quigg and J.94413 9.02265 10. L. (15.32025 4.e.51716 6.08795 5.En from Ai(£„) = 0 2. It is possible. Table 15.28) is really only valid for s large and s — oo.e. (15.03914 11.= ) = «.78671 7.82878 1 2 3 4 5 6 7 8 9 10 2.30) ! ^ / 2 .
Weisskopf [239]. i. The resulting masses agree in most cases very well with those extracted from experimental observations. have only a finite lifetime and decay into other particles such as electronpositron pairs.3 The ThreeDimensional Wave Function 315 15. (14. F. dr[u{r)f " f** «" (i £>i)4 Here we replace the oscillatory part cos (. e. The decay widths T. We saw previously (see for instance Eq. of course. See also J. in Chapters 24 and 26.34a) **The most frequently quoted reference for this result is R.** r ( # ^ e e ) o c *(0) 2 .e. In order to be comparable with experimental data these investigations. P. We indicate briefly here — without entering into further details — how these quantities can be calculated. In general these quarkantiquark pairs.e.e.g. . We shall encounter WKB normalization constants at numerous points in later chapters. ee. cos2()^— / 1 fr° 2 > w . e. Jackson [140]. r <53) 1 3  dr cos 2 () (15.g.3 The ThreeDimensional Wave Function The linear potential has in particular been used in the investigation of the spectrum of heavy quarkantiquark pairs. can be shown to be proportional to the modulussquared of the particle wave function at the origin. D. the charmanticharm meson ^ .* We can determine the constant N in the leading approximation by demanding that = 2 / Jo i. which are inversely proportional to the lifetimes. *Note that this makes N a W K B normalization constant.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.• •) by a mean value. i. van Royen and V.15. had to be supplemented by inclusion of relativistic corrections as well as other contributions arising for instance from spin and angular momentum interactions.
(15.316 In accordance with t h e relation CHAPTER 15. (15. so t h a t approximately l N2 / Jo k(r)dr .35) u T (•T=2TT/OJ dt cos 2 cot 2' (15. & (15.2) and (15.e. A = (15.39) in / ..32) we obtain to leading order u'{r) and therefore «'(0) = N N (1JU0) N (15. (15..2. Power Potentials this averaging implies a numerical factor like 1/2.38) N : sin V ^ Jr dr' 7T (15.l N n . (15.37) (15.34b) VMEvy Variation or differentiation of Eq.Jo dr' n M?) ~ 4 in sin sin JTCldrW2v{EV)j l \ 7T N 7W)sin = ( .rcu w X{r)dr = Kh .1 )n .35) N2 . i.36) \i dEn irh2 dn According to Eqs. From Eq. (15.4) we have for I = m = 0: ^(O) 2 = n'(O) 2 y O o(^0) 2 = ^ K ( O )  2 .40) VWY ..\ \ * n = l.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. dE a—— / dn J0 so t h a t with Eq.
. MiillerKirsten [147]. The case of arbitrary positive power potentials as above has been considered by R. G.2.15. S. (IM2) l*(o)r = M dEn dn irh 2 ~^WEn ~*T (15 43) ' Inserting here. K. the expression (15. § S. . . E. Hite and H. 3 .g.43) or the results of corresponding calculations one finds the following dependence of (^(O)!2 on the quantum numbers n = 1.l ) n . (b) for the logarithmic potential^ l*(0) 2 ex ^ v (15.41) I*(0)2=_L^)^.3 The ThreeDimensional Wave Function Since we are considering potentials with V(0) = 0. 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. J.45) 'See the references of Quigg and Rosner cited at the beginning of this chapter.37) we obtain: 1 (2M^)1/2 47r ft. S. Bose and H. we have ^'(0) = ( . for instance. 'For one such investigation which includes the Coulomb potential in addition to a confining potential see e. MiillerKirsten [28].38) implies 317 (15. Bose.1 ^ ( 2 ^ „ ) 1 / 4 . Kaushal and H. for various potentials: (a) For a confinement potential of power v. 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). and hence with Eq. K.23) for the energy levels of the linear potential. (15. J. Itt^pocn2*"1^"^. J. MiillerKirsten [29]. 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.44) (15. W. (15. W. Hence Eq. W. .* Using the result (15.
(14.c = —a//3. 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 . rewrite the equation as ^ z + [ .L 2 + U{z)]<f>. Example 15. Landau and E. ip = exp[(z — c)/2]. Power Potentials *(0) 2 oc . (11. *=^f.. D.46) and the classical Kepler period of Eq. U(z) =  dz zf3e2a/?e2*. .e. MiillerKirsten [29]. {S = ^ > 0.5.L2 h4 = 4/3exp[(2a — /3)/f3]. and there Eq. Lifshitz [157]." 1 11 L .133). Eq = i q . E' = E + gln(r0).— oo < z < oo.j . S. n = 0.1: Regge trajectories of the logarithmic potential Consider the radial Schrodinger equation for the potential V(r) = g\n(r/ro). = (I + 1/2) 2 .^ +•••. q = 2n + l. 7 With the substitutions r = exp(z — c). .. J. M.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. W.1. We observed there in the case of the Coulomb potential that the period is proportional to n 3 . i. (15.318 (c) and for the Coulomb potential^ CHAPTER 15.^ ( 3 ^ + l)55p^«(3^l) + . in agreement with Eq.. K.I l n f ^ p .75).g ^£+(<*Pl»riy = 0. Solution: Details can be found in the literature. . = 1(1 + 1).. Bose and H.2. (15.
in microscopic and hence quantum physics with the possibility of real and virtual creation of particles and hence a vacuum state. a charge as source of a Coulomb force leads to an effective polarization of the vacuum like that of a dielectric.Phenomenologically the screened attractive Coulomb potential may be written V(r) = g2pr/ro . it will always notice the latter's presence. it is. r where r is the distance from the source charge and the charges and other constants are collected in the coupling g2. that no matter how far the particle is scattered away from the source of the Coulomb potential.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 logarithmr contribution. so that like charges are repelled and unlike charges are attracted by each other. In reality. and in fact. This effect sounds unphysical. The exponential insures a rapid falloff 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 very similar potential is the Yukawa potential V{r) = g2 — . This implies. Historically this potential arose with the realization 319 pVr . The result is a screening of the Coulomb potential which thus attributes it a finite range ro. r in which /i has the dimension of an inverse length or equivalently that of a mass in natural units. It is clear that for ro — oo the > potential becomes of Coulomb type.Chapter 16 Screened Coulomb Potentials 16.
e. The realization that mesons and baryons are made up of quarks does not really change that picture at lower energies. In the literature reference is sometimes made to the socalled static limit of a relativistic propagator. Since. 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 KleinGordon equation which results from the quantization of the classical relativistic energy momentum relation of a particle of mass /x. A Green's function is effectively the inverse of a quantity called propagator which is intimately related to the expression in Eq. the mathematical expression with an exponential is not so easy to handle analytically. Heisenberg [133]. . i. What is meant is that the fourdimensional 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 .1) We encountered Green's functions earlier. and that. 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. (16.1) and is written in spacetimedimensional Minkowskian notation l/{jpvpu + /i 2 ). Yukawa potentials and superpositions of such potentials play an important role in nuclear physics.320 CHAPTER 16. who later — as is wellknown — deviated from this idea and invested his efforts into the study of nonlinear spinor field theory. Regge trajectories — which we encounter *The Smatrix theory of strong interactions was actually initiated by W.2 ) or (p = hk) 1 /" . from the relation Pi + P 2 + M2c2 = 0. and in view of their relation to the exchange of virtual elementary particles have therefore been objects of intense study in elementary particle theory. Screened Coulomb Potentials that the strong nuclear force is mediated by the exchange of mesons. in fact the parameter [i represents the mass of such a spinless meson.3) where for the real potentials we consider here. one frequently resorts in calculations to the expansion of the potential in rising powers of r.x j 36 x 4 (2TT) k + p? 4 2 (2TT) J d x e^xl 4TTX ' We observe that the Fourier transform of the propagator is effectively the Yukawa potential. (16. however.
(16. Thus in the present case it turns out that it is easiest to calculate first the expansion for the expression l + n+1.4) where E = k2 and h = c = 1 = 2m. See also the other references below. however. if>oo. S S.§ 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. . T.5) where K — ik and A n is an expansion in descending powers of K. Mandelstam [185].16. i.1.{l. where n is an integer and we referred to these already in Chapter 11 in the simple case of the Coulomb potential. £+**PVV t/. l+n+l = ^P.1 in the above. (16.3) in the radial Schrodinger equation for the partial wave ip(l. Bethe and T. f H. to be of the order of 1/k. "These were first considered by C. Cheng and T. bound state eigenenergies and the 5matrix. A. . . Froissart [223]. i.k. 2 .k.Q. 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. Regge trajectories were realized to play an important role in the high energy behaviour of hadronic scattering amplitudes. Omnes and M. only for the cases of n = 0. Wu [47]. Frautschi [97] and E. Lovelace and D. . 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.3) have been considered by various authors.r)=0. i.^ Regge trajectories arise as poles of the Smatrix in the plane of complex angular momentum. C. physical) values of angular momentum as functions of energy E. Kinoshita [21]. Potentials expandable as in Eq.r). Naturally it is easiest to familiarize oneself with these by studying solvable potential models.1 Introductory Remarks 321 in this context — are functions which interpolate integral (i.e. S. Squires [258]. ''The approximate behaviour of Regge trajectories for the Yukawa potential has also been calculated by H.mo being the reduced mass of the system. for the Regge trajectories or Regge or /plane poles of the 5matrixll / = ln(K) 'Standard references are the monographs of R. J.e.e. 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. Masson [180]. $ In the following we consider screened Coulomb or Yukawa potentials of the type of Eq. (16. (16.e. They are usually written I = an(E).
Under these conditions the S'matrix is meromorphic (i. A. In the following we follow mainly the last two of these references. < const. M. Regge [31] and E. Longoni and T. Muller [201] and [202].k. Muller [204]. J. *See in particular A. (16.9) the cut starts at E = k2 = 0.6) 2K{M°~ An(K))x+ 2K^(^K) MiX ' (16J) **H.2 Regge Trajectories Since our treatment here aims at obtaining an S'matrix as a generalization of that of the Coulomb potential. Screened Coulomb Potentials as a function of the energy. Squires [258]. W. o rV(r) I dpp\V{peie)\ regular at r = 0. one can assume an expansion of the potential V(r) in ascending powers of r. (16.19). W. J. 16. J. tf H .322 CHAPTER 16. W.15) and (16. 'In the case of the Coulomb potential (cf. .)/j. 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.** The energy is later obtained by reversion of the resulting series. Miiller and K. 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. Section 11. as may be seen from Eqs. in the present case at fc2 = 0 or k2 — Mi = 0.ndp.z) Then x is a solution of the equation VaX= (16. Schilcher [203]. The conditions for this have been investigated in the literature* and may be summarized as follows: /•oo V(r) = / d/xa(//)e^7r.e.k. for all n. Bottino. Proceeding now as in the case of the Coulomb potential we change the variable of Eq. has only simple poles) in the plane of complex angular momentum and in the complex &plane (E = k2) cut along the imaginary axis. H. starting with the power r _ 1 of the Coulomb potential.4) to z = —IKv and set TP(l.t Considering such superpositions of Yukawa potentials with /•oo / a(/j. < oo for all 0 < vr/2. J.z) = ez2/2zl+1X(l.
(16. anharmonic and cosine potentials have been derived in L. as in the case of the Coulomb problem. and all Sm(a.i 0) = 0.a + + l). K.z) is known to satisfy a recurrence relation which we write here for convenience in the form z$(a) = {a. b\ z) is seen to be a confluent hypergeometric function which for reasons of convenience we abreviate in the following as 3>(a).16.10): Sm(a.a + j)Smi(a.1). W.j +(a + j + l.9) (16.5) is equivalent to a = —n.7) is seen to be of order 1/K. MiillerKirsten [249].a+ l)$(a + 1) + (a.a + j)Smi(a.j) (16.b.z) = Ha). all other So(a. (16.j) may be computed from a recurrence relation which follows from the coefficients (16.11a) The coefficients Sm(a. (16. Eq.11b) ^ The associated boundary conditions are: So(a.2 Regge Trajectories where d2 Va = z^ and (cf.5)) a = + 323 .0) = 1.9) we obtain m l)$(a .b. J.j) = (a + j l.L ^d {bz)a l+ l + ^ ^ l 2K =  n &ndb = 2l + 2 = 2n^4^ K (168) The right hand side of Eq.j l) + (a + j.a)${a) + {a.j) for \j\ > m are zero. (a. {0) X = Ha. We also observe that the ansatz (16. (16. Sharma and H. The function <&(a. which means that the hypergeometric series breaks off after a finite number of terms. so that to leading order we have VaX{0) = 0. a) —b — 2a. By a repeated application of the recurrence relation (16.a — l) — a — l.10) zm$(a)= ^ j=m Sm(a. where $(o.* Recurrence relations for coefficients of perturbation expansions for Yukawa. j)Smi(a. a + 1) = a — b+ 1. .awhere (a. (a.j)$(a + j).
[ a . when n = 0.1] 2 .a\4.13) one obtains the quantity An(K) and hence with Eq. [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).12a) where [a. (16. a . We first cite the final result .j). 0 < \j\ < i. (16.7) and hence in Ra may be cancelled out by adding to x a contribution /x<I>(a + n)/n except. #(a + ra) fjL$(a + n) V. so that 1 1 2K— ' (2W[a'a]2+(2^F[a'a]3 1 r i .[a>a + J]i+i*(a + 3).aj 2 (2K)A _ [a. 1 .<&(a + n) on the right hand side of Eq.[ a + l. + (16. .An{K).a + j]i+l = MiSi(a.e.aj 2 \ [a. (16. of course. [a. (16.7). Any term /j. oo _ . the latter can be written .n.a + l ] 2 [ .' — .l. for the coefficients M^> of the expansion Details can be found in the references cited above.5) the Regge trajectories I = ln(K). Va+n = Va .12b) The usefulness of this notation can now be seen in the ease with which it permits the calculation of any number of higherorder perturbation terms. i R »] = 2K[a. This equation is seen to be 0 = 7n7L a ' a Jl + P a + n $ ( a + n) = 0. This follows from the fact that D a $(a) = 0. and hence Va$(a + n) = ra$(a + n). Screened Coulomb Potentials Substituting the first approximation x^0* = 3>(a) into the right hand side of Eq.a]l^a) + ^2(2K)i+1 Y.13) + ••• • One can now construct coefficients with their recurrence relations for the individual terms of this expansion.o]i = M 0 . (16. i.324 CHAPTER 16. Evaluating the first few terms of the expansion (16.
2.1. The result is AnW = M0'[n(n + l)M2 + MlM0] <2» + ^ ^ 1 + — ^ [ 3 M 4 ( n . we obtain the expansion for the Regge poles.^ The plots confirm the expected behaviour for strongly attractive potentials but exhibit also a superficially unexpected departure into the lower half of the complex iplane in the case of the first few trajectories (counting in terms of the quantum number n) for weak coupling.§ This is an important observation which indicates the equivalence of the methods. 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.e. (16. Ahmadzadeh. G.1) + 3M3M02 + M22n(n + 1) (16.l)n(n + l)(n + 2) + 2M 3 M 0 (3n z + 3n . .1) +6M 2 Min(n + 1) + 2M2M02 + SM^Mo] + g^s [3M 4 M 0 (n 2 + n . i. Analytical Observe there the quadratic form of the centrifugal potential! "See in particular A. Obviously the expansion — which.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.1) + 3M3M02 + Mln{n + 1) +4M 2 MiM 0 ] + 0(K~8). However. Tate [6]. .^ p [ 3 M 4 ( n . Inserting this expansion into Eq.15) The same expansion may be derived by the WKB method from the BohrSommerfeldWilson integral for three space dimensions as explained in Example 16.14) +4M 2 MiM 0 ] + 0 ( K ^ 7 ) . (16. With numerical methods one can achieve more.1 by evaluating the first two terms. 16.( 2 1 ^ 6 1 ) [ 3 M 4 M 0 ( n 2 + n . is an asymptotic expansion for large values of the energy — is not very useful in the very interesting domain around energy zero. Burke and C.1) +6M 2 Mm(n + 1) + 2M2M02 + SM^Mo] . including an exploration of the domain of small energies E. Thus the expected appearance is that shown in Fig. of course.l)n(n + l)(n + 2) + 2M 3 M 0 (3n 2 + 3n . P.5).
0) = (a. Screened Coulomb Potentials expressions of the behaviour of a Regge trajectory in the immediate neighbourhood of E = 0 are practically unknown. and the angle A#. Frautschi [97]. a] 2 = M i Si (a. satisfies the relation ajAO ~ h. (16. through which the particle orbits during the course of the resonance. ±1) = 0. [a. 0).11b) we obtain Si(a. For a bound state a j = 0 and the orbit becomes permanent. and the lifetime At of the resonance satisfies the relation TAt ~ h.1 Typical Regge trajectory for a strongly attractive Yukawa potential. C. However. .0). Hence Si (o. From Eq.^ .12b) we obtain [a.O) = Other terms vanish since So (a. 16.a)So(a. this is a particularly interesting domain since at the position soon after this point at integral I (as at I — 2 in Fig. .—" + In = — . one can expect a resonance with the lifetime determined by this imaginary part. The conjugate variable to energy is time. Solution: We evaluate the first three terms of expansion (16.326 CHAPTER 16. For a resonance with a long lifetime. 0) and Prom Eq. In fact. 16.1: 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." Re I 1 0 1 2 Fig. just as a decay width T represents the width of the resonance in energy. Similarly the conjugate variable to angular momentum is angle. (a.a) = b . (16. 114. a] 3 = M 2 S 2 (a.2 a = In See S. so the imaginary part ai of I — an represents its width in angular momentum.13). p. Example 16. where the imaginary part of I is very small.1). ai is small and A6 is large.
14.b)(a .13) we obtain 0 (Mo .11b). using the Cauchy formula (6.6 + l)a + A.1) + (a . 1). 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 BohrSommerfeldWilson integral in which 1(1 + 1) is replaced by (I + 1/2) 2 .An) • [ M i M 0 + n(7i + l)M 2 ] 1 A n = M 0 . 0) + (a + 1. (a.2 V~Az and z \ zz dr\ A 2B C dz V'A + 2Bz + Cz2 (6. a)Si (a.65) 2m dz V'A + 2Bz + Cz2 • Z— 0.n ( n + l Inserting these expressions into expansion (16.16.a + l)S0(a.1. Thus (with terms which are 0): Si(a.6)Si(a.0) (a . A" An An A" = = = (a.1 ) + (6 . (n+l)U+^pJ . .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 . Sec.^ 2 M " + ! ) M 2 + M i Mo] + • E x a m p l e 1 6 . i ) again from Eq. Solution: Ignoring higher order terms in r.00 B + Cz 2ni VA + 2Bz + Cz2 B 2niI —= + vC with ambiguous signs of square roots I.1) Si(a. We obtain the quantities S i ( a .al)S0(a.0) = (a . a)Si (a. (16.2 a ) . a)Si (a. 1) (a .0) + 0 + 0 = ( a .a)5o(a.1 ) + (a.(ra + 1) [n + (2K)3 A n+1+• n K K (M0 .6 + 1 ) . 0) + a S i ( a .0) Si(a.A „ ) H Mi 2An 2K if n + (2i^): r M 2 2K Hence 1 . .0)+ 0 + 0 = (a1).2o)Si(a. (a.0) + 0 + 0 = ( 6 .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". .2 Regge Trajectories 327 Analogously we evaluate S2(a. One sets z = 1/r so that. 1) Hence 52(a. we have to evaluate (cf.
±2HK in agreement with our expressions above. Thus we can write down the S'matrix for the present case by exploiting the limiting case of the Coulomb potential. Here we do not perform this procedure.. however. We know that corresponding to every Coulomb Regge trajectory we have a corresponding one in the Yukawa case.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. 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 .328 Thus here dr K' CHAPTER 16.3 The 5Matrix 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. t?{l + \f M0 ±2K 1/2 2TT 2J 2V^2 (n + l + l)h = 2TT Mo h.5) together with the expansion (16.(iO mp~ ~ ' ' ' '+ l)M2 + M0Ml] (16.* 16.1)1(1 + 1)(Z + 2)M 4 + 2M3M0(3Z2 + 3/ . The turning points in this case are at r = 0 and r = oo.16) = n n=o 1 2 With this inversion we obtain = M0^^[l(l + ^ [ 3 ( Z . Boukema [34] and [35].15). *J.1) +61(1 + l)M 2 Mi + 3M2M02 + 3Mx2Mo] + 0(K~6). I. In the second paper the second order WKB approximation is used and shown to yield complete agreement with the terms given in Eq.14) for An(K) in order to reexpress the latter in terms of I so that i+i+ A. we can use Eq. tSee Chapter 11 and V.e. (16. i.2. n = 0. Example 16. Solution: For details of the solution we refer to papers of Boukema. . by expanding the right hand side in rising powers of v and evaluating the individual integrals. Singh [252].t First. Screened Coulomb Potentials Mo . (16.l. This shows that 1(1 + 1) has to appear in the BohrSommerfeld—Wilson integral as (/ + 1/2) 2 .
17) and is therefore real for real values of the potential coefficients Mj. Miiller [211]..4 The Energy Expansion We observe that Ai(K) has the property MK) = MK) 329 (16.. which are precisely the expressions yielding the Regge trajectories of above.17).16.2. and could then have carried out the perturbation procedure not in inverse powers of K but in inverse powers of y/K2 + M\.e. 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.l)n(n + l)(n + 2) + 2M2M02 . the following *See e. H. Expanding the square root VfC2 + Mi for \Mi/K2\ < 1 one regains ln(K) with the expansion (16. the behaviour of the scattering phase in the domain of high energies. § See H.19) +2M 3 M 0 (3n 2 + 3 n . n = 0.18) of the Smatrix now to explore further aspects. 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 . i. One can use the explicit expression (16.0[(KZ + Mx)'6} (16.4 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\. such as. Miiller [204]. This has been done§ and one obtains .19) we obtain the energy. Paralleling the case of the Coulomb potential in Chapter 11.. W.g. W. J. Now reversing the expansion (16.l (two additional terms are given in the literature). . We observe also that the S'matrix is unitary as a consequence of the result (16. for instance.l.14). . J.* 16.
p.l)n 2 (n + l) 2 (n + 2) 1 + • • • (16. Warburton [279]. W. Sommerfeld [256].1) + 2M2M. Iafrate and L.f 24(2n + l)(Z + n + l ) 5 M4M0(nz M06 M$n(n ^ M9 + n . B.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 socalled SommerfeldWatson transform. N. 'According to S.11) + 2M22M03(9n2 + 9n . A.< 3 M 4 M 0 ( n . in the following expression with an "See in particular G.l)n(n + l)(n + 2) .330 CHAPTER 16. Screened Coulomb Potentials expansion.l)n(n + l)(n + 2)(n + 3) +2M 5 M 0 3 {5n(n + l)(3n 2 + 3n . Large coupling expansions derived in H. J. E. the transformation was introduced in the form given below by G.3 M 2 V ( n + l) 2 M(o I +2M 3 M 0 2 (3n 2 + 3n .* We recall from Chapter 11 the definition of the scattering amplitude F(9. Zauderer [286]. in which the leading term is the usual expression of the Balmer formula for the Coulomb potential: K2 = Mi+ 1 ° l4n(n + l ) ^ ( / + n+l)2 4(7 + n + l ) 2 Mio +4(2n + l ) ^ ( / + n + l ) 3 4(Z + n + 1)4 ( l) ^ — . 282. i.e.^ 16. k) as coefficient of the outgoing spherical wave in the asymptotic behaviour of the wave function tp(r). Frautschi [97].10) + 12} + 4M3M05 +2M 4 M 0 4 (6n 2 + 6n .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 . VahediFaridi [212] contain as special case the expansions of the first pair of authors. Analogous expansions for specific energydependent Yukawa potentials have been investigated by A. J. Watson [280] in 1918 and later resurrected by A.1) + Af3Mff + 1) ^ lOM 6 M 0 2 (n .20) Extensive investigations of the energy eigenvalues for the Yukawa potential can be found in the literature. C. p. . Mendelsohn [135] and E. 107. MiiilerKirsten and N.2)(n .
.kmcose).fe)P .2..k) 0ikr (16. \x\ <C l . (cos 9).k) = 2zfc L " v "'' v .23) as the contribution of the residue of a pole in the plane of complex I of some suitably constructed contour integral. (11. I = n + x. dn Iml 1 I c ( \ x o x 1 A 2 A 3 x4 X 5 Rel Fig.1.k) = ^T(2l + !)/(*..5 The SommerfeldWatson Transform 331 ingoing plane wave in the direction of z (here with h = 1. c = 1. ~:' sin nl (16. z=o in which (cf.187)) (16. Eq. 16.23) f(l.iJ 11 —[S(l..25) Setting Note the minus sign in the argument of the Legendre function.k)~l] = ~ k" PiSiW sin 5i(k) and S(l.184) to (11.24) Si(k) being the phase shift.187). so that sin KI = sin7r(ra + x) ~ (—l)nKX = (—l)lir(l — n). n = 0.21) The scattering amplitude determines the experimentally measurable cross section o given by da (16.TOO= 1/2): ip(r] r ~*°° Jkz elkz + F{0.22) = \F(9.k)\2.2 The integration contour C. For this purpose we consider the following integral taken along the contour C shown in Fig. Actually 5j lrr/2 as in Eqs. 16.k) — e 2i5[(k) ~" p—iirl (16. (11.''' The idea is now to consider each term of the expansion (16.2 in the complex Zplane: *<«•*> = s / c c K <u(21 + 1) vf{i. . The scattering amplitude possesses the partial wave expansion F(0.26.
16. we obtain m® = ± I dl^±^f(l.—) s = (p + q)2.28) . as indicated in Fig. Such a direct reaction and its "crossed channel reactions" are shown schematically in Fig.332 CHAPTER 16.3. a + c —> b + d.27) a(p) © b(q) Fig.—.n) oo {l)1 Pt{cosO) v ' (16. t = (qq')2. 16. For the momenta indicated we set (with metric +. Thus the direct reaction may.fc)P„(cos0). 16.3 Reaction channels and their respective Mandelstam variables. Screened Coulomb Potentials Then integrating with the help of Cauchy's residue theorem. for instance. u={pq')\ (16. d(q') (16.26) P((+COS0) = 2 ( 2 n + l)/(n. we have to digress a little and introduce a few simple ideas which played an important role in the development of particle physics. Consider the reaction of a particle a{p) with fourmomentum p^ colliding with a particle b(q) having fourmomentum q^ and producing a particle c(p') and a particle d(q'). In order to see the relevance of Regge trajectories in a reaction of particles.k) Zi Jc TT{L . in which case c = a and d = b. a + d — • c + b.3.—. 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. n=0 which is the usual partial wave expansion. be elastic scattering of a off b.
u = . if we had been considering the s reaction originally. S. Khuri and S. § See G.et) (16.^ (b) Chew's hypothesis: All (composite) particles lie on Regge trajectories.u) (if any of the external particles has nonzero spin. (16. there will be several such amplitudes which together describe all three processes). Symbolically we then have the situation shown in Fig.5 The SommerfeldWatson Transform 333 and for simplicity we assume here that all particles are spinless and have the same mass mo.32) t = 4(q? + ml). the variables t and u would describe momentum transfers in the crossed channels. Mandelstam [184] and R. s = 2qf(lcos0t)(16. the momentum k there would correspond to qs) s = ~4(ql + ml).u are not independent.u channels) are determined by one and the same relativistically invariant scattering amplitude A(s.c o s 0 s ) . u. etc. the dynamics is described by the exchange of quantum numbers. if s describes the square of the total energy in the s channel. e.. N. . 16.31) i l Taking all particles to have the same mass mo.The variables s.t) = VtFt(t. t = 2q£(i .(<fe)^(cos 6t).§ Since the three variables s. B.29) and with selfexplanatory meaning for the cross section of the reactions ~ = \Fs(s. A(s. one can write the amplitude as depending only on two. Thus. N. For a Lorentzinvariant normalization of the initial and final states. except for poles and cuts which characterize the reactions in the three channels. In a Regge theory on the other hand. T See e. In the Yukawa picture of a reaction the dynamics is described in terms of the exchange of mesons IT (called pions).g. the only kinematics we require in the following is given by the relations (e. M. the quantities carrying these quantum numbers are the Regge trajectories. one then has A(s.16. Ft = £ ( 2 J + l)F. Chew [48]. t. F.4. Treiman [25]. (a) Mandelstam's hypothesis: All three processes (described by the s. One now makes two hypotheses.u are known as Mandelstam variables.0s). which is an analytic function of the variables s.9s)\2. L. Goldberger. Blankenbecler.g.t) = y^Fs(s.t.2 ^ ( 1 + cos6 s ).30) with partial wave expansions (observe the noninvariant factor k has been removed) Fs = ^ ( 2 2 + l)F z (g s )PKcos 98).t).g.t. A(s.t. (16.t.
4 Yukawa versus Regge theory. Thus one is interested in establishing a representation of the amplitude in terms of ichannel Regge trajectories. 16.5 The Lehmann ellipse. . But > > wave expansion X)(2Z + l)F.5 for Yukawa potentials V{r) u: lm cose dr a(/j. For s — oo : cos 9t = 1 + s/2q^ — oo. the Legendre polynomial Pj(cos 9) does not possess a cut. However. 16.((fc)/Kcos0t) schannel ichannel schannel variables the partial converges only within the socalled Lehmann ellipse^ which is shown in Fig. So what can one do? 1 H. Screened Coulomb Potentials Regge oc(t) Fig. in the domain of schannel physical values of the kinematical of the ichannel). the partial wave expansion is not valid in the physical region of the (i. It is known from the Mandelstam representation that the amplitude possesses a branch point at cos 9 = 1 + TOQ/2(^. Lehmann [162].)e~ •fj. But for integral values of I.e. 16.r 1+m 2 /2q 2 o ^t "" Re cost: Fig.334 Yukawa CHAPTER 16.
Thus consider now the same integral but taken along the closed contour C shown in Fig.26) but for a different choice of the contour.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. and there is only a finite number N + 1 in the domain of 9W = . i.e.1 / 2 . Then / JDD' ••• + j JBB' • • • = 2TTI V ^ residues f5n.e. contour C / / D ^ ^ .6. 16. Again we have / Jc> • • • = 1ni y^ residues j3n „ For Yukawa tions A and f(l. sin nl [16. Moreover. to use the SommerfeldWatson transform.6 The integration contour C".t). 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.^ B 1 A Fig. N F(6. „ i.33) . B Iml 1 \ i All \ r\ 0 / closed .5 The SommerfeldWatson Transform 335 T h e answer is. 16. all poles of Ql > 0.16. the contour representation (16.
i.e. (16. J. ^M)^E i 2 a f: a „t ( t ) ^ m <^'F(s.32)) and finite. Bose and H. Foissart [223].35) This result is valid for s — oo and t negative (cf. Regge trajectories have. Apart from more refined details. In particular.e.336 CHAPTER 16.6 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. at high schannel energies. Eq. The > result demonstrates that the high energy behaviour of an schannel reaction is determined by the leading Regge trajectory in the crossed ^channel. we have (16. if ao(t) is the Regge pole with largest real part. Screened Coulomb Potentials Since for s — oo : cos Qt = 1 H n ~* °° > 2?* and (from Tables of Special Functions) farMoo: Pa(z)* ^+l]\(2zY. Potentials with a shortrange repulsion more singular than 1/r 2 have been considered by N. K. for plots and discussion see R. of course. The slope of the Regge function an(k2) is an " Simple cases are the harmonic potential and the squarewell potential. Omnes and M. p. 42. as we discussed briefly in the above. for high energies in • > the schannel). (IM4) This is the asymptotic Regge expansion of the invariant amplitude for s —>• oo. 16. Regge [229]. 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. this behaviour has been confirmed in high energy hadronic reactions.t)&B(t)sa°V.e. W. Predazzi and T. the amplitude can be represented as a sum over ^channel Regge poles. It follows that under these conditions. i. MiillerKirsten [29]. In subnuclear physics they played an important role before the advent of the quark idea and thus of quantum chromodynamics. . we have P\/2+IR{Z) ~ 0(l/y/z) — 0 for s — oo (i. The logarithmic potential has been investigated in a way similar to the Yukawa potential above by S. Limic [177] and E.
a composite particle being one with structure. 20. 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. For an overview see e. and also as socalled "fixed poles" .16. the latter are related to the distinction between "' elementary and "composite particles". The above treatment of screened Coulomb potentials is incomplete. MiillerKirsten [209]. The small imaginary parts of eigenvalues or lifetimes of resonances have not yet been calculated. like a meson or a baryon. H. J. called "Regge cuts".6 Concluding Remarks 337 important parameter in string theory. W.1. .** Whereas the former are related to absorptive properties of a reaction. Related aspects are discussed in Sec.g. and is composed of quarks. Singularities appear also in the form of cuts in the plane of complex angular momentum.
This page is intentionally left blank .
Taking in the ellipsoidal wave equation the limit k —> 1 (fc elliptic modulus) and putting tanhz = sin 6. depending on the parameter k. is effectively the Mathieu equation whose solution has for a long time been considered as being very difficult. the ellipsoidal wave equation reduces to the trigonometric form of the spheroidal wave equation. 2K or AK. See F.lP = 0 in ellipsoidal coordinates leads to the ellipsoidal wave equation with sn 2 z and sn 4 z terms. M. In essence these periodic Jacobian elliptic functions. K being the complete elliptic integral of the first kind. Thus Jacobian elliptic functions also pro*To be precise: Separation of the wave equation Aip\x2.Chapter 17 Periodic Potentials 17. Arscott [11]. and are themselves periodic with period e. separation of Laplace's equation Ai/> = 0 in ellipsoidal coordinates leads to the Lame equation (no sn 4 z term).* These elliptic equations involve Jacobian elliptic functions like sn(a. are not much harder to handle than trigonometric functions.or 27rperiodic trigonometric functions on the one hand and the nonperiodic hyperbolic functions on the other. the most immediate candidate is a trigonometric form like that of the the cosine function. 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 Schrodinger equation with this potential.1 Introductory Remarks Prom the beginning of applications of quantum mechanics periodic potentials of various types were immediately considered. since many analogous relations hold. 19 and p. p. The reason for these difficulties is. example 14.0 < k2 < 1. 25. motivated mainly by the regularities of the crystal structure of matter. called elliptic modulus. however. Apart from the discontinuous KronigPenney potential consisting of a periodic repetition of rectangular barriers.) which are functions that interpolate between the n.g. 339 . that the Mathieu equation lies outside the scope of equations which can be reduced to hyper geometric type.
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. Like the trigonometric functions sine. 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. those of integral order.340 CHAPTER 17. J. (17. and to relate these to the weak coupling bands or regions of stability.. Lame equation: H. W. such as sn 2 z + cn 2 z = 1 and double and half argument formulas etc. Periodic Potentials vide periodic potentials like trigonometric functions. J. In fact. 8. cosine and tangent. 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. Both of these domains are important for a host of other considerations. The Mathieu equation. or 5wave Schrodinger equation with the cosine potential is conveniently written d2v —  + [A . 17.1. k2 — 0 which means that in > > this limit the Lame polynomials degenerate into the periodic Mathieu functions. ''The reader who encounters these Jacobian elliptic functions here for the first time.2 . which is conveniently written ^ + [\2K2sn2z]y = 0.7. these periodic potentials may be approximated by series of degenerate harmonic oscillator potentials. the three Jacobian elliptic functions s n z . Here the Jacobian elliptic function sn z is a periodic function analogous to a sine function and has real period 2K} For n = 0. (17.. 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. i. See also the discussion in Sec. [205]. For large coupling. which does not possess any polynomial solutions. or Sw&ve Schrodinger equation with elliptic potential. We might mention already here the much less familiar but more general Lame equation.1. Muller [73]. each with its own characteristic eigenvalue. Dingle and H.4. B. This is a significant difference compared with the Mathieu equation.e. need not be afraid of them. Miiller. .2) where K2 = n(n + l)k2 and n real and > —1/2 and 0 < z < 2K.1) where h2 is a parameter and — IT < z < IT. the Lame equation possesses 2 n + l polynomial solutions. c n z and d n z are handled with analogous formulas. The most important formulas for our purposes here are collected in Appendix A. W. which one looks up in books like that of MilneThomson [196] or Tables when required.2h2 cos 2z]y = 0.
. as in Fig. Although the mathematics literature on the subject uses the physical concepts of stability and instability. we compare the For interesting related discussions see M. 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. Vachaspati [242]. i. these concepts are rarely explained as such there.17. The orbits of planets are stable in the sense that deviations from the recurring elliptic orbits are small.e.e. Salem and T. 17. the semiaxis A > 0.^ 17.1 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. since the solution there is of the form cos vAz.2. Looking at the Mathieu equation for h? —> 0. 17. y" + \y = 0(h2). the case of h2 small. we see that in a plot of h2 versus A. In classical mechanics stability is treated for instance in connection with planetary motion. a bounded function. Here an important aspect is the determination of the domains of stability of the solutions and the boundaries of these domains. h2 = 0 belongs to the domain of stability. i.1. Thus a bounded trigonometric solution is indicative of stability and an unbounded exponential solution of instability.e. 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. The instability of an orbit would be evident either from an unbounded spiralling away to infinity or from a collapse into the centre. i. we see that for sufficiently large values of h2 the periodicity or boundedness of the solution is destroyed and hence becomes one of instability. and one can imagine that a nontrivial parity factor exp(in7r) then turns into a complicated phase factor exp(iz/7r). Adding the negative term — 2h2 cos 2z to A.e. i.2 Cosine Potential: Weak Coupling Solutions 341 seen to rule out polynomials. Our considerations below are deliberately made simple and detailed because their treatment in purely mathematical texts requires more time to become accustomed to. 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).
2/(0)' yfr) .e. y~(z) = 6sin vXz. Setting z = 0. i.4) with (for convenience in connection with later equations) y'+{z) y'_(z) = avAsin\/Az = = bv\cos^f\z = —y(z). d2y(z + 7r + [A . a = const. ° ^ [1 + 0(h2)}. b: y(z) = A cos vAz + B sin \f\z and y+(z) = a cos yXz. y(2vr) 2 a = —7—^ = 1. (17. we obtain y(vr) y(2?r) a = — — a = —^^r.e. d(z + TT) cos 2z We see that there are solutions which are proportional.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. We set (here and in the following frequently apart from contributions of 0{h2)) with constants A. A solution of the second order differential equation is determined completely only with specification of boundary conditions which determine the two integration constants. — • .2h2 cos 2z]y(z) = 0. sin V Az or linear combinations. (with Tn as translation operator) Tlxy{z) = y(z + TT) = cry{z).2/i2 cos 2(2 + vr)]y(z + vr) = 0.B and a. (17. TT. Periodic Potentials equations ^ p + [A . 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).342 CHAPTER 17. i. y+{z) .
\y+>y\ = y+{ir)y'{ir) . whereas the solutions y+ and y_ have here been chosen specifically as even and odd around z = 0 respectively. Using Eq. y{z) = aty+(z) + 0y(z). (17. i. we choose these with the following set of boundary conditions: y+(0) = a.3) we obtain y(z + ir) = a[ay+(z) + @y(z)]. with constants a and /3. We can write therefore.y_] = y+(0)y'_(0) — y_(0)y+(0) = ab\f\.y{ir)y'+{K) = abVX in agreement with its value at z = 0 above.{ay'_{ir) + bV\y+(ir)}a Now from Eq.y+(7r)j/_(7r)} = 0. The roots a± are therefore obtained as 2bV\y+(ir) ± J4b2\yl(Tr)4. This means.oWX] . yV(0) = 0. [y+(v) .4). Setting in these last equations z = 0 and using the previous pair of equations and the boundary conditions. This expression is equal to the coefficient of a2 in the quadratic equation for a. we obtain V (TT) = ay+ (TT) + (3y _ (TT) = < [aa]. 7 y\z) = ay'+(z) + (3y'_{z). y'(z + n) = a[ay'+(z) + py'_(z)}. ay'~(ft) — bVXy+(n) and the Wronskian w + {y+(7r)y/_(7r) .17.W or fy+(n)aa J/_(TT) = W ^ ^ J a ^\ = n For linear independence of a and /3. y'_(Q) = bV\ with Wronskian W[y+.2 Cosine Potential: Weak Coupling Solutions 343 The solution y(z) is an arbitrary solution. 2/_(0) = 0. (17. 2/'(7r) = a ^ W + ^ .e.a2b2\ a± ~ 2a6\/A .y+(7r)y_(7r) = 0 or abVXa2 . and by'+(?r) = —ay/Xy{n).aa][y'_{n) . the determinant of the matrix must vanish.
(17. 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.6) We can derive various terms of this expansion perturbatively with the method of Sec. 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. i. this implies y+(Tr)/a = ± 1 . (5i/ 2 + 7)/i 8 vz + — .+ 2 2(i/ l) 32(i/2l)3(v24) ( 9 ^ + 58.455A2 + 1291A .5) is that for h2 = 0 we have v = A/A and hence more generally v2 = \ + 0{h2). .7. This is the easiest application of t h a t method since only simple trigonometric expressions are involved. i. T h e result for A is the expansion A = h4 2 .e. 2 _ hA 2(A1) (13A25)/i8 32(Al)3(A4) 12 ift (45A3 . Periodic Potentials g= 2aby/\. **This point is not immediately clear from mathematics literature.22)) the boundary conditions of periodic solutions.1169)/i 12 64(Al) (A4) (A9) 5 2 + 0{hw) r 16 (17.5) a The parameter i/ is known as Floquet exponent.e. <T__ + <T_ = 2y + (vr) . we have <7± — CHAPTER 17. (17. 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 . <7__cr_ = 1.0 This series may be reversed t o yield the Floquet exponent.. Eq.1.8) "Observe that for integral values of v (in lowest order of h2). Setting cr+ = e . 8. cr+ = 1/<T_.344 Setting / = 2by/\y+(Tr).= — . _ / ± v 7 ^ V . Prom this condition we determine later (cf. This calculation is demonstrated in Example 17. where the constants are usually taken as unity from the beginning.2 + 2 9 ) ^ 2 + 64(1/2 _ 1 ) 5 ( z / 2 _ 4 ) ( l / 2 _ 9) + u ^ ^ ^ . (17. (17. . ** An important consequence of Eq.
— 2/i 2 A and insert this into the Mathieu equation y" + [A — 2h2 cos 2z]y = 0. These cases therefore have to be dealt with separately. VA = ±n.9a) 4(1A)>/A 64(A4)(A1)3A^A One observes immediately that these expansions cannot hold for integral values of v or A. 8. With a perturbation theory ansatz for y+ around h2 = 0 as in Example 17.9a) and is therefore not derived here in detail^ ri/ = VA + rCOS7Tf = COS7TVA + /l 4 h4 7rsin7rVA = 4\/A(A .„„ x n.e. (17. It follows that in these cases i? = n2[l + 0(h2)]. (17.1) 15A2 .1 below one obtains the following expansion for cos7rz/ which is fairly obvious from Eqs.5) and (17. Meixner and F. n an integer. W. (17. = J—^ + o(hr). Solution: We write the eigenvalue equation A = v2. (17. For these integral cases of v we demonstrate in Example 17.jl2.35A + 15A2)/l8 /.2 how the expansions for small values of h2 may be obtained perturbatively. Schiifke [193].12) S e e J.7 to obtain the eigenvalue A as a perturbation expansion for nonintegral values of v and b? small. To lowest order the solution y is j / ° ) = yv = cos vz n where d2 Dv:=—~+u2. as will be done later. p. which can then be rewritten as Dvy = 2fe 2 (A + cos 2z)y. (17. Example 17. 124.2 Cosine Potential: Weak Coupling Solutions and so 345 (8 . cos TT\/A 32A(A1) 2 . . (17.9b) TT2 Next we observe that if we demand that the solutions y+ (z) and y_ (z) satisfy the condition (17.35A + 8 .3).11) or smvz or e±ivz.1: The eigenvalue A for nonintegral v and h small Use the perturbation method of Sec. we must have a2 — 1 and hence e 2iv/A7r = 1} i.17.10) A = n2. rr +h* 3 =7rsin7rvA 64(A1) (A4)AVA +Q(h12).
15) Hence a term fMy„+a on the right hand side of Eq.e. of course. v . = .h2).2. The recurrence relation can also be obtained by substituting the right hand side of Eq. i.3 and 19.v±2) = l. h2(u. v)yu + {v.2) which applies when v is nonintegral. (17. Schafke [193].v— 2) 2 ^ + yu+2 Vv 2 y (D = h _2(2u2) 2(2i/ + 2 ) ' Then up to 0(h2) the sum j / 0 ' +y^ is the solution provided the remaining term in Ri. Periodic Potentials The complete solutions of these cases are written respectively in selfevident notation cev(z. where except. (17.2) _ __ _ (y. v) = 0. v .(2) + . '. siniy + 2)z + sin(i/ — 2)z. or A = 0(/i 4 ). so that also Dv+ayv+a and hence Dv+a = Dv + a(2v + a) and Dvyv+a = a(2u + a)yv+a(17.16) directly into the original equation in the form of Eq.14) therefore leads to the following nextto leading order contribution (v. h2).2)j/„_ 2 + {v. (i/. .16) *These smallh expansions are convergent.h2).14) = = = cos(y+ 2)z + cos{y— 2)z.e. up to 0(h2) we have A = 0.v + 2)yu+2]. Next we treat terms yv+a in j / 1 ' in a similar way. W. since j / 0 ' = yv leaves unaccounted R\. . ("• ^ + 2) (o) 2(2^ . 17. see Example 17. as shown.2) — 2 "^ 2(2i/ + 2) " + 2 vanishes. We observe that in these cases 2cos2z cosvz 2cos2z smvz 2cos2ze±il/z i. in each case 2cos2z2/„ = 3/„ +2 + j / „ _ 2 . .!/) = 2A We now observe that DvVv = 0. for instance. We assume for the time being that a ^ 0 and 2v + a ^ 0 (the latter case requires separate consideration..2 — we can write down recurrence relations for the perturbation coefficients pa(2j) together with boundary conditions. when a = 0 or 2v + a = 0. The right hand side of Eq. sev(z. e ± i ^ + 2 ' 2 + e±i("2)z. se and me as* oo i y = „«>) + yW + i.346 CHAPTER 17. i. (17.11). (17. (17. Meixner and F.„ + £ h2i £ P2i(2j)yv+j (17. With our perturbation formalism — as demonstrated in cases considered in Sees. and (v.11) terms amounting to i?i 0 ) = = 2 h 2 ( A + cos2z)y„ = 2h2Ayu + h2(yl/+2 + y1y2) h2[(v. by J.e. (17. (17. i. so similarly yW leaves uncompensated / # > = ft2 ("• v ~ 2 ) R (o) . me±v(z.e. Proceeding in this manner we obtain the solutions ce.13) The first approximation y(°) = yv leaves unaccounted on the right hand side of Eq.11) and so in R}?' may be cancelled out by adding to j / ' 0 ' the new contribution a(2v + a)' = 0.3.
thus determining the quantity A.e. the solutions are not yet normalized. Example 17. Solution: We start as in Example 17.7). The higher order terms naturally require a little more algebra. This completes the determination of the solutions for nonintegral values of v in the domain of small values of h2. Hence we obtain 0 = /i 2 (2A + l ) f h4 + ••• ."2).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. Meixner and F. * Note.17) Inserting 1 for the stepcoefficients of Eq. For v = 1 we then have yi = cosz = y~\.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]. (17. J.17. i. fo^ + 2) ..2 ) v(y2.1 with A = v2 — 2h2A but with v = 1. „ . In the case of integral values of i>. .14).g. . (17. one has to deal with each integral case separately as in the following example. 2 ( 2 i / . + y~i) (17.. is set equal provided the coefficient of the sum of the contributions in 3/1 contained in R\ to zero. we obtain immediately 0 = 2Ah2+ 2(^T)+' which verifies the term of order h4 in Eq. 0 = h2(u.. . Following the first few arguments of Example 17.. . Schafke [193]. v) ' ( 2 v + 2) (17.2—r——— (i/ + 2.v)\ . 121. These expansions are actually convergent with a definite radius of convergence as explained in the mathematical literature.R{ .2 Cosine Potential: Weak Coupling Solutions 347 together with the equation determining A. (17. p..u) + h4 (". See e. W.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)+. however.18) Again applying Eq.
and v is to be found in ascending powers of h2. . The specific normalization here is taken as I /*7r rn 1 = — / dzce2(z. 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]. h4 (17. Meixner and F.e. Periodic Potentials h4 r ••• . the case of A an integer. see below!) one obtains the additional terms given by Meixner and Schafke [193]. 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.* The associated solution is ym+ym. m.n integers. 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 . In the above we considered the case of v an integer and calculated the eigenvalue A.19b) This expansion agrees with that given in by Meixner and Schafke [193] (p. The reverse situation is later of importance. (17.348 From this we obtain 2h2A = h2 CHAPTER 17. With normalization (not to 1. 123) (there the solution is called cei(z. We introduce a normalization constant c and rename the normalized solution y then yc. 120. Inserting this into A = v2 — 2h2A 8 for v = 1 we obtain (cf. J. We consider this case again in an example. 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. *" Jir Thus we have and uses / dz cos mz cos nz = — <5 mn . W.24c) below) .19a) This expression agrees with the result given in the literature (there the eigenvalue is called ai). Eq. Schafke [193]. h2). i. p.
o 1• • • . Manvelyan. Substituting these expressions into Eq. we obtain^ S 3 V 2 and therefore i/ = 2  iy/E ( h 3 \2J 7i .. MiillerKirsten. (17. . Gubser and A. 128A2(A1)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 . J . S.= l 2vA sin27r + (A — 4)(cos vA7r)x = 4—= + ••• = j= h •• • .17..9b) about A = 4. Hashimoto [124]. We thus obtain the expansions cos"\/A7r sinvAvr = = cos27r + (A — 4 ) ( — s i n V / A 7 T ) A = 4 — = + . (11A2 . J.Q . Liang and Yunbo Zhang [189].3 1 A + 8) ( A .1) (A .1)2/A TT2 +h8 (15A2 .3 1 x 4 + 8) 274233 (. fe47T2(A4) 4%/A(A .4 ) + .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. so that 2 2 cos nu = cos 7r(2 + S) = cos 27r cos TC8 = 1 and comparing with the above. on the left hand side of Eq.('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 . We set A — 4 ~ 4e and expand the cosine and sine expressions appearing in Eq. .4)Av A2v A ~ 32A(A . ^S.9b) v — 2 + <5. W.3 5 A + 8)TT2(A4) _ 3 / / 64(A . H.9b) and considering the approach A — 4 gives > (A4)2 + .2 Cosine Potential: Weak Coupling Solutions Example 17. (17.• 1 + 5n h 2 s 2932 Setting in cosm.1) . (17.l ) 2 (A .4) 2 7T 2 /l47T2 + • 8 h87T2 8A(A .
"These conditions.350 CHAPTER 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. (17.6. solutions for integral values of v.29. as also conditions (17. We now derive these boundary conditions.w^. We verify these for the large/i 2 solutions in Example 17.. Arscott [11].21a) and (17. Periodic Potentials Our next immediate aim is to specify the solutions for which Eq. .2*0 y~(z ± 7t) = 6 sin VA(z ± TT) = y_(z)cos VA7r ± 6 cos vXzsin v\ir. Fig. pp. M.10) applies. = and s+w 2tW±I.4) into account. 28 .1 Boundaries of domains of stability.21b). (17. concentrating again on the leading term in the even and odd solutions y±(z) for ease of understanding. Taking in particular Eqs.2. i. for instance. bV\ y+(0) 2/40) 07. we have" y+{z ± 7r) = a cos vA(z ± TT) = y+(z) cos vXir = TV{Z) F sm v<\7r.W4W=UtWSta±I(. 17. 17. These are defined by specific boundary conditions. may also be found.e. in F.
Then y+ y 351 U . f^\y+(j) /vr\y_(7r) and for the derivatives ^(i)^^'(D^ . ^\y±M /vr\yL(7r) . .17. Then y+ K 1 _ 2/+00 7T y V + (f)[i + 2/±Ml ^ ] y'il) 2y_(f )i/+(f.21a) Since (with the help of the Wronskian) *W4)MM?)^ we also have y+(v0 _ . /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. we obtain y+(7r) _ i  2y_(f). . o y+(fy(f)ob\/A _ = 1 + 2I&A/X . 1+ y+(f)^(f)y_(f)yV(f)' 7T Using the Wronskian (which we actually had above!) W[y+. /7r\j/!_(7r) y L l 7TJ .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) _ .4(f) (17.y V+ \ 2 7"^ V 2 fla^+U a cos V A  ] 6^A + a&( sin\/Aj ^A = abV\.
5) that the left hand sides of Eqs.' fh2 V2ceo = l . was treated in detail in Example 17. Finally we cite here some expansions of the eigenvalues along with those of the associated solutions (an explicit example. (17. 2K.352 CHAPTER 17. in this order. (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 . . Periodic Potentials and hence** a afc\/A For integral values of v (in lowest order of h2) we know from Eq. IT. and whether the function is even or odd. y > se2n+i=qo. The ordering of the eigenvalues which then results is a0 < bx < ai < b2 < • • • < bn < an < • • • for h2 > 0. of v is even or odd.• a2n(h2). (17. 17. These are defined by the following boundary conditions.. This is shown schematically in Fig.. A *• b2n+2(h2). y > ce2n+i=qo. (17. (17.2): hA 7h8 29h12 + Y l28^04.. We see that there are four possibilities and hence four different types of functions.21b) must be ± 1 . sei = sin 8 64 /i4 h6 a\ = 1 + h2 — . se2 = sin 2z 2 12 13824 ' h2 z —— sin 3z + • • • .22) * a2n+i(h )..+ . 27r. 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.. 17. These may be subdivided into classes depending on whether the integral value n — 0.1. There the second term on the right is nonzero. Mr7nA (17. .T + . ce± = cos 8 64 h4 5h8 b2 = 4 1 . A > b2n+i(h2). (17. y > se 2n +2s 9o +i The functions so defined have respectively period 7r. A 2 y • ce2n=q0i.2.— — — • • • .24d) v 12 ' **To avoid confusion we emphasize: In Sec. (17.1 for the first few eigenvalues which are boundaries of domains of stability (as discussed in Sec.24b) 8 h2 z —— cos 3z + • • • .21a) and (17.24a) ao = hA h6 b\ = 1 — h —— + — — • • • ..3.1). the case of ai with solution cei (of period 27r).23) We can now see how the domains of stability and instability arise.2 we require this relation for nonintegral values of v. 19.24c) 8 h2 sin 4z \ .
3.hz — 12 . We can see from the inverse of Fig.z3 ce2 = cos 2z . 13824 . 17. . 17.> (17. .3 Cosine Potential: Strong Coupling Solutions „ 5/i4 a2 = 4 H 12 763/i8 . Miiller [73].3 17. 17. In the case of the cosine potential. B. W.2 suggests that for very large fluctuations the potential can be approximated by a number of independent and infinitely high degenerate harmonic oscillator potentials.17„. Dingle and H. Consequently one expects the eigenvalues in the large h2 domain to be given approximately by those of the harmonic oscillator. . J.2 The cosine potential.17. Thus in Fig. 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 laigeh2 domain.1 that the boundaries of regions of stability merge to lines. 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.2 or by calculation that the potential 2h2 cos 2z has (harmonic oscillatorlike) We follow here R.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. 17.2cos4.1 Cosine Potential: Strong Coupling Solutions Preliminary remarks In the case of large couplings h2 we observe from Fig. 353 . 17. the walls are not infinitely high and hence tunneling occurs from one well into another.24e) 17. The cosine potential cos 2z depicted in Fig. of course.
4 (17. we therefore set 9 A A = 2h2 +2hq + —. but only approximately so. as in the case of the cosine potential. In the case of finite heights of the potential barriers. there is tunneling from one barrier to the next.25) This is a Weber (or parabolic cylinder) equation which we encountered earlier (cf.... 8.. n 0.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. Thus we rewrite the Mathieu equation as y" + A + 2hz cos 2 \z ± 7T y = 0 and take c o s 2 [ z ± . of course.2.1.354 CHAPTER 17.26) into the original Mathieu equation and obtain y" + 2h q + rrxr . Periodic Potentials minima at z — ±ir/2.3. (17. Taking the remaining terms of the cosine expansion into account. Ion (17.27) . Sees. Thus in this case the quantity on the left is not an exact odd integer.3) and is.2 The solutions We insert Eq. simply the onedimensional Schrodinger equation for the harmonic oscillator.26) 17. We therefore wish to expand it about these points. and we set this equal to q. where A / 8 is a remainder.2 and 8.2/i cos z V = 0. (17. This equation has normalizable solutions for X + 2h2 2h q0 = 2n + 1.
(17.3 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). can be . 16/i/ (17.17.29) is known.+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.28) A" ±4hcos zA' + 2h[ q =f sin z \ A. Choosing y(z) = A(z)e2hsinz.30) D A): COS « = 4z + 1 iq sin 2. 2 (17.27) we )A = 0. z). the linearly independent one. another solution.29) the equation can be written in the form D^A where 1 2 6 /i (16A" + 2 A ^ ) . 17.31) We make the important observation here t h a t if one solution of the type (17.3 Schematic picture of Mathieu equation eigenvalues. (17. (17. and substituting this into Eq.
2 /i D^Ag = 0. Aq+4 2Jh ._ 4 .36) Then ^4. when i = 0 — the terms not involving Aq in (17. With some algebra one finds that 16^ + where (9.4 ] . (17. ^ ) A + ( ? .9 + 4) = (9. (9. ? .33) (17. 2[(92 + l) + A].356 CHAPTER 17.94) = (9 + l)(9 + 3).(°) + A^ is the approximation of A to that order provided the coefficient of the term in (17.35) But since and so a term iiAq+n on the right hand side of Eq.30) amounting to Ri ] q = .9) (9.^ 6T [(«>« + 4 ) A ? + 4 + (Q. Periodic Potentials obtained by changing the sign of z or alternatively the signs of both q and h throughout. .g + 4) .9) = 0. (9.35) can be cancelled out by adding to A^ the next order contribution A& = 1 (g. 2 A A = («.35) not yet taken care of is set equal to zero.4 ) ^ .e.Q)Ag + ( ? . i.32) cos^'+i) [\K .34) The leading approximation A^ = Aq therefore leaves uncompensated terms on the right hand side of Eq.30) can be cancelled out by adding to A^> a new contribution — fiAq+4i/2i — except. D{q%Aq+4i = 0. (9l)(93). (17. of course. (17. 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. (17.94) \ — Aq4 (17.\z To lowest order the solution A is A^> = Aq. ? .4)A. 1 + 4 ) ^ + 4 + (^.
g4) 4.38) Aq8 + 1 2 Clearly we can continue in this way. . We note here again that if one solution y = A(z)e2hsmz 2hsmz is known. q.4 . except those involving Aq. (17. ) ^ 4.4.4) 1 1 (q. h) = A(z.3 Cosine Potential: Strong Coupling Solutions 357 This equation determines the quantity A in the eigenvalue equation to the same order of approximation. Then for A = A(0)+A(D+A(2) + . (17. q . the sum of the terms in Aq in Eqs. Thus 0 = (g. A(z. Thus in particular A(z.)+ & ! = « ( . In its turn the contribution A^1' leaves uncompensated terms on the right hand side of Eq. an associated solution y — A(z)e~ with the same eigenvalue and thus the same expansion (17.17.<J+4 Aq+& + 1 2 1 ..30) and hence y = Ae2hsmz to be a solution of the Mathieu equation. q . A.30) amounting to RW 9 = fag+ — 2?h 1 ^ 2™h? +< 4 ) R(o) \ .35) and (17.4) (q .g) 26/i L_ 2 13 /i 2 ^±A{q + ^q) + —1 ^A{q^q) '17.g + . +4.q + 4) (g. q.37) Here all the terms..37) and so on — left uncompensated — must vanish.q + 4)(q + 4. q .q 8)Aq+8 (o)' ^±V(q 1 4. h).8)A. h) = A(z. (q. Q . q.g + 4)(g + 4. (17._ 8 (17.4) (q . q. . (17.39) which is the equation from which the quantity A and hence the eigenvalue A is determined. (q.40) .q 4)Aq+4 q + 4 1 q + + + + + <^(. to satisfy Eq..39) is obtained by changing the sign of z or alternatively the signs of both q and h throughout. h). can be taken care of by adding to A^ + A^ the next order contribution A^2\ where A (2) 2h u 2 (q. . q (17.4.
e.45) {B) We recognize Dq as the differential operator of the Hermite equation. i. 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. this means the solutions are valid in particular around z — 0 since then l / \ / 2 3> l/Vh ~ 0. 2 w2 + H* (17. .41) so that the early successive contributions decrease in magnitude. where Hen{w) is a Hermite polynomial when n is an integer. (6.42) (17. (17. Before we study the eigenvalue equation in more detail." Hence in dominant order the solution B is JB^0) = Bq[w and it is convenient to set Hei Bq(w) = 1 (91) H (17. a solution Bq(w) of the equation Dq Bq(w) — 0 is Bq(w) oc He^^^iw). We return to Eqs.44) where D(B) 9 = dw2 $_ !)• (17.d2B w dw2 + w(l w dB_ dw d dw 1.46) 2s(«. We observe that for the large values of h we are considering here. it is found that D(B) z+ — ]B = 0.28) and write the upper equation for a solution y = Bexp(2hsinz) in new notation B" + 4/icos zB' + 2h( qsin Changing now the independent variable to w{z) = iVhcOS I 7T + Z I . and discussed there.43) 9 2eh . The differential equation there is Eq.i ) We encountered Hermite polynomials earlier in Chapter 6.) i ( Q . 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.358 CHAPTER 17. Note different definitions in use.67). (17.
e. y — Cexp(2hsinz) and Cexp(—2/isinz). with the equation C" + 4h cos z C + 2h ( q .sin z + by changing the independent variable to w(—z) = 4v/icos I 7r z (17.3 ) \He* ! (9+1) H. i.d2Bq + w{l dw2 "» 2 >^(» 2 + ^ = {q. i.47) where the coefficients are the same as before.51) . the expansion for the quantity A being identical with that of the previous case. 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..49) Then the dominant order solution is C^°\w) = Cq{w) with and — again choosing the multiplicative factor suitably — Cq(w) = 2i(" +1 ) ^ . i. as in Eq. The solution with contributions B^°\B^>. also the same coefficients in the case of functions C below. (17. 16hJ (17.. (17.. 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). A{wz (17. forms a rapidly decreasing expansion provided that \w (z)\ « Vh.4 (using the known recurrence relations of Hem{w)) the perturbation remainder can be linearized.50) C = 0.7 T H 1 2 Z 4 <1. which is an amazing and unique property of the Mathieu solutions — cf.q + 4 ) £ g + 4 + (q.17. (17. . q)Bq + (q. Finally we obtain a third pair of solutions.34). 1 COS I .qi)Bq_4.3 Cosine Potential: Strong Coupling Solutions 359 because then as shown in Example 17.e.48) This shows that this solution is valid in particular around z = ir/2. i.e.e.
(17. 17.54) where the proportionality factor is complex.A Hen . (17. .53) This solution is valid in particular around z = —7r/2.h) ocB{z.44) can be linearized with the same coefficients as in the case of the trigonometric solutions. (17. We have thus determined three pairs of solutions of the Mathieu equation. A closer look at that expansion is our next objective.52) and the coefficients are again the same as before. We note that C(z. we have a remainder •R:=w2He„+w(l Using the recurrence relations wHen(w) = Hen+i(w) + nHeni{w).4. q)Cq + (q.n2(n l)Hen2 . ie 1 COS I .q.4)C g _ 4 .w2)He'n . 2 J where n=(ql). q + 4)C 9 + 4 + (q. (17.(w2 + A)Hen. Periodic Potentials since then (using the known recurrence relations of He^) ~Mw2^ +w(l+w2)^ + (w2 + o dwz dw V 2 (q.7 T 1 2 Z 4 <1. Solution: Inserting the dominant Hermite function solution into the right hand sid of Eq. These domains of validity are indicated in Fig. show that the perturbation remainder of Eq.q.q.360 CHAPTER 17. He'n(w) = wHen(w) Hen+i(w).h). (17.4: Hermite function linearization of perturbation Using the recurrence relations (6. each of them associated with one and the same expansion for the eigenvalue. Again the solutions have the same structure as in the previous two cases and the eigenvalue expansion remains unchanged. 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. (9+3) (2n 2 + 2n + 1) + . Example 17.(ql)2(q3)HeX (95) i(« 2 + i) + iA Hei (91) .44).59d) of Hermite functions Hen(w). 2' the remainder 72. The solution y = C exp(2/i sin z) is a decreasing asymptotic expansion provided that w(z) ^ v^/i. can be reexpressed as 72 = (n + 1 ~(q + l)Hen+2 l)He.
.17.' 361 in 2*f*1>[i(gl)]! = (q.q4) (17. In addition (c) each move except the last has to be divided by the final displacement from q divided by 4..g) 1 1 and so on.l)(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).4) = (q .3). ~l)Aq4} 2h 1 { P 2 ( < Z 2 + + P2(<? lAq+i ) 8 A ) ' " ' +P2(q.j): y = = = e e2hsinzA{z) oo 2ftsinz r=0 V .)Bqi(w). r j=r. . the solution y = exp(2/i sin 2)^(2). q + 4) = (g + 1)(? + 3).q + 4) (q + 4. 2 + ^ M 3 + (17.57) . (q. in the following compact form with coefficients Pr(q.g4)(q4.3 Cosine Potential: Strong Coupling Solutions It follows t h a t with the definition of Bq (w) as Bq(w) that He K . We see that (a) every coefficient Mr results from a sequence of r moves from q back to q. 2)Aq_8} + • (17. 17.j^0 e 2ft sin z A + ~ww 1 7 i + ^7ir{ p i(9' ^ ^ + 4 + Pi(q. e.q 4.g. ^ 1 Hv" '^ ^' ^ ' .56) (94. ) .1i . The coefficients have thus been constructed in a way which now allows the formulation of a recurrence relation.q + A){q + A.i ) ( w ) 2l(.0 and —4. (b) The allowed moves are +4. (q.3. (?) 1 1 + (9. .q + 4)Bq+4(w) + (q. We can write the solutions obtained previously. q)Bq(w) + (q. (q. q) = 2[(q2 + 1) + A].l)[l(q _!)].55) M2 = M. no intermediate move to or from q being allowed. with (. l)Ag4 + P2(q. q .
g + 4) ^3(9. 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. j) = (g + 4j .i (g. M2 = P 2 ( g . i. l ) . g + 4j)P r _i (g. r ^JtPr^jOPr+lfejV^rfejO^r+l^j)] r 2^jPr(g.g + 4)(g + 4.j).g + 8)(g + 8.g + 4) + 1 1 1 and so on. p = (<?><? + 4 ) ( g + 4 >g + 8) . Thus we have. for instance.g + 8)(g + 8. (17.58) with P 0 (g. g + 4j)P r _!(g. and in general* M2r = J]j[P2(g.j).j)^2(9. g + 4 j ) P r . B. the factor j removing a duplication of factors in the denominator at the junction.59) One can now write down formulas for M^r and M<ir+\ in terms of the coefficients Pr{q. . Periodic Potentials PlM)=(M+i). J. W.4. Dingle and H.g + 4 ) ( g + 4. From the coefficients (17.P 2 ( g .g + 12) 1 2 3 (g. j .60) M2r+l = = *R.j)] = 2j]jPr2(g.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)Pr+1(g. x)_ (g._1) = Mzil.j ^ 0) = 0.e.j).j). as jP r 2 (g.j).3) ^3(9.g + 4)(g + 4.l ) . Miiller [73].g + 4) 2 1 (g. P2(q 2) Pl(g. .g + 4)(g + 4.1) + (g + 4j. j) +(g + 4j + 4.j) = (l)rPr(q. (17.34) we deduce that Pr(q. and the contribution from a sequence of r moves from q + 4j back to g.362 Clearly CHAPTER 17. 0) = 1 and all other P0{q.g + 4)(g + 4. 2) 1 2 ' ' ^ ' ^ 1 (g. j + 1) (17.
17. This means that the in approaching such a domain the appropriate solutions must become proportional in view of their uniqueness.4.3.^ . *E.3 Cosine Potential: Strong Coupling Solutions In this way one obtains the eigenvalue expansion A = 363 _2h2 + 2hq^(q2 1 . of course. h — —h.2 * +3) + ^ 2 (9g4 .^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.( 3 ^ . Its precise deviation from qo. as a result of tunneling between wells. Up to > > this point. These domains in the interval —7r/2 < z < 7r/2 are indicated in Fig. J. 17. S. [116].* We observe that the expansion remains unchanged under the replacements q — —q.62a) f R .(5q 4 + l)^Jiq(q2 + 3) + 410g2 + 405) + 34q2 + 9) . the solutions and eigenvalue expansions again have the same type of symmetries.4 The level splitting Above we obtained three pairs of solutions along with their domains of validity as decreasing asymptotic expansions.17. L. It will be seen in Chapter 18 that in the case of anharmonic oscillators. Goldstein [115]. In fact we demonstrate in Example 17. We observe that there are regions.92q3 + 70g2 .5 the following proportionalities there: B[w(z)] = aA(z). as pointed out earlier. remains to be determined next. B. [137].284g + 57) + (17. Dingle and H. . a= (8/i)3(9x) [1(91)]! 1 .61) +288161796g2 + 130610637) These are the terms given explicitly in the reference cited above. Miiller [73]. t Terms up to and including those of order l//i 5 had been obtained by others and by different methods. Ince [136]. where solutions overlap. W. q is only known to be approximately an odd integer qo = 2n + 1.
32).364 and C[w(z)] =a~A(z).62b) *z domains of overlap Fig.COS 7T 2!42 \4 2 COS /l 7T V4 Z H 1 2 Z H 1 2 . The first term of the expansion of A(z) is the function Aq(z) given by Eq. and determine the constants a and a.4 Domains of solutions and their domains of overlap.Z (g+1) 1 H 1!4 4 2 1 (g + l)(g + 5) . 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).7T H Aq(z) = = C O S ^ . Example 17._ ^ l CHAPTER 17.5: Proportionality of solutions in domains of overlap Show that in their common domain of validity B = aA and C = aA. 17. a .(17. We write and expand this function as follows: .1 ) ( 7T+ z 1 — COS 2 1 \l^+1> 2 Z C O S S ^ .1 ) ( 7T + .1 4 .
\ql)/2(w) 2i(9D[i(. with e.17. 17.i ) [(91)]! It follows therefore that over their common range of validity B = aA with a = b/a. Hence we have to construct even and odd solutions and impose the boundary conditions at z = 7r/2. s . These are the conditions of Eqs. 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 .6 at the end of this section we derive for these \axgeh2 solutions the conditions similar to those given in various equations in Sec.!)(. with (8h)li''1) (3g2 . (8.2g + 3) 1 [£(91)]! 2?h 215h2 (9g 4 .22).q) 2 7 /i 1 1 4(g + 3) 1 (g + l ) 2 ( 9 + 3) + 29ft ti(9l)]l i (8/1)4 ( 9 .g4)fa3) 1 1!4 / I V/i 2 = 1 (gl)(g3)2 2 9 /t Similarly substituting the asymptotic expansion of the Hermite function^ into the expression (17.284g + 57) + • The other relation is obtained similarly.g. Since the solution with A is obtained from that with A by a change of sign of z.g + 4) (q + 4.2 for the case of small/i2 solutions and thus insure that in both cases the same boundary conditions are obeyed.3 Cosine Potential: Strong Coupling Solutions 365 Inserting this into A = A(°> + A' 1 ) + • • •.46). i.92q 3 + 70q 2 . (17.61) for the connection and Eq. (17. these are conditions imposed on even and odd solutions at the point z = it/2.3) 1!(8™2) )[i(gl)]!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. As we saw there. For further details and explicit expressions of higher order terms we refer to Dingle and Miiller [73]. This is essentially the expansion of the parabolic cylinder function with the exponential factor removed — see Eq._i)]! w*te» 2l(91 (. In order to link our asymptotic solutions with the Mathieu functions of integral order defined earlier in the consideration of small values of h2.63) In Example 17.e.i ) 1 (g. A*1) given by Eq.2.53) for the expansion.36). we have to impose the appropriate boundary conditions. we have as even and odd solutions y±ocA(z)eZflsmz±A(z)e 2/isin. (17. (6. we obtain Bq[w(z)} H. the coefficient of the dominant factor c o s ( . _ 1 " 2 ( 7 r / 4 + z/2) in the result is seen to be a = 1 2 /i 7 fa.
W. Periodic Potentials We can extend the solutions (17.66).59d). Oberhettinger [181]. one obtains the first of relations (17. p. Thus y+ = ce and y_ = se. and then replacing 2n by i>.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. Here v is the not necessarily integral index of the Hermite function Hev{w) with Hermite equation ( jir w—+v)Hev(w) \ aw1 aw J = 0. 80. Then y± oc ^ M e ^ i n . He2n+1 (0) = 0.» a \dw _ C[w = 0}c_2h a w=0 = Q (17.22) and to correlate the solutions to those for small values of h2.g. He'u{w) = wHev{w)He„+1{w). With the help of the duplication and inversion formulas of factorials. a ± C[w{z)}c_2hsinz^ a (17.22) are therefore given by the following boundary conditions: 1T y'+ v+fao I'=° y a a\dwjw=0 B[w = 0]c2h a a 1h dw\ ! ' .62a) and (17. He'v(0) = Hev+1{0).e. Magnus and F.63) to the domain around z = ir/2 by using the proportionalities (17.e.66) ^E. i. The second of relations (17.66) then follows from the recurrence relation (6. (17.64) These are now the solutions around z — ir/2 which permit us to apply the boundary conditions (17.62b). The functions defined as in Eq.366 CHAPTER 17. . i. give the following expressions for polynomials with argument zero: ~19)"(^)!.
40g3 + 18g2 .136g + 9) + • • • .67a) and (17.( g .67c) 4 2 \dw)w=Q C[w = 0] [\(q .67d) Considering now the second of conditions (17. f^(16h)i/2e4h 2 [Uq1)]\ 3(g2 + l) + 2^h2 (9g4 .7.Qo) + 0[(q .69) If the right hand side of this equation were exactly equal to zero.g0)3]. .39 (17.1) 7T .! ) ] ! [ .3 Cosine Potential: Strong Coupling Solutions 367 With these expressions we obtain by insertion into B and its derivative.» 2 6 /i the condition reduces to the equation cot{ . (17.67c).82g2 .87 4 2 ¥w (17. Hence expanding the cotangent about these points.68) Inserting here expansions (17.106g . B[w = 0] 2TT _ _g_ ¥h+ g 24/* _ JL g4 .39 2~^h? (17. and using the reflection formula (8.11.67b) <f ~ 82g^ . from which one obtains "1 1) 4<*3) 2(9!)/2 2 ^ \b.17. the solutions would be given by q = go = 3. and similarly for C and its derivative.65) we have B[w = 0] _ a C[w = 0] ~ ~ ^ 4/i (17.14) from which one obtains 1 (q + i) TV cos{f(gl)}[i(g3)]!' and the duplication formula (2Z)\\/TT = z\(z — l/2)!2 2z .\4 q + m { 4« VJL ^[±(g3)]!sin{f(gl)} [J(ffl) "(^L^»{^': _ _q_ ¥h + g .67a) g 106g 87 2 14 /i 2 (17. we have cot j ^ ( g ~ 1)} = ~\<(l .
65) leads to the same result except for a change in sign.136g0 + 9) 2 13 /i 2 1 4 2 19 3f(9gg .73) We now return to the eigenvalue X(q) and expand this around an odd integer qo. (17. .40gg + 18gg .408g0 + 1089) 2 /i +• (17...136g0 + 9) + • Clearly the last of conditions (17.(17. The remaining condition (17. i.65) we have (dB/dw)w=0 (dC/dz)w=Q Proceeding as above one obtains tan{(ql)\ 7T .9.e.136q + 9) + . Periodic Potentials ' " Vn [i(gol)]' 3(g02 + 1) 26/i . Considering now the first of conditions (17.40gg + 18gg .120g£ + 467g . Thus we have 2 (16/i) qo/2 T2 4.70) except that now qo = 1.72) This leads again to the result (17. we use \(q)~\(q0) + (qq0)(^J .h qqo = * [£(*>!)]! 1 3(902 + 1) 26h +^^{9q^ .71) = 2 13 /i 2 fc{l6h)i/2e4h 1 3(g2 + 1) 2 [4(51)]! L (9g4 . = _a a 6 _4h (17.528gg + 3307Q ..70) + ^ 2 (9?o . .40g3 + 18q2 ..368 Hence we obtain qq0 a [2(16h)i°/2e*h CHAPTER 17.5.65) yields the same result except for a change in sign.
B. solution of the Schrodinger equation for the basic and nontrivial periodic potential is important for a thorough understanding of its quantum mechanics.3.^ ( 3 g 0 2 + 8go + 3) A ( Q 0 ) T (87r)V2[i ( g o i)]l L2 w ..74a) we obtain for the level splitting in dominant order* x r ^ \ / N 2(16/i)>+1 _4h A+(go) .. p.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. just as any solvable problem is important for the insights it offers in a concrete and transparent form. In the literature one finds frequently the statement that the splitting is a nonperturbative effect. 120.61). n = 0. plus sign) give the socalled level splitting AA(go) as a consequence of tunneling indicated schematically in Fig. The complete. . Arscott [11].1.e. (17. + ^ 2 (9<?o + 89o " 78g02 .3. quite apart from applications. *In Sec. The tunneling effect. M. Goldstein [116]. i.A_(g0) ^ . example 3. (17. results from the boundary conditions. 20. See also F.74b) The above results for the cosine potential are in many respects important. We see here that *R.e. n ' r 1 / rrre 4/l ^odd(go) Seven (<Zo) Wfaqom (16/i)T n\V2n e~4h.88q0 . this becomes 369 = (16/i)^+1e4h l . In the literature that we follow here* in all cases several more higher order terms have been given.17.t From Eq. Miiller [73].3 Cosine Potential: Strong Coupling Solutions Differentiating \(q) of Eq.2. and not the separation of these as a result of tunneling. and that for the odd Mathieu functions se qo+ i and se9o (lower. The dominant contributions were also found by S. (17.. Effectively the degenerate eigenvalue expansion results from the anharmonic terms contained in the power expansion of the cosine potential. In particular we saw that the perturbation expansion yielded only the degenerate eigenvalue expansion. 17. i.3 we call this difference 26E„n. W. minus sign). 9o = 2n + 1. (17. . though approximate. Dingle and H.87) = Azpteo). J. made evident by the splitting of the asymptotically degenerate (harmonic) oscillator levels.
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 we encountered in the semiclassical method where it has the structure of the factor exp(classical action/ft).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.2: y+(±ir) J4(±TT) y+{z±n) y(z±7r) = = ± —y+(z)± 2/+(0) . Naturally one expects the solutions there to be related to those of the periodic case considered here.§ The boundary conditions we imposed here are those of a selfadjoint problem which has real eigenvalues. Then we obtain first y±(z) = yA(z)±y^(z). y+(0) 2/_(0) Solution: The derivation requires a cumbersome tracking of minus signs. (17. 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.6: Translation of Solutions For the largeh 2 even and odd solutions (17. Periodic Potentials this nonperturbative effect. y^{z) = Aq(z)e~2hsin *. which one can then try to solve in order to obtain the behaviour of the coefficients of the late terms of the expansion. Example 17. We begin with the solution containing the function A(z).74b). in effect. 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. results from the imposition of boundary conditions.74b) is also important for various other reasons. The exponential factor represents.2. evident through the exponential factor exp(—4/i) in Eq.32) we obtain Aq(z ± TT) = (l)±i<9^'Aq(z). The explicit derivation of the level splitting (17. 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. (17. 17. . From Eq. The splitting can also be derived by the pathintegral method as will be shown in Chapter 26. . and one can compare the methods. T y'_(0) y'A±*) y{z). As mentioned at the beginning. the Schrodinger equation with the cosine potential is a special case of several other more general cases. The splitting can also be obtained with some methods of large order of perturbation theory. Thus we do not reproduce every step. _3/+(z) + ____2/_(z). M±TT) . ^ ± n ) = _ ( _ 1 ) T i(<JTl) A 9 ( 2 ) (the extra minus sign in the second equation coming from Aq(z ± 27r) = — Aq(z)).370 CHAPTER 17. yA{z) = Aq(z)e2hsin '.
where Aq(0) = Aq(0) = V2.4ft) ~ 8(q . (17.y]z=„ = [y+yL .VA DV 371 1 V±.4ft.( z ) From this we deduce for z = 0.g A ? ( z = ±TT).c o s  ~(<1 T 1) f y .31). the operator having solution Thus ^ ( 0 ) = (q/2)Aq(0) = A~'q(0) and A'q(z = ±TT) = ~qAq(z qAq(z = ±7r). we obtain — with Aq{z + 7r) = (1)^ ^ 2 A . and y+(z = ±TT) = V2(q .1 )  . Finally for the derivative of the odd solution we obtain y'_{ir) = K ( T T ) .4ft) sin j ^n(q . y'_ (0) = \/2(4h . = ±ir). 17. — j / + ( 2 ± 7 r ) = ± i s i n  ^ ( g = F l ) U + ( z ) . (l^ifr^i^O + Cl^ifo^j/aCO cos  .7 r ( g ^ l ) U + ( z ) q : i s i n { . which is easiest by comparison with the foregoing treatment of the Mathieu equation.4 17.( 2 ) . Replacing sines and cosines by the functional expressions. A' (z = ±ir) = (.Ah) ~ 32ft.7 r ( q = F l ) } j / .) cos < 7r(q 1) [ = .2hAq(n)] .[A'qM + 2hAq(n)] = iV2(q .%/2(g .( ± 7 r ) = c ° s  .• Then y'+ (0) = 0.17.z ) . but also to enable a brief familiarization with the Lame equation.q)..4. ( . Aq(z) The derivatives can be handled with the help of Eq. y+(0)~2V2. Similarly we find y(z±ir) = = and by putting z = 0: V.4ft) cos i TT(<J + 1 ) 1 . 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+. A ?q( Z = ±TT) = . etc.T ( ? =F 1) p + (0). The Lame equation has not been a widely known equation of mathematical physics so .3 The Ellipsoidal Potential Replacing VA. since A q (0) = \/2 = Aq(0): y+(±n) = ±isin^(9=Fl)J3/+(0). we obtain the conditions stated at the beginning.1 Elliptic and Ellipsoidal Potentials Introduction Our intention here is partly to deal with other periodic potentials.yV+\z^ = [j/+(0)] 2 (<? .
the new stage of the development of the investigation of the Lame equation was really initiated by Ince^ around 1940. Periodic Potentials far. Liang. But more recently it has been observed to arise in various contexts. Erdelyi [85].5 < z < 5. k) for k = 0.0. H. This line of investigation. 19. M.K2sn2u . MullerKirsten and D.0.0 and .98. Ince [138]. but also cubic potentials.372 CHAPTER 17. H. and was continued by Erdelyi. inverted double well and cosine potentials (J. 194) remarks to parallels with the Mathieu equation:".75.fi2A. f E .. § F . however. Arscott [11] (p. L. 17.75) ' T h i s means double well. one arrives at three equations of which one is the ellipsoidal wave equation pt + [A . the nonoccurrence of the Floquet exponent. 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 ... as alluded to at the beginning.4sn4u]y = 0. In particular the Lame equation was recognized to arise as the equation of small fluctuations about instanton solutions for practically all basic potentials.. *A. (17. . The most conspicuous difference compared with the Mathieu equation is. 194) remarks... Arscott [11]. If one separates the wave equation § V 2 * + J2* = 0 in ellipsoidal coordinates (which we do not need to consider here). p. for general n the solution of Lame's equation is not singlevalued owing to the singularities in the finite part of the uplane." Fig.5 The potential sn2(z. Tchrakian [165]).Q. W.5.* As Arscott [11] (p.* and the work of both prepared the ground for presentday investigations.. so that Floquet's theorem cannot be immediately applied .1. J.0.
(17. for very > high barriers (harmonic oscillator approximation around a minimum of the potential).3 The Ellipsoidal Potential 373 Here K2 and A are separation constants and U2 = l2(a2 — c2). (17.a > b > c being related to the lengths of the three axes of the ellipsoid in a Cartesian coordinate system. K being the complete elliptic integral of the first kind. W.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.e. the usual factor —h2/2mo. 17. J. The function sn2u is plotted in Fig. (17.K2sx?u}y = 0. in front of the second derivative has to be kept in mind (mo being the mass). Eq. (17. we refer to the potential consisting of the two terms with sn 2 n and sn 4 n as the ellipsoidal potential. The range of the independent variable u is 0 < u < 2K. the equation ^  + [A .\k\ < 1.77) into Eq. If we put fl = 0. The second step is to insert (17. in the case n2 — oo. 17.2 Solutions and eigenvalues In the following we sketch the main points of the method of deriving asymptotic expansions of the eigenvalues and eigenfunctions.17. In comparisons with the Schrodinger equation. .76) and to write the solution y = A(«)exp  f Ksnudul = A(u)[f{u)]K. i. cmx and dnu. In order to distinguish the above equation from that with the Lame potential.K) = qK+^±. is the elliptic modulus of the Jacobian elliptic functions snu. .e.2 = n(n + l)k2.4. 2 .1. (17. MiillerKirsten. N = 0. Jianzu Zhang and Yunbo Zhang [206]. i.5 for several values of fc. The first step is to write the eigenvalue A as A(q. For barriers of finite height the parameter q is only approximately an odd integer qo in view of tunneling effects. and of the derivation of the level splitting'. Although the results have been calculated for the ellipsoidal wave equation.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.2k. . we consider mainly the Lame equation.77) where q —> go = 2iV + 1.78) We follow in this brief recapitulation the description in H.
/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. dnu± cnu K Thus one can construct solutions Ec(u).82 .e.Es(ii).81) Since these expansions are not valid at the extrema of the potential (where the boundary conditions are to be imposed). (17.q.80) The domain of validity of these solutions is that away from an extremum of the potential. dnu =F cnu . A second solution is written y = A(u) exp < / nsnudu >. _ Solving the resulting equations iteratively as before. which are respectively even in u (or snu) or odd.e.374 where CHAPTER 17. . ^ (17. determine their proportionality factors) in domains of overlap (their extreme regions of validity). i.K). Thus in the third step two more pairs of solutions B. more precisly for dnu = cnu F 1 » .K — > A(u)=A(u + 2K). . C replacing yl. B and C. i. / z J . the other in terms of those of an imaginary variable.e. C which are valid for dnu±cnu <1.K)=A(u.. one obtains again the same expansion for A. k' = Vl~ k\ \dmi±cnu/ 17. A into equations in terms of the variables z(u) = ^—{k' . one pair in terms of Hermite functions of a real variable. one has to derive new sets of solutions there and match these to the former (i.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. ll^ ^A(u)[f(u)]^k±A(u)[f(u)}^k. A are derived.77). A. \/8K (dmz = cn« \ ' F . by transforming the equations for A.q. (17. (17. A(u. Periodic Potentials . but solutions B.
3 T h e level splitting In the fourth and final step one applies the appropriate boundary conditions on these solutions. (17. (17. W.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. £2 = 0).6(5g4 .85) The first three terms of this expansion were first given by Ince [138].e.3 The Ellipsoidal Potential 375 In their regions of overlap one can determine the proportionality factors a." 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) . Miiller [205]. W. Tricomi [87]. a of _ _ B = aA.^ ^ { ( 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£. Magnus.75). .63)} + ••• .e. F. G. This expansion is found to be in the more general case of Eq. who obtained this expansion for the eigenvalues of Lame's equation (i.384n2k4(q2 + 1)} q r {(l + A.38g2 . (17. 17. p. *A. Erdelyi. Es(2f0 = Es(0) = 0.17. one sets at it = 0 and u = IK altogether** Ec(2A") = Ec(0) = 0.4.86) \duj0 o h r 2K > \duJ =[sr 0 =0\duj (1787) H. 64. as well as \duJ2K 11 (17. J.2)4(33g4 + 410g2 + 405) 24/c 2 (l + k2)2(7q4 + 90q2 + 95) + 16fc4(994 + 130g2 + 173) +512f22fc4(l + k2)(q2 + 11)} . C = a~A. Oberhettinger and F. the ellipsoidal equation. i.
4k2(q2 + 2g0 + 5)} + • • • ].g0 is obtained by expansion around zeros.78g^ .Es^ 0+1 .91) tt H.87) (17. 2K and 4K respectively.:J {3(1 + k2)2 (9q* .go)« qo(l + k2 22K {3(1 + k2)2(q2 + 1) .Ec£ 0_1 and Es2° of periods 4K.l n 2^+ i J + 0(«i) (17.g 0 ) ( ^ . Finally expanding A(g) ~ A(g0) + ( g .88g0 . (17.2 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 (lfc)*. from which the difference q . 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 Hqom 25. .j 26K2 (1 85) = A(g0) + (q . Periodic Potentials These conditions define respectively functions EcX°. J. W. go being an odd integer. and the lower to Es*>+1 or Es*5.2 /c +256fc2g0(g2 + 5)} (17. One obtains with go an odd integer QQo =F2 2fl + ir\lkj 3(g02 k\K/k 8K qo/2 \lk ) 2 1 [i(g0l)]!L + l)(l + fc2) 25 K ^ 2 J J .90) + 3.89) one obtains the eigenvalues from which the level splitting can be deduced.i Here the upper sign refers to Ec*> l or Ec£°. + fc )(3g + 8g0 + 3) k2)2(9q% + 8g^ .1 / 2 (2TT) 1 /22« l + (l_fe)^«( i .136g0 + 9) n 3. 2K.376 CHAPTER 17. Muller [205].40g03 + 18g02 . 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).
+ E. Rao [79]. however. A(g)>A(g0)T4/i\/e ^ [Hqo ~ !)]•' in agreement with Eq. (17.4 h ( i e M 9 0 7 2 \. h .* Apart from the choice of notation.Ah2 sin2 u}y = 0.* M G . (17. L.92) reduce the periodic ellipsoidal wave functions and their eigenvalues to corresponding Mathieu functions and their eigenvalues.2h2 . the conditions (17. K2 = h2 (say). i. 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. Hence the conditions fi = 0. that calculations with the Schrodinger equation are not only easier.90) one has in this limit: /l±±\ ~K/U = e . this equation becomes y" + {A . Miiller [206]. (17.2h2 cos 2x}y = 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. Dunne and K. but also more easily generalizable. K = 2/i.f ln[(l + fc)/(lfc)] _ e 4h) ( 8" \ ^^ __ (16h)«. . in which they dubbed the classical configurations Lame instantons. 17. W. . A = 0. One can see.3 The Ellipsoidal Potential 377 This result agrees with a result of Dunne and Rao** who calculated this expression using instanton methods. Ince [138]. it = x ± ^ . (17. Thus in the case of the dominant contribution of Eq. Replacing u by x ± IT/2.87) agree with those of Ince.4. V.86).J1 O.74a).4 Reduction to Mathieu functions Under certain limiting conditions the ellipsoidal wave equation reduces to the Mathieu equation. Thus if k — 0 and n —> oo in such a way that > K2 = n(n + l)k2 ~ finite. *H.e./2i and hence A^ A/ A AU I2 .17. J.2h2 = A. then snit — sin u and Lame's equation (f2 = 0) becomes > y" + {A . A .93) (17.
S. MiillerKirsten and D. Tchrakian [165]. Thus. See e. for instance. Khare and U. Kan and K. Sukhatme [148]. W. i. § . all basic potentials. J. Gu and S.** The associated Mathieu equation appears also in string theory in connection with fluctuations about a L>3brane (see Chapter 19).5 Concluding Remarks The potentials we considered above will reappear in later chapters. M. thus revealing an unexpected significance of this not so wellknown equation of mathematical physics. Ouvry [276] and Jianzu Zhang.^ The equations considered above also appear in diverse new problems of physics. .378 CHAPTER 17. Ganguly [103]. MiillerKirsten and J.Y. A host of related elliptic equations has recently been discovered and studied. In particular we shall encounter the Lame equation as the equation of small fluctuations around classical configurations associated with cosine. **Y. 1 1 See Z. A. Liang. J. Cho. Quian [123]. 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. W.Q . This is an interesting problem. H. N.W. See Chapter 25.§ This is therefore a very important equation which in the limit of infinite period becomes a PoschlTeller equation. inverted double well and cubic potentials. de Veigy and S. A generalized associated Lame potential has been considered by A. 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. Rana [287]. H. Periodic Potentials 17.g. Shiraishi [49]. H. A. D. double well.e. J .
Chapter 18 Anharmonic Oscillator Potentials 18.t which related analyticity considerations to perturbation theory and hence to the large order behaviour of the eigenvalue expansion attracted widespread interest. J.1 Introductory Remarks The anharmonic (quartic) oscillator* has repeatedly been the subject of detailed investigations related to perturbation theory. This seemingly simple problem revealed extremely rich internal structure . Liang and H. D. Wiedemann [4] and a revised version of parts of this reference by J. In the cases treated most frequently in the literature the anharmonic *We follow here largely P. Lee [98] referred to it as a "long standing difficult problem of a quartic potential with symmetric minimd'1. W. [19]. ' T h u s A. In particular the investigations of Bender and Wu. H. both of which make the calculation more difficult. J. W. MiillerKirsten [163]. Priedberg and T..* The fact that a main part of their work was concerned with the calculation of the imaginary part of the eigenenergy in the nonselfadjoint case which permits tunneling. T. Weinstein [281]. T. 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. demonstrates that derivations of such a quantity are much less familiar than calculations of discrete bound state eigenenergies in quantum mechanical problems. 379 . Bender and T. Turbiner [273] remarks: "It can not be an exaggeration to say that after the seminal papers by C. Achuthan.". MiillerKirsten and A. There is no end to this: An entirely new approach to anharmonic oscillators was recently developed by M.Q.. Wu. M. f C . Bender and T. in nearly a thousand of physics articles the problem of the anharmonic oscillator was touched in one way or another. Wu [18]. The recent work of R.
V(z) = \\h*\z* + \\<?\z*. These contributions may be given different signs. Anharmonic Oscillator Potentials oscillator potential is defined by the sum of an harmonic oscillator potential and a quartic contribution. (3) Complex eigenvalues with tunneling: In this case. V(z) = \\h*\z2 I 2i 4 \cr\z . *~z (1) (2) (3) Fig. To avoid confusion we specify first the potential V(z) in the Schrodinger equation dz2 + [EV(z)]y(z) =0 for the different cases which are possible and illustrate these in Fig. 18.1 The three different types of anharmonic potentials. 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.380 CHAPTER 18. which are nonetheless linked as a consequence of their common origin which is for all one and the same basic differential equation. described as the case of the double well potential. with the potential described as an inverted double well potential. Case (1) is obviously the simplest with the anharmonic term implying simply a shift of the discrete harmonic oscillator eigenvalues with similarly .1. and thus lead to very different physical situations. 18.
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). The boundary conditions are nonselfadjoint and hence the eigenvalues are complex. The shift of the eigenvalues is best calculated with straightforward perturbation theory.e. 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 17.18. Thus we are mainly concerned with the double well potential and its inverted form. that the general applicability of the method becomes evident. We do not dwell on Case (1) since this is effectively included in the first part of Case (3). We therefore consider in this chapter and in Chapter 20 in detail some prominent examples and in such a way. 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. The result is an expansion in descending powers of h?. since the potential decreases without limit on either side of the centre.86) below. 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. We begin with the latter.175) below. 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. This type of potential allows tunneling through the barriers and hence a passage out to infinity so that a current can be defined. Case (3) is seen to be very different from the first two cases. however with decay as a consequence of tunneling. and this behaviour is that of asymptotic expansions our treatment largely terminates this muchdiscussed topic. (18. If the barriers are sufficiently high we expect the states in the trough to approximate those of an harmonic oscillator. as is sometimes hoped. 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. i. In this case our aim is to obtain the aforementioned complex eigenvalue. except for a change of sign of c 2 . (18. Calculations of complex eigenvalues (imaginary parts of eigenenergies) are rare in texts on quantum mechanics. . it will not be possible to obtain the exponentially small contributions with convergent expansions. The imaginary part will be rederived from path integrals in Chapter 26. The eigenvalues of this case are given by Eq.1 Introductory Remarks 381 normalisable wave functions. Case (2) is also seen to allow only discrete eigenvalues (the potential rising to infinity on either side). The eigenvalues of this case are given by Eq. The level splitting will be rederived from path integrals in Chapter 26.
a point which has to be kept in mind in comparisons. 18. .2) We adopt the following conventions which it is essential to state in order to assist comparison with other literature. The potential in this case is given by V{z) = v(z).z Fig. v(z) = ^h4z2 + ^c2z\ (18.382 CHAPTER 18. we can pass from one case to the other by making the replacements: 4 E1/2 = 2Ei.2 18. If suffixes 1/2.1 and more specifically in Fig.1 refer to the two cases.2. and the Schrodinger equation to be considered is C L^ + [E + v(z)]y = 0. h 4 2 2h1> ~1/2 2c?. We take here h = 1 and the mass rriQ of the particle = 1/2. Anharmonic Oscillator Potentials 18.2 The inverted double well potential with (hatched) oscillator potential. 18. (18. 18.1 The Inverted Double Well Potential Defining the problem We consider the case of the inverted double well potential depicted as Case (3) in Fig.3) "1/2 harmonic oscillator h8/25c2 . 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 .1) for h4 and c 2 real and positive. (18.2.
n = 0.18.7a) The problem then reduces to that of the pure harmonic oscillator with y(w) a parabolic cylinder function Dn(w). The problem here is to obtain the solutions in various domains of the variable.5) becomes . then to specify the necessary boundary conditions and finally to exploit the latter for the derivation of the complex eigenvalue. /i).. and a variable w defined by setting E= b* 1 + A A and w = hz.1. i. \h?\ — oo and c finite. The result will be that derived originally by Bender and Wu.2. y{w) oc Di. (18. (18. (18.. and h large the eigenvalues are essentially perturbatively shifted eigenvalues of the harmonic oscillator as is evident from Fig.4) we can rewrite Eq. 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. (18.1/7  Vg(w)y(w) = o(J^j. the harmonic part of the > potential dominates over the quartic contribution and Eq.e.8) h8 v"{z±) = h and V(z±) = 5 2 2 c ' 2 2 Thus for c > 0 and relatively small. to match these in domains of overlap. _i\(w) and q = qo = 2n + l.. (18..2. although our method of matched asymptotic expansions here (which parallels that used in the case of the cosine potential) is different. 4 . 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.d2 .2) as Vq(u >)y(w) = with T)Jin\ 1 (A + c2w )y(w 0 (18.5) (18 6) "^'"dw* ' * 2' 2 In the domain of w finite.2 The Inverted Double Well Potential 383 Introducing a parameter q and a quantity A = A(qr.
14b) .2 We are concerned with the equation d2y(z dz2 where ^ 1 .) exp Z ifdz[ifdz[ h4z2 4 h"z2 + + cV cV 1/2' (18.+ ^ .9) z =0 these Thus these c2z4 1/2 + • 2 (18. y 4 2/i Here again q is a parameter still to 2 determined from boundary conditions. Inserting (18. as encountered and explained earlier. (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 ) . •A~2 hrz" 2^\ c^z V2 —— + (18.2. Before we return to solutions we derive a new pair which is valid in the adjoining domains.10) E= qh2 ^ j. Anharmonic Oscillator Potentials T h r e e pairs of solutions 18. be and A = A(q. (18. This observation allows us to define the pair of solutions yA(z) = . This construction is simplified by the consideration of symmetry properties of our solutions which arise at this point.14a) 1/2 VA( ) = A(z)exp (18.4(2. these solutions are not valid around z = 0. as we shall see.9) we obtain 1 h4z2 c2z4 A L2 (18. We observe — before touching the square roots in Eq.13) — that one equation (of the two alternatives) follows from the other by changing the sign of z throughout. In order to arrive at solutions we set in Eq.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 = 0.11) y(z) = A(z)exp ± z dz h4z2 4 + E hAz2 + cV y(z) = 0 (18.10) into (18. h) is obtained from the perturbation expansion of the eigenvalue.2 A (18.384 CHAPTER 18.12) Then A(z) is found to satisfy the following equation r /iV c V V/2 A"(z)±2<^. 1 d r A!{z)±iA{z)±{ 0.
15) where A{z) is the solution of the upper of Eqs. /H 4 Z h +~C Z 21. We take the square root by setting z2h^1/2 +.4 . 1 (18.16) For large h2 we can write the Eqs.34) below) — to be derived in detail in Example 18. / c z Aq{z)+0 1/2 .13) TzA'(z) T ^A(z) + qA{z) = O We define Aq{z) as the solution of the equation zA'(z)(ql)Aq(z)=0.17) Aq(z) 1(91) Z2 (Z )V4 exp 2 We define correspondingly w dz (z2)l/2 (18.18) M*) zf(9+l) A_ g (s) = (^2)1/4 exp ?/" dz (z ) / 2 1 2 (18. (18.18. Thus we now have the pair of solutions yA(z) = exp yA(z) = exp dz I^4 + l„2„4 1/2. 1 2 4 2 Aq(z) + 0 (18.20b) .e. 385 (18. Since these higher order contributions are of little interest for our present considerations.19) We see that one solution follows from the other by replacing z by — z. (18.1 and as a verification again in connection with the solution y# — as the other solutions.zh2 =()*2"(18. i. One finds that these solutions are associated with the same asymptotic expansion for A and hence E (given by Eq. (18. Clearly Aq(z).2 The Inverted Double Well Potential with A(z) = A{z) and yA(z) = yA(z).Aq(z) approximate the solutions of Eqs. we do not pursue their calculation.13) and we can develop a perturbation theory along the lines of our method as employed in the case of periodic potentials. (18.13) and A(z) that of the lower of these equations..20a) These expansions are valid as decreasing asymptotic expansions in the domain \z\ >0 h. (18.
z)).[yA(q.:=*£ +i(5l). 1 . X » / A q + 4 i = Mg+4i . 128 8 16 A = i.20b). * = — .! ) ^ . (18.20a).386 CHAPTER 18. Anharmonic Oscillator Potentials i.18) we obtain A » = H^rL^z) 2 z . ft = .21) Example 18. z) ± yA(q.e. 2 . 04 = 5 .l)A(z) 1 = tfA"(z) A .^A(z) °° /1r2z2\i + J2 { fir ) \**A'(*) 1 + jA^Wl. (18. i. A = . A A _ 3 . h2. h2. zA'(z) where the expansion coefficients are given by ao = l. ..e.13) in the following form with power expansion of the square root quantities and division by h2: 1 + (q . c*i = 1 1 . W i t h proper care in selecting signs of square roots we can use the solutions (18. (18.. away from the central minimum. (18. procedure. i. .2 ( * ) 2 and from this or separately A"(Z) : 1 (q . Vq+ii = Vq + 2i.1: Calculation of eigenvalues along with solutions of type A Use the solutions of type A. one observes that with £>.• Using Eqs.e. to obtain in leading order the eigenvalues E.20a) and (18.v ' = o (« . h2 —> —h2).20b) to construct solutions y±{z) which are respectively even and odd under the parity transformation z —> — z (or equivalently q —• —q.f . we write > V±(z) = .17) and (18.. „ .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).. „f 1 \ S o l u t i o n : We rewrite the upper of Eqs.2 . a2 = 2 1 . As always in the In this way Rq is expressed as a linear combination of functions Aq+4i(z). 03 = . (18.
^f( g + i)(±H2^ + 1 > \ — 1. c V + It follows that l) + 0 (i The same result is obtained below in connection with the solution of type B. We return to Eq. (18. those in Aq(z).18.. = •>— r. [1(9 + 1)]! . when i = 0. Hence in the present case In its turn A^ leaves uncompensated terms amounting to A „ /2c 2 \ 2 9 „ M 2ft4 2 { 4/i2^ . Hence to the order we 0: {^)^) [{ 1){ 2 ^ ^ ^3)^ 0 +1 ^ + 2^ + 3)]+0 {h) 2ft6' A = 24 1 2 ^ + 1>2^+°G9A= ..22) is invariant under the combined substitutions q — —q. give an equation from which A is determined. w = hz.(q + 3)(q + l)Aqvri(q '^ ' . of course. i.w — > > ±iw) or functions R B w q( ) . _1J±w) and D_\.Aq+4i in Rq 387 can be taken care of by adding to A(°> the contribution — _ ^ 4 ' except. In this way we obtain the next order contribution A^1' to A^°>.'2H9i) ^faDfr™) and .. (18.5). The solutions yq(z) of the equation Vq(w)yq(w) = 0. Rq '. r ( CJw) = ^ .2 The Inverted Double Well Potential Thus a term /j. (18.. 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.e. +1J±iw) (observe that Eq. and the coefficient of terms with i = 0.—77 [i(9l)].23) . (18. 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).22) are parabolic cylinder functions D i .
Oberhettinger and F. W. Actually these factors can also be extracted from W K B solutions for large values of q.24) The extra factors in Eq.0) = = = ±(q±3)(q±7). Comparison with our notation is easier if this reference is used.23) have been inserted to make this recurrence relation assume this particularly symmetric and appealing form. See Example 18.( < ? . As an alternative to Bq(w) in Eq.q}=A + c2S4(q. (q±2)(q±S). S2i(q.Aj)(18.0). since Vqyq = 0. (18.28) (A + c w )yq(w) E 2 4 1  2 ^ [q. (18.25) where in the case i = 2: SA{q.±8) 5 4 (g. and for j ± 0 : [q.^ For higher even powers of w we write i w2l V<i = Y. 115 .29) ' A .q + 4j] = c2SA{q.26) The first approximation y(w) = y(°>(w) = yq(w) = Bq(w) therefore leaves uncompensated terms amounting to 1 4°) = where [q. 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 . pp.±4) S4(9. and so Vqyq+Aj(w) = Ajyq+Aj(w).27) Now.4j)yq+Aj. (18.23) one can choose the solutions as Bq(w) with B QM = ~ 77[(93)]!24(91) These satisfy the recurrence relation w2yq{w) = (q+l)yq+4 + qyq + . G.123. (18.3)y g _ 4 . Magnus. (18. Tricomi [86].388 CHAPTER 18. j=i (18. .l)2/?4. l(q2 + l).q + 4j}yq+4j. Vq+4j = Vq + 4j. (18. we also have Vq+^yq+tj = 0. Erdelyi.3. F.
(18.34) is the expansion of the eigenenergies E of Case (1) with q — qo = In + 1. . (18.3 °) For the sum y{w) = y(°)(«.e.) + j.35 ) .—v*+*j ( 18 .• q. p i(9»9 + 4 J'WK7».e. h ) = qh . A = ^(q2 + l)c 2 + o(J^.34) We observe that odd powers of q arise in combination with odd powers of 1/h2.^ g ( 4 g + 29) + O(^ 2 1 2 Qr2 2 4 2 / 1 \ J. h2 —> h2.27) can be removed by adding to T/°) the contribution (—^/4j)y<j+4j. . and c 2  replaced by — c 2 .(!) («. We can now write the solution y(w) in the form y{w)=yq{w) + Y. (18. Equation (18.) + j/ 2 ) (W) + • • • with the corresponding equation from which A can be obtained. This type of invariance is a property of a very large class of eigenvalue problems. i. ( 18 . (18. 2 . Thus the next order contribution to yq is y H = ie 2^ —z. In Case (3) the parameter q is only approximately an odd integer in view of tunneling. n — 0.[rp ) Y. and even powers of q in combination with even powers of 1/h2.q]=0. so that the entire expansion is invariant under the interchanges q .—A(q + 1) .1.32) Evaluating this expansion and inserting the result for A into Eq.18. i.) ty^^w) to be a solution to that order we must also have to that order [q.31) Proceeding in this way we obtain the solution y = yW („.10) we obtain E(q.2 The Inverted Double Well Potential 389 Hence a term (iyq+4j on the right hand side of Eq. . (18.
—.g±4] ±4 ±4 [g. q + At] (18.q±A) = MM!i±MM+[M±8][g±8.37) For further details concerning these coefficients. Again we can write down a recurrence relation for the coefficients Pi{Q.)]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." Since our starting equation (18. J.390 where for instance v J CHAPTER 18. there is another solution y(—z). (18.argz=0 (18. i. We observed earlier that these are obtained by making the replacements q —> —q. H. 2 4tPi{q.e.38) [yB(z)]argz=7T = [?/s(2. 5 + 4j) = 0 for \j\ > 2% or \j\ > 2i + 1.11) is invariant under a change of sign of z. Our third pair of solutions is obtained from the parabolic cylinder functions of complex argument. q + tt)= J2 i=2 p il(?> Q + ±3 + 4*)[q + 4j + 4t.36) with the boundary conditions p o(q. q) = = 1.i 1 + 4j) in complete analogy to other applications of the method. and for j ^ 0 all other P0(q. W. we may infer that given one solution y(z). g ± 4 ] T4 ±4 and so on. / l \ 1i i( ) + J2 ( ^e ) J2 P ^q + 4j)Bq+lj{w) iu=/iz. ? T 4 ] [ g T 4 . MiillerKirsten and A. their recurrence relations and the solutions of the latter we refer to the literature. Anharmonic Oscillator Potentials ±4 ±4 ±8 ±4 P2{q. Achuthan. We thus have the following pair of solutions VB(Z) VB(Z) = B w . w — ±iw. ^VO(<L<?) ^i(?. o. Wiedemann [4] . > "p. q + 4j) = 0.
(18.2 The Inverted Double Well Potential These solutions are therefore defined by the following substitutions: VC(Z) = [yB(z)]q^q. Thus in the transition region some become proportional.g.4)). so that the eigenvalue expansion remains unaffected by the interchanges q —> —q. (18. h2 — — h2 as long as corrections » resulting from boundary conditions are ignored.h^ih 391 (18. In order to be able to extract the proportionality factor between two solutions. in fact.5) are known in some mathematical literature as "stretching variables" and are there discussed in connection with matching principles. We add parenthetically that all our solutions here are unnormalized as is clear from the fact that the function in the dominant term (e.2 _4 A } 2cVl1/2 h4 8c 2 1 in the solutions of type A which are not valid around z = 0. like w of Eqs.The solutions yc. Mauss [192].q + 4j) as in ys. see e.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. h2) in which odd powers of q are associated with odd powers of h?. J. with normalization since the normalization constants are also asymptotic expansions.g. one has to stretch each by appropriate expansion to the limit of its domain of validity.g.h^ih. the solutions of type A being valid away from the minimum. Integrating and expanding as follows since h? is assumed to be large. 4 . Vc(Z) = [V!B(Z)}q^~q.z2 . Such contributions arise. we obtain: exp 8c2 3 < 2cVl3/2 /i4 h2z2 „(zA exp 12c2 (18.39) are with the same coefficients Pi(q.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.** First we deal with the exponential factor 1/2 exp exp dz z z h* + c z J2c z 2 2 . In this bordering domain the adjoining branches of the overall solution then differ by a proportionality constant.2.18. Aq(z)) does not appear in any of the higher order terms. .40) "Variables like those we use here for expansion about the minimum of a potential (e. 18.
.0x ! [ £ ( g .44) In the solution y~g(z). > n. Thus from the literature [86] we obtain Di_{q_1){w) = w*{ql)e\w* but D Wi) ^=.2)i (27T) 1 /2 e f(9l) 1 eb Uq + m (18.43) (since z — — 2 implies > argz = ±7r).. we would have to substitute correspondingly the expression (18.44) we see that in their common domain of validity yA(z) = ~yB{z) a (18. (18. y^(z) we see that in the direction of z = 0 (of course.z). with z in ys(z) replaced by — z. Comparing the solution VA{Z) of Eq.4 i .l ) ] ! ( . 2 ^i![i(g4il)]!(2«.(«.2 « . We do not require this at present.42) we obtain for the solution ys(z). 2 ) i ' ir! i iHim 1 3 j argw < 7T.ir_I + M + 1)]! (2w*y U i\[~W • i 5 with 7r > argw.392 CHAPTER 18. not around that point) VA(Z) VA{Z) = ehVl2c 2 e\z*h* z^) + o(± zfr+i)+o(±\\. 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.45) . eh* /12c* e\z*h? (18. 4/T 4 Prom (18. (18.42) H W2 (ql)e$W I „ . Anharmonic Oscillator Potentials Considering the pair of solutions ju(. The function Di. ^Jw) has a similarly complicated expansion for 1 5 — — > argw > .43) W + 1)]! w^+V A .w yB(z) ~ B.) ( ^2 „22)\ z £(</!) e 4re z 4 4 = hz: [i( 9 l)]!24(«D l+O h? (18. (18.7 r .41) with the solution ys(z) of Eq.
we see that in their common domain of validity yA(z) where a = =Vc(z).2.50) and y+(0) ^ 0.18.y'_(0) ^ 0. We proceed similarly with the solutions yc(z). (18.2 The Inverted Double Well Potential with a 2 (K ^ ( 9 .yc{z)pansion (18.46) However. The first of the conditions (18. the ratio of yA(z)iVB(z) i s n ° t a constant. (18. (18.41). we see that at the origin we have to demand the conditions yV(0) = 0 and y_(0) = 0 (18. Recalling the solutions y±{z) which we defined with Eq. and the second qo = 3.49) d yc{z) Again there is no such simple relation between yA(z) 18..9. hence we have to . Looking at the potential we are considering here — as depicted in Fig. — For instance qo = 1 (or n = 0) implies a ground state wave function with the shape of a Gauss curve above z = 0. 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. large probability for the particle to be found thereabouts.2 — we see that near the origin the potential behaves like that of the harmonic oscillator in fact — our largeh 2 solutions require this for large h2.50) will be seen to imply qo = 2n + 1 = 1.11. ^^+i)Ah^ze* [_i( g + l)]!24(9+D Inserting the ex l +O h2 (18.42) into y~c(z) w e obtain Vc(z) = (h*.e.21) as even and odd about z = 0.5.48) (h2y 4(«+i) [i((? + l)]!24(^i) l +O /i2 an (18.. 18.. i.1 ) e 12c 393 Kgl)]^"1) l + O h? (18. At z = 0 the solutions of type A are invalid.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.47) yA(z) Comparing this behaviour of the solution yc(z) with that of solution of Eq.7.
4W >S [ I ( ( ? _ 3 ) ] ! s i n { f ( g + 1)} and hence =^= 1. (18.(iu)].51) and 0 = V(0) = Jim i»xM .' '. [ .F ? .(0) (18. A.58) diu [  ( g .54) dio q(w) .394 CHAPTER 18.l)(°) Thus with the help of the reflection formula cited above: D B.1)('») + ^ r + l)]! ^. Anharmonic Oscillator Potentials use the proportionalities just derived in order to match these to the solutions valid around the origin. We leave the detailed calculations to Example 18. C.fc(0) Thus we obtain the equations (18.52) # 1 = ^ y' c (0) a and 2/B(0) = a (18.(0) = . Then imposing the above boundary conditions we obtain 0 = y'40) = limJ[y'A(z)+y'A(z)] = i»i.(0) + ^ ( 0 ) l z—>0 2. Z (18.! ( . Stegun [1]. ( 9 + 3)]! s i n { .(«.u=o follow with the help of the "circuit relation" of parabolic cylinder functions given in the literature as (q r — ) Di(."*(«!) ^ ' ~ [ . IT 4 dio In fact.g.^{(9+1)}.57) [i( 9 _l)] ! 2 i(9i) B.( g + 3)}. y c ( 0 ) .i ( g + 3)]l (18.i ) .5 ( 9 + !!)(*"')• i ( 1) . .. [C.55) Solution: Prom the literature. Abramowitz and I.. we obtain i V7T24( g . + 1)]!' D ' j . e.(0) = and htoi)(°) ^—'. = = V^ >—= = — i i ^ i [1(93)]! ^_ sin{_(q + i)} (18. 1 [7(93)]! n «(°) = ..(0) Cq(w) y' c (0) r^ Show that — with w = hz — the leading terms of the quantities listed are given by B n\\iam' 4(«Z + l)]l[i(8l)]!' i\ sin{(q3)}. one finds that [ 1 IT (q + 1)1! = .l ) ] ! [ .i)H = e.2.2: Evaluation of yB(0). Example 18.( g + 3)}. Expressions for C 9 (0). M.56) D 4(./o^Tpif (q+i) . — C.53) yc(0) a' Clearly we now have to evaluate the solutions involved and their derivatives at the origin. y'B(0)..) [i(«l)]![i(?+l)]! 2TT V5F[i(gl)J! V (18.oi(.s i n { .SxW) = \ i»fl(0) . with the reflection formula (—z)\{z — 1)! = IT/ simrz.
3)}.i ( « + i ) [ _ i ( . isin{f(g + 3 . using the above reflection formula and the duplication formula ^/TT(2Z)\ = 22zz\(z .+i)(°) [ _ I ( . Starting with the derivative expression we obtain (apart from contributions of order 1/h2) lsin{f(g + 3)} i Sin{(g .sin{^(g+ 1)} = J .i . (18. we obtain 2 i(»+i)[_i(.sin{^( 9 . + i)] V* H(?+l)]'[£(9l)]l Similarly we obtain [°'i(.53) in dominant order and insert the appropriate expressions for a and a from Eqs. (18.18. _ l)]!2l(91)(/J2)(9+D We rewrite the left hand side as 1 sin{f(g + 3)} r„.46).6 ) } ^ — i t a n < — (q + 3) >.)[i(g3)]! and ^i(.60) (1861) [C>)]o = iV2i [1(9 + l)]![i(9 " 3)]! (18.62) . 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) _^_ .2 The Inverted Double Well Potential From this relation we obtain 395 ^^^i^w^vy ^(9i)(°) (18 59)  Inserting from (18. ~ ^ 1 5(9+1) 2i(''._l) From this we derive C. W .+1iH]«=o = . 4 r T (18. 7T 4 V 7T 4 (B) Evaluation of the boundary conditions We now evaluate Eqs. + i)]!2i(«+ ) 1 [(«3)]! V5F[i(9l)]! sin{(q + l)}.3)} (h?)faV[\(q [Uq U + l)]\2*(q+V _£ e 6c^.1/2)!.56).49). (18.(0) = *>i(„+i)(0> 2 .
65) Expanding similarly the left hand side of Eq...53) becomes 8 m(I(. the energy E is real.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) about go — 3. ^ . (gg 0 )^±^ y e5?.. With a Taylor expansion about go the left hand side of (18. 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.e.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.5. we again obtain (18. (18.^ >^_>m e J*. Anharmonic Oscillator Potentials Then the derivative relation of Eqs.9. the vanishing of the wave functions at infinity). .11. and the left hand side of (18.( g 0 + 3)  + • • • ~ (g .53). In fact the left hand side of (18.. + 3)} = . z2 > z2. 18..63) vanishes for q = go = 1. 7.. Recall > the original Schrodinger equation (18.63) becomes (? . (18.9. equivalently. z^ ein/2z. Thus we have to determine these conditions first.. This is Case (1) of Fig..63) Proceeding similarly with the second of relations (18.1.1).64) the right hand side is an exponentially small quantity. by the rotations E > einE = E. 18.396 CHAPTER 18.63) and (18. The analytic continuation of one case to the other is accomplished by replacing ±c 2 by =Fc2 or. we obtain cos((g + 3)) = ._ e =*.go) ^ (1) It follows that we obtain for the even function with q — go = 1. this is the case of the purely discrete spectrum (the differential operator being selfadjoint for the appropriate boundary conditions. . (18. i.^ . For c 2 < 0 and the solution y(z) square integrable in —oo < ?Rz < oo. 7 .2) with potential (18.i .5.2. 1 1 . (18. . .64) In each of Eqs. (18.9o) ^ cos  . (18...64) for q = g0 = 3..L ^ i . .
i. It is then necessary to insure that when one rotates to the case of the purely discrete spectrum without tunneling.e. 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. if TT/3 < 6 < 0. + i s i n 3 6 0 i y ) ~3~^ c2\l/2\z\3 oc exp <M — J — .2 The Inverted Double Well Potential 397 One can therefore retain c2 as it is and perform these rotations. i. (18.e.sin 3(9 This expression vanishes for \z\ —• +00 if the angle 0 lies in the range — ir < > 36 < 0.e. i. (18.cos 36. —}.68) . in the domain — TT/2 < 36 < > 7r/2. Then c 2 y/vi ^p^My) yf = exp _ r HvcV2!*!3 cos36.+00 in arg z € (!•" * 0 for 5ftz — —ex) in arg z € » N> (18.e. we see that the solution with the exponential factor is exponentially decreasing for \z\ — 00 provided that cos 39 > 0. 2 c2>0.e. 6 6 In the case of the inverted double well potential under consideration here (i. the resulting wave functions vanish at infinity and thus are square integrable. replacing sin 30 by " 0 + ^) 1 IT = . 1/2. we set z w r = \z\eie. Now. r / c 2x 1/2^3 = exp{ ± i — —\.66) z3 = \z\3em. the case of complex E). or — < 6 < . i. we therefore demand that for $lz — +00 and • —7r/3 < arg z < 0 the wave functions have decreasing phase. Thus exp exp (<£ff} [+<£ )%} sin * 0 for $lz —s.67) Rotating z by vr/2. Thus in our case here the behaviour of the solutions at infinity has to be chosen such that this condition is satisfied.18. y{z)~exv\i[) 7T dV'V.
68) on y±(z) (by demanding that the coefficient of the solutions with other behaviour be zero). This is not the asymptotic behaviour of a wave function of the simple harmonic oscillator. (18. We have to remember that we have various branches of the solutions y(z) in different domains of z. This is the boundary condition also used by Bender and Wu [18]. Our procedure now is to continue the even and odd solutions (18. and there impose the boundary condition (18. 18. For c 2 < 0 we have correspondingly y(z) ~ exp I ± ( — 1 — y c2 < 0. Equating to zero the coefficient of the term with sign opposite to that in the exponential of Eq. 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.21)) were defined in terms of solutions of type A which have a wide domain of validity. for z > Too. Our even and odd solutions y±(z) (cf.e. (B) Evaluation of the boundary conditions The following considerations (usually for real z) require some algebraic steps which could obscure the basic procedure.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.Z\.68) for c 2 > 0. Looking at Fig. we match the WKB solutions . 18. We therefore explain our procedure first. (18.3 The inverted double well potential with turning points ZQ. We do this extension with the help of WKB solutions.21) to + infinity and to demand that they satisfy the condition (18.68) 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.398 CHAPTER 18. Eq. i. Anharmonic Oscillator Potentials V(z) Fig.Z\.
We begin with the exponential factors occurring in Eqs.^ 4 1/2 + icV 7T Jz + (18.V 2 iL44 + <?z / „2 4 X COS < f / il/2 G?Z g/l 2 1 2L4 z 4 M 1 2i. z2h4/A > c2z4/2.18. . z\ ~ . i. A J24 1 ? 2 • "^ •z h \—c z ~ — q h . B.69) The W K B solutions have been discussed in Chapter 14. 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. 18.=c z 4 dz 1 L2 9 + 2 4 1/2 x exp . 291.71) We now come to the algebra of evaluating the integrals in the above solutions. h2 (18.2. 291.4) for E and ignoring nondominant terms) t o . Sec. ^ 2 4 •K + —(r. The distant turning point at z\ as indicated in Fig. ^2? 1 2qc2 /2? .zi)/ \ 1/4 ywKB\ ) z — g/i* . (18. p. I [195].2 The Inverted Double Well Potential 399 to the left of z\ to the solutions of type A. f R . Vol. Dingle [70].4.B. 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\.e.70) where z < z\.. *R.70). Messiah.4 h + c z 2 1/4 < .^ T rz\ 1 + yh* .*l) Z 1 2L4 1 qh? + zzh* 1/4 _2i4 VWKB\ ) . (22) or A. p. equations (21). 4 2 T ' i. Dingle [70]. In using these expressions it has to be remembered t h a t the moduli of the integrals have to be taken.z\) ( \ y^NKQyz) qh2 z 4 dz h H—c z 2 x sin ^ .3 is given by (using Eq.e. 6. 1 2i4 1 2 4 / ~ 2 4 ~ 2 C * (18. (18. t To the right of the turning point at z\ these solutions match on to (r.
Anharmonic Oscillator Potentials E± = exp = exp .2^3/2 2_(h^\ll2 2czz 4 N 1/2 8?3\ 2 ~hf ' =F9/2 ^ 2 ^ . We obtain this factor by expanding the expression in powers of zjz\ (since in the integral z < z\)..2. (18.2 \ 1/2 .±9/2 = ^ V M z^l2 21 2c2 exp ± i dz 1 4 1 1 V2 2 (18. zi 2q 1 c 2 [1 . Eq.72) Here the first part is the exponential factor contained in 2/A(Z)) J / A W respectively (cf.400 i. (18.40)) E± = exp _/^_2 f ± In A 2c2 .73) . CHAPTER 18.i c V 8c T 2 1/2* ti>_(2 ~ exp 8c \3 2 r< 2cV 2c2 z2 i24£Y'2u h* 3/2 >. Thus the above factor yields '2c2\l/2 h4 so that (cf. (18.69)) 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 2dependence contained in this expression.211/2 1 \ /j. Eq.2. In the remaining factor we have (looking up Tables of Integrals) 21 2 dz h 4 2 2c.^ i ^ ^ c 2 .40)).^ I 2 2 1/2 zV = F 8c2 Jz [i . (18.\qh2 + \z2h± .^ ] V 2 and so /^2f 8c2 3 1 2cVl3/2 h* f q fZ\ ck h^y/ E± = exp T 2 jJ Z ^ [ g .4 In 2c 2 J + 1/2 ^1 2c2 1/2 21 and hence (with use of Eq.e. .
20b).76b) apart from factors [ l + 0 ( l / / i 2 ) ] .75) 9/2 1/2 0= or '(2c )V2 2 and (3 = — 2 2 (fr2) (2c 2 ) 1 /2_ 9/2 (18.Zl) ( \ yWKBV^J \z h* 2 1/4 and ywKB(^) \z h* 2 1/4" (18.74) Comparing these solutions now with the solutions (18.76a) P P r 2/1 2 " 9 ( .79) +S_(±)exp .\z2h4 + J Z\ nl/2 \M fZ .2Z 4 C S ' + ( ± ) e x p I i\ cz / cz3 V3\/2 U^2 h6 12c 2 12c2 J (18.18. (18.71) and these into (18.21 1/2 . In these expressions we have chosen the signs of square roots of h4 so that the conversion symmetry under replacements q — —q. > » Returning to the even and odd solutions defined by Eqs.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.20a) and (18.78) Inserting this into the solutions (18.77) 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) we now have V±(z) 1 2 \yA{z) ± yA(z)} = \W^(z) WI ± _(J.(J*i) VA 0 ) = PVwKB (Z C.*l) (18.1 ) 9 / 2 _(2c 2 ) 1 /2_ 1 (18.*l) (3y^(z) (18. we see t h a t in their common domain of validity VA (z) = /3ywKB 0)> where .77) Now in the domain 2 — oo we have > \qh2 . CZ2 JZ1 cz 3^2 V2 h6 12c 2 ' Ac2z2 cz" 3y/2_ (18. h2 — — h2 is maintained.
18.81) Inserting expressions (18. we see that we have to demand that 5 + ( ± ) = 0. Inserting this into the latter equation we obtain (the factor "i" arising from the minus sign on the left of Eq. (18. (18. Anharmonic Oscillator Potentials where 5+(±) 5_(±) = = Q / ? ± ^ e x p ( ^ ^ T ^ ) e x p ( .68). E(q.34).6 The complex eigenvalues (1883) We now return to the expansion of the eigenvalues.^ j . (18.82b)) /j6 \ 9o/2 («*) = W? mam ''*• with qo — 1./i2) = E{qQ.h2) + (qqQ)h + (18. h2) = \qh2 . .e. i.. (18. (18.3..80) Imposing the boundary condition that the even and odd solutions have the asymptotic behaviour given by Eq..82b) This is our second condition along with Eq.402 CHAPTER 18.. i / 3 ± .^(q2 + 1) . this equation can be rewritten as or as the replacement + / u6 \ 90/2 (/i 2 )W 2 =^(_) 2®"1 f — J ..e./ ? = 0.h2) + {qqo)(^pj = E(q0.85) + . (18. (18. (18.76a) for (3 and /3. i.65).^ § ( 4 g 2 + 29) + O ( ^ ) .84) Expanding about q = qo we obtain £(g.2.5. Eq.
~p{q. 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 + 29) + o ( ^ ft6 \ 2qoh2[ —~ 2 ( ) < • 9o/ 2 ' 2c J ft 6 /6c 2 '(27r)V2[i(gol)]r • (18 86)  The imaginary part of this expression agrees with the result of Bender and Wu (see formula (3. where the superscript {r. [19]) for ft = 1 and in their notation 90 = 2 ^ + 1.g. ft6 ^ = e. o 2 + l)^(4.86) will be dealt with in Chapter 20.yA{z).36) of their Ref. ^ 2 4 y qh 2H ""{f' z h H—c z 4 2 1/2 1 ^ 4 + IC2Z4 dz qh2 4 2 .18. !*WKB(Z) ~ 2 . Solution: The turning point at ZQ close to the local minimum is given by qh2 2* 4 z2hA c z ~ 0.e.4 1 2 4 1/2 ft + i c V 2 1/4 + f}' z.g ( . above) (1.h6/6c2 E = E(q0.2 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 .l ) ] ! > + = ^ 2 . The large order behaviour of the eigenvalue expansion (of Case (1)) which Bender and Wu derived from their result (18. h2).h2) The final result is therefore E () i ( 2 7 T ) V 2 [ l ( g o .3. h2) of the relations KWKB (Z) =PVA(Z). Since this is not without interest (e. in comparison with the work of Bender and Wu [19]) we deal with this in Example 18. 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.2 . * _ qh2 1/4 *• 2 j _ 4 i 1 2 4 z h H—c z dz qh2 2 1 2L4 •{/. i. zo} means "to the right of the turning point ZQ" .3: Matching of WKB solutions to others at ZQ Determine the proportionality constants p(q.ZQ) . S W K B ( 2 ) = PVA(z).
f x Zn 4) h2 2.1/2 h? \z{z zlf' \z \n\z 2 2 2 + {z2z2Y/2\ 2q h2 4 • In (V2zh V s/q Then to the same order of approximation ..ic224 2 4 qh2 + i^fe 4 . Schwind [247]. and we have 1/2 dz h?_ 2 J /i2 qh2 + iz2h4 _ ^ 4 J dz zn qh2 + z2^ 1/2 .404 CHAPTER 18. N. 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 . the (approximate) solutions* 1/2 exp • ± i ( . Anharmonic Oscillator Potentials where zo > z. 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. cf. 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.O.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. . 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." / 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./ ^ 2 X 1 / 2 exp(ifi.g In other words. 2 z 2 ) / 2 e \ .zo)/ \ qh2 2 ZO + z2hA 4 2 .zo) % ' K B (Z> 1 / 4 2 l / 2 h z V Q / 2 7i/2 4 2 e x p ( ^ e .
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. (18. i. h4 > 0.3 18.23)) appear quite naturally.5 "l _ o f . we shall employ basically the same procedure as above or.kyil(. 4" ) (h?z2)^q+1) for q j£ odd integers.89) . J/J Hence ^ and h so that \]q] \2"'2 ~ ( 2 7 r ) 1 / 2 e .yA{z) with expressions 27WKB i J'WKB ' that the factors p. ~p requested at the beginning are given by /1y/2[j(gi)]i2^c^ h 6 /12e3 /ly^ih+iwi"" w 18. 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.41) for yA(Z).zo).3. We can reexpress these relations with the help of Stirling's formula.^ 1 . in fact._ .88) + [EV(z))y{z)=0 (18.z ± ) = .DfJ»v we see Comparing now the expressions (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. as in the case of the Mathieu equation. in Eq. We consider the following equation ^f& with double well potential V(z) = v{z) = jz2h4 + ^c2z4 for c2 > 0. „ / M 1 / 2 i .(r. (18. (j ." / 4 ( ? « / 4 .3 The Double Well Potential 405 It may be observed that 3/Y/KB ( z ) corresponds to solution (3.g. Then . But there are significant differences.87) The minima of V(z) on either side of the central maximum at z — 0 are located at h2 h8 z± = ± with y(.18. (18.13) of Bender and Wu [19].e.1)! = ( 2 7 r ) 1 / 2 e .J ' i J ..
z±) y = 0.91) (18.4 The double well potential.i>5.z Fig.92) (thus we sometimes use h4 and sometimes Ii±) and EV{z±)=lq±hl With the further substitution + ^. (18. (18.91) becomes T>q±(w±)y(w±) = o(^)y. (18. 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 .94) (18.93) w± = h±(z . .z±).406 CHAPTER 18. Anharmonic Oscillator Potentials and V{2\z±) V(4)(z±) = = h\ 12c2. (18.90) V(i)(^)=0. 18.95a) . the previous equation (18. vW(z±) = ±Qch2.V(z±) l{z z±fhA + 0[(z .
95b) with the equation of parabolic cylinder functions u(z) = D„(z). (18.1. where U{z) = and near a minimum at z± 4_ 4 jp[V(z)V(z±)]. Thus again.96).. i. n = 0. we obtain another by replacing z by —z. Eq.97a) U(z) = (zz±)2 + 0[(zz±)3}.87). we conclude that in the dominant approximation q± is an odd integer. (18.93) for E into Eq. d2u{z dz2 + v+ 2~r u(z) = 0.95b) By comparison of Eqs.100) . (18.96) dz2 + 5 « 4 +  ..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)TV2Lz2I^\A'(z)TV2czA(z)+(^q±h ±+^r)A(z) = 0..3. we obtain d*y (18. (18. (18. Inserting the expression (18.3 The Double Well Potential where ^ 407 d2 •« 2 2  d8.95a) and (18.98) 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. (18.97b) Our basic equation.i 4 f W » = o.2.18.2 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.. qo = 2n + l. given one solution.e. 18. is again seen to be invariant under a change of sign of z.
h2.101b) To a first approximation for large h2 we can neglect the right hand sides. h2 h4 2c' with g+ = q. z) = yA{~q.1 6 ) . z) may be obtained from the solution yA{q. ~h2.c W + 1) . we observe that the solution y^iq. (18.105) "8 + 4 ^ .^ A"{z) + 2c V2 *A(z) (18. (18. Vq = {z2+z2)— + (qz+z).408 Since CHAPTER 18. yA(q.102) We observe that a change of sign of z in this equation is equivalent to a change of sign of q. Integration of Eq.h2.yA(z) are associated with one and the same expansion for A and hence E. The result is given by A and E(q.4.e. (18. but the solution is a different one. h2.104) Both solutions yA{z). We leave the calculation to Example 18.103) Looking at Eqs.106) . Aq(z) = Aq(—z) = Aq(z).h2) . The dominant approximation to A is then the function Aq given by the solution of the first order differential equation VqAq(z)=0. n± = V2h2. i.^q(l7q2 6 4/i + 19) + (18. 8hw (18.101a). z) = yA{q. Anharmonic Oscillator Potentials and selecting z+ (z2+z2)A'(z) h z+ = —.102) yields the following expression M*) I z2 _ z2 11/2 *+ z + z+ q/2 z+ 1(91) \z + z+ 1(9+1)' (18. z). i. (18. these equations can be rewritten as + (qz+z)A{z) = .101a) {z2+z2)A\z){qz+ + z)A{z) = ^ A"{z) + £A{z) (18.z) by either changing the sign of z throughout or — alternatively — the signs of both q and h2 (and/or c).e. h2. (18.101b).c W + I ) 2C yfi 2h 5 2 A V2c* g(17g2 + 19) + 0 ( / t .
114a) .101a) with A replaced by A .107) KV) ~ Aq(z) {z*z\y Aq{z) (z 2 z\f (z . (18.{z2 .3 The Double Well Potential 409 Example 18. for instance by componendo z z et dividendo.. we reexpress A'q in terms of functions of z multiplied by Aq.105) and (18.qz+) [2z2 .qz+)2 .106). First.4zz+q+ z\{q2 + 1)]. i.18.112) From these we obtain. The first such step would be to reexpress the right hand side of Eq. (18. J = 0. such as periodic potentials.2 z\) • Aq$4 £Aq + Aq—4. as a linear combination of terms Aq+2i.*l) (* ~ "+> ) A (z)exp q /2 \ 3 4c J (18. . ± 1 . (18. A Differentiation yields » = ^^w (18. 92 ZZJf (18.e.103): . ± 2 .(l + + 2  z y* . Similarly we obtain _z+_ Z _ Aq + 2 — Aq Aq+2 + Aq 12 ^g+2 A ± ± _2A 2 ± 6 + (18. . . .+ 2 ^ ± l + 2^±i + .111) {Aq+2  Aq2)2 (4zz+)2 {z2  2 .110) Aq+2+Aq2 9+2 A. <4q+2 — z — z+ Aq+2iAq+2j (18.102).Aq 1 + 2^± 2 Aq+2 (^M^f (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. . Thus it is natural to explore analogous steps. however. . z • .jf2i(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 . + Aq+2 + Aq Aq+2 .113) 2 .z2 ) N t. (18. (18.z\) + 2z(z . (18. We know the first derivative of Aq from Eq. = Aq+2i+2jAq.103) for Aq{z) is very similar to that of the corresponding solution in considerations of other potentials.109) We observe some properties of the function Aq{z) given by Eq. Solution: The structure of the solution (18.108) We wish to rewrite this expression as a sum y coefficientiv4.
(18.9=F4) = (g=Fl)(?=F3).q + 2)Aq+2 (18. .q2)Aq2 + (q. Vq+2iAq+2i=0 and Vq+2i = Vq + 2iz+. q) = 0.l ) 2 A . i.4 4+* + .(0) V2 ' 2c K + h4+Ao . (q.<? +{q. QA?+4 _ 10 A + 6 .q)Aq + (q.+6 + • • • ] . (18.e.q + 6)A<. Anharmonic Oscillator Potentials A 1 .4)A„_ 4 + (q. Eqs.4]j. (18.4 ( g . (17.3 ) A .4A ? _2 + 6Aq .114b) .101a) the contribution R . where the lowest coefficients have been determined above as (9. cf. q + 4)Aq+4 + (q. _ 2 + 6 ( 9 2 + l)A ( ] + (q + l)(q + 3)Aq+4.115) and (18. •••]. (18.118) 2s/2c ~^j[(?.117) 4(q + l)2Aq+2 Here the first approximation of A. . (q.115) (z*z%) 2 i2 (z 2 .2 ^9+2 „^9+4 + \4z \4z+J z+ Finally we have also [A q _4 — 2Aq2 + 2Aq+2 — Aq+4 (18.l ) ( g . leaves uncompensated on the right hand side of Eq.q + 2) = 4{q + l ) 2 . (18. ' T h e reader may observe the similarity with the corresponding coefficients in the simpler case of the cosine potential. In particular the dominant approximation of A is obtained by setting (q. we obtain 5 K = ( ^ f ) [ ( 9 .410 and from this = Hence with Eq. (18.4Aq+2 + Aq+A Inserting (18.z\Y Z2 + {~)\Aq+42Aq 1 224 Aq. <?) = 6(<j2 + 1) + z: 2 2A +72"' (18. _ 4 .34).112) CHAPTER 18.119) It is now clear how the calculation of higher order contributions proceeds in our standard way.112).116) into Eq.2Aq + Aq4] 1.108). A(°~) = Aq. Since VqAq = 0.2\2 (z* .z%) 1 n z+z 2 \4z+J [Aq+4 [Aq+4 .aAg+S 12+ 16 (18.2A g + A g _ 4 ] .116) [Ag4 .33) and (17. 1_4^±^+8^±±12^±^ +.
h2.h2. (g. for i = 0.z)±yA(q. i. h2. 9 + 2) Aq+2Az+ (9. g .g) 4z+ {q q)+ 2r (9. y'+(q. Terms jj.0)=0.124) .g + 2)(g + 2.105).2 ) (0) (18.e.g2)(g2.94)(g4.0)=0. g + 4)(g + 4.g ( 1 7 g 2 + 19). A' (0) = q/z\) y+(q.g) 8z+ (g.e. (18. i. where Rq is the sum of terms left uncompensated by AW.101a) the contribution Rq . h2. in the domains away from the minima. 9 + 4) ^+48z+ (18.92) ig4 ' H ~ 23/X4 iz+ An2 H + (9.4 ) ( 0 ) ( g .3 The Double Well Potential we have T>o 411 2iz+ J Aq+2i except.{^ h2.(0).g) Az. \ZZ±\ > 0 ( y We can define solutions which are even or odd about z = 0 as y±(z) = = £ fe^te' h2> z) ± y A. h2 2y^ Age? h4 (18. g) 8z 4 .94) 8z+ (9. (18.0) = ^ .h2. (i) _ / V2c\ [ ( g .h2.h2. g .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. The first approximation A(°> = Aq leaves uncompensated on the right hand side of Eq.yA(z) derived above are valid around z — 0.122) The solutions yA(z). *)] (18.121) This coefficient of Aq set equal to zero yields to that order the following equation 0 £) ' {^) (g.Aqjr2i in this may therefore be eliminated by adding to A^0' the contribution A^1' given by (9. y'_(q. z)\ = 2^A^ h2 '>z)± VA(Q. thus yielding the next approximation of A as given in Eq.123) ^[yA(q. which reduces to 2 \/2c 2 0 = 2(3g 2 + i) + — A + ^ g .28.0) = y^(q. (18. of course.z) Considering only the leading approximations considered explicitly above we have (since Aq{0) = l/z+ = A.
18. Bq[w±{ ^ .95b) — 1 T>q(w±)y(w±) = j£ Thus we write the first solution yB(z) = Bq[w±(z)] + 0(h±2).3.Evaluating the exponential factor contained in VA{Z) of Eq.We draw attention to two additional points.2A y(w±).95b) that the solution there is of parabolic cylinder type. (18. + 0(hl2).125a) = VB{~Z) = Bq[w±(z)] + 0(h±2) = Bq[w±(z)] + 0(h±2) (18. Since w±(—z) = —h±(z + z±). (18. we > . Inserting (18.^ % [i(?l)]!24(«. in view of the factor "i" in the argument of Cq the solutions VAJVC have the same exponential behaviour near a minimum. (18. 2(9+) . Anharmonic Oscillator Potentials Our second pair of solutions. we have w±(—z±) = —2h±z±.412 CHAPTER 18. This means.125b) It is clear that correspondingly we have solutions yc(z).91). (18.3 M a t c h i n g of solutions Next we consider the proportionality of solutions yA{z) and VB(Z).126a) hjW + l)]! VM = Vc(z) = Cq[w±(z)] (18.126b) These are solutions again around a minimum and with yB(z). (18. but w±(z±) = 0. (18. in this case we use the Schrodinger equation with the potential V(z) expanded about z± as in Eq. yB(z). and another VB(Z) ± 25^ch3wl + c2wi .23) with appropriate change of parameters to those of the present case. We see already from Eqs.1 ) D (18.yc(z) providing a pair of decreasing asymptotic solutions there (or correspondingly y~B(z)iyc(z)).yB(z) is obtained around a minimum of the potential.99) for z — z±.yc(z) with complex variables and Cq(w) given by Eq. the equation is — with differential operator T>q as defined by Eq. Thus yc(z) Cq[w±(z)} = = Cq[w±(z)]+0(h±2).95a) and (18.93) and setting w± = h±(z — z±). Moreover.
Later we calculate the coordinates of turning points ZQ..18. ein+z+e in+(z 22+(<?+i) . 2 An+z+e 4w+ (18.z±)dz .frW.127) Recalling that around arg w± ~ 0.146). .172) obtained later.ZI and find that these are given by z+ + 0(l/h) for finite q (cf. .147)).172)) we add here comments on this approximation. the dominant term in the power expansion of the parabolic cylinder function is given by Dv(w±) ~ wv±ew±>4. Thus for h very large these turning points are very close to the minimum at z+.129) . (18.164)) only in a nonleading contribution and hence implies the equivalence of the exponentials in the relation (18.^/w + ± 1 .• • " .97b)) exp 413 '\hlJQ'uV\z)dz = exp = exp (18. . we have \z ~ Z+\^Q ^ l f e 2 2 2 _kh2( ~ i . (18.« + ) 2 (9+1) e (18. We observe that for z — z± the approximation yields exp[±/i? t z 2 t /4].97b) ~ exp . Eqs.3 The Double Well Potential (cf.• )2 (^)itoi) ( 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.125a) we see that (considering only dominant contributions) in their common domain of validity e ^+z+ l + O 1 yA{z) = yB(z). . and comparing with Eq.i*l(«^ 4'"+^+ 4"'+v ( « . (18. 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. (18. a= ^ . 24(9D[i(gi)]! hj (18. (18.128) Similarly we obtain in approaching z+: VA(Z)  ( 2 ^) (9 " 1} e. Allowing z to approach z+ in the solution IJA(Z). (18. .^(±>(^*4 ^\h\{zz±f exp 7>2 r 2 ^n±z± For later reference (after Eq.
We have to impose boundary conditions at the minima and at the origin. D. K.414 and CHAPTER 18. P. The pair of solutions of type B is defined around argz = 0. 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. The involvement of these WKB solutions leads to problems. S. g .130) Therefore in their common domain of validity a a = dte+vM' {2z+)^. since. Various investigations^ therefore struggle to overcome this to a good approximation. K. Bhattacharya and A. they assume large quantum numbers. Rau [22]. Thus 1 E .7r and the solutions of type C around argz = ±7r/2. H. Bhattacharya [23].\h\)^+l\\{q l + l)]\ 1+ 0 J_ . J. basically. 18. 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. but it is clear that none of the above solutions can be used at the top of the central barrier. we naturally expect the wave function there to be similar to that of the harmonic oscillator.(18. Anharmonic Oscillator Potentials D_h{q+l)(iw+(z))2i(o+V Vc( ) z 2i(g+1)ei^(z_. R. and this means at both minima. Furry [102] .+)2 [1(1 + I)]' [i(q+ l)}\[ih+(z . The solutions in terms of parabolic cylinder functions have a wide range of validity. The next aspect to be considered is that of boundary conditions. even above the turning points. de Deus [66] and W.z+)]fr+V ' (18. Since it is more probable to find a particle in the region of a minimum than elsewhere.3.131) We have thus found three pairs of solutions: The two solutions of type A are valid in regions away from the minima. S. Thus it is unavoidable to appeal to other methods such as the WKB method to apply the necessary boundary conditions at that point.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.
5.Zii yB{z±) a ± a yciz±) (18. ^(*±)#0 (18.18. and y'±(z±) V'B(Z±) ± =y'c(z±) (18. y{z±) £ 0. 18. as indicated there.135) Hence the conditions (18.5.136) . 18.3 The Double Well Potential 415 the most basic solution would be even with maxima at z±.132). We have y±(z±) = ^VA{z)±yA{z)]z^. y'_(z±) = 0.132) Fig. (18. 18.5 Behaviour of fundamental wave functions. as indicated in Fig. However.133) imply yB(z±) yc(z±) « a' and y'B(z±) y'c(z±) a a (18. and y+(z±) = 0. as indicated in Fig.134) y'+{z±) ± 0. The odd wave function then exhibits a correspondingly opposite behaviour.133) y+(z±)^0. We have therefore the following two sets of boundary conditions at the local minima of the doublewell potential: y'+(z±) = 0. (18. At dtz — ±oo > we require the wave functions to vanish so that they are square integrable. y(z±)=0. an even wave function can also pass through zero at these points.
136) as  M .141) and cos<J^( 9 + l)> 4 ~ cos  ^ ( g o + l)  ~^(qqo) (_l)*fo>i)( g _ g o )?[ 4 for sin   ( g 0 + 1) J + go = i.„*{!<. also (18. . (18. (18. Thus sin {^ +1) } = ^^>^30fe^i. . 1 1 ..416 CHAPTER 18.134) this equation can be written Now sin  ^ ( 9 + 1 ) 1 oi sinl j(q0 +I) \ + ^(qq0) ~ (_l)^(9o+i) ( g _ g o ) for cos I ^(q0+ !)[ + ••• 50 = 3 .54) +l j 2^[I(Ql)]![I(g3)]!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. _ 8)} s .142) (<Z9o)J(l)^°+1)(l)1/2 .136) the dominant approximations we obtain (cf. 7 . (18.«*{=(. . (1. +1)} = 4 m.m Using Eq. Anharmonic Oscillator Potentials (B) Evaluation of the boundary conditions Inserting into the first of Eqs.2 we can rewrite the second of (18. (18.5. Eqs. .138) Using formulae derived in Example 18.
6 Turning points z$. y'_(0)^0.5 (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.3. 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. We emphasize: We needed the solutions of type A for the definition of even and odd solutions.18. y'+(0) = 0. y+(0)^0. 18. we must also demand this behaviour here along with a nonvanishing Wronskian.1 " <813 1 4)  In Example 18. Since the type A solutions are not valid at the minima. 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. We deduce . We do this with the help of WKB solutions. V(z) — • z E=V Fig.z\. Hence we have to impose at z — 0 the conditions j/_(0) = 0. (18.3 The Double Well Potential Thus altogether we obtain 417 ^ • .» ^ B ^ ' .
We also note that the height of the potential at the turning points is h8 qh V(z) 20. one finds that . As a consequence. Anharmonic Oscillator Potentials from Eq.146) and Iqh* 1/2 z\ Ac" + +o ^ (2g2)V4 2c + /i ' (18.145) \qh\ + \zW Using z+ = h2 /2c. ZQ) means "to the left of 20" .148) ("f * "The superscript (I. h8 25c2 (18. to the left of ZQ where V > E. the dominant terms of the WKB solutions are* 1 yWKB\ ) Z 1 1 ~l/A 1/2 ~ qh\ x exp + h%U{z) .\qh\ + \h%U{z) (18. + £\h%U(z)=Q. 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.^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.e.96) that the two turning points at ZQ and z\ to the right of the origin are given by \qh\ i.24 C Z O 1/2 ZQ = Ac2 + + °(r A_ 2c 1 2 2q\ 1/2 2 / cq 5 + h± + 0(h~ ) h%) (18.418 CHAPTER 18.e. In the domain 0 < z < ZQ. in leading order the integrals to be studied below from ZQ to z = 0 are equal to those from z+ to z = 0.147) Since the minima are at z± — h2 /2c with /i 2 very large. (18.144). (B) Evaluation of the boundary conditions We now proceed to evaluate the boundary conditions (18.
^ + ) 2 + igln2(zz+).18.148) and (18. .149) In order to be able to extend the even and odd solutions to z = 0. Hence we have for ywKB the exponential factor exp ("f * Z+ \*l*e\*(\h% 1 1/2 \qh\ 9/4 + \h\U{z) e\{zz+?h\_ (18152a) .149) and consider both types of solutions in a domain approaching but not reaching the minimum of the potential at z+.y^(z) to 2/VVKB(Z) anc ^ ^ W K B ^ ) . 20 dz \qh\ + \h%U{z) .^ e therefore have to consider the exponential factors occurring in (18. )+^*+)2lr} rnD h\z'2)}z Q )\ = .(29/^)1/2 1 ^(zz+){(zz+) 1/2 2 2q_\ 1/2 hJ 2 l l q+ q  ^ ( ^ .150) Evaluating this we have ±h ~ h2 GMln 1 In 2g q In +4 1 (z .i (2q ln (18.^ j O e .S2 ) dz rzo 1/2 'Ihl + ^Uiz) I 1/2 / Jz dz (z~z+)  w + 1/2 1 = ±lh+ 1/2 ZQZ+ Z — Z+ 2 i^$) 1/2 ~4 >'^. we have to match yA(z).3 The Double Well Potential and 1/4 419 ywKB\ ) z — \qh\ x exp I / + \h\U{z) 1/2 I.151) In identifying the WKB exponentials we recall that y^KB is exponentially increasing and ywKB is exponentially decreasing. Thus we consider in the domain \z — z+\ > {2q/h2+)l/2\ h 1.* 4+ . (18. (18.
of course. 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. using Stirling's formula (and not the inversion relation) — we have —(1. 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. \ yWKB\ ) z 1 fl(g3)]! V*(h%)V*\zz+\h(*+V\2 q/A h + e* \(zz+)W+t (18.e. t h a t the Stirling approximation is amazingly good even for small values of the argument).125a).126a).154) . the results necessarily require adjustment or normalization there in the q—dependence. Since correspondingly 1/4 2 l/2 \*% we obtain + 4/1W*) (*^)i/'(M)i/*' 27T 1 / 2 {h\)^{\{q1)]\\2"+ *+ 1(91) ~hi l2 q/A e \{zz+)*hl for \z — z+\ > m 2q \ (1 ' (18. In a corresponding manner — i.125a) and (18. This is the aspect investigated by Furry [102].153) This expression is valid to the left of the turning point at ZQ above the minimum at z+. Thus using the Stirling formula z\ ~ e~(z+1\z + l)z+2y/2~rr. However. 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.ZQ) . l/2 ~h\U{z \(zz+?h\ (18.420 CHAPTER 18. We see therefore. since there is no way to obtain an exact leading order approximation with Stirling's formula for small values of q. 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.
158) Applying the boundary conditions (18.5 for the calculation of the tunneling deviation q — go by using the usual (i.3 The Double Well Potential 421 where [\q}\ was written as [\(q — 3)]! for q large. (18. (18. However. as the results also seem to support. this result is somewhat imprecise.103) and (18.e. . we obtain in leading order (i.156b) Using again the duplication formula^ the ratio of these constants becomes 1 7 (h2+)i/2(2z+)i [faW e 2n+z+ — Aft 2 72 (18. it is our philosophy here that the factorials with factors occurring in the perturbation solutions are the more natural and hence correct expressions. .K?+1).159) 7 i(«i) r(9^3) ^2~ j(9l) .123) we have V±(z) 1 [yA(z)±yA(z)} = ^y^l(z) ± ^j#&(*) (18. (18.18. The relation (18.144) we obtain fi(0) Tin the present case as 1 (91) 1 — 5 (18.99). Comparing for z —• z+ the W K B solutions (18.e. \hl.143) T^^*)) 27T (18. multiplied by (1 + 0(\/h\))) the proportionality constants 7 .e.153).143) which was obtained with our perturbation solutions from the boundary conditions at the minimum. linearly matched) W K B solutions. (18.157) Since the factorials [^(q — 1)]!.157) is used in Example 18. 27T1/2 9/4 1VA{Z). Returning to the even and odd solutions (18. 7 of t h e matching relations y^NKB\z) i.104). and is shown to reproduce correctly the result (18. [\{q — 3)]! are really correct replacements of [\q}\ only for q large.154) with the typeA solutions (18.155) (2z.n('>2o) 3/WKB(Z) =1VA{Z)I (18.156a) and 7 _ i\(Qm •K •/2(M )l/4 ^ 2 n« q/4 (2*+) •(9D e ^^4.
iV2 ! i . D. 213. n + uY 2c „ ^6 8V2c2 (18.149) near z = 0. .162) T h e integral appearing here is an elliptic integral which can be looked up in Tables.148).05.. Anharmonic Oscillator Potentials Thus we have to consider the behaviour of the integrals occurring in the solutions (18. 3G2 12V2c ^See P.163b) The integral hi?) .z2)(b2  z2).161) we can rewrite the integral as r I2{z) = = dz^{a2 .k'/2. Friedman [40].19.422 CHAPTER 18. 60 and 361. formulae 220.164) / 2 (0) z ca [(1 + k2)E(k) . F.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. (18. Byrd and M.163a) (18. (18. G2 K [i 4.3 evaluated at z = 0 is then given by = y/T+u[E(k) uK{k)} (18. K(k)} 3 ^ 2 y/T+u[E{k) uK(k)\. p.160) Setting 6 =4?^T"z°' rb Ac2 + VQ—> (18. p. 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. ...
1. It then follows that (again in each case in leading order) '2_hl)9/4 ^ ^ * > l * = ° ^ 23g/S(_g:)t/4(!)g/4eg/4 "1^Using Stirling's formula we can write this e ( 18 . ^ ^ . (G\ .( ft 7.1)]' Gfo.2.§ Since the integral is to be positive (as required in the WKB solutions) we have to take the upper signs.r1)/10//2 ( f g 1^ yS» /L ( ^ ) l^4 .( 2V + l)ln V 4 / 4( 2 n + l ) l nV 2 n ... where the result is shown to be (with q ~ q0 = 2 n + l .1.± ' 3G 2 ' T ( ^ (18.) = 2 \. n.166 ) ^^iffil/^(2*)1/affiii£ Correspondingly we find ^ (18168) S W I 2 = o .3 The Double Well Potential 423 where K(k) and E{k) are the complete elliptic integrals of the first and second kinds respectively.1 " e " O T (18 169)  Y^V / 4 tfo. which implies small G2. n = 0..18.e T ^ 5 U W ( 2 „ ) i / 2 ( ^ i W4C • (18 170) (18 1?0)  ^The expansion of I2 used in Ref. ± . . /2n + l 2 ± — . ..2 A( L ) 4 s / and z dz yWKB\ J and d^WKBW 2=0 (*?) a ( .165) in agreement with Ref. . The nontrivial expansions are derived in Example 5. Here we are interested in the behaviour of the integral in the domain of large /i 6 /c 2 . [167]. [4] misses an n—dependent power of 2 in the result. .3 )l ! .
157). (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. h2) the split expressions 29o+1/i2(—W2 h6 (18.173).i ) ] ! e ~ s A ' .172) into Eq.3.143) the replacement: {hlY'2{2z+fe^zl + 2 9 ( ^ y Z e"A. (18... (18..»W.175) ~E0(q0.h2) + (qq0)^=. . we can therefore impose the boundary conditions at z = 0 by making in Eq.h2) and hence the splitting of asymptotically degenerate energy levels. and obtain ( * . (18.174) .^ W ^ / 4 t o . (i*™) We obtain the energy E(q. 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. .n = 0. (18. (18. 2 .3 concerning the exponentials.106). . Anharmonic Oscillator Potentials With these expressions we obtain now from Eqs.172) A corrresponding relation holds for h+ and z+ replaced by /i_ and z. 18.6 Eigenvalues and level splitting We now insert the replacement (18. (18.424 CHAPTER 18..1. with the help of Eq. Expanding this around an odd integer qo we have ( 8E\ h? —J Inserting here the result (18. 18.3.143) with qo — 2n + l. we obtain for E(q0. we see that the relation is really an identity (the preexponential 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). Inserting the expressions for h+ and z+ in terms of h.
7 =( e N n+1 . 1 u2 c 2 (3 9 2 + l) ^ « b .^ 2^ h* . 2h4' i. can be given by In the work of Bhattacharya [23] the WKB level splitting is effectively (i..2 = 2 . The result (18. (18.fr2) is given by Eq.V2^\ \n+ i . for finite h2.. 1 7 425 2 . i.n = 0. which evidently supplies some correction terms. in the WKB restricted sense) defined by ^( n +2^KB(0) 2 Here the derivative dE/dn corresponds to the usual oscillator frequency. the level splitting. Combining Eqs.e. Bhattacharya [23] the "usual WKB approximation" of this expression is — with replacements h4/4 <» k. 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 / » : = \ .176) is described by Bhattacharya [23] as a "modified well and harried result AjfWB. c 2 /2 <> A — given as E0 = 4\ h(2fc) 1 / 2 o y J H q2 4/c > ^— + q ^ 2s c2 V2 . (18.178) "In S.3 The Double Well Potential where £0(<7o. T \ *n ! 1 ' (18. K.2.e. .1 7 6 ) (2n+9)/4^/ft V7T«! V C / \ eiI72^.143) with (18.^ e 21/^6c2 fe6 ( 1 mass mn = (1 8 .157)..q2 7. i n U n / ! Thus AE(q0.e. i. the following expression for large q ~ 2n + 1.18. the difference between the eigenenergies of even and odd states with (here) qo = 2n+l. the pure WKB result of Banerjee and Bhatnagar [14] (i.e.174) and (18.h2)E+(qQ..h2) / h 6 \ —K />6\n+l/2 q o / 2 2«>+2h2 — —^= m 1.1.106).e.h2) = E_(q0.
** Had we taken the mass mo of the particle in the symmetric double well potential equal to 1 (instead of 1/2). Furry [102]. (24) there is k 2 (i0n+i3)/4 //c3/2\n+1/2 V2fc3/2 6XP n\ ft \/nn\ in agreement with AE(q0. as we do here.2X2 V(z) = \z2tL) . K. (18.^ 2 + ^h2 If in addition the potential is written in the form A / ." Of course. Eq. 2/z4 and 2c2 respectively (see Eq.4/4 here. Anharmonic Oscillator Potentials which is unity for n — oo with Stirling's formula. (66). The factor set equal to 1 represents a somewhat "mutilated" Stirling approximation. that if this symmetry is taken into account from the very beginning. h = 1. h4 and c2 replaced by 2E. (18.2 / 7. A>0.6) to yield an improvement of WKB results for small values of n. 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. This.176) together with Eq. (18. we would have obtained the result with E. we identity the parameter k there with ft.3)). the Furry factor corrected WKB result follows automatically. The Furry factor represents effectively a correction factor to the normalization constants of WKB wave functions (which are normally for mo = 1/2. Bhattacharya [23].178). H. Then E^h2) and AE becomes 2^+U 2 (^)W 2 h6 = Eo{q0. Then the splitting A ^ W B of Eq.426 CHAPTER 18.180) h8 E0(qo. 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. ** Comparing the present work with that of S.h2 <> — h2. and A there with c 2 / 2 here. h2)\mo=i = . however. . these deriva> tions do not exploit the symmetry of the original equation under the interchanges q <> —q.h2) * ^ 2 q o / ^ q l _ m ^  (18179) l~o o».. (18.181) "W. We see therefore.h )/2 2 2 ( — ' 3 6\/2c A h6 n+l/2 h6\ 2{2n+5)/if \ exp 2 \ c / of Eq. as is also explained by Bhattacharya [23]. supplies the bridge to the perturbation theory results and removes the unnatural appearance of factors e and nn.6 \9o/2 e~h&l^\ with (m 0 = l) (18.
first paper.3. ^To help the comparison note that in J. ZinnJustin and U. Banerjee and S. §J. Friedberg and T. however. Lee [98]. We considered above only the symmetric twominima potential. D. Gildener and A.180) agrees similarly also with the result for arbitrary levels obtained with the use of periodic instantons* and with the results of multiinstanton methods. Jentschura [293].c2 = A/2. Patrasciociu [110]. M. J. Equation (18. Eqs. m 2 <• p 2 / 2 . 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. carried out along the lines of the corresponding calculations for the cosine potential and thus of the wellestablished Mathieu equation.h>) * 2V2^^^y/2e^3/3A. (18. 'This will become evident when the perturbation theory result of Chapter 17 is compared with the path integral result in Chapter 26. W. See R. P. D. Liang and H.11). so that by comparison with Eq. H.7 General Remarks In the above we have attempted a fairly complete treatment of the largeh2 case of the quartic anharmonic oscillator.34) and (E. Mansour and H.15). 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. Jentschura [293].181) therefore implies the correspondence rj1 <> A/2.3 The Double Well Potential 427 a form frequently used in field theoretic applications.88) h4 = 2/x2.^ Thus for the case of the ground state Eq. In principle one could also consider the case of small values of h2 and obtain convergent instead of asymptotic expansions.182) implies AlE(l.W. MiillerKirsten [186].* A similar correspondence is also found in the case of the cosine potential. . J. Bhatnagar [14].§ 18. (18. provided their result is multiplied by a factor of 2 resulting from a corresponding inclusion of antipseudoparticle (antiinstanton) contributions. (2. Miiller—Kirsten [167] the potential is written as V2 VW = \ for mo = 1. these are presumably not of much interest in physics (for reasons explained in Chapter 20). The comparison with Eq. D. J. and g2 is given as g2 = r]2/m3. formula (4. K. ZinnJustin and U.18.^ *See E. (18. M.—Q.
has also been the subject of numerous numerical studies. The ground state splitting of the symmetric double well potential has been considered in a countless number of investigations.[131]. 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]. Perturbation theoretical aspects are "See e. as some relevant references with their view we cite papers of Harrell and Simon [129]. Mahapatra. 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. can be found in an article by Coleman [54]. 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.428 CHAPTER 18. The double well potential. There is no reason to view the anharmonic oscillator differently. since — as we shall see in Chapter 20 — these nonperturbative contributions are directly related to the largeorder behaviour of the perturbation expansion and determine this as asymptotic. Santi and N. B. As references in this direction. In fact.g. . The wide publicity given to the work of Bender and Wu made pure mathematicians aware of the subject. as he discusses. A reasonable. in both symmetric and asymmetric form. B. N. like those for anharmonic oscillators. [132] and Loeffel. we cite papers of the Uppsala group of Froman and Froman [100]. though not exclusively numerical. Very illuminating discussions of double wells and periodic potentials. Martin. P. Ashbaugh and Harrell [12] and Harrell [130]. Simon and Wightman [178]. for instance. Sophisticated mathematics — like that of the extensive investigations of Turbiner [273] over several decades — seems to approach the problem from a different angle. Tables of properties of Special Functions are filled with such expansions derived from differential equations for all the wellknown and less wellknown Special Functions. Anharmonic Oscillator Potentials Every now and then literature appears which purports to overcome the allegedly illnatured "divergent perturbation series" of the anharmonic oscillator problem. Pradhan [182]. There is no way to obtain the exponentially small (real or imaginary) nonperturbative contributions derived above with some convergent expansion. [19] — that the expansions considered above are asymptotic. 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]. mostly in connection with instantons. in his recent paper on simple uniform approximation of the logarithmic derivative of the ground state wave function of anharmonic oscillator potentials.
185) (1 + u)^2 = (1 + GVW/2 = 1 + GJ±.183) S o l u t i o n : We have k2=1^. We obtain the expansions from Ref. 2 a.3 The Double Well Potential 429 employed in papers of Banerjee and Bhatnagar [14]. Saxena. Bender./4 E{k) = 1 + • l n In (18. fc ~ 1 . Srivastava and Varma [24] and Bhatacharya and Rau [23].2G^2q + 4G2q . q ~ 2n + 1.192) 'F 12 16 .u)1'2 = V2u{ 1 . Biswas. (18.186) (18.188) 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 . Example 18. X +2 + ' ^ (18.G2q + .5: Evaluation of WKB exponential with elliptic integrals Show that (l + u)^2[E(k)uK(k)}'.184) Hence we obtain for G close to zero: 1Gy^g l + Gy/2q~ and 1 . + (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. T h e r e f o r e it is convenient t o deal w i t h this here. Thus k' and hence k' = v ^ ( l . [122] as 13 1. 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. We have 4 ~k' 2u(l .\G\ + • and 1k2 =2Gv/2q4. (18.190) Hence ln]=ln : l n ' 7 2 ^ + 2 (18. which is assumed to be small.187) We now reexpress various quantities in terms of u.189) =2u 2u2 (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' .. Gutschick. Cooper and Strottman [221]. Datta. 1+ u : Gy/2q.f + • • •) 2u\ .18. k! (18.
3 « .u ( l .6w2 + 13u} + 12u 3 (« .3 y .195) With (18. (18.u){4 + 3u(l .u){24 .195) we obtain E(k) 1 + ln uK(k) .3) .{l u ( l u) / W 2 + 2} .3u(6u .u ( l .6 in the denominator and in the last step pick out the lowest order terms in u (i.« 2 + 3u] In [ .u){24 .194) ln '^J + 2 In i/2u )+H l + 2u(l . ) J 4 (18. 8 + 6 u 2 .13)} + ~u3(u .3).1)(1 .. . Anharmonic Oscillator Potentials *« = M £ H Consider first E(k): E{k) = 1 + .3u(6u . up to and including u2): u(u .1)(1 .u)} + — u ( u . 8 + 9u 2 .u ) (^){« + 5^ 1 )} + 5.^ L ) H[(i _ U ){4 + 3u(l . 4 \ u In —= I + V W 2 1 + ln x(l u) 31/ +.u)} .18u 2 + 39M} + 12u 3 (u .13)}. 1 3 « 2 = .4.u) x2 2/1 u2(lu)2 H K 42l We can rewrite this as — Analogously we have uK(k) 6J ' — {4 + 3u(l .u 13 \ 1 ln(± k> fc'2 +  (18.u) 2 {8 .«) + 2} + V [ .« ) ^ { 2 4 .1)(1 .u) + 2u} + i u 2 [ 2 + (1 .u) + 2}] 1 H 1 o u(u . (18.194) and (18.2{«(1 .u) + uf\ 2 + ju2(l 16 + (u W 2 12 4u2(lu)2 +  l ) u ( l .u) + 2} + i u ( u .3) .3) 1 + In ^ . 8 .{ M 2 ( 1 .3 u .197) .13)} + ±u3(u .3 u ( l .— ) .193) 2u(l .e.13)} + 12u 3 (u .2)1 4 uA"(fc) In ( ^ = ) ^{«(1 .196) Now consider the last line here without the factor 2.u) 2 {24 .u)} u{u(l .3u(6u .u)(u .a = (l«)(jl 2u/ 2 ) .u 3 ( u . (18.3) = = ~ .3u(6« .u){24 .3 .1)(1 .430 and CHAPTER 18.
3) .2 l_ln(1 JL\ V {2.156a) we obtain a and 7 and hence the ratio (always in the dominant order) T2~ ) w i t h h += ^ h • Since (cf.196) and (18.e.4u . Inserting here from the above expansions the contributions up to and including those of order u2..197) we now obtain 4 1 • E(fc) .18.195) with (18. we obtain: h ^ —^ 1 + 3G 2 V 2 8 2 /„ u 1 . ' 8G 2 q = 4 2 q ln f 26/2 \ o .e.2{w(l u) + 2}] = u(3u2 . Eqs. From (18. 2 . / to + 3GH~ \ 8 .6: Normalization of WKB solutions Show that the normalized form of the WKB solution (18.^HfHGH Example 18.J? dz^\E V(z)\ \ 87T2 / lEVtz)]1/4 Solution: From Eqs.2 (18.«){4 + 3u(l .zo) HWKB normalized _ ( h* y^eM.2 /o 2 431 (18. (18. VB(z) = y\hTB(z) with VB(Z) ~ Bq[w±(z?  ^ [±(q  J l)}\2^V . (18.196).. . l2 = ^ ( l + uf/2[E(k)uK(k)}.4u .128) and (18. we have.199) We now return t o the expression on the left of Eq. (18.173) t o be evaluated.u H I . [(1 . i.] in (18.3).198) 9 «.200) Remembering that u = Gy/2q. + iu^Zu2)^~u+ — +0(u3) l .128) and (18.148) (for particle mass 1/2 and in dominant order) is given by H. i. this becomes _2 2 u 2 / j M 2 / 3^2s 2 ~ 3G 2 2G2 H \ / 2 H / . (18 201) = Thus 3ffl« (2i/aGi/a2i/V/V"4 3ffl2 (Gii7»J4  '• . V^LJ 4 2 16 ^ .155)) <*VA{Z) = VB{Z) a n d VWKB(Z) = 1VA{Z).uK(fe) ~ 1 + In I — = Mu2(3u. (18.. where w±(z) — h±(z — z±).it)} .3 The Double Well Potential Now consider the bracket [.
143).g. normalized  V4exp[ ~ f*° dz\/\E~ \EV(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 [ .e. we obtain (for q reasonably large) 1 (91) and hence Joo Hence with h+/V4n = (h4/Sn2)1^ W2TT SI {[UqIWH"1)}2 2TT h2+\1/2 ~VWKB \ ~^T f 4^ \V4ir ^g^ J VWKB . (18. oo . We start from Eq. W. and obtained as Eq. zo) meaning t o the right of ZQ and note the asymmetric factor of 2) 11/4 y±(z) ^ W K B (z) ± ^ = % K B ( Z ) : \qh\ 1/2 \h\U{z) ±— cos 7 **See e.7: Recalculation of tunneling deviation using W K B solutions Determine the tunneling deviation q — go of q from an odd integer go.158). pp.35)). (18.3)]!23<«D [5(91)]' >A. Anharmonic Oscillator Potentials With the standard normalization of parabolic cylinder functions. Then ((r.hA/2 = h^/4 this implies . Magnus and F. 93. Solution: The turning points at zo and z\ on either side of the minimum at 24.f hi\ VWKB.** oo / dwD\. by using the periodic WKB solutions below the turning points.432 CHAPTER 18. i. 1 7(l.Aw) = s/2i 2("!) V (41) and [(9 . (18. y±(z) = IVA(Z) ± yA(z)\ = — i M O O ± 1 Ji.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+.l)]![i(0 . (18. ° \Edzy/\EV(z) V(z)V4 VWKB. i.147).204) . Eq. normalized = We see that the normalization constant is independent of a quantum number. Example 18./ . are given by Eqs. The constant obtained here will arise in Chapter 26 in the evaluation of the normalization constant of the WKB wave function (cf.*o)/ ^WWKBOO (18. (26. 82. Oberhettinger [181].zo).e.146) and (18.
212) tt See e.205) We now apply the boundary conditions (18. from z\.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..e... from ZQ. of course.209) "27 In the present considerations we approach the minimum of the potential at z+ by coming from the left.g..J^Wj +(18.e. approach the minimum also from the right.132) and (18.e.18. i.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.V(z) = r J zo I ZQ dz[\qh\ \ Z h%U(z) (27V + 1 ) . Sec. JV = 1.g.3. \h\U{z) 1/2 .208) and cot L />G + 1/2 ^(ig4^+^))"" + J 1/2 2qh%h%U{z)\ 27 +  :± (18. Messiah [195].207) (18.. i. .. 7T\ + ^ 4 7r\ dz .r + ^=co S [r.Jz0 sin ^o J LJzo 4 Jzo 2 y •••4y ••• = . A. (18.2. z{) we expect tt (18. ^(7(z) r t J z0 •I Zi I V2 dz \qh\ 1.211) 7 z l .i ) N c o s / •••+! provided the BohrSommerfeldWilson fZ1 dz^/E J zn quantization 4 condition holds.6. 6. Then at any point z € (zo.iK h\U{z) + / 4 (18.^r + + . i.210) Choosing the point z to be z+. 1/2 . We could. this implies sin cos sin cos Thus e.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 .2.( .
Banerjee and S. 7r"y in agreement with Eq. 5. (18.149).434 CHAPTER 18. 1 1 . Landau and E. D.217) exp ( V [ ° dz^/E JZQ + V{z) This expression represents the WKB result for the level splitting and may therefore be called the WKB level splitting formula. The prefactor 2\/2h2/n is. 2IO/K.146)). . in fact (with our use of qo).216) Inserting this into Eq. (18. The latter give 7 7 1/2 exp dz dz \qh\ \qh\ + + \h%U{z) 1/2 exp I — / \h%U(z) (18. P. .3.215) we can insert for 7 / 7 the expression given by Eqs. Anharmonic Oscillator Potentials Similarly under the same condition (l)Jvsin J Zn where it is understood that qh2/2 ~ Ey(z±). We now see that Eqs. One may note that the exponential factor here is not squared as in decay probabilities (squares of transmission coefficients). In Eq. (18. (18. 2Sm (2 g /ft2)i/2 in agreement with the right hand side (the last step following from Eq. We note here incidentally that this agreement demonstrates the significance of the factor of 2 in (18. .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.e. K. Bhatnagar [14].210) assume a form as in our perturbation theory. and cot< (q + 1) '!}» for q = 50 = 1.h2) = = h2 h2 1 y 1/2 2{EEo)=2^{q~q0)=2^±l V2 V2h2 .215).205)._ £/>(£<•>•) tan{(g + l Since tan ) ^ ^ .213) The integral on the left can be approximated by 1.209) and (18. .5. The formula thus agrees with that in the literature.143).27 9 . the level splitting is given by AE(q0.148) and (18. Bhattacharya [23]. and L.exp ( \/2h2T V2 7r 7 dz l / qh\ + \h\U{z) (18. Lifshitz [156].2 .90 — =F— for q0 = 1. (18. they become c o t { ( . S. where u> is the oscillator frequency of the wells. ' /^i . 9.215) we can expand the left hand sides about these points and thus obtain . M. K..Zi / r.215) which results from the factor of 2 in front of the sine in the WKB formula (18.. W "Cf.214) {«+"i} 0 for q = go = 3. (18. i. (18. 7 . + l ) j } ^ 1_ 27' (18.
also in P. since no probabilistic interpretation is available for the field theory matrix elements in virtue of creation and annihilation processes during the interaction. W. in an early investigation Case^ showed that potentials of the form r~~n. are not as troublesome as one might expect. Some decades ago — before the discovery of W and Z mesons which mediate weak interactions — and before the advent of quantum chromodynamics. In particular Case pointed out that for a repulsive singular potential the study of scattering is mathematically welldefined and useful. H. 'For a review from this perspective with numerous references see H. The centrifugal potential oc r~2 is generally considered as exceptional and is treated in detail in wellknown texts on quantum mechanics. Aly. M. 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. n > 2. 435 . Feshbach [198]. Giittinger and H. MiillerKirsten [9]. J.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.Chapter 19 Singular Potentials 19. 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. 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.* However. Case [45]. M. W. f K .* The physical analogy between singular field theoretic interactions and singular potential scattering of course breaks down at short distances. Morse and H. However.
W. D referring to Dirichlet boundary conditions) in 10dimensional string theory. § 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. In the concluding section we refer to additional related literature and applications. MiillerKirsten. We shall see that not only can the radial Schrodinger equation be related to the modified or associated Mathieu equation (i. Gubser and A.2) § C . Singular Potentials theories in mind.2. S. our presentation below is deliberately made elementary and detailed so that the reader does not shun away from it. W. J. . J. K. In fact. 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. roughly speaking a brane is the higher dimensional equivalent of a membrane visualized in two dimensions from which the DS brane derives its name. This chapter therefore gives the first complete solution of a Schrodinger equation with a highly singular potential. the attractive singular potential 1/r4 was found to arise in the study of fluctuations about a "brane" (the D3 brane. It is therefore natural that we study this case here in detail. N. that with the hyperbolic cosine replacing the trigonometric cosine). MiillerKirsten and Jianzu Zhang [226]. Singular potentials arise in various other contexts. Maldacena [41]. S. Recently.2 19. but also that the Smatrix can be calculated explicitly in both the weak and strong coupling domains. (19.e. some also permitting further research. Hashimoto [124]. for instance. H. Liang and Yunbo Zhang [189]. Callan and J. D. This property singles this case out from many others and attributes it the role of one of very few explicitly solvable cases. Tamaryan. In view of the unavoidable use of Mathieu functions expanded in series of Special Functions.436 CHAPTER 19. this particular singular potential plays an exceptional role in view of its relation to the Mathieu equation.Q. S. G. Manvelyan.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  (191) The radial Schrodinger equation with this potential may then be written 2 z y + k W + l) 92 V = 0. J. Park. R. 19.
using the method of Dingle and Miiller of Sec. The following substitutions are advantageous: y = r1/2cf>. Liang and Yunbo Zhang [189].2 The Potential 1/r4 — Case of Small h2 437 where E = k2. It may be noticed that since I is an integer. R. (19. Eq. In the following we study the solutions of Eq. (19.19.7 that we employed also in previous chapters. This is rarely possible and therefore this case deserves particular attention. J. and so =4 5 and n ez = r = ~ = e^4fir. It is clear that we draw analogies to our earlier considerations of the periodic potential.3 below) g2 is negative and hence h? is real for E > 0. .4) dz2 + In the literature this equation is known as the modified Mathieu equation or associated Mathieu equation. 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. The radial Schrodinger equation then assumes the form 2h2 cosh 2z . (19.3a) In the case of the attractive potential (as in the string theory context referred to above. This has advantages compared with higher dimensional cases.Q. W. h2 = ikg. h = ei^y/k^. 8.mo = 1/2 and h = c = 1. r = 7 e z . J.^ where \f\ can be an integer and thus leads to singularities in the expansion of v (see e. \[\ in the correspondence is nonintegral.4) first for small values of h2 and then for large values of h2.28) below). (19. and as we have in mind in Sec. MiillerKirsten.3b) 7 = % j. We begin with the derivation of various types of weakcoupling or smallh2 solutions which we construct again perturbatively. Our ultimate aim is the derivation of the /Smatrix for scattering off the singular 1/r4 potential. [l+2) ~ * A> ~ the equation converts into the periodic Mathieu equation of Chapter 17. Subsequently we derive the same 5matrix from the consideration of largeh expansions and calculate the absorptivity in the case of the attractive potential. We observe that with the replacements p2 z^iz. H.( I + 0. 19. Manvelyan.g. ig h \j g (19.
11) We follow here largely H.Nv(w).7. The zerothorder approximation = Zv[w) (19.438 19.1. (19. The solutions Zv(w) are written Jv(w). .Hv (w).i l . (19.7) We shall see that this parameter is given by the same expansion as in the case of the periodic equation.2.. where £>„ : = —_ + i _ + J l . (19.4) becomes dw2 w2 w2 \ dw2 w dw \ j Next we define a parameter v by the relation V2 v2 = (/ + . — Zv+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. H. H. MullerKirsten and N. Aly.g) This equation is wellknown as Bessel's equation or as the equation of cylindrical functions. VahediFaridi [10]. 8. and H„ (w).I . (19. (19. First we make the additional substitution w = 2/icosh z. where these are the Bessel function of the first kind. w 2 . Proceeding along the lines of the perturbation method of Sec. dwA w dw { wA ) (lg.5) so that Eq.9) leaves uncompensated on the right hand side of Eq.e. W.2 A / i 2 .^% dw1 1 1 —— = dw = ][Zv22Zv 4 w Zv + Zv\ = (Zv\ 2 + Zv+2]. —Zu — (Z„_i + Zu+i).6) terms amounting to 1 „ 2 d2Zu A V (19. (19. J.2 CHAPTER 19.6) becomes to zeroth order </>(°) = <j>v of <f> in h2 Dvcf)v = 0. Neumann or Bessel function of the second kind and Hankel functions of the first and second kinds respectively. Eq. i.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.
10) is now particularly simple: i?i°) = h2[(v.u + 4fZu+A}. v)Gv + (y.v + 2)*{u + 2.2 ) * ( z / .14) (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. ~a(2v + a)~ w2 * a> Zv+a = a(2u + a)Gv+a. where the starred coefficients are defined by (19. The next contribution to (^°) + t^1) therefore becomes 0(2) = /i4[(z. it is more convenient to use these relations in order to rewrite Eq.21) +(i/.15) (19.2.u (19._2 + {v.20) 2)*ZV_2 + {y. v . We assume in the following that 2u + a ^ 0 (the case of 0 has to be treated separately). (19.2)*Zi.18) Now </>(°) = Zv left uncompensated Rv . (19. wz The expression (19.6). (i/.i/4)*Z 1 / _ 4 +(u. .2 ) * i ^ 2 + (!/. v .13).17) Thus a term fiGu+a on the right hand side of Eq.10) as a linear combination of various Zv.13) (19. (19.12) (19. However. v + 2)Gv+2].2)G„_ 2 + (y.19. v + 2)*(v + 2. I / ± 2 ) = 1. . where We observe that DUZV = Q. and this means in (19.i/) = 2A. v + 2)*Z„ +2 (19.10) in terms of functions G„ defined by Gv+a = —nZ v+a .. = §.u + 2)*Z„ +2 ].2 The Potential 1/r4 — Case of Small h? 439 we can rewrite the expression (19. therefore 4>^ leaves uncompensated i#> = / i 2 [ ( i / ^ .13) therefore lead to the first order contribution jfU = h2[{v. The terms (19.16) ( I / . z . when a or 2v + a = 0.!/+ 2)*4°J 2 ]. of course. v2)*{y2.
j=i.2 + 2 9 ) ^ 64(^ 2 . (19.2. i/ ± 2)* (i/ ± 2.23) These coefficients may also b e obtained from t h e recurrence relation P2i(2j) = p2i2(2j .1169)/i 12 . i/ + 2j).j^0 (19. etc. Evaluating t h e first few terms. . v + 2)*{v + 2. (I/.I/±4)*. .440 CHAPTER 19.i/±2)*. P2*o(0) = 0. Reversing the expansion and setting a = (I + 1/2) 2 (not to be confused with a e. p 0 (0) = 1. and p 0 ( 2 j ^ 0) = 0. i/)] + • • • . + 2(y2\) — + 32(i/ 2 l) 3 (^ 2 4) —^ (9. ^ 64(a_1)5(a_4)4(a_9) + 0 ^ )• (1928) 1 6 .17)). we obtain 0 = h2{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.l)(z/ 2 . v) +(i/. v — 2)*(v — 2. v + 2j) (19.g. p4(±2) = (i/. Ri. i/ ± 2)*.v) + hA[{v.4 + 58.455a + 1291a . Adding these terms and setting the coefficient of Gv equal t o zero.v + 2)*(i/ + 2. .v2)*{v2.26) iy — 2. Singular Potentials Proceeding in this manner we obtain the solution ^ = 0(0) + 0 (1) + ^(2) + .v)] (19.22) where p2(±2) p4(±4) = = (i/. Zv + Y^h2i i=l J2 p2i{2j)Zv+2j. we obtain t h e expansion . (19. . v ( 4 5 a . in Eq. (19. we obtain 9 h4 = a 2(al) 3 2 (13a . ..25)/i 8 32(al)3(a4) .v) + {v. 1 1 \ 2 + 2) ~ ) s X 2 A = v 2 hA (5v2 + 7)h8 v .I/±2)*(I/±2.v + 2y{v + 2.25) Finally we have to consider the terms in Gv which were left unaccounted for in Rv .2)(v + 2j .24) P2i2(2j + 2){v + 2j + 2. subject to the boundary conditions P2i(2j) = 0 for  j  > i. .4)(»/2 . i/ + 2j) + p 2 i _ 2 ( 2 j ) ( ^ + 2j. .
and is therefore real for both cases g2 > 0 and g2 < 0. 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 . so that J2 (19. v . Thus to O(0) in h we have (jr ' = 4>v — cosh vz.7) into Eq.An(u + n). v)(f>v + {v.2.cosh 2z)4>.4).^ + e±(v~2)z.19. .3 Small h2 solutions in t e r m s of hyperbolic functions For later purposes we require yet another type of solutions. ±2. 2 . (19. apart from a normalization factor which we have chosen (so far) such that the coefficient of Zu in <f) is 1.34) .29) sinh vz or e±uz.i .30) £>„&. (19. The solutions 4> of the modified Mathieu equation are now completely determined.v2(j> = 2/i 2 (A . = 0. — We return to this question later but mention here that this problem has already been dealt with in Chapter 17.2)<t>„2 + (v. (19.h2[(j)u+2 + 4>v_2] h2 [{u. £>„ = — .31) It follows that Dv+2n4>v+2n = 0.v + 2)<t>v+2]. we may say that the first approximation <^°) leaves uncompensated terms amounting to R^ = 2ft2 (A . ±{ v+ z e = sinh(v + 2)z + sinh(z/ — 2)z. we can rewrite the latter as j ~ .cosh 2z)4>v = = 2h2A(/>„ .33) = = cosh(^ + 2)z + cosh(^ — 2)z.32) (19. 19. Thus v is real for small values of  h21. (19.2 The Potential 1/r4 — Case of Small h? 441 We note that this is an expansion in ascending powers of h4. We are still left with the question of what will happen if 2v + a = 0 or v — ± 1 . (19. Substituting (19.
J7t0 p2i(2j)<f>u+2j. The solutions in terms of cylindrical functions are written *J. 8. 19. F. .h).35) The form of i ? ^ is seen to be almost identical with that of the corresponding expression for solutions in terms of cylindrical functions. we prefer to adhere to one equation with different solutions.h2i *=1 J2 j=i. 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. CHAPTER 19.I/±2)*. In fact we could have obtained the same Ru by starting with the modified Mathieu equation for h2 replaced by — h2.7 — we refer to Meixner and Schafke [193]. In order to avoid confusion arising from the use of different equations. Singular Potentials (I/.* The solutions of the modified Mathieu equation which we are considering here are written with a first capital letter. Arscott [11]. The use of the symbols (v.1 .2. Schafke [193]. Meixner and F.J/±2) = . (19. > These solutions correspond in the periodic case respectively to the solutions and me^. v ± 2) etc.h).h). 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. etc. 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. • (19. W.442 where (I/.37) where p M (±2) = (I/. (19.4 Notation and properties of solutions We now introduce standard notation as in established literature.I/) = 2 A . Denning (19 36) (">" + <*) =o^7Ta)> we now have the solution  p(z.h) = <(>„ +Y. > (f> with </v = sinh vz — Seu(z.38) (f> with (\>v = exp(vz) — Meu(z. M.
e. p. Also from the expansion of Jv{z) in rising powers of z.g. i. whereas the — as yet — unnormalized solution (19. 4>v = Hi.h) and oo i M^\z. 17. . (19. h). we obtain$ Ju(2hcosh(z + inir)) — Jl/(2hcosh z exp(m7r)) — exp(inuTr)Jv. For the relation of Hankel functions used below.19.44) implies. as Mev{z.(2hcosh z). W. h).43) and hence similarly M^\z + inir.h)±Sey{z. introduces the dominant order function into higher order contributions''' — oo i 2i Meu(z.h). — oo cv2r{h2)e^+2r>. h) = av{h)M^ (z. h) = exp{vniri)Me„(z.h) = exp(vz) + Y. (19.h). which. 16.h). (19.MP(z.e.40) Cev{z.44) (19.42) The solutions Meu(z. i.h = £ p2i(2j)(l)j exp[(^ + 2j)z] (19. Schafke [193]. <f)v = N„(2hcosh z) MJ?\z. Mi.h).41) i=l We see immediately that Meu(z + niri. h) = exp(inwn)MP(z. see this reference p. 'See e. this normalized solution possesses contributions exp(i/z) in higher order terms. W. (19.h2) oo = Y. Oberhettinger [181]. h). {2h cosh z) • MJ?\z. ' (z.39) Writing two solutions out explicitly we have for example — apart from an overall normalization constant.42).40) does not.h) = Jv{2hcosh z) + ^h 2i J^ j=iJjLQ p2i(2j)Ju+2j(2hcosh z).45) T Below we frequently write the normalized solution as in J. Meixner and F. 4>v = Hv (2h cosh 2) * M^\z. h). (19. 443 (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) . As emphasized in Chapter 8. (19. as we remarked earlier. h) are therefore proportional to each other as a comparison of Eqs. Me„(z. Magnus and F.
MJ>>(0. h) = M™(z. h) . the explicit evaluation of the S'matrix element) we elaborate here a little on the computation of coefficients of normalized Mathieu functions. s (19. 178) to be convergent for  cosh z\ > 1 but uniformly convergent only when  cosh z\ > 1 for otherwise complex values of z. i.49) Inserting this expansion into the modified Mathieu equation (19. These points are important in our derivation of the S'matrix below.e. p.8.exp(TM"r)MW(. we use the notation of Meixner and Schafke [193] and so write the expansion of the function Mev(z.47) With this equation we obtain for nonintegral values of i/:§ ±isin WK MJ)3A\z. . Equating the coefficients to zero we obtain [{y + 2r)2 + \]cv2r = h2(cv2r+2 + c£ r _ 2 ).g. Meixner and Schafke [193].4) we obtain oo £ r=—oo [(i/ + 2r) 2 4. i.444 CHAPTER 19.e. the first relation is obtained as follows: 2M<£1 = M^l+M^l. The functions Me±v{z. Singular Potentials where clearly MeJQ. r=—oo (19. h). Meixner and Schafke [193]. we have the following for a change from v to — v. H™(z) = e^HfXz). 2M™ = eiuvMi3) + e~ivlrM^\ .50) E.h) on the other hand can be shown (cf.h) can be shown (cf.43).48) The series expansions of the associated Mathieu functions M„ (z. ^ .\c\r + h2(c^2 + 4 r + 2 ) ] e ^ + 2 ^ = 0.e. 130) to converge uniformly for all finite complex values of z.h2): oo Mev(z. Mi 3 . For later essential requirements (i.h) Using further properties like (19. As mentioned earlier. (19.h) . oiv{h) = (19.h2) = ] T cv2r(h2)e{y+2r)z. 4) = exp(±^7r)i\4 3 ' 4 \ M^3'4) = M™ ± iMJ?\ (19.46) H^z) = e^H^(z). p.
51) we have 2r (19. + o2„r 2 l2 ]  C 2r 1 2r c 2r2 or 1 c J =^[A(.2 The Potential 1/r4 — Case of Small h? which we can rewrite as fc2^±2 c 445 = [A _ („ + 2r) 2 ] . eg = c~v = 1.^ c 2r 2r 2T~2 or 2r ^2 /l2[A_(I/ + 2r)2]£2l± 2 . Meixner and F. see J. + 2 r ) 2 ] .54) 2 C2r 4 ^ h.19. .2 [ A .[X(u 2 + 2r2) } " c 2r2 This continued fraction equation can again be used to obtain the explicit expressions of coefficients of normalized modified Mathieu functions. 117. . 2r (19>51) 2r For ease of reading we give the steps in rewriting this. h~2[\ .53) /i2[A(^ + 2 r ) ] . W.* For 'Actually.^ C 2r+2 (19./ i 2 C ^ . 122.^ p 2 2 ^ or c 2r+2 2r2 1 (19.T22r[\ A . p. (19. Thus now c 2r . Eq. Alternatively taking the inverse of Eq.(/^. (39). \ . p. \) /.(v + 2r)2} / z .52) in agreement with Meixner and Schafke [193].^ + 2r + 2 ) 2 ] . Schafke [193].
and also R. Manvelyan.( " + 2)2] ~ h2 h. 128(i/+ l) 2 (i/ + 2 ) ( ! / . MiillerKirsten and N. W. (19. Liang and Yunbo Zhang [189].Q . Then di 2 1 t" + 2(^1) + • • • .2. + 0(/.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.12) ^ 384(^ + l)(z/ + 2)(z/ + 3)  " ^rW^hy. H. .55). VahediFaridi [10] (this paper contains several misprints which we correct here).446 CHAPTER 19. W. (19. Spector [257]. t For this purpose *We do this in the manner of the original derivation of R.5 D e r i v a t i o n of t h e S . J. M. Here we insert for A the expansion (19..C + l) 3 (i/ .52) and obtain c% 1 eg fc"[Afr + 2)»)h_a[A_(.l ) 19.+0^"> < 19 .55 ' Example 19.1: Evaluation of a coefficient Evaluate explicitly the first few terms of the coefficient Cj/cfJ given in Eq. . H. i/2 + 41/ + 7 7i6 + 0 ( / i 1 0 ) .^ ^ + 2)(i/ + o> l)(i/ 3) 0(^10). MiillerKirsten. Singular Potentials nonintegral values of v examples obtained in this way are (see Example 19. Aly._{u+m h2 M ( " + D + 55^iy] + i ^ 5 j /i 2 4(. but follow our earlier reference H.27) and truncate the continued fraction after the second step.1 for the evaluation of a typical case): hb + 0(hw). Solution: We put r = 1 in Eq.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. + i). J .. H.[v^. J.+4)!ll_. 8 4 ( i / + 1) 128(i/ + l)2(v + 2)(vl) + 5 10 32(i/ + ^ l)(i/ + 2) + 768(^ „„ .
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 . Magnus and F. (19. this wave function near the origin is the wave transmitted into this region (as distinct from the reflected and ingoing waves). Oberhettinger [181]. I / + 2 2 l\TT +E ^ i\ E j=i. The asymptotic behaviour of the Hankel functions H„ ' (w) for \w\ ^> \u\. and then the continuation of this to infinity. (19. In the case of the repulsive singular potential here.. We obtain the regular solution by choosing Zv(w) = H„ (w) for 5ft(z) < 0.h) 1/2 w) (19. 2 ) 2 1/2 ex Z/7T IT H=i —J TTW P l+O 10 (19.2 The Potential 1/r4 — Case of Small h2 447 we require first the regular solution yreg of the radial Schrodinger equation at the origin. p.57) ^ 1 . we have W r = ip + ut — 7r/4' .3a) and (19. and considering the propagation of this wavefront.56) where by Eqs.59) which tends to zero with r.j7L0 P2i(2j)exp i[v + 2j + 2 2 . 22. \w\ S> 1 and — IT < arg w < IT is known to be given by* (19. The timedependent wave function with this asymptotic behaviour is proportional to (here with ui — k) —iuit—g/r+iTr/4 Fixing the wavefront by setting <p = —lot + ig/r + TT/4 = const.19. W .5) w — 2h cosh z = [ kr \ r Thus r —> 0 implies \w\ —> oo. . Then y reg = = r rV2M®(z.
j^Q p2. '(z.3b) ln • 7T Q"*4' r fg ° Vk (19. h) through the entire range of !&{z). 17. Magnus and F.3a) and (19. Singular Potentials so that when t — oo : r — 0.#. AV . We require therefore > > the continuation of Mi. In a similar manner we can define solutions y± by setting for 9fJ(z) > 0: y± = rl'2Ml?>*>(z.61) In fact. but uniformly convergent only when  cosh z\ > 1 for otherwise complex values of z./l ~ 2 ~ 2z + " 4 > 0./i).h) can be shown (cf.A1"—2 z '+"2 A~~ n or rA . p.57) the condition  cosh z\ > 1 implies I cosh z\ kr + ig/r >1.h) oo j 2 — rV flp^ + ^fc* £ i=l j=i.j¥=o Ki(2j)exp(Tmj) (19. '2"4 .60) Using the above asymptotic expressions for the Hankel functions. Now from Eqs. 2h and the square of this expression implies 2 2 ~ ' A. we can derive y_ from y+ since one can show from the circuit relations of Hankel functions^ that M^\z + m. Meixner and Schafke [193]. 4%r g >0 4r^r ' rg See W.V + § 9 2 > 4hh* . 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. We now require the continuation of the regular solution y reg to solutions behaving like y± at r — oo. ^ 1 ) (e i 7 r z) = Hi2!>(2) e^H^iz). The series M^ (z. 178) to be convergent for  cosh z\ > 1.448 CHAPTER 19. Prom the relation r = (ig/h)ez we see that r — 0 > corresponds to $l(z) — —oo and r — oo to $l(z) —> oo. (19. .h) = exp(Mr^)il4 4 ) (z.62) Also with Eq. This means that the origin of coordinates acts • > as a sink. Oberhettinger [181]. (19. p.A14A.(2j)<gH • (19.
3a).e. which has to be bridged by using another set of solutions. This interchange in the > solution My interchanges y reg with y+.62). .1.38) or (19. 19. A suitable set is the pair of fundamental solutions Me±v defined by (19. These solutions converge uniformly for all finite complex values of z (cf. p. (19.19.r2_) > 0 with r£ = rg(2 ± >/3).rl)(r2 . i Im z z = In r/r0+i7t/4 .1 The domains of solutions.e. Originally we chose the sign of m / 4 as in Eqs. r 7r for 0 < r < ro. is satisfied only for r < r_ < r__ or r_ < r + < r.ijt/4 Rez . 449 This condition. (19. z = In h i— r0 4 and 7T z = In • ro { 4 for ro < r < oo. The one further point to observe is that the variable w = 2/icosh z = (kr + ig/r) is even under the interchange z — —z. the domain r_ < r < r+ — as illustrated in Fig. 19.2 The Potential 1/r4 — Case of Small h2 Hence (r2 .3b) and (19.ft/4 _Z z = In r/rm/4 0 r 0 =(9/k) 1/2 Fig. i. we have to assign different signs to the imaginary part of z in the two distant regions. (19.49). Thus we choose .63) Thus there is a gap between the two regions of validity — i. Since the real part of z changes sign at r = ro. (r — r+)(r + r+)(r — r_)(r + r_) > 0. 130). and in order to maintain this symmetry. Meixner and Schafke [193].
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_. 131).v{z)\ . (19. we express in the domain r_ < r < r+ the regular solution as a linear combination of solutions Me±v with coefficients a and /3. .l/2Me_i/(T. i.z=—in/4. At r = ro the real part of z vanishes and then switches from negative to positive. Hence we match the right hand sides of Eqs. at this junction we require also the imaginary part of z to change sign. + ^[r'^e^z)] dr z=—iw/4 + (31 ^[rll2Me.)](1964) The right hand side of the first equation now represents the regular solution in the domain r_ < r < r+.[ r l/2 M. Singular Potentials Then starting from the region r ~ 0. as just explained.v(z)\ —m/4) + ^[r^Me.450 CHAPTER 19. (19. (3 = a .B^0. i.e.66) a = (3 . (3) (r)] aA[rl/2Mei/(r)]+/3i:[r. and we determine these coefficients from this equality and the additional equality obtained from the derivatives.65) in/A a>d[rl/2Mei/{z)]+f3>d[rl/2Me_Az)] Since Mev(z. these relations can be reexpressed 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. a4[r1/2Mev(z)] dr a'^lr^MeAz)} It follows that (19.(2)]z=w/4 + 0Me. Thus. h) (as one can check or look up in Meixner and Schafke [193]. p.67) Next we have to continue the solution beyond the point r+ to a linear combination of solutions y+ and y_. Thus we set rl/2M^\r) = = rll2[aMev(r) + (5Mev{r)}. to r^2[AM^ + BM^\ A.^)] dr z=in/4 dr (19. h) = Me^u(—z.
(19. the nonderivative parts (from the first term on the right of Eq.g]:=f(z)*&g(z)WZ) dz dz Thus. (3.3a) and (19.71) We now determine the coefficients a.64) the relations M&Hz) jz[M^\z)] and from Eqs.19.65).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.70) by Me'_u (the prime meaning derivative). we use the notation W\f. A and B.) d_ _ 1 d dr r dz Hence for any of the functions Mv:  [ . for instance. dr BM^(r)]. multiplying the first of Eqs. (19. Then.68) From Eqs. = aMev(z) + 0Mev(z). (19. with the replacements of Eq. For the Wronskian W of two independent solutions f(z) and g(z) which is constant and can be evaluated at any point.68) \j3Me„(z) + aMev(z)] P±[Me„(z)}+afz[Me„(z)] = = [AM^\z) + BM^(z)}.2Mei/{r)]+d[rl/2M] dr dr = = rxl2[AM^\r) + Ad[ri/2M(z)(r)] dr +B^[rl'2M^{r)]. (19. (19.70) by MeU and subtracting . (19. (19. (19. ^ ] = ^ + ^ ^ .65) and (19. (19. Thus we obtain from Eqs. (19. (19. and the second of Eqs. (19.3b) we infer that (z = l n r + const. (19.67): rl/2[(5Meu{r) + aMe_u(r)} pd[ri.69) Thus if we replace in the derivative relations of Eqs.68) the derivatives with respect to r by this relation.69)) cancel out in view of the nonderivative relations. and we are left with relations expressible only in terms of z.70) = a±[Meu(z)}+P^{Me_„(z)} A^[M^{z)] +B^[M^(z)}.
4] = . Singular Potentials the second resulting equation from the first resulting equation. h)/M^\o. we obtain immediately W[Z. we use the following circuit relation which can be derived like Eq.3] = [1. we obtain the first of the following two equations.MiA\z)} = 2(i)(2/Trw)(dw/dz) = (4i/Tr)(2hsinhz/2hcoshz) ~ 4i/>. M^4)] + W[MJ>3). ..j] given in Meixner and Schafke [193] (p. W[MJ?\ M^] W[Mev.Me^} (19.M^] J™aV. (19. 170/171).M^]= IT e~iv*av. 169) M0) = e«^Ma) _ isin U1X M(A) _ (19. h).M^] W[Mev. (19. (19.4] = .(0. W[M^\Mlf)]W[Meu. TV W[Mev. B = J 2 L _ e . = IT <*„.46)) W[Mev.M^} W[Me„.76) . av{h) = Me.58) and considering large values of \z\. p.71) we obtain in a similar way A B = = W\MP\ Me. and Wronskians W[M^\MIJ)] = [i. using Eqs.48) (or cf.73) We now use Eq.u.75) With these relations and Eq.u}W[Me. Meixner and Schafke [193]. 7T [1. (19. With these expressions the quantities A and B are found to be A= f au ^ 2isin VK \ a_u 1 a.452 CHAPTER 19.^ ^ a „ Q _ .A] = W[Mi3\z). .. or obtainable by substituting e.e. 7T W[Mev. (19.\ \ ( av 2iwKav ^ v .47) we obtain (av defined by Eq. and the second is obtained similarly: W[M?\Mev\ W[Mev.Mevy W[M?\Mev] W[Mev..* « * r r z H . Me. (19. i.Mev]' P l ' From Eqs. the leading terms of the respective cylindrical functions.e. i.^ 47 9? MJ>4)] MeV] [3.46).Mev] = = = av.] W[Mev. (19.57) and (19.77) av I 2ism VK \ ctv av ^Thus.74) Moreover.g. (19. 7T (19.
46) together with the relation Mev(z. (19.f c . Z ) e ^ ] .h)./ ( . B=tfD'"". (19.19.g. eii/7r/2 (19.(l)le^}Pi(cos 9). Then f' .l) ' We note already here that with the substitution R = ei7r7. (19. Z ) e .78) we obtain Ay reg « _ . in Chapter 16). (19.85a) . The superposition of an incoming plane wave and an outgoing radial wave is written and reexpressed in terms of partial waves for r — oo » with z = rcos 6 as: 1 tCT" "1 OO e *« + /(*)!_ ~ 5 > ' c < f c P .h) = Me_u(z.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.79) where the second expression follows from Eq. f c r .83) The Smatrix element is therefore given by ifc*"* .82) The S'matrix is defined in the following way in the partial wave expansion of the scattering amplitude /(#).(ite™)" 1 { 8 J ' f(k._ [ / ( f c .80) 2^fc (1981) where \7r/ i t sin VK /(*. (19. 2zitsm z/7T zzitsin I/TT Inserting these expressions into Eq.h) + BM^\z. (19.h)] ( m / 9 \ 1/2 {L\ {Aeikrei(V+l/2)*/2+Beikrei{v+l/2)*/2^ g) We set fl = ^ = MM). as we recall from earlier considerations (e./) = (l)1/2R2^2i^)ei^)U^.
we can define respectively as amplitudes of the incident wave.454 we can rewrite the . we can write this (multiplied > by 2ismvTT. and recalling that yreg is proportional to a function Mc> {z. cf.56). R — — = 2i sin ivy. We can rewrite Eq. 9?. Thus in terms of the (3) variable z.81) in another form from which we can deduce the amplitudes of reflected and transmitted waves. (19. h) (see Eqs. and with the left hand side following from Eqs.z—>oo 2r \l/2 i(is+l/2)ir/2 2hir cosh z r> \ „2ih cosh z RU + i [Re —2i/icosh z R Thus. R 2i sin TXV. V(r)=g2/r Fig. (19. 19. 19.2. (19.Smatrix in the form s i n 7T7 Si = CHAPTER 19.2 The repulsive potential and the various waves.85b) We shall obtain the 5matrix in this form in the case of large values of h2 later.56) to (19. Singular Potentials r(i+k) sin 7r(7 + u) (19. Fig.86) . (19.58)) in the limit r — 0.58)) 1/2 2r e~ 2hir cosh z. the reflected wave and the transmitted wave the quantities: A Ar At = = Re1' R 2zsin 7r(7 + u). (19.
178). (19. i. (105) to (108). J.88) the power expansion of the Bessel function one realizes soon that the expansion is inconvenient in view of its slow convergence.h) is (cf. unitarity is preserved.39)) the associated Mathieu function expanded in terms of Bessel functions of the first kind. Jv+2r. 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. (19. S.6 E v a l u a t i o n of t h e S .In Meixner and Schafke [193] (p. y We observe that this relation remains valid if the real quantity R = e and the pure phase factor eW7r exchange their roles. but in the 10dimensional string theory context.2COS2VTT = (R.„„„X (1 Q Q'T'l l \Reiun e~iuw/R\2 ~ \Reiv* .79). W.m a t r i x Our next task is to evaluate the expression (19. (^)l4lu{h2)Jir(hez)J±u+i+r(hez). (19.88) where the factor Me„(0.e. J. / i ) = Y.** 19. (p. Hashimoto [124] and R.e.84) of the ^matrix.. if R becomes a pure phase factor and eil/7r a real exponential." \R~Ji\2 1  tO _  2 i s i n VK\2 [  .h) The function Mo (0.28)) and R = ey real.2. A derivation is beyond the scope of our objectives here.19.2 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. i.Q.e. 180).h) = 1 Me V\ 5 / r J2 __ 0 O c»2r(h2)J„+2r(2hcoShz). MullerKirsten. H. therefore we cite it from Meixner and Schafke [193]. unity minus reflection probability = transmission probability. in which v is complex. i. Eqs. l=—oo 1 1 lite*"* . Eq. " S e e S. By inserting in Eq. (19. The latter is what happens in the S'wave case of the attractive potential. (19. The expansion is: 4 r ( / i 2 ) M S ( z . this expansion is given as 1 oo MJ> Hz.89) . This implies basically the evaluation of the expression R of Eq.( i J e ^ ) " 1 ! 2 = R2 + ^ . Manvelyan. h) serves the use of the same coefficients as in the other expansions. i. and then exploit it for the evaluation of the quantity R. M^l)(0.e™*/R\2 ' ' . (19.±)2 + (2sinj/7r) 2 . Liang and Yunbo Zhang [189].e. however not here. Gubser and A.
55). The formula (19. h2 is complex but v2 is real (cf. one obtains *.6 2 ) m 6 (^_i)3(z.e. setting r = 0 in Eq. r = 0 and 2. e.94) .91) Inserting here the coefficients given in (19. (19. we obtain M&(0. + 2 u l\2j 2 fh\2 "2+2 2 MX4 {v l)(v 4)\2 2 2(^ + 4 ^ .90) is in some sense amazing: It permits the evaluation of one and the same quantity M±J (0. by allocating different values to r./i) with the help of Eq.456 so that in particular for z = 0 oo CHAPTER 19. As a matter of introduction we recall the definition of the coefficients with Eq.h) v 2 = ^2 (l)lc%ih2)Ji~r(h)JM+r{h).90) Here the coefficients c2"{h2) are the same as those we introduced earlier for the normalized modified Mathieu functions in Eq. Singular Potentials 4r (h )M$(0. Here in the case of the repulsive potential.49) from which we obtain for z = 0: Me„(0. Eq. u»^. i.28)).g.90) and choose r = 0 (one can choose e. Next we evaluate M^1}(0..90). (19. (19. (19. ^ (19. (19./>) = Co> 2 ) £ 7 " = — OO j ± > .g.3 9 ^ .h) = Jo(h)J±Ah)^^Ji(h)J±„+i(h) c0 {n ) ^ ' C^r^J2{h)J±l/2{h) cr(h2) + •••.2_4)(l/2_9)iv2y) + <AA. h) in many different ways. (19. r = 2 as a check). 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. /=—oo (19.49). /n ^ .
111) / / i 5 4 3 1+ 4i/ 2 + +' (z.7) F " (i/±l)2(^^l)(i/24)V2.2 The Potential 1/r4 — Case of Small h2 457 Inserting the power expansion (19.IV 2 2 2(v =F 3i/ . one obtains M£!(O.l) 5 (19.1) 2u(4u + 15i/ .4 ) V 2 > 1 (I/)!V2/ [1+ 2_Mx2 2(i/23«/7) (h)4 (iy+l)2(i/l)(^24)' .2 _ ^ 4 ( ^ 2 _ 4)2 (19. 4(z/ T H ^ 3 . .12i/2 + 64u .96) With this explicit expansion we can evaluate the S'matrix.19.95) These expansions imply for the quantity R: I/! 2v ri1 i . From expansion (19.4)(i/ ./I) h u .93) and the coefficient expansions (19.] 2 \v i + {v2 .1 ) 2 ( I / + 1 ) ( V 2 . + # "+" ( I / ..2^ 2 ± 59^ .32i/ . _2_(h\2 .96) we extract for later reference the relation sin7rz/ R IT h h 2\ 2 Av [v\(viy..9) V2 1 fh ±v 1+ 2 + • (19. In this procedure our attention is focussed particularly on the quantity called absorptivity in the case of the attractive potential.55) and collecting terms in ascending powers of h? (here for the case of nonintegral values of v).23) (hxQ 2 3 2 2 V ± l) (i/ T 1) (^ .!' or with the help of the inversion formula of factorials R = v\{y — 1)! smKv f h 7T 2v (^ .97) + This expansion will be used below in the low order approximation of the absorptivity of partial waves.. We observe that this approximation involves v and h4 and hence is real. 2 (h\2 ^ l/2_1l2^ 2(i/ 2 +3f~7) /feu .
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/? }]"' (19.104) We now consider the case of a — 0 implying that g of Eq. This is the case of real . Then cos 2TT/? = 1 .28)).R*2 ^ * #2 \ 11 tf2 R* (19.^ i 4 2+^ # 1 / •. 9 / « i(2yr/3)2 ss i(sin 2vr/3)2 ~ 2 sin2 TT/3. Thus we set v = n + i(ot + i(3) = (n — 0) + ia.28) that v is real.100) Here we set 2^/3 = 1 + j / ? e 2™ = l + gt (19.7 CHAPTER 19.9 / . In this case R — R* and so 1/i?2 ~ Ofji4). (19.98) (cf.458 19.101) is > zero. (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.101) where / is complex and g is real.eiun/R2)(eiu*n ~ (19. (19.103) Then SiSt (1+5) i . _ i?* 2 2+ g # 2 i?* 2 2 3/(i + s)(4 + i?*2). and sin 2TT/3 = ft/.2. (19. Eq. (e iun e~iv*^/R*2)' (19. where n is an integer and a and (3 are real and of 0(h4) We have with Eq. 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.102) (19.84): Si ST 1 (11/R2)(11/R*2) . (19. + ( i + s w (E 4 .
Q. Hence with Eq. "This is the result effectively contained in R. 2 /l4 '•i Al=0 2[(l + 1 / 2 ) 2 .61n2 . Hashimoto [124].2 The Potential 1/r4 — Case of Small h2 parameters v = n — (3. (19.86) and obtain in leading order for I — 0: Ai = *[l + 0(h% Ar = ±[l + 0(h2)]. 1 29/ ^ (1 . p.* )  14(^V = 1 T h e absorptivity A is therefore given by A = 1 .107b) = Ah2[l + 0(h4)}. Manvelyan. Oberhettinger [181]. We can also evaluate now the amplitudes (19.19.577.106) and evidently violates the unitarity of S.105) sin7r/3 (19. one can calculate the next term of the expansion so t h a t At=Q = 4 ^ 4h4 {7 .l ) 2 2 \) +0{h«) (19. At = 2i[l + 0(h2)}.28).1)\} 2 h \2 iv 1 + Au {y . Liang and Yunbo Zhang [189] (there Eq. f W . Magnus and F. (19. . J.e.61n/i} + ~9~ 0(h8) (19. W.97) this can be written A 4TT2 [v\{v . With t h e help of Eq.107c) Here h2 is actually Vh4.3b)) h2 is real for E = k2 > 0. ( I l l ) ) in agreement with a result in S. = 1 .102) we obtain Sis. i.^ ) ~ 2 2 459 1+ 9 / • & ( ! . Gubser and A. J. 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 threedimensional case we are considering here.SiSf » 4 sin7rz/ R (19. Using the derivative of t h e gamma function expressed as t h e psi function ip(z) = T'(z)/T(z).1 ] Oih* For S waves this implies an absorptivity given by ((—1/2)! = 1/7F) (19.3.6C . (19. Here in t h e case of the attractive potential (as remarked after Eq. and t h e values of ip(\/2) and ip{2>/2) as given in Tables^ with Euler's constant C ~ 0. H. (19.107a) In this result* we now have to insert the expansion of the parameter v given by Eq. S. MullerKirsten.
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.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 5matrix obtained above by using the asymptotic expansions for large values of h2. W. i. MiillerKirsten and Jianzu Zhang [226]. Tamaryan. h) for large h and upwards I = 0. Park.460 CHAPTER 19. H. .h) (19. N. S.e.h) 10.4) which — for some distinction from the smallh2 treatment — we rewrite here as 2 + [2h cosh 2z . VahediFaridi [8] and G. (19.3 19. completely different solutions. W. Aly.26) and hence set a1 = 2hl + 2hq + A{q. J. This is therefore a highly interesting case which found its complete solution only recently.$ q(l. K. Miiller and N.1. H.3 The function q(l. 19. Singular Potentials 19. Wannier [278].3.3. We are again concerned with the associated Mathieu equation given by Eq.108) For the large values of h2 that we wish to consider here. 2 2 a = [l + 2 (19. we again make use of the replacement (17. where the part of interest here is based on H. J. 2.H. H.109) ^We follow here mainly D.4 Fig.a }tp = 0.
with an essential mathematical step. With the boundary conditions y+(0) = l. We also found the boundary condition (17. and a is its normalization constant. however.111) we have W[y+.yJ\ — 1 and cos TTV . we defined the Floquet exponent v and observed with Eq.110) a where y+(7r) is the solution evaluated at z = TV. . (17. Therefore the second term on the right is nonzero. See also Eq.3.l + 2y+(l.§ a W[y+.19.21b). We begin. 19.e.h*y_(l.113) 'We are considering nonintegral values of v.h*y (19.y^] = ab\f\ is the Wronskian (in leading order) of the solutions which are even and odd about z = 0 respectively.112) In Chapter 17 the unnormalized even and odd large/i2 solutions of the Mathieu equation were given by Eq. which therefore cancels out or can be taken to be 1. (19. y± oc A{z)e2h sm * ± A(z)e2n sm z .5) that this is given by the relation cos iris = ^ M . (19. to which we have to return in this subsection.21b). i.109) in this way enables us to obtain asymptotic expansions of solutions very analogous to those of the periodic case. (17. since it is clear that large h2 considerations require a knowledge of the large/i2 behaviour of the Floquet exponent v.e. i.y} where VF[y+. y(0) = 0. (19. and A / 8 is the remainder of the laigeh2 expansion as determined perturbatively. (17. The behaviour of q as a function of h for some values of / is shown in Fig.63). y'_(0) = 1.2 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. which is even about z = 0.3 The Potential 1/r4 — Case of Large h2 461 where q is a parameter to be determined as the solution of this equation. The replacement (19. 19. J4(0)=0.3.
118a) "We emphasize again: For integral values of v the normalization constants are different.117) a a Referring back to Eq.114) at z = 7r/2. (19. Dingle and H. i.462 CHAPTER 19. Hence we require (obtained from Eq. J.46) together with the expansion of the Hermite function from Tables of Special Functions" ) B[w = 0] 7T 4 [lfol)]![l(g+l)]! 2 /i 5 82q2 . Dingle and H. 1 1 Or see R. (19. K [ (z)}2hsinz (19. in the second expression 2h is obtained in addition from differentiation of the exponential factor).A(z)e2h sin 2 sin z sin z (19. Miiller [73]. (19.111) we have^ iVo = 2 3 /2 1 + °{k and K = ¥frh l +O (19. B.115) Setting z = 0 we obtain in leading order y+(0) = 2iV0A(0) and y'_(Q) = 4hN0A{0).114) a a Our first step is therefore the determination of the normalization constants No. (17. Singular Potentials and their extension to around z — TT/2 by the relations C[w{z)\ ^_2h sir Mz)}„2h. Miiller [73]. W.39 2 14 /i 2 + (19. J. and hence with Eqs. (19.e. B. Kin B y± oc y+(z) = N0 A(z)e2h A(z)e2h sin z + A{z)e~2h . which is \/2 at z = 0.112) from evaluation of Eqs. (17. (17. We now obtain the expressions needed in Eq. 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 . This paper contains the normalization constants of large/i 2 periodic Mathieu functions.116) (in the first expression we have A(z) ~ Aq(z) with Aq(z) given by Eq. W.32). .43) we see that z = 7r/2 implies W(TT/2) = 0. See R.
62a) and (17.h . 210.\ { q + 3)]! 7re 4/i 24/i + (2/i 1 /2) ( 8 / l ) 9 / 2 [ _ l ( g + !)]. (19.2g + 3) 27/i ( 3 g 2 . With the help of the duplication and inversion formulas of factorials the result can be reexpressed as cos( g 7r/2)[£(gl)]! .[_ l ( g 2h and hence 5 + 3)]! 'All Ah 1+2 (3g2 . MiillerKirsten and Zhang [226] — is cited without proof by Meixner and Schafke [193]. a in Eq.3 The Potential 1/r4 — Case of Large h2 and 463 as dz )z=n/2 dw (dw\ o\dzJz=n/2 1C%2 .! ) ] ! [ .62b) (see also the explicit calculation in Example 17. (19.1Q11QM cos irv + 1 ~ •==— e .19. Tamaryan. Thus in particular (8h) i(?i) a (3q22q + S) 27h + l\(qW " ' Inserting these various expressions now into Eq. (19.118b) The factors a.119b) COS TTV + l ire This result gives the leading contribution of the lavgeh2 expansion determining the Floquet exponent v. p. (19.3). . Presumably they extracted it from "between the lines" of a sophisticated paper of Langer's [159] which they cite together with others.117) are known from the matching of solutions in Eqs.87 2 14 /i 2 [ i ( g . (17.8 g + 3) +0 ^2 (8h)^[~l(q + l)]\[l(q + 3)}\ 2eh ' ~ (19.i ( g +3)]! 1 24/i {2hl>2).! ) ] ! [ .119a) This formula — derived and rediscovered in Park.112) we obtain COS TTU + 1 7T '2Wa[\(ql)]l[\(q+l)]\ 02h 1 ¥h + 2TT ' 2^ha [\(q .
121) We let Aq(z) be the solution of this equation when the right hand side is replaced by zero (i. dA 1. each with a specific asymptotic behaviour.3 Following the procedure of Sec.. tp(q. of course. —h.h. it is still a solution but not with the symmetry property ip(q. z) can then be written as . z) = ip(—q. . h. Vcosh z \ 1 .h. —h. —z).122) Correspondingly the various solutions ip(q. for h — oo).4.464 CHAPTER 19. h.z).h. i = 1. With the solutions as they stand. We define these solutions in terms of the function Ke(q. h.120) to (19.z).3. .s i n n z ± iq)A = ± K qj dz 2 Aih d2A dz2 (19.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.z) := = exp[i7rg/4] j===^Aq{z) exp[2/wsinh z i(77r/4 V2ih k(q. h.h.109) into the equation (19.z)exp[±2hismh z\.2.123) We again make the important observation by looking at Eqs. we can obtain the linearly independent one either as ip(—q.120) The resulting equation for the function A(q.2 we insert the expression (19. (19. the expression (19. Singular Potentials C o n s t r u c t i o n of largefa 2 solutions 19.108) and set ip{q. —z). h. z) = ip(q. z) are TP(q. 17. 1 / l + isinhz\T9/45R^oo e^ / Ag(z) = ===[r^— ~ . h. h(q.? s m h z) Vcosh z (19. .h).h. (19. „ 1 A „ —A cosh z——H .h)iP(q.3.z) or as tp(q.124) Since this function differs from a solution ip by a factor k(q.e. V— 2in (19.3. z).109) remaining unchanged. Then straightforward integration yields > inq 4 „. h.122) that given one solution ip(q. In the following we require solutions H e ^ ( z ) .h.z) i/>(q.h) =?==.z) = A(q.
(z.fi2) can be reexpressed in terms of Hankel functions./i 2 ) M (19. This is. 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.e. * ** = > rojKe( ". we h a v e ( u s i n g cosh 2rj = 2 c o s h 2 77 — 1) ry Jo ' = dr. (19. >• * F $ 0 * e". (19. or — equivalently — by the appropriate expansion of the Bessel functions as given in Tables of Functions. 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 . 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]. 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 ./h). (44) of W a n n i e r .126) For large values of the argument 2/icosh 2.19. .3 The Potential 1/r4 — Case of Large h2 465 Instead.45). 2j<. we have to match a solution valid at Kz = — oo to a combination of solutions valid at $tz = oo. as we saw earlier.h2) = OvMPfrh2). : 7r/2.^ s i n h 7?)]g. Me„(z.127) sin z T /'1 M TT\ — <>{ 4 .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 .h2).e. the Bessel functions contained in the expansion of the associated Mathieu function M{.fc 2lqk A + 1/2 + 4fc2 c o s h 2 77 A / 8 P . av{h2) = ^(0J ^ Mi i .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]. Ju(z) TTZ VK 7T cos z T~ 4 1 + 0 ^V. we h a v e 77 = 0—> 6 = 0.e. in performing this cycle of replacements the function picks up a factor. i. We observed that these satisfy the relation (19. achieved with the help of the Floquet solutions Me±v(z. (0. It follows t h a t S e t t i n g s i n h 77 = t a n 0 . . In order to be able to obtain the Smatrix. / 2i<jfc + A / 8 \ ~ 2fc / drj cosh 77 1 + * _' Jo V 8fc2 c o s h 2 77 7 2fcsinh y + [ t a n .<5±1) (19 125)  (the expression on the far right in leading order for h2 large). i. [ \ . One can show that the quantity < o of Wannier* (see above) is related to q by $o = & iqir/2 + OO.
130a). h.h) Be^(z.q. (19. HeW(z.g.q.h.q.4 T h e c o n n e c t i o n formulas We now return to Eq. 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. (19.466 CHAPTER 19.129) The solutions so defined have the following asymptotic behaviour (where e(z) = (2/icosh z)~ll2 and h2 = ikg): H.130b) r_0 rll2exp\g/r (ig) ' 19. He (z./i) Be^(z.h).q.h) (3) = = = = Ke(q.q. (19. 3te»0.h) He (z.h) = e{z) exp ihez + iexp[ikr + i7r/4] kr 7T .128) and consider the cosine there as composed of two exponentials whose asymptotic behaviour we identify with that of solutions of Eqs.q. exp[—ikr — in/4\ kr Re{2)(z.130b).q. z): Re(2\z.130a) Re^(z.h). (19.h )~ \ ot±v cos(2/icosh z = vir/2 — 7r/4) F V2/tcosh z (19.3.q.h) = e(z) exp ihez . (19. In this way we obtain in the domain . KeW(z.z). 3te»0.h) = e(z)exp exp ihe^ 4 + i\ 12 i4 3^cO. (2) (19.128) We now define the following set of solutions of the associated Mathieu equation in terms of the function Ke(q.vr .eW(z.g./i).
129)) He (2) (z. q.q.h2) Me„(z. q.g. e W 2 H e ( 3 ) ( ^ ? j ^ _ e M/*/2jj e (4) ( ^ q j ^ e . / i ) + sin7r(7^)He(4)(2. (19.134) in the region where the function He^(z.g.i ^ / 2 H e ( l ) (z> ?) (19. we see that the solutions (compare with Eqs.h).g. h.125) the proportionality factor can be seen to imply the relation exp •ffa + 1) sin7r(7 + "} (19./i) — sin 7T7 Re^(z.* ^ / 2 H e ( 3 ) ( Z ) g? Mev(z. g .130a).132) These relations are now valid over the entire range of z.19. (19./i).(z. Changing the sign of z.q. these are exponentially increasing there.135) sin irv 'This means. Eqs. (19. e x ^ r / 2 H e ( l ) ( 2 > g> />) _ e . and He(*> exponentially decreasing.133) In a similar way one obtains the connection formulas — sin nv HeW(z. .131) where the second relation was obtained by changing the sign of v in the first. (19. z) are proportional there.3 The Potential 1/r 4 — Case of Large h2 fc>0 the relations: = 2TT 467 Msv{z. q.130b)).q.132) into Eqs.134) Considering now the first of relations (19.h2) = 2vr i a_ 2TT />) _ e ^ / 2 H e ( 4 ) ( Z j ^ Q (19. h) = Ke(q. q. and thus can be matched in this region. h) = Ke(q.h2) a. — sin 777 H e ( 3 ) ( 2 .131) and eliminating He") and setting again exp[m7] := we obtain the first connection formula ozwr — sin irv He (4) (z.h) is asymptotically small (cf. (19.h) sin irv = sin7r(7 + ^)He (3) (2. h) = sin 7r(7 + v) He (1) (z. (19. (19. h. h) — sin 7T7 He^(z. we obtain in the domain tfcz <C 0 the relations: Me. (19.« * r / 2 H e ( 2 ) ( ^ g> ^ ft) _ e W 2 H e ( 2 ) ^ ^ ^ = a_ 2TT e ./i 2 ) = Me_j.li 2 ) Oil. z) and He (3) (z.(z. t With Eq./i). Substituting the Eqs.
Wannier did not have the relation (19.85b) obtained earlier. (19. which in our case is the solution He^4^. (19.e.119b) as „4/i 1 + COS TXV ~ rrl ( g . (60) in the work of G.138) TXVICOS TXV We obtain the behaviour of the Floquet exponent for large values of h2 from Eqs. sin TXJ. (19. The latter is defined by the following larger behaviour of the solution chosen at r = 0. ' cos TXV ± v 1 + elin sin2 TXV =p v'cos 2 TXV .136) ^ jrfl+l/2) (19:I35) sin 7r(7 + v) sm WKljrtla/2) S m TXV (19.135). H.139) ''Equation (19. Squaring Eq. (19.5 D e r i v a t i o n of t h e ^ . .l ) ' s i n 7 T 7 ^VK /2 ikr sin 7r(7 + v) _l)leikr i5i„—ilir/2 SxelKr . is only the dominant term for large h2} 19.(19. (19. Singular Potentials The factor on the left.l ) ] ! c o s ( W 2 ) 2 2 (8h)i/ 2TT29/ 2 +0 (19. Wannier [278].( .Sin TXV. exp 2^(9 + 1) and solving for sin TXJ one obtains the expression sm 7T7 sm TXV = ~iei7rq/2sin le iqK • COS 7T7. h) ~ — sin TXV (1) sin 7r(7 + v)e kr ikr in 4 rl/2eg/r+m/A / (ig)1/2 ( . and therefore remarks after his Eq.119b). i. of course.1 .l ) (l„—ikr e From this we can deduce that Si (19.135) is the equation corresponding to Eq. where 8i is the phase shift. (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".3. The quantity 7 is now to be determined from Eq.133) we can now deduce the 5matrix Si — e2t51. Thus here the S'matrix is defined by (using Eq. q.e'i7ri 1.135).137) We see that this expression agrees with that of Eq. (19.468 CHAPTER 19.133)) — sin TXV He^4' (z. cos TXV + sin TXV.m a t r i x From Eq.119a) and (19.119a) or (19.
19.3 The Potential 1/r 4 — Case of Large h2 469 Since the right hand side is real for real h. (19.141) COS TTU irvj — i s i n WR s i n h TTUJ. T h i s is r e a l for UR a n integer. i.e.19. h) of the Zth partial wave. Using Stirling's formula in the form z\ ~ e~zzz+1/2V2^.h):=l\Stf ^cos rr(uji + ivi) = cos Tn/jicosh COS ^7TQ N 2 1 (19.137) together with (19.e. for the attractive potential. 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 (lp/2)n (19. . i. (19. This is different from the behaviour for small values of \h2\. we can approximate the equation for q ~ h (i.8h COS From Eq. A{l. (19. as a comparison with Eq.139) could actually be neglected) implying  cos m>\ = cosh nvi.109).4 The absorptivity of the S wave (attractive case). 9/2 (91) with the approximation q ~ h obtainable from Eq. irrespective of what the value of I is) by cos™ + l = c o s ( T j^J y/e1. since cos TVU is large. imaginary part uj (so that 1 on the left hand side of Eq.e. (19.140) From this we obtain the absorptivity A(l.28) shows.138) we obtain. the real part of v must be an integer.§ 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.
in further contexts beyond those already mentioned. and. (19. Jones [78]. A. E. 19. D. small h2) case have been considered by some other authors. Dombey and R. H.k) in the approach to unity are too small to become evident here. In essence. H. Rosendorf [225]. N. I r a n [61] and M.^ A highly mathematical study of the potential ~ r~2 as an emitting or absorbing centre has been given recently. Esposito [88]. 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. Apart from the papers already referred to. Challifour and R. Apparently this is one of the very rare cases which permits such complete treatment. F. Paliov and S. L. H.107b) and (19. This is sketched schematically in Fig.4 (there are tiny fluctuations in the rapid approach to 1). Limic [177]. one may expect corresponding results for other singular power potentials. Treiman [271].4 . . Pope and T. Cvetic. D. such as phase shifts.142) we can see the behaviour of the S'wave absorptivity as a function of h.P. VazquezPoritz [61]. Y. B. 11 J. Lii and J. S . Furlan [20]. Tiktopoulos and S. (19. **A. Handelsman. Shabad [248]. Lii.4 Concluding Remarks In the above we have considered the solution of the Schrodinger equation for the potential r . "G.139) 2Tr(16h)q /32\ f e Thus with Eqs. YuanBen [285]. Fubini and G. **R. see also M. 19. N. Bertocchi. The diminishing fluctuations of A(l.e.' Rudimentary aspects of a singular potential with a general power n have also been considered previously. C. Partial or other aspects of the weak coupling (i. N.470 CHAPTER 19. J. A. Masson [191]. Eden [46]. W Some of these sources can be helpful in further investigations. Singular Potentials with near asymptotic behaviour (for h2 — oo) > Afi . and G.** 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. S. A. Pao and J. Cvetic. in fact." as well as other aspects. Lew [128]. 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.
B.* In fact. *R.* However. M. B.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. the asymptotic solutions in different domains all have the same coefficients. f R .g. J. the cosine potential is more suitable for such studies than the anharmonic oscillator since its case is simpler. Dingle and H. e. We concentrate in this chapter on the large order behaviour of asymptotic *C. Bender and T.Chapter 20 Large Order Behaviour of Perturbation Expansions 20. T. This is indeed a remarkable connection. also convergent expansions are known and a lot of literature exists on the equation. Muller [74]. Wu [18]. 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. Dingle [71]. [19]. how the large order behaviour of the expansion of the eigenvalue is related to the discontinuity across the latter's cut. 471 . W. 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. 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. We shall also see explicitly.
Presumably this has never been done so far. This approach leads . 20. 8.2 and below.472 CHAPTER 20. Large Order Behaviour of Perturbation Expansions expansions using in particular the recurrence relation of the perturbation coefficients obtainable with the perturbation method of Sec. 10 12 14 16 18 20 Fig. as evident from Example 17. Fig. 19. 8.2.2 The function ur = r\/pr vs.1 and cases in Sec. one could naturally also explore the large order behaviour of these in a similar way and expect the behaviour discussed in Sec.6. 5. 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. r for p = 4.2.7. r for p = 4.6. 20. Since the method is also applicable to convergent expansions.8.1 The function ur = pr jr\ vs.
in the power expansion of the exponential function). In some cases the answer is affirmative. The formal perturbation expansion of a quantity E(g ) .20. 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. the function E(g2) is given only approximately by Yln=o En92n. i. even if each En is known. . (20. and the expansion is said to be Borel summable.1) is an asymptotic series if for any N lim 92^o 1 g 2N N E(g )^2En92n 71=0 2 = 0 (20.2) or. 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.1 Introductory Remarks 473 automatically to asymptotic series.1 and 20. For an authoritative discussion of these aspects we refer to Dingle [70]. If the function E(g2) can be written as the Laplace transform E{g2)= dze'z/g F{z)=g2 dte^FigH). 20. this term represents the size of the error. The behaviour of the rth term (r large) of the convergent expansion is generally of the form of a power divided by a factorial (as. The question arises: What is the information about F(z) that can be obtained from the asymptotic expansion of E. It can be seen from these.2.3) n=0 Thus for any given value of g . T V lim h2N 2 n EQ?) .e. The traditional method of using an asymptotic expansion is to truncate the series at the least term. and the integral E(g2) is called the Borel sum. for instance.J2 2n h E (20. E(g2) = f2 Eng2n = Eo + J2 n=0 n=0 E 2(ra+l) ^9 (20. i. equivalently for g = l/h. as we discussed earlier in Chapter 8.e. whereas in the case of the asymptotic expansion it is the opposite: a factorial in r divided by a power.4) Jo Jo its Laplace transform F(z) is called the Borel transform of E.
(20. (20.4) can be evaluated and E(g2) thus recovered from the asymptotic expansion by Borel summation. (20.n E(9 )~ 2 n=0 ^ E0 +1) = Ea^2( °° oo /»oo = J2ang2^ „n 'O n=Q dte~H\ a_i = E0. In fact. . Large Order Behaviour of Perturbation Expansions Eq. . n=0 (20.6)).1)? Asymptotic expansions originate from integrating a series (i. that of F(z)) over a domain that is larger than its region of convergence (see Chapter 8 and Dingle [70]). In this ideal case the integral of Eq. In any case. .n _ En+\ n! En+1 ^ (!)"«!. 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. In general this is not so easy. The function F = 1/(1 + z) is an analytic function which can be continued beyond this circle to the entire positive zaxis as required by Eq. (20. Thus if CO F(z) = aiS(z) + J2 anzn.474 CHAPTER 20.6) The power series of F(z) converges inside the unit circle about the origin at z = 0.e. We can therefore write down a dispersion relation representation choosing the cut from 0 to +oo: . many asymptotic series do not even alternate in sign (as in the case of the series in Eq. .4). . n=0 (20. 1 we obtain oo F(z)£(*)" = — .5) then with the integral representation of the gamma function F(z) = (z — 1)!. We observe that a function E(h) with asymptotic series ^ r=\ A has an essential singularity at h — 0. J we obtain Now if />oo dteHz\ °° oo ' 0 dteH*\ TEn+1g^V . (and E0 = 0). El . . Q.
Comparing the last two equations we obtain (by expanding in the latter the denominator in powers of h'/h) I jQO Ar = 7T JO tir~l%E{ti)dti.2.e. if the potential is V(r) = +grN..2. (20. N = l. we have a quantum mechanical problem where probability can leak away. S.. if the potential or a boundary condition is complex such that the problem is nonselfadjoint. The potential (20./ \p\dr\.10) r as shown in Fig. .. M. § where 7(ff) = To exp  . i.3 and g < 0. This is the case if the real potential has a hump of some kind. and E represents an eigenenergy. (20. and with g = 1/h and it is found that§ En = 4 n ) ~ \lis). or can we determine Ar independently? Fig. Popov and V..10) is considered in detail in this work. (20. Weinberg [228].. e. $s E can be calculated by the WKB procedure... for < ^ 0.g..11) V..3 The potential V(r) gr». ? When ?s E ^ 0.20. 20.9) The question is therefore: How can we determine 9 E(h).N = l. 20.1 Introductory Remarks 475 except for possible subtractions.
^ The logarithmic factor in Er is a novel feature of this case. 8. of course. the cosine potential being t h e most completely investigated will also be seen in the following. M. (20. the perturbation method of Sec.113)). Eletsky. the eigenvalue expansions of the cosine potential and of anharmonic oscillators. the BreitWigner formula (10.76) (see also the comment on the comparison with the Kepler problem before Eq. Popov and V. Popov and Weinberg.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. (20. (11. Example 20. Equation (20. Thus in the following we apply these methods to our typical examples. Here n is the principal quantum number of the Coulomb problem (see Eq.2 Cosine Potential: Large Order Behaviour It is instructive to investigate various methods for obtaining the large order behaviour of an asymptotic expansion. Straightforward RayleighSchrodinger perturbation theory usually does not enable this in view of the unwieldy form of coefficients after very few iterations. 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.11) is. S.1. as we expect on the basis of Eq. . However. Evaluating the integral show that 00 r=0 v ' where n is the principal quantum number of the Coulomb problem as in Eq. Weinberg [83].9) for Ar. (14. since this is effectively the Mathieu equation for which — for comparison purposes — extensive literat u r e was already available. Solution: The solution can be found in a paper of Eletsky. V. An interesting application is provided by Example 20. In the first such investigation it was sensible to consider the Schrodinger equation with the cosine potential. 20.7 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.476 CHAPTER 20. as was (and still is) not the case for t h e equation 1 V . 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 .11).73) in the W K B or semiclassical approximation.134)). (11. L.
J.13) represents effectively a partial difference equation.60).j) a n d M2i can be deduced.j + 1).58) and (17. B. (17. Eqs.13) In the special case j = i the boundary conditions stated after Eq.l ) +2[(q + Aj)2 + l + A]Pl.55)) X = E(q. j . i. (17. This has been done" and it was found that (we cite the results here since we rederive them below in detail by other methods.20.Cq. Miiller [74].l ) / 2 ] ! i![(gl)/2]! • We have seen earlier (cf.26) and (17.58) eliminate the last two contributions and one obtains the exact expression ^ ' Z j _ 4<[2» + ( g .34)) that the coefficients Pi(q. Dingle and H.j) of functions ipq+4j.J) = (g + 4 j . (17. arising in the ith order of the perturbation expansion of the wave function oo 1 i ip — const. % = Aq.M2i+1 for large i.l(q. Thus in the case of the Schrodinger equation with periodic cosine potential we saw (cf. h).j) are known.e.12) (20. (cf. W. .2 Cosine Potential: Large Order Behaviour 477 with an anharmonic oscillator potential.j) +(q + 4j + l)(q + 4j + 3)i^_ x (g. The recurrence relation (20. 3)? .61)) that the coefficients M2i of the expansion of the eigenvalue A = E(q.2 A = Mx + ~M2 + 7^WM3 + •• • . (20. or equivalently of the quantity A.Bq. to avoid discussion of difference equations here) 2 l{q j) ^ (16)«(»l)!(2i)«/° / _ 2q 2q + 3\ f i \ ' [\{qmwm\ sr ){j)> (2(U5a) R. Eqs. (17.3 ) P i _ 1 ( g .l ) ( g + 4 j . M2i = 2j2 J[Pi(q. (17.h) = 2h2 + 2hq+j. Eqs. (20. With some simplifications for large values of i this equation can be solved approximately and the large order behaviour of the coefficients Pi(q.14) Thus these coefficients can be obtained once the coefficients Pi(q. obey the recurrence relation jPi(q. are given by i .
. Large Order Behaviour of Perturbation Expansions where the bracketed expression at the end is the binomial coefficient. and from this for i even or odd it is found that Mi ~ (16)^! 1 2g2 + 3 1i . Using this relation and the duplication formula lw('\W±^.l ( 9 + l)]![(Q + 3)]!}2 i!^" 1 1 or (with the help of the duplication formula.•91 {[l(? + i)]![(? + 3)]!}2 (20.478 CHAPTER 20. (20.17) We see that assuming i is very large compared with m is here equivalent to assuming m is approximately zero. In the first method we do this by first obtaining the imaginary part of the eigenvalue with the help of nonselfadjoint boundary conditions.th term iW~^q 1 {[i(gl)]!} 2 2 7 r(8/ l ) i . also written lCj.15b) The zth 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) il 1 2g2 + 3 2i )[{[\{q 7.18.19) ^nOW =2(9"1) We observe that successive terms do not alternate in sign.l (20.*.1 22 the result becomes i t h term (i + 2n)!2 2 " 1 where n 1 (20. see also below) .91 l)]![i(g3)]!} 2 { [ . Hence exact Borel summation is not possible.16a) (20. In the following we concentrate on the cosine potential and rederive the large order behaviour obtained above by other methods.16b) With the help of Stirling's formula one can derive the following relation: i large 1 (i + m)\ \o (20.
We retain the previous boundary condition at one minimum. in the calculation of the level splitting as a result of tunneling. t To enable comparison with the work of Stone and Reeve we shift the cosine potential considered previously by 7r/2.5 A i 1 1 r f \ \ M\ \ \ / ! \ \ 1 \ \ I 3 \ \\ \ /41 i \ \ 0. . Stone and J. this difference was found to be real. l . 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. What changes as a result of different boundary conditions is the deviation of q from the odd integer qo. \ 1 \ \ \ \ \ 0.* However. remains unchanged. J. The potential we consider here then has the shape shown in Fig. . . This is a very instructive exercise which shows how the eigenvalues change with a change in boundary conditions.e. 20.4.3 20. W. but now change the boundary condition at the neighbouring minimum in such a way that the difference q — qo becomes imaginary. when this is g0 = 2n + l .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/?. f H .20. A means to achieve this was found by Stone and Reeve. 2 . i. we follow here a formulation in the context of our earlier considerations of the Mathieu equation. . Reeve [262]. .3 Cosine Potential: Complex Eigenvalues 479 20. MiillerKirsten [200]. we emphasize a few points *M. At the risk of repeating what we explained earlier. Previously. From our previous considerations it is clear that the formal perturbation expansion of the eigenvalue X(q) in terms of the parameter g. 20.4 The cosine potential shifted by TT/2.5 \ \ / \A Fig.3. n = 0 . but with different boundary conditions.2 solutions obtained earlier. the difference q — q$.
and correspondingly we assume alternately even and odd oscillator eigenfunctions there. 20. 20.4.21) <u> + ( i « . (20. Large Order Behaviour of Perturbation Expansions for reasons of clarity (since some considerations here may otherwise not be sufficiently clear). but t h a t with complex argument there. Effectively we consider the tunneling from the left well in Fig. but achieves exactly what we want.2h2 cos 2z]ip{z) = 0. Now we change this situation at the minimum on the right. X + 2h2 = 2hq+—. To be more precise we recall the Mathieu equation we had earlier: ip"{z) + [A .20a) As stated. Thus around z = 0 the potential behaves like t h a t of an oscillator well with minimum there. The physical wave functions there are therefore the solutions which are real and exponentially decreasing. Setting w = 2h1/2z. i.4 (around z — 0) to the one on the right (around z ~ ir).Ah2z2 • • • }^{z) = 0. 20.480 CHAPTER 20.4. (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. which means replacing z by z±ir/2.T K ^ a D_i_{q+1){±iw). a boundary condition which at the face looks somewhat abstract.e. Previously this situation was the same at the neighbouring minimum. and thus obtain ip"{z) + [A + 2h2 cos 1z\i>{z) = 0 (20. Thus one chooses not the solution with real argument at the minimum. We can distort the potential there or alternatively impose a different behaviour there on the wave function. the degeneracy of eigenvalues of neighbouring oscillators being lifted by the finite height of the barrier separating them. Here we choose t h e second way.20b) with potential V(z) = — 2h2 cos 2z as shown in Fig. . as we found these for this potential. 8 the equation can be approximated by d2ip(w) (20. and allows its imaginary part to tend to infinity. Around z ~ 0 w e can expand and obtain i)"{z) + [A + 2h2 . we shift the argument by n/2.
23) . 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. 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. (20. 17. . i. Thus we write the boundary condition at Kz = ir: tp(±iw) — 0 for > w — 2h}/2z K z— > i. we naturally constructed the even and odd solutions with these (cf. A derivation is given subsequently.24) Considering t m O i n Eq.e. We consider here explicitly only the case of the ground state around z = 0. whereas here we start off with a solution even or odd about a minimum. the even or symmetric solution about z = 0 can be taken as *• = liDfrriVhWz) +Dh{q_qo)(2h^z)}.> 0 for z > ±ioo. (where for the ground state q = go — 1) D_i{q+1)(±i2h1/2z) ' ± zoo (20. (20. . 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. so that the deviation q — qo results from the finite height of the barrier taken into account with boundary conditions at the next minimum. by a different method. whereas here we construct these from oscillatortype solutions. considering around z = 0 the ground state of the harmonic oscillator for which go = 15 and replacing go by g. 2. . we see that ips ~ cos \j2qhz for z ~ 0.e.3 Cosine Potential: Complex Eigenvalues 481 As explained above.20. A (which we can loosely dub solutions of WKBtype). . we now choose an harmonicoscillator type of real solution around the left minimum of Fig. Thus. This somewhat abstract looking boundary condition is effectively simply one imposed on a complex solution of the equation giving the oscillator eigenfunctions.22).63). .4. however. 3 . Since the large/i2 solutions valid away from the minima were the solutions of types A. 20. 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. that at z = 7r. such that this deviation becomes imaginary. Eq.
(20.w) = e T Dv{w) 2TT n») AvViDv^iw).2 that one parabolic cylinder function is exponentially increasing in h and the other exponentially decreasing.26) Thus here we can approximate ips by expanding it about E2 i)s~DQ{2hll2z) + E2 ~{DE2/4h(2h ^ z) 1 2 + DE2/. tp(v). 92. The arguments of the functions appearing in Eq.24) are linearly independent around z = 0 for q ^ qo.29) Equation (20. p. or logarithmic derivative of T(^) for which:§ ^(1/) r» (20. W. p.27) requires the differentiation of this expression with respect to the index v. we know from the solutions of types B and C in Sec.27) The differentiation of the parabolic cylinder functions with respect to their indices is nontrivial. 17. However. this can be handled in a few lines.25) qo) E2 0: (20. §See e. As before we set A = 2hz + 2hqA where in the present case Ei = 2hz + 2hq0 + — and A & = Ex + E2 (20.30) 'See e. From Tables of Special Functions'. 2.482 CHAPTER 20.3.(. Magnus and F. Oberhettinger [181]. The differentiated function Y{y) is in Tables of Functions expressed in terms of the socalled psi function. Oberhettinger [181]. Thus the only nonvanishing contribution comes from differentiation of Y[—v).h(2h1/2 J £2=0 (20. (20. Magnus and F. Large Order Behaviour of Perturbation Expansions Although the two parabolic cylinder functions in Eq.g. W.g. and then evaluation at u = 0. . Setting v = 0 in T(—v) = (—v— 1)! gives infinity.we obtain the following circuit formula of parabolic cylinder functions Dv(w): Du{w) T DV{.•w) + 2ir T{u) <"+!)% Dvi(iw). (20.28) Extracting Du{—w) we have A. (20.27) differ by an angle IT. We see this as follows. 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.
)e" w /4 2 l ^Dx(iw) (20.e.27) exponentially decreasing contributions. . we obtain the formulas (X) / \ 1 OO I » J ] (l+)e. Oberhettinger [181].29) v=0 (in + In w)D0(w) + v2ire (m + lnu. 1E2 + ± 0(e~hz2) 2 Ah Ihir ehz* E2V'< 16/i2 V2^ehz2~ 2hM2z (20. . those of type A. (20. pp. Magnus and F. Thus calculations for qo ^ 1 are not given by M.34) This is the behaviour of the symmetric ground state solution away from the origin towards the limit of its domain of validity. W. . .3 Cosine Potential: Complex Eigenvalues From formulas given in the literature we deduce the limiting behaviour^ 1.29). We now go to the WKBtype solutions. This result is » not easily generalized to <jo > 1. (20. dropping from the derivative part of Eq.31) dT(u) r*(v) dv 1 i/>(v) ^ o l.20.32) which we can use here since all other problems have been resolved: d D {w) dv v (20. "See e.33) = IV2~TT. i. which are valid around z ~ 7r/2. Stone and J. Reeve [262]. 3 .e. we obtain 4>s D0(2h^z) D0(2h1/2. 5 7 7 . (20. and "Prom W.+ v ^ v k^x k ^ + ^ where C = Euler's constant = 0 . Oberhettinger [181].31). i/^0 483 T(v) It follows that d dv [T(u)\ (20. including the derivative of DE2ufl(2h1/2z)). Differentiation of this equation and subsequently setting v — 0 yields. p. with the following expression for the largez behaviour'! (here v = E2/AK) Dv{w) ~ wue~w2/4 = evlriWe~w / 4 for  arg w\ < ^TT.4.g. 92.//n and ij>{y) = C . r(i/) We now return to Eq. For u — 0 one obtains the result (20.+w2/i IW Inserting this result into ips and retaining only the dominant contributions (i. Magnus and F..
We observe that in the matching domain (i.cos z — 2/icos z ipA^ Tcos  and i n A —..39) Now.e. Eqs. c2 the WKBlike linear combination e2hcos ^ W K B = ClipA + C2lpA ~ Ci z _—2/icos z — + C2 rT (20.29) and (17. 92) we take the formula: Dv{w) = T { y ^ \eiv^l2D^v^{iw) + e'^D^iw)] .36) we can identify the dominant large/i2 behaviour of the constant C\ as c\ = e~2h. (20.34).41) 2TT . Then V'WKB becomes (with cos z ~ — 1 + z' /2.e. (20. Thus in the present case and with q ~ qo = 1 we obtain „2ft. y COS c\ J (20. near z = n the boundary condition (20. (17. We now return to the WKBlike solution (20. cos z/2 ~ (20. (20.484 CHAPTER 20.40) From Magnus and Oberhettinger [181] (p. we obtain the required solutions from the earlier ones. (20. x .38) in its limiting domain towards 2 = 0. (20.36) cos  . smf (20.e.23) demands (with constant A) ip = AD_1(±i2h1/2z') > 0 for z' + ±ioo. that in approaching z = ir. where we can then match these to the solution ips.sm I We now want to match this solution to that of Eq. Thus the WKBlike solution becomes p — 2/lCOS 2 V'WKB ~ D0(2h1/2z) D0{2h ' z) l2 + c2 + c2 ^Tsinf 2/i hz2 _z_ 2 (20. Recalling that we have shifted the potential by 7r/2. i.40). Large Order Behaviour of Perturbation Expansions extend these to their limit of validity to the left.37) Comparing Eqs. 2 c2e2hehz UJ2 . with expansion of cos z in the exponential) D0{2hll2z) = e~hz ~ e~2h / „2/i cos z \ — ).34) and (20.35) We construct with constants c\. by there replacing z by z + ir/2.— 721 . To this end we set z' = z — ix.36) and consider this at the other end of its domain of validity. i.36) V'WKB cie~2hehz' 9h  .
34) we obtain E2 = i%\ = ±ih22Qe~8h. i. ^'mliw <2a47) For qo = 1 we recover the result (20. A = ±i4/l1/2e4h ^ t c2 = T i2(2/ivr) 1 /2 e 6/ l . . 1 2 4 1 2 8h ^^2 2 (20.complex conjugate] (2Q42) (2032) y /  Z e _ ^ + 2(. (20.39) and (20. J Hence Eq.^.1/V)_le^_ Z With this result Eq. (20. (20.e.20. with c\ = exp(—2h) from above.45) Comparing this result with tps of Eq. (20. (20.46). and hence (adding and subtracting equal terms) Di(i2hV2z') = ^[D^h^z') + D_1{i2h1/2z')} Z + hD_1(i2hl''1z') = \V2^D0(2h^2z') Z D_l{i2hll2z1)] D^(i2hl/2z')} + hDiWh^z1) + [D^i(i2hl/2z') Z ^D0(2hl/2z') Z .38) can be written TAWKB (20.3 Cosine Potential: Complex Eigenvalues Thus for v = 0 we have DQ(w) = —j=[D\(iw) V 27T 485 + Di(iw)] = real.40) can be written ^ ~ AJ^e~hz'2 + A~(i2hl^2zTlehz'2. (20. We derive the general result in the next subsection using our own method.43) we obtain. for both even and odd harmonic oscillator wave functions around z — 0 with qo > 1) is nontrivial.46) Repeating these calculations for the general case (i.44) = DQ{2h / z) ± (2h7r) / e.e.43) Comparing now Eqs.2/. The general result stated by Stone and Reeve [262] is therefore presumably guessed.
20. i.3. Considering the original equation. MiillerKirsten and A. i.z. J.486 20. i.5.e.46) we consider a half period of the cosine potential as illustrated in Fig. At the minimum to the right of the barrier shown in Fig. •_(:)*<>.()#0. (20. V(z) . On the other side of the barrier at z = —TT/2 we consider different boundary conditions. We have to impose two sets of boundary conditions.e.*. 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.48) These conditions provide the required quantum number qo = 2n + 1.e. and expanding cos 2z around the point z = — TT/2.2 CHAPTER 20.5 The cosine potential from — TT/2 to n/2.20a). at z = TT/2. Large Order Behaviour of Perturbation Decaying excited states Expansions For our derivation of the generalization** of Eq. we impose the usual type of boundary conditions as for the minimum of a simple harmonic oscillator there. (20. Fig. H. (20. Achuthan.5.Ah2 ( z + . Eq. .) + 7T V> = o. Wiedemann [4]. this equation may be approximated there as d2ip ~dz^ + E + 2h2 . the result is formula (334) there. 20. •+(=)* V(f) 0. 20.E curre it \ j. W.
(20.21) into the Mathieu equation we obtain. (20.r/2 ~ e i(2hq) V 2 .Hz^+Ol^ Bq[w(z)}+0 h1'2 (20. i.z).e. Recapitulating these from Chapter 17.51a) where H Bq[w(z)} w(z) = h(«v (w) Bq[w(z)} = Bq[w(z)}.50b) which are valid as asymptotically decreasing expansions in the domain away from a minimum.21)).h. Proceeding as in our previous cases and substituting the expression (20. (A) Boundary conditions at the right minimum We begin with the evaluation of the boundary conditions at z = 7r/2. For the continuation we use again the WKB formalism.h. we demand that ^(*)L<. we have IPA(Z) „2hsinz Aq{z) + O il>A(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. for COS I Z ± 7T 1 1 2 4 >o^\ The second pair of solutions is ipB(z) J2hsmz B.50a) V'AfeM) = where (the second expression is needed for later comparison) Aq(z) cos§(9i)(iz + lyr) + W) [ t a n (±z + W)]«/ 2 [sin(i* + ±TT) C O S ( ^ + ±vr)] V' ' sinfr+Vaz (20. B and C. 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. three pairs of solutions of types named A. (20.' 4 / 1 1 / 2 COS ( Z + 7T \2 4 .z)=tpA(q.51b) 2i(«1)[i(9l)]'. (20. as shown earlier.3 Cosine Potential: Complex Eigenvalues 487 (using Eq.20.
48) and proceeding as in Chapter 17. Large Order Behaviour of Perturbation Expansions the function H!/{w) being a Hermite function. we obtain 1>±W = \ a a (20. We return to this condition after derivation of the .55) Extending these solutions to the domain around the minimum at z = 7r/2. the relation q . \vw).54) The even and odd solutions are defined as iP±(z) = tyA(z)±iPA(z)}.52a) = Cq[w(z)}. 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. and the second around z = 7r/2. 5. Here the upper sign refers to the even states and the lower to the odd states. ^ W H . i.52b) The first solution is valid asymptotically around z = —vr/2.^ i .53) where a (8M («!)]« 1 + 0 { \ a = 1 (8/i) K9+ ) l + O (20. (20.56) Applying the boundary conditions (20. 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 ) .488 CHAPTER 20. (20.qo . (20.=F2 2\1/2{28h2y/4e~Ah l + O (20. 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) •((73) 1 h1/2 1 hW> (20.57) where go = 1) 3. we obtain our first transcendental equation.e.
These then provide complex exponentials like (20.58) Ignoring the correction A / 8 . Jz+ ' T In principle the integral could be evaluated exactly — see P.3 Cosine Potential: Complex Eigenvalues 489 second transcendental equation from the boundary condition at the other minimum.e.2hq] / dz) We know that the solutions I/JA{Z). (20.67). (B) Boundary condition at the left minimum The solutions of type A have been matched to the oscillatortype solutions at the right minimum.49) and we can apply our boundary condition.Ah2 cos2 z ~ 0.[ [Ah cos z . as we shall see with the result of Eq. 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. The solutions of type A are valid to the right as well as to the left of the maximum of the barrier. p. i. Inserting the approximate form of the eigenvalue into the Schrodinger equation we obtain dz2 + 2hq+ — . we see that the turning points at z± are given by 2hq . (20.5 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 .20.01. and we obtain the proportionality by appropriate expansion. D. Some nontrivial manipulations are again required to uncover the generation of important factorials.Ah2 cos2 z ip = 0. e x p ( . Friedman [40]. 163 — but here we are interested in the WKB solution in a domain where it is proportional to a solution of type A.61) [Ah2 cos2 z2hq]1/2dz. r ywKBlZJ ~ ~ — [4/i2cos2 2/i ]i/4 2 9 z 2 2 e x p ( £ [Ah2 cos2 z  2hq]ll2dz) ' 1 2 . Byrd and M. Thereafter we extend the latter across the turning point down to periodic WKB solutions. C O S 1/2 ^" (l) ± ' 2<2+<2 (2 °. formula 280. Our procedure now is to match these to exponential WKB solutions. . F. I^A(Z) overlap parts of the domains of validity of these solutions.
» _ i y v . using the relation 1 + sin z = 2sin2(2i/2 + 7r/4). (20.) +0 z+ IT Inserting this into the integral I.64) . 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 (20.62): 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 (20.490 CHAPTER 20. 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 /•«"(.cos2 ( z+^) = (z+ .62) as follows. Large Order Behaviour of Perturbation Expansions Since cos(z + n/2) = — sin z. We have for the following expression contained in (20. we write cos2 z = l .62) where we set — in accordance with our expansion *+ + 2 IT 1/2 <? + (20. 2/i In the same spirit we can set h[z + l 2h 1 — cosi I z + — 7r 2/i[l + sin z\.sin2 z = l .59) 2h COS Z+  ~ 0. (20.63) We handle the logarithmic terms in Eq.
(4^2)i/4 [ i( g _3)] ! V/ l (20. on using in the second step Eqs.66) The WKB solution therefore becomes (approximating the denominator of (20.65) [Z[4h 2cos2z2hq]1/2d< e2he2hsin ^eg/4(g/2/t)g/4 2<?[tan(§ + f )}i/2 (20. ./ * [Ah2 cos2 z IZ+ 2hq\l/2dz) + f )]«/ 2 (4/i 2 COS2 z)V4 e2he2hsm zeq/4(q/2h)i/4[tan(% /2^~Q/4 (4/i 2 )V429{2sin(f + f ) c o s ( f + f)}V2 ^27*291 (4/l )V4 e ?/4( g / 4 ) 9 /4 ^ J e" ^^(2TT) /2 2 1 2 ^ ) / 4 /2\"  . . (20. <$&?(*) e2he2hsinzei/4(q/2h)i/4[tan(% + l)]'q/2 2?(4/i 2 ) 1 /4{2sin(f + f)cos(f + f ) } 1 / 2 e2heq/A(q/4y/4 /2\q/4 . but mention that for q an odd integer: j(9D l + O and 1 r(93) 4(9+1) ! = ±7I .^ Proceeding similarly with the other WKB solution we obtain rWKB (z) ~ e x p ( .20. . we obtain I ~ 2h + 2hsm z Hence exp 491 V^m 2h\~'I z TT 4 — tan .3 Cosine Potential: Complex Eigenvalues Inserting the relations (20.50b).63) and (20. 2*4(4tf)V« W ^ e[l( V D]! 1 /2y^ (20.62). We ignore this here.60) by (4h2 cos2 z)1!4).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. (20. 1/2 (20.64) into Eq.67) (27r) /22«+5(4/i2)V4V^y The last expression is again obtained with the help of Stirling's approximation.+ qj V2 4 A.
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+. Eqs. io(8/*)g/V^i6ft2)V4[±(g3)]! .T 7 77 r )V2 /4(2 .69) l(8M9/4e2^ \2 [ i ( g .492 CHAPTER 20.50)).e 2[J(«l)]l[j(?3)]! i l[(g3)]!e (20. (14. i 27r(8^/ 4 2/t /4 ' r(±) . i. (20. S1H (2/ KZ ) / (z f 2 (2^).67) and (20.i ( 2 / i g ) 1 / 2 ( z + .?/ 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.+) + B^) HAh^/\ir.1 ) ] ! 1 (8/Qg/ e4 2fe l[(g3)]!e^ .z+)/ VWKBW 2\2j (2^72 l(8^/4e2fe(27r)i/22V2(4fr2)V4 2 < nz+) ^WKBW [\{ql)]\ 1 (8/i). we obtain l(8/i)g/ 4 e. where _ 1 + [±) *) + £ (20.U ^K^.l^l !T ^^* T2 7 r ( ^^ /. We now return to the even and odd wave functions defined by Eq.KW4^8 4 .2 f e (27r) 1 / 2 (16/t 2 ) 1 / 4 2 (2/wz)V4[i(9_i)]! cos (2M1/2(^^)+4 7T .49).—^ (2^)^(16^2)1/4^ ^ _ _ [T+(±)exp{i(2/ig)1/2(z+_z)} 2(2/ig)V4 + r _ ( ± ) e x p { . Thus with IPA(Z) and TPA(Z) taken from Eqs.68) we obtain ^±(z) = obpA{z)±t/JA{z) l/^y/4e2fe(27r)1/22^^(4/i2)1/4 1 « llfh\^e^[l(q3)]\2^ ± /(r. (20. 1 2 1 ±2 . (14.2 [ ? ) Imposing the boundary condition (20.™) . Large Order Behaviour of Perturbation Expansions again using the Stirling formula.49)..z)}]. we obtain T _ ( ± ) = 0.
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. Eqs.57) by the left hand side of Eq. Eqs. We now have the two equations which have to be satisfied. (20.qo) El Ah' (20.20.h)+2h(qq0) (16h)i°+le'Sh E(qQ.72) we obtain qqo q0fiSh 2(16/i)*>e {[!(<?<)I)]!} 2 1+ 0 (20. the result (20. h) of Eq.h) + i l + O '4{[^ol)]!}2 E(q0.21) expanded about q = qo. (20.72) We observe that with the square root as on the right hand side both sides are complex. (20.25)) E = A = Ei + E2. In both cases we wrote (cf. (20.3 Cosine Potential: Complex Eigenvalues Using the duplication formula \{1l) this becomes 7rV 493 \ ^ .47). Combining these equations by replacing (28/i2)<?/4 in Eq. for q an integer both sides are imaginary.h) + (qq0)( BE — 90 E{qQ. (20.57) and (20. (20. 20. (17.3 ) I = }(»« !7rl/22I(.73) Inserting this expression into A = E(q.4h (16 2 /» 2 )9/ 4 (91)]! = (16 2 /i 2 )«/ 4 = (16/i) 9 / 2 .26). Equation (20.h) = = E(q0.l)j /2 e4fe(162/t2)^4_±l or ±i 7r\ 1/2 (16 2 /i 2 )9/ 4 e.72).75) .3. in fact. i.e. we obtain E(q.h)+i%Ego. (20. who derived it for qo = 1 and guessed the form for general values of qo. Ex = 2h2 + 2hq0 + A 1 (<? .74) This result agrees with that of Stone and Reeve [262].72) is our second transcendental equation.
W [£(901)]! < (20. i.* The existence of a formula of this type was conjectured without an explicit derivation like that given here and only for the ground state. Achuthan. Bogomol'nyi and V. MiillerKirsten and A. we recall that we obtained there for the difference 25Eqo between levels with the same oscillator quantum number qo. Setting ng AEqo:=2iSfEqo. by others.80) *P. cf. f E . For definiteness we recall the appropriate expansions together with the definition of coefficients Aj. The same formula can be derived in the case of the double well potential. J. we can obtain the imaginary part calculated above from the formula 4 (H) A ^ o = 27ri(<5£90)2. we can recalculate with the help of Eq. (20.3.I g  ± l ( 2 „. (17.76) he Thus SEqo is the deviation of one of these levels from the harmonic oscillator of level. the discontinuity across its cut from zero to infinity. Fateyev [26]. We had A= or _2ft2 + 2ft. Eq. B. Wiedemann [4] and Example 20.2 below. 7 8 ) i=l V .e. (20. its tunneling properties and the large order behaviour.4 R e c a l c u l a t i o n of large order behaviour Now that we have obtained QE.+  . * It would be interesting to see a derivation of the formula — reminiscent of an optical theorem — from first principles.) {I6h)™/2e4h 1 + . i=l We can calculate Ai for large i from the relation Ai = * Jo 1 P°° ti^QE^dti (20.9) the coefficients Ai of its perturbation expansion. .74b).77) One can say that this simple relation summarizes the intricate connections between the discontinuity of the eigenenergy across its cut. H. A. A = . W. Large Order Behaviour of Perturbation Expansions Referring back to the calculation of the level splitting with selfadjoint boundary conditions.494 CHAPTER 20. 25Eqo=8h[ 2\1/2 . 20.
13) out again: jPiiQj) = (g + 4 j .14). MullerKirsten [200]. f H . i ^ = ^ + E?27MT E ^ifoj^W (2082) The coefficients Pi(q. (20. J). (20.l ) +2[(q + 4j)2 + l + A}Pi_1(q. J..19) for the large values of i under consideration here.81) We observe that this result agrees with the result of Eq. there is nothing special about this case. the integration is seen to give (20. (20.20. *R. But as Dingle remarks. so that an exact Borel summation is not possible.j) + {q + 4j + l)(q + 4j + 3)Pi1(q. W. (20. Using again n = (go .4 Cosine Potential: Another Method of Calculation 495 with $$E(h) given by Eq. B. The integral is recognized to be of the type of the integral representation of the gamma function.l ) ( < ? + 4 j . Dingle [74]. We also observe that these terms do not alternate in sign.74).j + 1).3 ) P i _ 1 ( c ? . For convenience we write Eq.j) satisfy the recurrence relation (20.t We consider again the periodic cosine potential with (unnormalized) eigenfunction expansion oo .83) One now sets ftH = ( 2 % P * ( 9 . 20. and the method can be applied in many other cases. (20.13) and the coefficients M2i or M2i+i of the eigenvalue expansion can be expressed in terms of these as in Eq. j .1).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. a = J + \<1> (2084) . We sketch this method here slightly modified and extend it to the final formula for comparison with our previous results.
496 CHAPTER 20.83) can be written j .z)WaA/2(z) Setting V.(a) oc . we regain Eq. A. 1  1 / 2 (4 (20. Pi(a) i.h). formula 13. (20.e. i.88) satisfy among other relations the following recurrence relation which is of significance in the present contextt {2Kz)WK>li{z) + WK+1. (with a = j + g/2 in the coefficients) 8/? —Pi(a) = P i _x(a . (q + 4j) 2 ~ 4j(4j + 2g).90) See M. p(a.g. i.85) Considering now the generating function of the perturbation coefficients.1) + 2 P i _i(a) + Pi_i(a + 1). (20.89) Replacing here K by a and \x by 1/2.4.l/2l Wa. .86) one can convince oneself that Eq.l.h) = ^2pi(a) with p.91) + Wa+lil/2(z) = (1 .31.e. we obtain (2a .ll{z)= L . Abramowitz and I. z2 (20. 4j(4j + 2g + 4) + / . 0 4j(4j + 2g) pi_i(a . Now.^(Z).1) + 2 (a — 1) a . which are solutions of the differential equation known as Whittaker's equation.87) Taking terms of 0(h~l+1) of this equation.a R . nx pi_i(a + l).1/2(z) (a1)! (20.4 ) ^ . d2WKh dz2 + 1 4 K z zM2n WKtll = 0. (20.85) implies p(a . Taking dominant terms. (20. the recurrence relation (20. Whittaker functions WKjfl. a.e.27h pi(a) a = 4j(4j + 2 g . (20.85).h) 2+ 8h a p(a.K + ^ ] L + K^)WK.h) +p(a + l. e. g < 4 j . Stegun [1]. Large Order Behaviour of Perturbation Expansions and considers large values of j for large values of i. (20.
94) Hence to insure that p(a.(« .86) have to be chosen such that p(a.(« .87).4 Cosine Potential: Another Method of Calculation we obtain Vai.e. > > ^See W.g(h) are functions of h which in view of Eq. Oberhettinger [181]. Thus the approximate largez asymptotic behaviour is WKtp(z) ~ exp •FH V dz z e ^' 2 exp(/clnz). h —> —h.93) (a1)! ' where f(h). (20. (20. The more precise form obtained from Tables of Special Functions' is w*A* z 2 e / z" n=l [f . Dingle discovered the solutions^ p(a. (20. ^Observe that if one solution of Whittaker's equation (20. v .84)) {8h)q/2( Ah for j < 0.. Magnus and F . i.„ +1)2] 7 n\z \n which with K = a and /z = 1/2 is: WAz)~e^z«Y. (20. another solution is obtained from this by changing the signs of K and z since this leaves the equation unchanged. 89. . h) is expressed in descending powers of z = 8/i.j — —j.88) which reminds us immediately of a radial Schrodinger equation with Coulomb potential.20.Tl ss\{a)\{al)\ (z) s=0 s — a)\(s — a — 1)! for z — oo. > (20. g(h)(20.92) with Eq.92) Comparing Eq.h) = \~\Pi(&) with pi oc h l. such that p(a.i/2(z) + Va+lA/2(z) = 497 2+ Va.h) ~/(fe) ' (a1)! . h) is expressed in descending powers of h.\)V .88) is known. p. In the present case this means the replacements q — —q.(« 1) 2 ]. a (20. We can infer the approximate behaviour of the Whittaker function for large values of z from the differential equation (20.i/2(z). we have to remove a factor (recalling a = j + (7/2 of Eq.
(20. h\ oc (8h)q/2e~4h ( _ . n = 0..94) with s — s — j : > x = J + y u ( 1)J ~ f (*+&!(*+*?pi i_ Using the reflection formula (for q = 2n + 1. This expression can be obtained from the recurrence relation (20.1/2(8^) ( . W. i).. Dingle and H. (20. this becomes / 1 \ (1)5'" ^ U + iq)<(s + iq1)\ 1 Picking out the term in hrl. g — —q. (20. B. Muller [74].95a) Inserting (20. x 4i[2» + i(« — 1)1! il! [ ^ ( 9 . we obtain the result of Dingle: (20. ^ ..l ) ] ! 11 R.96) and a similar expression for j < 0 with /i — —/i.1. _ i s.j _ g / 2  1 / 2 ( . .8 / t ) for j > 0. Comparing > > > this result now with Eq.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.+ i f or 3 < 0 (20. j — —j. Correspondingly we set for j > 0 / 1 A pla = j + q.).498 CHAPTER 20.94) here we have a series in descending powers of h. J.95b) Then for j > 0 we obtain from Eq.83) and is given by 1 1 . Large Order Behaviour of Perturbation Expansions and write / i \ pl<x = j+ 2<7>h) « (8h)~q/2e4h v ^+9/2. 2.
99) In the next step we use for both factorials [2i + ^(q — 1)]! in Eq. W. 5. (20. p. .98) the duplication formula 22zl / y P*l>'7J<*l)'(»j»and obtain Mi% + \{q^)W + \{q.l ] ! {[.97a) we obtain (observe that as required this expression vanishes for j > i) Pi(q.l ) ] ! 2 M [i + \q)\ i\\l{ql)]\(J + \q) [j + lq]\[ij]\ (20. M.14).97d) Then the coefficient M<n of the eigenvalue expansion.100) **R.J)} 2 i=i 2 [2i+l(ql)]!22il2 i![i(gl)]! i (20.l ] ! (20. A special case of the formula can be found in I.{j + \<l).j) (i + lg)[2i + i ( g . Ryzhik [122]. B.20.+i?_l]!}2 (20. (20. formula (30).3)]!2 2 '+^. note the misprint there: (r — s) ought to be (r — s)\.e. E* J'=I i+M 2 [2i + 9 . l{Q. S. of Eq.157(4).1 )/ 2 2 2i ' ^i\{\{ql)]\ [2i + g . Miiller [74].0) = 1 as desired (although not really enforcible for large values of i). Dingle and H. (20.Then [2i + ±(ql)}\2* Pi{qJ) [i + \q]\ (20. Thus ° 9 (i + \q) 2«(i+Igl)! p .l) (i+y) 2*%(ql)}\ [2i + §(gl)]l i\[i+\ql]\ ' Inserting this expression into Eq.4 Cosine Potential: Another Method of Calculation 499 We observe that Po(q.98) We evaluate the sum by approximating this to a formula given in the literature ** r \ f R\ __{r + Rl)\ J (rl)\(Rl)\' E j=0 i. is M2l = 2j2j[Pi(<l.97c) We assume that in a leading approximation we can cancel the factors (i + \q). Gradshteyn and I. formula 0. i.e. J.
Eq.e. i\ M i large 1 .l.17).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.l l ! with the large i behaviour in agreement with the result (20.o i • .e. [^ + i ( g . We refer back to Eq.e.. i. i.81) for q — qo = 2 n + l .34) that £ is an expansion in ascending powers of h 6 /c 2 .Am i . (18.3 ) ] ! _ 1 Hence M or. (18.4)) £ = ^ = ^+ w and y= ^ > ( 20 .' 1 i • (20. Eq. (18.5 20.101) This formula allows us to cancel all factorials [i.500 CHAPTER 20.e..86) and set (cf. the result of Bender and Wu [18].i ) ] ! [ ^ +  ( g .105 ) the latter because we see from the eigenvalue expansion (18. replacing 2i by i: 2 ^2 8 ^ % + *"*}' (20. n = 0..e. 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. [19]... The imaginary part we obtained is interpreted as effectively the discontinuity of the eigenvalue across its cut from h2 = 0 to infinity.2. i.1. i.. since \{q — l) + \{q — 3) = \q. Large Order Behaviour of Perturbation Expansions In the following step we use the approximate formula (20. . 20.5.]\.102) Thus the result is M m 1 2«Fi + g . i.86).
and hence inserting here the imaginary part from Eq.l ) r .5. 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.5 Anharmonic Oscillator The Borel sum can now be formally written 501 (20.(20.e. and then the relation (20. We leave this to the following Example. (18. 20. h2 _ h222K+llHr+K+ll2{l)r+1 r+ K . (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. This can then be compared with the level splitting we obtained for this potential.106) With £= E r=0 Mr (20.107) we obtain (i)r r 71" JO yirZ£{y')dy'. . i.77) can be verified. the discontinuity across its cut. in terms of K where go = 2ftT + 1. (20.2 The double well It is clear that one can now proceed and apply a nonselfadjoint boundary condition also in the case of the double well potential and thus obtain the imaginary part of the eigenvalue.20.110) This is the result in agreement with the large order formula (4.86). i.109) or.4) of Bender and Wu [19].e. one obtains the result Mr 2io+hr+1+f (l)r[r + f ] ! 7T3/2[1((?01)]! (20.108) Proceeding in this way. in fact as one with Borel summability in view of the factor ( . The result establishes the series unequivocally as an asymptotic series.
T. Wiedemann [4]. A relation similar t o (20. Harris. Silverstone [62]. MiillerKirsten and A. An approach to arrive at the relation via instanton considerations has been examined by ZinnJustin [292].Q . Damburg. J. S. . + 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. 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). J. Wiedemann [4]. K. [19] can be traced back from articles in the Proc. R. 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. Silverstone [52]. R. V. V. J . H.* A more recent status evaluation is provided by the book of LeGuillou and ZinnJustin [161]. R. . (357) to (371). *R. J. J. MiillerKirsten and A. J. G. M. Achuthan.77) was conjectured without explicit derivation and only for the ground state by Bogomol'nyi and Fateyev [26]. The relation (20. Damburg. H. K. of International Workshop on Perturbation Theory at Large Order (Florida. Large Order Behaviour of Perturbation Expansions Example 20. MiillerKirsten [164]. J. 1982) [139]. M. E. J. W. it can be used to obtain the other. Propin. *Some of the literature in this field following the work of C. Harrell. Graffi. Nakai. 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.77) and its possible generality has been referred to by Brezin and ZinnJustin [36]. Harrell. S. as in later chapters. S. M. Solution: For the solution we refer to the literature. t p . Grecchi. Gram. Wu [18]. Eqs. Cizek. J.^ 20. W. Grecchi. Its usefulness has also been demonstrated there. J.6 General Remarks The study of the large order behaviour of perturbation theory has become an individual direction of research.77) has been obtained in the Schrodinger theory of the molecular hydrogen ion H^ . W. in t h a t when the quantity on one side is known.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. Propin and H. " P . Liang and H. Paldus. J. Cizek. t The relation (20.502 CHAPTER 20. Bender and T. J. Paldus and H. E. Achuthan.
since in such cases the probability of this system choosing a path far away may be considered to be small. Thus we are in particular interested in rederiving the level splitting formulas obtained in earlier chapters with this method.Chapter 21 The P a t h Integral Formalism 21. A standard reference is the book of R. which explains also difficulties. P. Kleinert [151]. unknown). Chapters 2 and 5. ^ Our interest here focusses on quantum mechanics. Feynman [94]. Felsager [91].1 Introductory Remarks In this chapter we introduce the path integral formalism developed by Feynman. The path integral is. P. Hibbs [95]. Feynman [93]. S. The formalism is not so useful for boundstate problems (in this case the initial wave function is.* 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. Probably the most readable introduction. 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. 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. However. however. *R.e. A brief introduction is also contained in R. This formalism is particularly useful for evaluating scattering or transition amplitudes. P. in general. 503 . Feynman and A. Coulomb problem) we refer to existing literature. A further wellknown reference is the book of L. R. 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. Schulman [245]. We illustrate the method here by rederiving the Rutherford scattering formula with this method. For the application of path integrals to the hydrogen atom (i. particularly useful in cases where the behaviour of a system is to be investigated close to its classical path. However.
*In Dirac's notation we write this equation later (xf. The Lagrangian L of a particle of mass mo moving in a potential V(x) in onetime and onespace dimensions is given by L = TV= m0x2V(x). i.x fc _i) 2 2e2 V{xk) Thus: Given xk(tk). (21. the initial time. 2mo dx2 (21.1) Let the wave function i/> at time t = ti.504 CHAPTER 21. =J dx(xf.tf) = / dxK(xf. (21. we wish to know* ip(xf.e—»0 Sn.ti).tf\ip) . ti).ti)ip(x.x). We are interested to know the wave function tp at a final time tf > ti and position coordinate x = Xf.e. The Feynman Path Integral Formalism 21.ti\tp). (21.ti){x. Obviously K(xf. we can compute the action S1.tf\x.x.3) We divide the time interval tf — ti into n + 1 equal infinitesimal elements e such that £() ti = = hi ti + e.5) Sn = X / fc=l mo (zfe .2) where K is a Green's function. */ = U + (n + l)e.t).4) so that the timesliced action S becomes rtf «+i S = / 'ti n+1 dtL = d£L = n—»oo. ti) = 5(xf .e—»C lim V e m0xk fc=l  V(xk) = lim n—>oo. (21. U. be ip{x.2 P a t h Integrals and Green's Functions The onedimensional 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.tf. x.
as will be shown to be possible below. The repetition of the quadratic factors in the argument of the exponential with indices differing by 1 suggests the construction of a recurrence relation. 21. of which one is illustrated in Fig.ti) to its final position at (xf.6a) .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. The usefulness of the method therefore depends on the fact that some paths are more probable than others. The multiple integral involves individual integrals like f dxk exp Xk)2 + {xk Zfci) 2 } with Xk contained in the step up to this point and in the step away from it. 21. 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. Here and in the following we need one or the other of the following integrals (cf.21. Thus there is a countless number of possible paths that the system may choose in propagating from its initial position at (xi.1 A path in (x.2 The Path Integral and Green's Functions 505 Above we considered time intervals of length e.tf).t). the Fresnel integral (13.28)): f dr\ exp ia a) ' (21.
ie f me\ 2a\a~) 1/2 (21. (21. one has j ^ dr1eifa+i^ /£ = .6c) In the evaluation of these integrals we replace a by a + ifi..s f c _i) + (xk+i .8) Note t h a t in spite of the n integrations..mQf Xk L ~2e^ ^ . tf > U.x f c _ i ) 5 \m0 J (21.ir 1/2 £dxeXp[i^{2(3 dxexp\i—<2lx+ 72 iV(e) exp Z7TT.xk_i) dx exp i — { 2 a . dxneiSn/n.506 CHAPTER 21. 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. 2 + 2a.z f c _i) 2 } } .xk+x oo —oo oo .. dx exp z—{a.(x fc+ i .2 + (x + x f c + 1 . with f™ dze"™ 2 * 2 / 2 = s/2ir/w2.tf.xk+i) exp .t ~Xk+l> + xk+i 2 ) +{xk .Xi.37)) dxk exp d(xk .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 dx2.6b) —c ia o drjr] e x p—V € (21. (21.Zfci) 2 } I +{xk+i x*.ti) '•= f g m ) (N(€\]n+i dxL. We then integrate and finally let /x go to zero.5)): G(xf. One therefore removes the troublesome factor by defining (Sn being the sum in Eq. n > 0.* W i t h the first of these integrals we obtain (compare the result with the Green's function (7. for example.m0 .0 ^7 (x f c + i . The Feynman ia Path Integral Formalism f J—c J dr) r\ exp v = 0. J—oo dxneiS/h — • 0.
U) = / JXi=x(ti) = K(xf.tf.5). which all vary from —oo to oo. i. p. i. These integrations are indicated by the horizontal arrows in Fig. (21. 11 For discussions see e. B.7) for n — 1) to give a time interval. 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.26)) that — ignoring the potential — the multiple integral (21. 21.9) T>{x{t)}eiS/h = ^ww^IdxlIdXnelSn/h(2L10) This is the formula originally given by Feynman. it may be noted that with a different philosophy concerning the summation over paths. and so on.ti) for tf > U. Felsager [91]. (21.Xi. § See e. It is neither a variable of integration in T){x} —> Ylndxn. however prefixed with a new factor. we arrive at a path integral without the measure factor N(e) above. See the square root in front of the exponential in Eq. B. (21.Xi. in fact. (21. and will therefore not be attempted here. summation over all possible paths.24).g.21.10) can be looked at as the product of a succession of free particle Green's functions (or propagators).2 The Path Integral and Green's Functions 507 Eq. G(xf. We shall see later (with Eq.2). Different constructions of these families of paths usually lead to different measures.e.g. (21. A mathematically rigorous definition of the path integral is difficult and beyond the scope of our present aims. as in our discretization in Eq.Xi.§ The factor [iV(e)]_(n+1) associated with the multiple integration is described as its measure. nor the limit of integration of some variable Xi. (21.tf.' These factors N(e) were introduced to cancel corresponding factors arising from the Fresnel (or Gaussian) integration. p. (21.t In the following we demonstrate that the function constructed in this way is.tf. In the expression rx" / ?){x}eiI{^ Jx the quantity x" is only a symbol to indicate the endpoint of the path. the Green's function K of Eq. Thus one first integrates over x\ from — oo to oo and obtains an exponential involving (x2 — xo)2.1. However. . 183.ti) For tf > ti we now write* G(xf. 183.5). Felsager [91]. The additional factor is needed later to combine with the number n contained in the extra factor y/n + 1 (in Eq. Then one integrates over X2 from —oo to oo and obtains an exponential involving (x^ — XQ)2.e.
one with the full action S.508 CHAPTER 21.1 and allowing n to go to infinity.ti) = 1 j^p: f°° cte n +iexp } • fm0. Each of these linear pieces is completely characterized by its endpoints.74)). the other with that of the action of the free particle SQ.Then we can construct a recurrence relation as follows.exp 2e n ieV(xn+2) G(xn+2 + V.Xi. 187 . We next determine the equation satisfied by G in order to verify that this is indeed the Green's function for tf > t{.xi. Then the measure drops out. The Feynman Path Integral Formalism The factors N(e) can be avoided by instead summing over piecewise linear paths. and one is calculating the Feynman amplitude or kernel relative to that of the free theory. Then we have on the right the factor G{x+rj). so that the piecewise linear paths fill the entire space of paths. Performing the sum now in the sense of summing and hence integrating over the vertex coordinates in Fig. I We employ this method later in Example 23. pp.xi. we have See.ti). Consider a time t = tf + e for position Xf = xn+2. (23.188. we have G(xn+2. the discussion in B. we obtain the same as before except that there is no factor N(e).tf. (21. We have G(xn+2.Xi. 16 x2 Sn+l) \ 7T2~(Xn+2 V(xn+2) Setting now xn+\ — xn+2 = n. however.72) to (23. . One way to avoid this problem with the integration measure is to consider the ratio of two path integrals.11) For reasons of transparency in the next few steps we replace xn+2 by x and suppress temporarily tf.U) imo 2 G(xn+i W)Idr. Expanding this in a Taylor series around x. 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. Felsager [91].tf + e.ti. 21. It is for these reasons that — unless required — the measure is frequently ignored.tf + e.3 in the evaluation of I a specific path integral (see Eqs.
tf.ti) +V(x)G(x. Xi.tf.tf. Retaining only terms up to those linear in e.tf.xi.G(x.13) We integrate with the help of the integrals (21.tf.Xi.U + 0(e ). U) .tf.2 The Path Integral and Green's Functions so that G(x.Xi.U ieV(x) 2i7re\ 1 / 2 m0 ie d2 G(x.ti) 2 —l€ — ^2 + V(x)G(x.ti)\ . U) = e^G + 0(e 2 ) dtf 1 d2G(x.xi.U) +••• \ 509 G(x.tf''Xi.ti)ip(x.Xi.ti) = 1 d2 + V(x) G(x. It follows that ip(xf.ti) = j^rr dnexp I ^rj1 .Xi.ti) N(e) or G(x.tf.tf + e.15) We see that G(x.16) .tf + e. U) + .14) = G(x.xi.21.t Thus + 0(e2 (21.tf.6b) and (21.xi.tf.tf. ti) satisfies the timedependent Schrodinger equation.ieV(x) + ri< —G(x.tf.tf.6c).Xi. 2mo dx2 (21.ti) for tf > ti (21.tf)= dxG(xf. . d and this means d i—G(x.Xi.ti) +\if(^G{x. tf.xi. tf.6a). U) + Q(e2 = (lieV(x) + 0(e2)) G{x. we obtain: G(x.xi.xi.U). (21.ti) = G(x.ti) . tf.xi.ie 1 d2 2nV0dx^G{X. tf + e.— p G ( j .tf + e.Xi.Xi.Xi.(21.x.
We have therefore established that the Feynman integral (21.1 Configuration space representation In the case of V = 0 we have from the previous section for the action. 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. now called So.tf. The function G which satisfies the timedependent Schrodinger equation (21.Xi. (21. (21. (21. Eq. K0 = 9(tf . 21. or K(xf.510 CHAPTER 21.ti) = = 9{tfti)—G + G5(tfti) . (21. denoted by KQ.U)G0. as may be verified by inserting this into Eq. as this is also called.20) .U.1)). The Feynman Path Integral Formalism also satisfies the timedependent Schrodinger equation.tf.U). Hence our original function K is the same as G. Eq.x.*<)' ( 2 L 1 8 ) 0(tf ~ li)QfG + 5(Xf ~ X)5^f since G(xf.x.ti) = 5(xj — x). X U) = i6{Xf ' ~ X)6{tf " U)' (2L19) Thus K is the Green's function for the problem of the Schrodinger equation.3 The Green's Function for Potential V=0 We now calculate the Green's function K for potential V = 0. 21.U) = / T){x} eiS\ [fdt \mx2. In a problem where V is small it is convenient to calculate K with a perturbation method.x.x.U)6(tf .17) Differentiating K with respect to tf we obtain —K(xf. is the Fourier transform of the nonrelativistic propagator. and the Green's function KQ\ S0= Here (cf.U) = G(xf.15) is a wave function (cf. It follows therefore that {{W ~V+2m~0~^) K{Xh t/.16). The result.tf.10) is indeed to be identified with the timedependent Green's function.8)) Go(xf. (21.3.tf.
n + i .t) = ( m0 \2mt 1/2 irriQ and writing x = Xf — Xi..24) ( s / .i ) 2 .1 we show t h a t oo /.ti) = = G0(x./ .Z i ) \2KieJ \m0 J y/n + 1 1/2 m0 «mo (21.Xi.3 The Green's Function for Potential and hence Go{xf.tf. exp 2 (21. 2t Go(xf.21) fe=l and t = (n + l)e = tf — U and h = 1. Then Go(xf. (21. (21..z / c .23) for Go and obtain Go(xf.Xi.x 0 )" n + 1 (in agreement with Eq. (7.7) for here n = 1 (n+l)/2 lim ( —^~ K2meJ m /n A = 0 V 2e (21.ti)= iX (o. we have Go(xf.22) • \ n/2 .Xi.Xi.tf. (21. oo dxi 71+1 / J—c /_ dxn exp oo irriQ oo ~27 ^](a.21.37) — by explicit solution of the differential equation which the Green's function satisfies.ti) = V = 0 511 m0 lim n—>oo \ 2irie oo (n+l)/2 /oo oo /*o / dx2.tf.t) is the configuration space representation of the free particle (V — 0) nonrelativistic Green's function.Xi) exp \2Kie{n + 1) 2e(n + 1) Note the square root factor in front which originates as discussed after Eq.Xi.ti) Ko(xf.tf. . In Example 21. We obtained the same expression previously — see Eq. But t = tf — ti = (n + l)e.25) T h e quantity Ko(x.oo /*oo / oo cfcci / J—oo cfcc2 • • • / J—oo + (x.ti)9(t). f c .8).23) We now insert our expression for In into Eq.t) K0(x.Xi. exp where Xj = xo and Xf = xn+\ and there k = 1).ti) lim n^oo lim n>oo m0\(?1+1)/V27rie\"/2 1 z —— exp 2e(n + l )(a.tf. TJ+1 dxn exp[i\{(xi 2 — XQ) + • • ^n) }] (21. (21.tf.
as / JXi f r D{x{t)}eiS°lh . ••• 2 / \ 1/2 \2iriheJ dxldx2•••dxnK0(xn+l.xo)2 + (x2 .tn)••• K0(xi. n+l = lim / n ^{a^jexp fc=i ^°° Jxi = lim / • • • / dx\dxi • • • dxn I —1 nKx J J \2mneJ / \ 1/2 .22) oo Indxn+leiX<x"+*x»+^ 1/2 f°° dxn+ie^:. or else by induction. (21.xi)2 = 21 xi ^2o(„.25) we can rewrite the Feynman path integral or kernel (21. In our proof of induction we assume the result is correct for n._*R±HY+l_{xo_X2f J —! r dXieiM(xlxo)2 + (x2~xi)2 ei^(x2x0) and using Eq.10) in a product form.e. ^ (xi . we have instead of 5*0 and KQ expressions S and K.6a). (21. Solution: The result can be obtained by cumbersome integration.tn+l•xn. (7.7) by direct integration with ^ .5) is carried along. Eq. (21.x „ + i ) 2 = ( x n + 2 Xi + Xi .54)..t0) (21.2y(xn+2 . Then oo 2 / (21.ti. i. Example 21. e 2he ^{Xn+lXn)2 = \2iriheJ lim ••• n^ooJ J x(^*) e^() .512 CHAPTER 21. (n + l)A"_ We set y := x n +i — Xi = x n +i — XQ and (x n + 2 .)+(1"+2..x ^ 2 + y2 .^(Xn+1~X0^2+iX(Xn+2~Xn+1^'i.22).x „ + i ) 2 = (x„ + 2 . 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.)!} J — OO .1: Evaluation of a path integral Verify Eq.f_^y e.an) Then n I 1/2 (n + 1)A" "I 1/2 fX r ioo ^n + dxn+1eiX{^+v2y2Hx"+2x"+1 l ) 2 } J —OO (n + l)A n «ix„+1ea(Sl»22»lI«^I.x0. (21. If the potential V(x) of Eq. the latter being a Green's function with interaction like that for the harmonic oscillator.26) with n integrations and n + l factors KQ. For n = 1 (see below) h This result can be verified like Eq.
) H — (xn+2Xi) n+2 n+l n+2 y2 .6a).a .2y(xn+2 . The first.25): Ko{xJ) = = (S) e{t)(^X'2 y J .x2rriQ 1/2 toij_.2.i) H —3/ . i. (n+l)A n n j(a. This quantity is also described as the nonrelativistic free particle propagator. Using Eq. 21.max exp i 2t 6(t) 1/2 \2mtJ f°° 2KimQ J dp exp J—c ipx P t 2mo (21. iir(n + 1) 1/2 jn+1 n+l L(n + 2)A'n + l I 1/2 4 (n + l ) A .i j ) H n +— z 1 . a dp exp [ipx + iap2] = exp — i(21. We pass over to momentum space with the help of the following two integrals. (21.3.2 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. and + 2 2 l n + 2 ( x n + 2 .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.21. 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.27) we have (with a = t/2m0) . n + 2 .i) .29) Hence with KQ from Eq.n+2a. n+l T h u s as r e q u i r e d 1/2 Jn+l ^ e J — oo 1/2 .28) exp (— V 2irimo J —( dp exp ipx — i p* 2mo (21.3 The Green's Function for Potential V = 0 Next we set z : re+1 (x )2 — Xj). it .Xi) + ( x n + 2 Xi)2. (21.Q . A(n + 2) 21.2 y ( a .. e > 0 .e. rc+2 rl 1 / s2 513 so that n dz = dy = d i n + i . s 2 n + 2 2 „ / ^ n + 1 / \2 (x„+2a. The second integral is an integral representation of the step function. (21.30a) —6(t) / dp exp ipx — to Joo 1 i—t 2m.
*' where 2mo — r. (21.28) this becomes K0(x. an expansion in rising powers of the coupling constant (contained in the potential).51)) is fairly selfevident.e.t) i(2 h?L L E iv 00 dr exnlipx T '2mo t + itr] — te ~U2^LdPJ ^E+£. Thus in first .Q (21. 21. The Feynman Path Integral Formalism Im t \t>0 Rex Fig.31) is the nonrelativistic propagator which describes the propagation of the free particle (V — 0) wave function. the second order two factors of V and so on.514 CHAPTER 21.2 Contours of integration. The first order correction contains one factor of the potential V. i.4 Including V in First Order Perturbation We now proceed to calculate the first order correction in a perturbation expansion. The generalization to three space dimensions (required below in Eq. ( 3 b) ° The expression E+ V — it 2m. This is the type of expansion most people describe as perturbation theory. 21. Replacing 9(t) by its integral representation (21.
(21.xfc_ OO J— OO J—OO L J..xi.Xi. ^ = k ) = lim (S^ A H*\ nKX) *—J \2irieJ r "+1 />oo X / dsi / dx2.32) Since — we recall this for convenience G(xf.)} (21. .21) this can b e written as Coo oo /•oo />oo / oo dt / J —oo ^Go(cc/. .i. li f f 515 order (cf..4 Including V to First Order: Perturbation Expansion h=l. Eqs.^.i V e / D { a ._i By regrouping factors this becomes Gl n+l lim ^ e n—>oo V{xhle) dxn / i ci) n+l dx z+i / * * / d / _ i . dxi ( m0 V V27rie/ m0 \^He \ exp fe=. . x (n+l)/2 \ lim .tf. dxr.t)\. (21..ti.tf..20)) and with elS = exp i S i o / hi dtV(x.t)] n+l .+i 6XP L — {xk XkiY {iV(xhti)} 1/2 X Em fc=l — 0 (xkXki) W i t h Eq.U).33) xG0(xi.i. (21.U) = — lim fv{x(t)}el —— js (n+l)/2 I dxi dx2.(xfc .t) the first order correction is (G\ ~ G — Go) Gi(xf.Xi. . } e °° f l f •/ oo /"oo F ( a .5) a n d (21.t) JSo dtV(x. dxn exp i ^ —. \2meJ n+l J.21.tk) J_( x exp fc=l = ie(—2(xkxk 1>{x(t)}exp iSoi dtV(x.t/.ti)= n+l ri_> f V{x(t)}eiSo „ iSo f ' dt[iV(x.^ J_00 V(xk.^){iV(a.
Jti Consider the term of second order.t)K0(x. .x2. The first two terms lead to identical contributions and therefore cancel the factor 1/2! in the expansion.t) (21. The factors of "V" originate from the series expansion of the exponential exp ftf . i.t2)V(x2.ti). OO /"OO / oo dt / J—oo dxKo(xf.t) =li dtV(x.ti).37) Jti The last two contributions do not describe the time evolution from ti to tf.t) ! Jti r*f dt'V(x.xi. (21.t.t)V(x. It can now be surmized that the next order contribution to KQ + K\ is K2 = (~i)2 / dti / dt2 I dx\ I ti.x2.ti)K0(x.i / dtV(x. The Feynman Path Integral Formalism and hence correspondingly for the first order contribution Ky. t2. / 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'.tf. t') + • • • .xi.tf.tf. They must therefore be deleted.34) We observe that we can extend the ^integration to — oo since KQ contains the step function 9{t).516 CHAPTER 21.t2)V(x2.U) xV(xi. (21.e. Xi.xi. t) Ju Thus the term containing n times V contains a factor 1/n!.36) {if •j.ti) t2)K0(x2.I dtV(x.x. dx2Ko(xf. This is cancelled by n! different time orderings.t2. (21. in writing down the expansion exp i dtV(x. i.35) We also observe here that the problem of timeordering arises only in perturbation theory. = i dt2 / dx2Ki(xf.t2) xKo(x2.e.
E) which is given by (in onetime plus onespace dimensions) V(x. This can also be written in the form of an integral equation which when iterated yields this expansion: K(xj. (21. 21.Xi.t) ) K K K0 'V X K o Fig. (21.tf.t)K0(x.ti). These may be formulated in configuration space as below.3 along with the diagramatic representation of Eq. we are concerned with quantum mechanics. and hence the diagrams representing perturbation terms here are Feynman diagrams corresponding to those in field theory.ti) i dt = K0(xf.t. tf.Xi. For various reasons — such as illustration.t. U) (21.Xi.38) . or in momentum space as mostly in field theory.x.3 Feynman rules and representation of the Green's function.39). x"^ X x"2.ti) (21.i ] P I dt J dxKn(xf.21. called Feynman rules.t)Ko(x. Here.tf. of course.tf.x. 21.xi.Xi.t2 K o (x 2'l2 .39) dxK(xf. The diagrams represent mathematical quantities and hence are designed on the basis of rules.ti) = K0(xf. .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.t)V(x. transparency or brevity — it is useful to introduce a diagramatic representation of individual perturbation terms.t) is V(p.Xi.tf. Such diagrams in the context of field theory are called Feynman diagrams.tf. The Fourier transform of V(x.t)V(x.E).40) The rules for the present case are given in Fig. x r l i 'V(x.ti).t)= f dpdEeipxiEtV(p.
here with E — k 2 /2mo.7 = e i k ' x _ i : l i with f i/}*i/jdx.518 CHAPTER 21. (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. We recall for convenience from **Note that from the wave function we obtain.ti). The plane wave function normalized to one over a box of volume V = L3 is.ti)il)i(xi.tj)K{xf.ipf have to be properly normalized. . and dn = dnxdnydnz.Xi. The Feynman Path Integral Formalism 21. mo V since ip is normalized to one particle in volume V. (3) The transition probability per second is = \A\2 . (21. 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). i.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. There the wave functions ipi. The transition amplitude A describing this process is defined by the Green's function sandwiched between the inand outstate wave functions. 2 The expression A 2 /T can be related to firstorder timedependent perturbation theory in quantum mechanics. to one particle in all of space.t) = .= l.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 threemomentum interval** = V/(2irh)3. (21.k = 2wn/L. i. for the allowed values of k: kx = 2imx/L. in onespace plus onetime dimensions by A= dxf dxiil)*f{xf. with p = Kk. dn = (L/27r) 3 dk. T = duration of time for which potential is switched on.tf. ^(x. * k 1 • .41) We have to pass over to three space dimensions.e.e.
47) . Then H'ki can be taken out of the integrand in Eq. (21. # = ^an(i)VnM) n = a^(t) + \a^(t) + .e.T)\2 f • (2L46) We take the following expression for the timeregularized Coulomb potential.t) = e~^lT\ r where r = Ixl.44) = u(Ek. Eq. Then for stationary states M^t) and itm^ = f HUt'V^t'dt'. u = ^\H'kfp(k).. 47T 1 3 7 (21. Hence (h = 1) atot = f J j2^~k dp V Vm0\A(p. «= — . and with these the following perturbation ansatz: ih^* with an(t) = HV.44) and f_oo can be replaced by fQ.4). H = H0 + \H'. (21.117) and Example 10..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. Proceeding in this manner one obtains Fermi's "golden rule".k. The transition probability per unit time is .r)eiE^\ 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. 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.— . (10. in which T is a large but finite time.21. P(*) = j r (2145) (cf.
t. (21.f.x.e.ti). x.t) = ^ J*ie*<*'*><<*'*>. (21. In our previous onedimensional considerations we had with Eq.tf.tf. i.tfx. The lowest order approximation of A.44) by contracting the propagators to instantaneous point interactions.44) (here in the scattering problem with initial and final energies equal).30a) K0(xf.). is now in three space dimensions A = Ai= / dxf / d^ty*f{'x. (21.tf. we have to replace the onedimensional quantities dxf etc. (21. Thus with the ansatz Ko(x.51) Inserting this into Eq. (21.i x (i)V(x. dt dxip*f(xf.tf)Ki(x.* / . ti) = co5(t — U)5(x — Xj) etc. we obtain A1 = clS(ti tf) / dxf(i)V(xf. £. t.tf)ip*(xf. (21. We can relate this amplitude to the amplitude ark of Eq. t.^ / . T a large but finite time.49) i.49) together with the following wave functions (of asymptotically free particles) 1>i(*i. (21.Xi.t) (21.50) Here x — xj — Xi and t = tf — ti. (21.tf.x.tf)ipi(xf.f. so that 0(t) can be dropped) K0(xf. Xj. This now has the form of the amplitude (21.520 CHAPTER 21. t)K0(x.U) = ~ e ^ l ^ \ ^(*/>*/) = _ L e * P .t)V(x.yLi.tf)K0(xf.>/>. The threedimensional generalization is (taking tf » T and ti <C —T.. by threedimensional dxf and so on..39) OO /"00 / dt / OO J —OO dxK0(xf.52) .tf). also called Born approximation.ti)'ipi(xi.xuti) = ^ J°° d p e i p x _ i ^ * .t)K0(x.48) where from Eq. (x*.ti). Ai = dxf / dx. The Feynman Path Integral Formalism In calculating the Rutherford formula with the help of our previous formulas.
i ( k 2 .r) = = y . Then we require the following integral. 2 „ x v~ m o (21. .21.2: piqx / Y iukawa •" dx e~Mr 47T q + M2' 2 (21. £). At2 dteiat~^ a2T2 f°° =e " J dre^lT2 p2k2 2mo = (~)l'2ea2T'' / 1 6 .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.54) Now inserting the expression for the potential V(x.5 Rederivation of the Rutherford Formula we have (for ingoing momentum k and outgoing momentum p A(p. Then integrating over q and q' we obtain . this becomes A(p.iaT2/8: L With 0 0 • . (21. k2 ipx + i—t + ikx — i 1 2mo 2mo (21. which is evaluated in Example 21. (21.k.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 .p 2.k.56) (21.T) = x exp L z/ q/2 k2 521 f dx.T) = JdtJdx(^\V(x.fdxe^rt* dxe< p) x fdte1 ' (4) 2m0 T2 e "^^J2" 16 .~ .k.).k.t)exp A(p.T) = ^JdXJdt%*t2/T2exp i ( k . P2 .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 .p ) . (21.59) .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.
k.60) Now. k2 . y 2 [ ( k . dE = PjV = ^ E ± . Using the result (21. i.^ . 66) o . k .62) We want to consider the limit T — oo. <rtot.59).< ^ £ \M VTTT2. Using Eq. we obtain . ( 21 .59) we can reexpress A as I /7rT2\1/2 e (P 2 k 2 )T 2 f Ajr ~) p * P . where M has the meaning of mass (in the case of the Yukawa potential the mass of the exchanged meson). otot contains the factor \A(p. integrating over the square of the ^dependent factor in V we have /"°° dt(e4*/"1*}2 = (^p) ' 2 instead of T.3.e.61) We can now Hence in our case \A\2/T has to be replaced by \A\2/(nT2/8)1/2.63) We establish this result for T — 1/e in Example 21. The Feynman Path Integral Formalism This means: The result of Fourier transforming the Yukawa potential ~ exp(—Mr)/r is the propagator (21.65) But E=*. . / / P 2 _ .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 ' kP 2 (21. { I i _ F T l } .63) we have: Probability per unit time = 77777." + M ] p) ' "•"• ^2 . For this we require the following • representation of the delta function: —k2T2 rp —k2T2/8 S(k) = lim T . (21. (21.64) Inserting the result now into our expression for the total cross section.„ 2.T2f 8 V/ 2 .522 CHAPTER 21. mo m (21 .T)2/T. In our case this has not been done. 2 with 2mo J 2ir a2 Ct 16TT2 LU7I F [(k . This expression assumes that the integration over time t is normalized to 1.„. e (21. e 8{k) = lim 7=*—^ (21.51 — 2 /w2 y ~ —. T ) ~ .\ 2?7lo / 1 (21. write down the probability per unit time which is ( 8 \1'2 _a2 16^ .P ) 2 + M2]2 4 VvrT2.a ( — ) ^ .
M £ M * = / m0kdn = m0^2m0E (k . setting M2 = 0. and inserting the expression thus obtained into the final result of Example 10. 2mo / dft. 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.65) with the cutoff M > 0.5 Rederivation so t h a t of the Rutherford Formula 523 M^f) .70) . Then (x = r) foo „2jj.21.3: A representation of the delta function Determine the Fourier transformation of e~ax of the delta function <J(fe) = lim — / dxe~ax a*o 2K Joo for a > 0 and hence the following representation eikx = lim — .j . (21. giving the potential a coupling constant. (21. This result may also be obtained from lyukawa of Eq. Example 21.68) into Eq.4.67) and (21.68) Inserting (21.e. esfn (21. with k = y 2moE / we obtain the differential o?m\ 4fc sin 4 (0/2)' 4 cross section (21.^ . p 2 = k 2 . so t h a t (21.59) by replacing there q 2 by ( k — p ) 2 .p ) 2 = 2 k 2 ( l .cos 9) = 4k 2 sin 2  .* 9 ' d r e . i.e.67) T h e delta function here implies energy conservation.2: Fourier transform of the Yukawa potential Show that [ j / j eiqx dx—e . /Yukawa = ) Solution: We choose the 2axis along the vector q.^ Jo 47T q 1 M — iq 47T i q2 + M 2 Jo G— 9 Example 21. (21.69) This is the wellknown Rutherford formula.= e " f c a ^ o 2v^fa /4a = lim £^o —..^ .j . 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 .e . —Mr 4w = .
We consider again nonrelativistic quantum mechanics.* .72) co From (21. 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 ) . a simpler method suffices. In the present case.71) with respect to k yields the same as °° 2 . 7 r°° d 9 xdxe~ax elkx = — dx — (e~ax oo 2a Joo dx Partial integration of the right hand side implies oo /. Direct integration /4a .75) . The Feynman Path Integral Formalism o° 2 / oo dxe~ax eikx. '<*> = . 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. We write the state vector of a particle of threemomentum p in the momentum space representation: p).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. 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). Since dxe" 1 1 1 = « / . a. In configuration space the state of the particle is written x). Differentiation of (21.^0 a^C 2^/TTO. Va we obtain g(k) = Je~k V a = 0.x) (21. however. (21.524 The function required is CHAPTER 21.73) 5{p) = (W / dxe"P'x' 5{x) = J^rI dpePx '' (2L74) The normalizations (qp) = (27r)3<5(p . ' i a = lim e —0 pk 2 /e2 e^TT 21.oo J / )eikx. (21. (21. (yx) = 6(y .71) we obtain a representation of the delta function: <5(fc) = lim — / 1 /"oo a ^ O 27T J _ 0 0 dxeax 2 eikx = lim —g(k) o ^ 0 2?r ^ O 2K 1 i 1 = lim k 9 .q).70) and (21..
The relation between Heisenberg picture states and Schrodinger picture states is \ipt)s = e~lHt\ip)H.We now introduce the Dirac bra and ket formalism.e. so thattt d — x. t\ipo)s. (21.e. (with ft = 1) d/dt\ip) = —iH\tp).t) = +iH\x. M^ip(x. (21.81) Comments on signs in Eq.The time development in the Schrodinger picture is given by \tpt)s = e~lHt\4>o)s.21.If '/'(x.80) (21.p). 4) is to satisfy the Schrodinger equation. i. t) = (x.6 Path Integrals in Dirac's Notation therefore imply 525 (xp) = e ipoc .. 159 . i. Then ( X  ^ . t' — t. so that \if>t=o)s = H>)H.We define x. i.82) (21.t'x.t) = e* H t x).t\ = iH{x. We consider first an infinitesimal transition.i) = e iH4 x. ) H = (x</>o)s. Thus (qPp) = p(qp) = p(27r)3<5(q .t). (21.83) (21. The time development is given by the Hamilton operator H.0) = e""x). i. Hence for the infinitesimal transition (x'. Then the states {x. at initial time ti to its final position xy at time tf.i'x. d/dt\x.i) = K(x'.. we must have d/dt{x.x.t)} constitute a "moving reference frame" in the sense that ip{x.t).79) Feynman's principle is expressed in terms of the propagator or transition function (x'. pp. . x. as had to be shown.160. (21. (21.t'.78) Pp) = pp) is to be interpreted to say: The quantity p is the eigenvalue of operator P with eigenvector p). Similarly we have a position operator X with Xx) = xx).£) = ( x V ^ ' .t) = —iHip(x.e.t).^ l x ) .82): The general Schrodinger equation has the state vector to the right.t> =iH\x.e.76) These relations imply the completeness relations 1 = J dxx><x. x' — x infinitesimal. which describes the propagation of the particle from its initial position x.t\. i. (px) = e~ipx.77) (21. The equation 1 = J ^p)(p. H.t).e. Ryder [241]. See also L.
* .y<&(p'ix')(x'ivix)<Xp) = f dxe. Pq) = J 7^3Pp)(pq) = J dppp>*(p ..P V<P e l )  P > e P X <2184 = /(w/(^ ' ' *" '"'' *=£Uv«.85) the amplitude or expectation value or position space representation (x'#x) = . i.91) .i p '. (2187) (21.p').90) (P'IVIP) = y<&.x = V(p .x') + V{x)5(x' . 2mo Since (21. 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.x y(x)e i p .88) Thus P has the operator representation P= /dxx)V(x.e. "  » We suppose that for a single particle of mass mo moving in a potential V(x) the Hamiltonian is (21.V 2 < 5 ( x .89) Using this and previous relations we obtain for the Hamiltonian (21.85) The momentum operator P satisfies the relation P = /(2^3PP)(P' (2L86) so that it projects out the momentum q of a momentum state q). (21. (21.526 CHAPTER 21.q) = qq>We also have by mapping onto configuration space Pp) = = P /*dxx)(xp) = P /"dxx)e i p x = p /"dxx)e i p x f dx\x)—eipx= /"dxx)V(xp).x).
527 (21. We now consider the transition of the particle from Xj.92). EQ — Hi . tf subdivided into infinitesimal transitions (xj+i.e. t).82)).21.tj).6 Path Integrals in Dirac's Notation we have in momentum space the representation (p'ff p> = ^(27r) 3 <5(p . i. But e'ipx + 0[{t . (21. Thus with •"•/ — ^ n > *•/ — ^ni ^ 0 — Xj.p'). i. (x'.92) We now return to the expression (21.j. the signs are reversed when these operators are applied to the ket vector \x.t) (cf.tj+i\x.p) + i{t . We have <x'.95) when applied to the bra vector (x.i'x. ti to x j .94) In the above we have chosen the signs such that p = —iV and H = i— ot (21. for the amplitude of an infinitesimal transition.t') \^(27rf5(p 2mo +V(pp' where we used (21.93) + ipx P r ip(x'x) iH(t't) (2TT) 3 (21.84).p') • • • j W < 5 ( p ' .p') + V(p .fx.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 '. (21.t> d f dp' P' r_dp f dV rivf*!ini\rmtf) J (27T)3y (27T p e p) e ipx +i(tt'){p'\H\p) + }eipx .
t n  x n _ i .t 0 ) + H(p2.tf\Xi.ti) rx/=x(i/) d x i dx2 . .e. = 1 X X<jPfc We now replace H(pfc.2.tf\Xi. ..i /Xj=x(*i) f J [ ^ 1 dp2 3 3 J (2TT)3 (2TT) . We now insert for each of the infinitesimal intermediate transitions the amplitude (21.. T h e integration over the n—1 intermediate positions.99) .i/xj.528 CHAPTER 21.xic) by fc — Xfej H(pk. x i ) ( t i .ti) = x 0=Xi n^n fc=1 ? <^Pi (2TT)3_ exp / ^fa .tixo. i. dx. x exp [ . .ti.. .t0>(2196) We define a p a t h in configuration space by a succession of n + 1 points x o .tj) = / ••• / d x „ _ i d x n _ 2 .i { f l " ( p i . zmo T h e n t h e integration over all p ^ can be performed by using Eq.x2)(t2 We can rearrange the factors here in the following form c„=x/ n—1 (Xf.97) H hH(pn.x i ) H h pn • ( x n .94) and thus obtain a phase space expression. t n _ i ) t„2><xi.xn)(tn in_i)].x 0 ) + p 2 • (x 2 . The Feynman Path Integral Formalism we obtain (it will be seen below why we do not introduce normalization factors here) (x/. The resulting relation is m0 2irieJ 1/2 m0x exp ze ) " / 2TT exp Pk 2 mo 1 (21. then corresponds to the sum over all paths. d x i ( x n . x i .xk (21.27) in which we make the replacements x —> eifc.a — —e/2mo and divide b o t h > sides by 2TT.dpnl dpn (2TT) 3 (2TT) 3 exp [i{pi • (xi ..tn respectively. (Xf. x n which the particle passes at times tQ. . . x f c ) = ^ .tfei) fe=i ..X n _i)] h) (21.e. i.. d x n . d x i ..+ V(x f c ). .. (21.98) tk — *fcl # ( p f c . d x „ _ i . .
102) We see t h a t the factor (with h = 1) m0 2m(tk 3/2 3/2 [N(e)}< = tki) \2irie) is precisely the normalization factor inserted earlier for the onedimensional case in Eq. Truman [254].100) over V{p}= L J ^ rwooll V (27T)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.H(x. m& ze ) J (2vr)3 exp it p fc • xfe IP* 2 mo (21. Smolyanov. i.103) is another form of the pathintegral for which the integration with the help of Eq.98) can be rewritten as 3n/2 2^") x / n dxfe exp * ^o** ~tfcV rnl <i 2 m ° x fc ~ ^ (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 threedimensional case. > Thus the phase space representation of the pathintegral* (21. G. to / m0 \ 3 / 2 las. .100) Hence with all tk — ifci = e we obtain /n dpi i _(2vr)3 exp •.p)]\.. G.ti) = J D{x(t)}D{p(t)}expi [' dt\pi.98). (21.] ex H ^I (.tfci) fe=i £(**'**){ p***^} —tki m0±2k (21. A.8) as N(e) so t h a t the integral does not vanish for e — 0.fc=i 3n/2 r n m0 exp 2_^i{tk 2m(tk . / {xf.e. Tokarev and A.101) Then Eq. i. (21.tf\xi.21. (21.e.
V(xk) = m0±l . pfc) = Pfc • xfc . We know from classical mechanics that this relation is a Legendre transform which transforms from variables x. (21. Note that we have n factors [iV(e)]3 but only n — 1 integrations dxk.tf\xi.e.8).x) . t to x. Thus finally we have Feynman's principle: (Xf. operator (21. The Feynman Path Integral Formalism yields automatically the correct normalization factor or metric in the configuration space path integral^ in agreement with Eq.530 CHAPTER 21.ti) = ^lim^ N(e)3n / * JJcbc fc exp i / fc=i * dtL(x. Garrod [106]. p . .tf\xi. x. We have now obtained the path integral representation of the matrix element ( X / . £ /  X J . pk) = L(xk. Consider the following expectation value with primed and doubledprimed states denoting some intermediate states between initial and final states denoted by indices % and / : (xV'l x(<) \x'.106) In order to be able to deal with this expression we first require the momentum operator representation of the operator x(t).ti) where nl D{x(£)}exp '. We check below that this is given by x = / ( ^  p > i  ( p   <2L107) See also C. nl (xf.8).105) D{x} n ^ o o [7V(e)]3n Udx kk=\ In this notation the measure factor is contained in ©{x}.H(xfe.x) (21. Clearly for a better understanding one also wants to explore the case of sandwiched operators in the path integral representation. (21. this agrees with the corresponding discrepancy in the onedimensional case of Eq. t.104) where for the particularly simple Lagrangian L under consideration here m0±l .t'}. i.iJ(x fc . xfc). (21. £ J ) with no operator in between. I hi dtL(x.
6 Path Integrals in Dirac's Notation For the verification we use Eq.t'} = = fdx5x5(x"\eim"\x5)(x5\eim'\x'} f dx5x5{x".76) we can rewrite the factor (x4p)^(px5) as (x 4 p)x 5 (px 5 ).t"\x5){x5\x'. ••• 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. (px).108) (x 4 p)^^(px 5 )(x 5 e^ t 'x 6 )(x 6 p / )e.t"\x\x'. (21.77) and thus have 531 *!*') = *M = * 7 ^ I P > « . .21.t'}.* * f dp = j (2^ p) ^ d . . so that we obtain a cnumber position coordinate which we can shifted across other cnumbers.*') = / I F / (2?FeiP"'x" / ^ 5x5 (p // K^> 5 )(.107).v ' x ' and thus {x". With contractions and integrations over delta functions the expression for the matrix element then becomes <xVxx'. . Thus <xV'xx'./ ^ I < « .0 = J'^ j'^e^''(p"\e^''xe^'\p')e^'. x fdxG  ^ P <p"x3)<x3e^V) x (21.i p '• x .109) .5 l^ t 'lp / > e .* * . (21. With the wave functions (21.
t"\x\x'. We now consider the transition of the expectation value of operator x from Xj.tj+i\x\xj..112) where x" is a cnumber. (21. tixx 0 . ] .t') dp .x') .£ n xx n _i. . we obtain with (21.tf subdivided into n transitions (xj+i. We now proceed from there along parallel lines.i{t" . Inserting here the integral representation (21.0 = and hence /dx 5 x 5 [(x"x 5 )(x 5 x') .£/xxj.£'> = x"[«5(x" .f ) i f e i p .t n _i) x ( x n _ i .iit!' 1') 2mo + vtf) + <5(x"x').t') (xV'xx'. .i(t" .111) 1 .110) Recalling Eq. (21.90) we can rewrite this expression as (x".532 CHAPTER 21.x') . t0).74) of the delta function. (21.it / / x / (x"ffx / ) + it'x"{x"\H\x>) + • • • .t'){x"\H\x') = X + •••} (21.tj) at equal time intervals e.i t " ( x "  #  x 5 ) ( x V ) +.i ( t " .85) (xVxx'. we obtain (here and in the following the limit n —> oo at the end being understood) (x/.tf — tn and x$ = XQ. The Feynman Path Integral Formalism In the second last expression here we expand the exponentials and obtain (xV'xK.£j to Xf. Thus with x^ = xn. This is the expression corresponding to that of Eq. x / / e . £ n _ixx n _ 2 . tn2) • • • (xi. (21.*/<x"x5)(x5tfx'> + .113) Inserting for each of the infinitesimal intermediate transitions the amplitude .x ' ) (2^ i^+vM + D «p(x"x') (21.0 dp x„ / (27T): 1 .ti = to as before.• dx1 (x n . ( x " .94) for the case with no operator in between.£j) = / ••• / dx n _idx n _2 . = x"6{x" .
t"\x'. .t') (21.tj) = / JXi 2){x(i)}x(i) exp i I I Jti dtL{.x .114) x exp [i{pi • (xi . tj) = xf (x/.x 0 ) + p 2 • (x 2 . U). ...x2)(t2 ~ h) + • • • + H(p ••• + p „ .. Xn _i X„ xi dx.112). x n _ i x n reduces to the single factor x^. We demonstrate first that the operator representation of P given by Eq.t"\x'.117) "•"One may observe here that if x = x^ = x„ in the integrand of Eq.xi) + f f dpi dp2 dp n _i dp n The momentum integrations can be carried out as before and we obtain finally (x/.« • > We reconsider therefore the transition amplitude (x".x 1 )(t 1 Z & ) +H(p2. tj \ xt. Before we can consider canonical quantization in the context of path integrals we have to investigate in more detail the role played by momenta.tf\x. (21.y (27F (27F •" (2^(2^)3 exp[.21. dyi l xi x 2 .114).t"\x'. (21.105).t') = (x" + 5x".ti) 533 rxf=x(tf) = / • • • / dXidx.89) is consistent with the conjugate expression dL (2L116 * . For simplicity we consider the case of one spatial dimension. x (21.2 .x n _ i ) ] . we obtain* (xf..( x „ .2 . the string of factors xi X2 .t"\x'.\Xi.iyxxj.7 Canonical Quantization from Path Integrals (21..t') .and hence may be put in front of the integrals to yield (xf.{x". n—>oo *•»• 21. . tj  xj1 x»...3 " n d x f c .t') sider the variation S(x". (21.{^( P i. dxnn _i Xi X2 .115) where (there is no integration over x n ) nl 0){x(t)}= l i m [ i V ( e ) ] .10) or (21.7 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.t') = and con 5x"Vx„{x".. x n _ i x n i JXi=x(ti) /Xj=x(tj) 7 .t"\x'.
120) However: In our case here 8x ^ 0 for t — t" (although 5x — 0 at t = t'). <9L' dt —dx + — 5 x ox ox f . t".t"\x'. dL d5x / oft—5x + / dt dx dt Jt' ox Jt t' and hence <9L —oxdt + ox ox nt t.118) JI(x)/h . jt/ 37 ( £T.t"\x'. dt\Ox w(x) = r dt d^_dL(dL dx dt \ dx °L 8x + ^ T!5 X 1 ox (21. V°{x}[i6I{x)/h]e* Jx' The following steps are familiar from classical mechanics: yV 51{x) S /1 dtL(x. (21. with N = N(t" .tt).534 CHAPTER 21. " (t). The Feynman Path Integral Formalism with 8x" = 5x(t") ^ 0 and 5x' = 8x(t') = 0. For the purpose of enabling some formulations below. x) Jt Jt dL. (21. .t'): 1 fx" 6(x". Hence w « = (t). dx dt \ dx _ (21. Then. dL r /•*" .t').10) and set D{x} = D°{x}/JV(i/ .t') =  6— / 1 V°{x}t .i/(x)/fi ) j ^ ) dtL(x.x).t i_(dL_ x't' h\dx J t„ (21.122a) = 17) 5x"Vx/. (21. and we demand the validity of the equation of motion.(x". " M{x)/h &(t ) = fa (L2) 2 11 and — using Eq.119) The classical equation of motion follows from Hamilton's principle for which 5x = 0 at t — t'. Thus in this case of classical mechanics dL_ d fdV.l&cdi.118) — = (21 5x"(x". we separate the metric factors from T){x} in Eq.
S.t"\p(t")\x'.1 (A / ).t').125b) and the measure property (3) N~\A + A') = A^.115) in three space dimensions. The following properties are assumed to hold: (1) / Jti V°{x} = ^ x. Here f*f D°{x} indicates that the integral is to be taken over all x with boundary conditions Xi = x(ti).124) where with A = tf — ti. (21. Streater and A.125a) (2) the completeness relation holds. 535 (21.x).89) we see that Having explored the appearance of canonical momentum in path integrals we can proceed to extract the canonical quantization. (21.Xf = x(tf).1>) = ^{x". (21. We consider this here in quantum mechanics in one spatial dimension but parallel to field theory. ^2\x'. . x' (21. 6 (21.126) R.t'\ = l.. we now have in one spatial dimension {x^tflxlxuU) = J^ff'DQ{x}x{t)eiIWh.7 Canonical Quantization from Path Integrals or JL(x»y\x.21. § The evolution of the system is expressed by the Feynman functional integral (xf. F.122b) By comparison with Eq.1 (A)iV.U) = T777TT / &{x}eaW\ IXi=x(ti) ) iV(A) JXi=x(U (21. I(x) = / JA dtL(x.t')(x'.tf\xi. / Jxi V°{x} / Jx' V°{x}. Wightman [264].125c) Corresponding to our relation (21.
t') (21.ti) f ' V0{x}xlX2eu^lh. tn lie correspondingly in consecutive order between U and tf. (2) and (3) above. (21.tf\x(ti)\xi.127)) (x".129) = ^ y With Eq. Eq.tf\xi\x'.t"\x"\x'.t"\x'. (21. But for t = ti we have (and equivalently for t = tf) as we can see from Eq.t>). The Feynman Path Integral Formalism The time t is in general a time between ti and tf. Then the relation follows with the help of the properties (1). (21.t').t"\x'.536 CHAPTER 21. We can rewrite this expression as the commutator relation (ITU*" x"^](x". (21. Taking the functional derivative 5/iSx" of this equation we obtain 'ilx7'^' '^"^x"\x ''*') = (21.ti). We consider the case n = 2. Thus if T denotes such ordering.tf\T(Xl • • • xn)\xi. (21.130) ~i5x7jX"{x"..t"\x'..t'){x\t'\x2\xi.t\x'. we have the relation (xf.132) The generalization to higher spatial dimensions proceeds along parallel lines.122b) we can deduce the nontrivial canonical commutation relation arising here.128) where tf is later than ti and t\.ti) = Xi(xf.t').t')=x"(x".t"\X'...tf\xi.t\x'.127) We can now write down an expression for a timeordered product of operators.t"\x'. (21.t"\x'. (21.t') = ih(x.t').tf\T(xix2)\xi.114) (xf. .t') + x"^J^{x".x](x. (21.t') \ i ox" i ox"J or in commutator form [P.ti) = x' 'Y^{xf.133) = ih{x".131) = ih(x". We have (cf. Thus (xf.ti) = —!— p V0{x}xx • • • xneu^h.
102. An essential singularity is really an ambiguity rather than an infinity. in the sense of rigorous convergence tests. z = a + i/3 with a — oo or f3 —> oo or both. nonzero radius of convergence in the domain of the coupling parameter. Whittaker and G. as we have seen in Chapter 8. in the context of quantum mechanics are not convergent.e. so that successive contributions to the first approximation do not invalidate this approximation. If the expansion is mathematically well defined. such as the fine structure constant of quantum electrodynamics. i. Such expansions are strictly speaking. integrated over) beyond its radius of convergence.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. 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.g.g. Watson [283].e. > 537 . the series converges. the modulus of the ratio of the (r + l)th term to the rth term is larger than one. W = e x p ( l / z ) . although the first few terms indicate a convergent behaviour. Such an expansion parameter is usually known to be or assumed to be small in some sense. it was known from quantum mechanics that RayleighSchrodinger perturbation theory by itself does not yield e.Chapter 22 Classical Field Configurations 22. Consider e. the exponentially small level splitting which occurs in "The expansion involves both positive and negative powers of the deviation. i. i. E.1 Introductory Remarks In field theory a perturbation expansion is generally understood to be an expansion in ascending powers of a coupling constant. The vast majority of perturbation expansions discussed e. it has some finite. N. but asymptotic.e. Expansions of this type arise. divergent. It was therefore natural to search for alternative methods of expansion.g. see e. Besides.g. In physical applications such cases are rare. T. p.
e. specific boundary conditions have to be implemented in order to yield these effects which are therefore frequently termed "nonperturbative". The study of expansions of this type has turned out to be extremely fruitful.h. and has led to insights which previously seemed unimaginable.. In searching for other means of expansion. The fundamental extension of the method of canonical quantization (i. These methods are often described as methods of collective coordinates.) l/h2. There. cnumber. A further challenging task was the development of methods of quantization of theories which incorporate classical. 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. nature of the dominant approximation does not imply that this describes classical motion.e. although this terminology is not precise.. In particular it enabled nonlinear problems to be studied and led to a consideration of topological properties. 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. This type of expansion is decsribed as "semiclassicar.e. Classical Field Configurations the case of the symmetric double well potential or in the case of a periodic potential. 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).h°. such as soliton theory.h2. 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. higher order corrections are of orders ±. 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. the dominant contribution is singular (has an essential singularity) for vanishing semiclassical expansion parameter (h — 0). such > a series begins with a contribution proportional to (e. the procedure discussed thus far) to systems with constraints was developed by Dirac [76] and is introduced in Chapter 27..h3.538 CHAPTER 22. and in general this change is accompanied by certain constraints. A method which achieves this is in particular Feynman's path integral procedure. cnumber first approximations. However. One therefore faces the problem of quantizing a system which is subject to constraints. i. i.g. The classical.. On the contrary. and the corresponding expansion is called the "loop expansion". Many of these aspects are interesting by themselves . the latter's topological properties if any. In the following we consider various typical examples.
1) ' T h e reader who wants a highly advanced overview of instantons.* assuming that this is acceptable to a reader with some familiarity of the basics of the more complicated field theory of electrodynamics. monopoles. vortices and kinks. to keep this case in mind. d^] = d^d^ . 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. Tong [272]. It is evident that symmetries and their violation by the classical approximation play a significant role in the entire consideration. a static (time independent but spacedependent) solution of the classical equations of motion.V(4>). and also to see these in a somewhat broader context (by comparison with higher dimensional cases). may like to consult the article of D.g. 22. 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. Thus references to field theory will be exceptional and hopefully do not irritate the reader. V{</>) = m§0V + A0(<^)2.2 The Constant Classical Field 539 and in any case deserve detailed study. A main aim of this chapter is to generate some appreciation of the distinction between socalled topological and nontopological configurations (later referred to as instantons and periodic instantons and bounces respectively). (22. modified by going to imaginary time). . This means theories with field densities which are therefore infinitedimensional. We shall not be concerned so much with the case of a constant first approximation.22.2 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. However. or even a solution to a modified field equation (e. 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>. The important aspect we want to draw attention to here is that of "spontaneous symmetry breaking". t The classical first approximation in the procedure outlined above may be simply a constant (time and space independent) quantity. it is important for a better understanding of the considerations which follow.
p. Classical Field Configurations where d^ — (—d/d(ct).e.e. V[<t>] m 2>o o Fig. and hence here [d^ + m20}(l) = Xo(f (22.S£/<90(x.3) See e. m l + \04>*4> )$ = <).g. (In models of the early universe one often considers V(0) with m\ > 0 at the early stage. £)) is bounded from below we must have Ao > 0. An even more familiar example is the GinzburgLandau theory [111] of superconductivity § in which the phase transition to the superconducting state implies the transition to the double well potential shape).£). For UIQ > 0 the potential V((f>) has a single well with minimum at </> = 0. 22. The EulerLagrange equation is d„ dC d(d. for m§ < 0 the potential is a double well potential with two minima at \<f)\ ^ 0.2) We ask: Are there classical (i. 22.1. i. cnumber) solutions <j)(x) = 4>c = const. 433.540 CHAPTER 22. Felsager [91]. . as illustrated in Fig. V ) . § (22.1 Different potentials for different signs of m§. 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. B. of the equation of motion? For these we must have all derivatives of <j) zero.
(22.e. whereas the field (bc becomes (a) A oi(0+<x) V~2 + Every new phase defines a different solution. But for Ao > 0. Thus the classical solution 4>c violates the U{\) symmetry of C and of the equation of motion. the only such solution is the case we wish to consider. 22. Fig.2 The spontaneously chosen phase. TT = —J. It is important to observe that <bc does not possess the rotational symmetry of the Lagrangian density C or of the EulerLagrange equation in the plane of complex fields <f>. the equation has this invariance.ir]=7r<fiC. we have 541 0. H = —C = V. 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. w A iP 3ml 5* "XT" (22. with 0 = 0. 22. i.</>2)plane.2 when allowed to fall chooses an unpredictable phase parallel to a radius in the (0>i.5) .22. mg < 0.2 The Constant Classical Field For Ao > 0. not the solution x(t)). The U(l) phase transformation <b — exp(ia)((> leaves both £ and the EulerLagrange equa> tion unaffected. and so ft[<M=m6:(^* 1 + Ao(0*0)2. We can convince ourselves that <fic is the field associated with a state of minimized energy.4) Here (3 is a spontaneously chosen phase like a stick held upright along H in Fig. aq> For (b = const.e.
22. . (22.Then m\ m\ = ml + A0A2 = m2. where (f>c = — T](x) = —= [Vl (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. We examine this in more detail by setting <f>(x) equal to the classical cnumber configuration plus a fluctuation field n(x) which is again complex like <f>(x).3) obtained above. as indicated in Fig. V?) (22. e. i.g. in a transverse direction. This is the condition (22.3).7a) (22. i. For every value of the phase (3 we have a different constant cnumber field configuration 4>c. we set cf>(x) = <f)c + ri(x). one obtains the equations (d^ and (d^ + ml + ^ A 0 A 2 V = 0 ( ^ 2 .tp2.7a) vanish on account of Eq. Suppose we choose one such configuration.e. i.542 CHAPTER 22.e.7b) In Eq. (22. &H/d(fi = 0) if [m20 + ^>]</> = 0.7b) the constant on the left and the coefficient of A on the right of Eq. >° for m 0 < °> . ^ ) o + m 2 + 1A 0 A 2 ) ^1 = A (± A0A2 + m 2 ) + 0 ( ^ 2 .e. and we wish to investigate the behaviour of the field cf> in its neighbourhood. 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. along the trough of V (which we can call a longitudinal direction).6) Inserting this into Eq.e.2) and separating real and imaginary parts. (22. Clearly these are the field configurations which trace out the circular bottom of the trough of the double well potential. and partly by climbing up the parabolic wall on either side. (22.2. the one for (3 = 0. Classical Field Configurations the density Ti is minimized (i.3ml = 2ml = ml + TA 0 A 2 = 0.
We can see the problem here by looking at Eq. The elimination of the . 22. i.3 The spontaneously chosen phase seen from above. Thus the Green's function is similar to that in electrodynamics. which.r')S(t .e. 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 £. then a massless boson exists. the existence of individual perturbation contributions).22. t') = d(r . Like the Goldstone mode in the above Higgs model it leads to a divergence in the Green's function of the theory. t. The appropriate Green's function G is the inhomogeneous solution of the the equation [8^ + m20]G(r. (22.7b). irrespective of the question whether the perturbation series as such converges or not. though not identical phenomenon. of course. (3=0 Fig. 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. 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. has to be removed in order to allow a well defined perturbation procedure (i.e. tangential to the classical path.t'). There the wave function with an associated vanishing eigenvalue is called a "zero mode". 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 fourvector potential A^ are independent. In our later discussion of theories with nonconstant classical field configurations we shall encounter a similar. ^ = h eip / \ ^2 l .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. Goldstone's theorem applies to fully relativistic field theories. whereas ipi(x) is y ' " " " " " ^ . r'.
22. we shall need only at a later stage. equates to zero the coefficient of this wouldbedivergence. t Potentials of this type have been discussed by M.544 CHAPTER 22. instead we shall consider static and timevarying solutions in important models of one spatial dimension. Lohe [179]. Khare [17]. The Lagrangian density is taken to be C[^. W. MullerKirsten [291].e. *S.d^} = ~d^d^V[^) = ^2^(^j V[$]. 7711 = — 1 ( 1 6 Minkowski manifold). C. A. .* Potentials of higher polynomial order in $ than $ 4 can also be considered. We see therefore that the classical cnumber field configuration and the attempt to develop a perturbation series in its neighbourhood leads to intricate connections between symmetry properties. To insure that the energy (see below) is positive definite we require V[$] > 0. We make one more assumption concerning V which. see e. spinless field $(x. zero mass configurations and constraints even before the question of quantization of the fluctuation field T){x) can be considered. Later we wish to develop a perturbation series about a classical configuration of $.t In particular the sextic potential can be considered along parallel lines. N. and we choose the metric ?7oo = + 1 . so that we are interested to have a parameter *For a collection of many informative papers on solitons etc. In the following we will not be concerned with classical cnumber solutions of the EulerLagrange equations which are simply constants. Soliani [237]. The principal aspects will always be very similar so that it really pays to study simple models in considerable detail.8) where V[<&] is a selfinteraction or potential of the field (to be specified later). however. Classical Field Configurations other components results from the vanishing mass and the constraint called the gaugefixing condition which can also be looked at as the condition which removes from the Green's function G the divergent contribution. Rebbi and G. J. i. Behera and A.3 Soliton Theories in One Spatial Dimension We consider here the two basic soliton models known as <J>4 and sineGordon theories.g. (22. Zimmerschied and H. F.£) in one spatial dimension.* We consider the theory (later: quantum theory) of a real.
.«{x) = V'(<p). The other condition which we impose (and study in detail) is that of stability. but in such solutions which are subject to (ignoring the dimensional aspect) reasonable physical conditions. i. (22.3 Soliton Theories in One Spatial Dimension 545 which serves as the expansion parameter of the series.13b) We are not interested in just any solution of this equation. The first such condition to be imposed is that the energy of the solution of interest must be finite.13a) We write such fields <j>(x).§ Unstable classical configurations are also important and will be considered in later chapters. For this purpose we assume that V[$] depends on a scaling parameter g such that Vm = V[t>.e.22. (22.l].g] = \ v m For instance. From Eq. (22.l}. 1]. 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). at least in the present case. (22. 9 9 The EulerLagrange equation is seen to be (with c = 1) D * + V'[*] = 0.11) (22. such that J > — — — = 0. — m$ 1 — cos Im 1 I ^ $ (22. the reason for this condition being that we visualize the classical solution as representative of a lump of energy. In a quantum theory the field $ is an operator defined on a space of states. (22.12) we see that they satisfy the Newtonianlike equation <j. In fact to begin with we restrict ourselves further and consider cnumber fields <> —• (f) which are static. the socalled ^theory = ^V[g$. Before we consider quantum aspects we study classical cnumber fields which satisfy the EulerLagrange equation.10) n*]  m4 r 25 i m2 and the sineGordon theory with cosine potential by ±V[g$} = \v[g$.12) where the prime denotes differentiation with respect to $ . i. Since $(x. The field $(x.9) with quartic potential is defined by ^V\g$\ = ^V\g*.t) has an explicit timedependence it is a Heisenbergpicture operator.e. ^ ^ * + V'[*]=0.
The energymomentum tensor T^ and the conservation law it satisfies are given by the equations dC T^ = n^C + Q^^d^.e.14) where the overdot implies differentiation with respect to time t. d. 22.T = <I>.c = <£z + .4 The wellknown soliton of $4theory.17) Since we have to integrate over all space. dC • Too = r?oo£ + =r$ = C + vr$ = H. (22.2\dx 2 ' 1 fd& + V[$]. HI tanh m(x>k) 9 Fig.e. the classical solution will therefore have to be such that this integral is finite. We observe here already that the integrand of the integral in Eq.T^ = 0.17) is reminiscent of classical mechanics if we interpret x as time and V(<f>) as the potential energy of a particle at location d>.15) The zerozero component is equivalent to the Hamiltonian density H. (22. Classical Field Configurations The energy JE?[$] of the system is defined to be the spatial integral of the Hamiltonian density 7f($. <9$ i.7r) where IT is the momentum conjugate to $ defined by dC ? r = .546 CHAPTER 22. (22. (22. i. .16) In the static limit we obtain therefore and write this as Em = H[<t>\ + E(4>) = fdx 1 2 \dx 2 + V(</>) (22. n = & .
the $*theory. m ( x . O±m(xxo) (22. — XQ). Rubinstein [240]. < = — tan [e >  ° ] 9 2 rrm/g — Fig. The sineGordon theory differs from the KleinGordon theory in that it possesses invariance under the shift rf> — <f> + 2n".22. around <j> = 0 the sineGordon potential behaves like the KleinGordon potential proportional to 2 4> . Again XQ is an integration constant.4.5 The soliton of sineGordon theory.3 Soliton Theories in One Spatial Dimension 547 As we shall see below.18) is depicted in Fig. (22. (22. 4m . The energy is correspondingly obtained — as will be shown — as 4 m & 9 The typical shape of the configuration given by Eq.10) and (22. one obtains by straightforward integration (but see also below) the soliton configuration m 4>c{x) = ± — t a n h 777(2.13b) for the specific potentials (22.20) (22. 22. (22.1 .11). See J. 22. it is not difficult to find classical cnumber solutions 4>c to Eq. in the case of the first.777.18) Here XQ is an integration constant. 22.21) This sineGordon configuration is depicted in Fig.e. .x n ). > .5. "The name sineGordon is derived from the analogous KleinGordon equation or theory. In the sineGordon theory^ the classical or soliton configuration is (see below) found to be (x) = ±4— tan" 1 with energy 777 . i. In fact.
Inserting the potential (22.11) we obtain r<Kx) Jd> 4>{x0) m d{g<t)/m) 9 ml g V h(lcos&) m or m(x — XQ) I. 4m In tan tan 5 l/e±m(*xo)tan^£^)U. or \(<P'? = V(4>)+ const. 0(x) = ±4—tan" 1 j e * " 1 ^ " * 0 ) ! .9 2 4>{x) = i.* ( m' m 2 r)] ^2" .o) =m7r.1 "»JtKxo) 9 1 .20) are obtained from Eq. (22. . Thus J<t>(xa) \/2V(6) JXOJxn U(x0) y/2V{</>) dx = x • XQ. In the case of the sineGordon theory defined by the potential (22. <^>(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 . Classical Field Configurations The solutions (22.548 CHAPTER 22. Here the constant is zero for V(<p) and <j>' zero at x = ± o o . ± 1[tanh.e. and hence 4/71 9<t>{x.18) and (22. With we obtain g(/>(a.^^! m ^m <p(x0) m 4>(x) = ±—tanhm(x — XQ).13b) which we can write d \W? = !"(•».10) we obtain f<t>{x) x — XQ i 9 ! 1 = ± / J<j>(x0) dtf> 1 .
This is called more precisely "classical stability J The EulerLagrange equation is obtained by extremizing the action or Lagrangian. 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. Then we have to demand that the functional derivative of the energy E(<p) be zero.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) . (22.23) See for instance R. instead of doing this now. 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. in the onedimensional Ising model.22.19).e.4 Stability of Classical Configurations The concept of stability can be complex. we obtain these expressions later (cf. 22. A state of stability is therefore associated with minimized action and/or energy. the discussion of Bogomol'nyi equations) in a simpler way. . (22. (22. for instance. Suppose now that we demand stability in the sense of minimized energy.17).4 Stability of Classical Configurations 549 Prom their monotonic behaviour the solutions <f)c in these cases derive their name as "kinks". the curves rising monotonically from negative to positive values are known as kinks. SE((f>) 5cb(y) 0. However. The relation (22. however. this does not require the action to be minimized. i. The expressions (22.22) and that its second variational derivative be positive semidefinite. (22. the others as antikinks. Jackiw [141].21) for the energies can also be obtained from Eq.
i. Since it arises from the violation of translational invariance by 4>c. (22. i. of course. The original Lagrangian was.27) shows that the zero mode results from application of the generator of translations to the classical cnumber field configuration.550 CHAPTER 22. Assume that S[U] is invariant under the transformations of some symmetry group G with elements g = exp(i\T) € G.e. 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.24) For socalled "classical stability" the differential operator of this expression has to be positive semidefinite.13b) retains the invariance although the solution (f>c does not. dx (22. if we apply the generator of spatial translations d/dx to the classical equation. A theorem on zero modes Let S[U] be the action of some field U defined on Minkowski space. = 0. (22.A. and the EulerLagrange equation retains this property. if 4>c{x) is a solution. we obtain 4'(x) = v"(4>c ^ or d2 dx2 4>'c(x) = 0.25). Thus the vanishing of the first variational derivative yields again Eq. in fact. Eq.25) Before we begin to study this equation in detail for specific potentials we can make a very important observation on very general grounds.13b) with classical solution 4>c.e. (22.a / 0 is not the same. its eigenvalues w\ > 0.e. dx2 + V"{4>) (22. Proceeding to the second variation we obtain at 0 = <fic the relation 52E((j>) 5<j){x)5(t)(y) S(xy). This is. with eigenvalue w2. Classical Field Configurations Since y ^ ±00. namely (p'c. i.13b). it is also called a "translational mode". in fact ip0(x) const. written down in Lorentzinvariant form. the function 4>c(x + a). (22. The eigenfunction 4>'c{x) is for this reason called a "zero mode". (22. the integrated contribution vanishes. we see that the latter has one eigenfunction.26) Comparing this equation with Eq. The Newtonlike equation (22. a very general phenomenon which can be established as follows. In fact. Thus we have to investigate the eigenvalue spectrum of the Schrodingerlike equation d2 + v"{<t>c ipk(x) = wkipk(x) dx2 (22. in particular the invariance under spatial translations.28) .
30) Differentiating this with respect to A (applying —id/dX) and setting A = 0. (22. We shall see later that these zero modes lead to undefined Green's functions for the semiclassical perturbation expansion unless one or more suitable constraints are imposed.33) x—>±oo . = 0. Invariance of S[U] under transformations of this group implies S[U} = S[gU].]. We can choose the potential V such that the constant is zero.] = 0 or x—>±oo lim V{(/)c) = 0 for const. As stated above. we are interested in classical cnumber field configurations of finite energy.22. and we have seen that Eqs. i.27). Thus TUC is a zero mode. this integral has to be finite. / oo Thus for E(4>c) to be finite. (22. in fact. and that S[UC] is finite (or equivalently the energy).20) are such configurations in two particular models. Then the vanishing of the first functional variation implies that 5S = 0. lim [V(<f>c) + const. (22.32) = V'(</>). 0 = S'[UC] = S'[gUc}.4 Stability of Classical Configurations 551 where T is a generator and A a parameter. (22.18) and (22. i.e.e. We see therefore that the occurrence of these zero modes is a very general phenomenon. (22. 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. (22.29) Suppose now the EulerLagrange equation possesses the classical cnumber solution Uc.13b) as ~d~4> > ' The first integral of this expression is \(tfc? = V(<l>c) + const.17) we obtain oo dx[V(4>c) + const. (22. In the onedimensional case under discussion we can rewrite Eq. we obtain 0 = i^S'[exp(iXT)Uc]\x=o = S"[UC](TUC). it is an expression like (22. (22. We return to our considerations of stability.31) Inserting this into Eq.
Classical Field Configurations V Fig.6. (22. n = 0 . For a configuration to have finite energy.552 CHAPTER 22.34) jfc" = e^dv(l since d^k^ = 0. .c o and another for x — oo. in view of the antisymmetry of eM^. the potentials have degenerate minima (i. . The solutions can be characterized by an integral number called the "topological (quantum) numbcf^ "topological charge" or "winding numbed (for an illustration see Example 22.11) we must have correspondingly for x — ±oo: » 771 9 2vrn. » In the <fr4theory defined by Eq. the classical field (bc must approach one minimum for x — . Thus the classical energy or vacuum configuration is not unique. 22. . ± l . (22. (22. This number is defined like a charge in field theory by the spatial integral of the time component of a current.10) we must therefore have that for x — ±oo: 4>c — ±m/g as indicated in Fig.6 The minima of V in $4theory at x — ±oo. The charge Q is therefore given by dxk° = / dxe01di(bc = [ ^ ( s ) ] ^ = ±—. In the case of the 3>4theory we can define a conserved current k^ by (22. Solutions which approach > > the same minimum for +oo and —oo are the constant solutions <f>c = ±m/g in the $ 4 theory. 22. V{4>) = 0 at different values of </>).35) and it is seen that in this case q can be ± 1 or 0. .e. ± 2 .1). In the sineGordon theory > • defined by Eq. . oo J—oo y The topological quantum number q is defined by 0.
± l ± 2 . continuous values. Again we see that E(\<f>c) is finite only for A = ± 1 .38) 4 V(\<t>c) >m ~2[l . The difference Tfl 4>c(oo) — 4>c(—oo) = 27T—An .4 Stability of Classical Configurations 553 Since we call q a topological quantum number.cos(27rnA)]. We can see this as follows. in general. if it remains unchanged under continuous deformations of the (field) configuration while preserving finiteness of the energy. termed topological. . and we call this "topological stability. x—>±oo so that under such deformations the configuration remains in the same topological sector defined by the boundary condition. In the case of the sineGordon theory the minima of the potential (22.22. n = 0 .A2)2 + 0 except for A2 = 1. Then for x — ±oo: > > > V(\4>c) (2 ^ 0 ) ? 4z( 1 . and exp(ix(0))<pc reduces to 4>C when x(#) = 0. Such a smooth deformation (in the sense that exp(i%((2))<^c depends smoothly on 6. 2g Thus even without varying E with respect to A we can see that E(X(f)c) is finite only for A = ± 1 .7r) does not change a boundary condition lim [\<t>c{x)\ = (/>c(±oo). A property is. . First. one may ask how topology comes into the picture. . zero is not a meaningful topological number. Thus we consider the family {X(pc(x)} of (field) configurations where A is some parameter which assumes real.36) oo Consider the $ theory with 4>c — ±m/g for x — ±oo. Thus we expect the constant solutions (which are characterized by topological number zero) to be nontopological in some sense. 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. Tfl \ (22.37) for the onedimensional case here.7T (22.11) are given by ( so that for x — ±oo: > — J. For such configurations we have the energy oo 2 2 dx ^ A ( ^ ) + y ( A ^ / (22. .
g. J. (c) the time component k° is a spatial divergence.40) (22.41) dxy/2V{$)4>'{x).13b) or ^ ' ( * ) ] 2 = V(</>).5 Bogomol'nyi Equations and Bounds We have seen previously (cf. (22.e.554 CHAPTER 22. 00 (22. (22.33)) that the classical configuration cf)c is the solution of the nonlinear second order differential equation (22. Eq. where n±<XJ are the integers n corresponding to the minima of the potential at x = ±00. . 425 .9 the foregoing arguments in a higher dimensional context. Eqs.43) **See e. i. MiillerKirsten [215].** (b) the time component k° depends only on <p but not on momenta (again in contrast to the case of Noether currents).17)) f 00 > 0. pp. the factor A = exp[ix(#)] varies over a complete circle). the energy is minimized.31) to (22.^ = N. 22. 00 (22.34) is conserved independently of the equations of motion (other than Noether currents whose conservation follows from the equations of motion). W.42) Thus the energy has a lower bound given by the right hand side of this inequality. The inequality is saturated. its integral is nonzero only on account of nontrivial boundary conditions. From this we construct the inequality [</>'{x) = Vm^)}2 F Since (cf.39) E(cp) = J — 0 0 we can write / dx oo ~1 2(0'(x)) 2 + y(^) (22. (22. We observe (a) the topological current k^ of Eq. Classical Field Configurations involves the difference An = rioo — n. Thus with one space dimension higher there. (d) the topological quantum number iV arises in a completely classical context. 22. H. (22.428. if oo / dxy/2V{<l>)<f>'{x). (For a broader view we consider briefly in Sec.
40)) this occurs when 4>'(x) T V2V(4>) = 0. .n > 1. In the sineGordon theory we can consider. for instance. 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 1cos— .5 Bogomol'nyi Equations and Bounds We observe that (cf. (22.19). (22.22. Here (V(c/)) > 0. Bogomol'nyi [27]. B. n E .21).39). 555 (22. the set of configurations (fic which interpolate between 0(—oo) = 0 and <f>(oo) = 2nnm/g. We discuss this in the following Example. In the case of the <fr4theory 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. 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.40) is called the "Bogomol'nyi equation" since it was first introduced by Bogomol'nyi. In order to obtain a better understanding of how topology comes into the picture we recall that the sineGordon theory has an interesting analogue in classical mechanics. (22.44) This first order equation which saturates the inequality (22. Eq.^ Clearly it solves the second order equation or (22.
Initially the pendulums are suspended freely in the gravitational field. rotate) only in the (y. } n= — oo roc —> / L J . Fig.g.7 The pendulum analogy.7. xn+\ — xn = a) 00 a —> dx. Solution: Consider a string along the xaxis as described. . Classical Field Configurations Example 22. ka —> K. 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) measured e.t)\2 +mgr{\ . 22. Construct an expression for the energy of such a classical system and its Lagrangian.cos</>(x„. so that oo > u.g.e. V dt „ 1 (d4>_ dx • i^gr(l — cos(j>) (22. with mass m and connected with short strings to neighbouring bobs. H. z)plane and only such that their elastic strings remain taut. 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. Assuming that the pendulums can move (i. * / dx OO + 2^ V dt 1 K (d<t>\ 2 \7T I + ^9^(1cos 4>) (22.1: Classical analogy to sine—Gordon theory Consider a string along the xaxis.556 CHAPTER 22.t) 2 V dt The energy of interaction with neighbouring bobs is (recall "fcx 2 /2") k[(f>(xn+i. all being identical.ij\2 Thus the The potential energy due to gravitation is (maximal for <j> = 7r) mgr(l — COS <f>{Xji.t) 4>{xn.2 V dt + k[(f>(xn+i. t)). rotate) only in the (y. Assume that the pendulums can move (i. from the yaxis as shown in Fig.45) We can therefore write the Lagrangian L=TV• r dx fir P 1 2 . Chapter XI. At equidistant points xn = na along this string we attach strings with pendulum bobs.i)) The continuum limit is obtained by t a k i n g ^ (since xn = na. 22. total energy of the system is with the number of pendulums allowed to become infinite *= £ i ''( "' ) .46) **See e. We can now write down an expression for the energy of such a system of (say) n pendulums.e.t) <j>(xn. Goldstein [114].
12) we obtain d2 d2 \ Since (j)c{x) obeys Eqs. (22. (22. we see that ipk(x) has to obey the equation 92 dx2 +V"{(t>c{x)) ^ fc (x) = w^k{x).25) obtained earlier.t) = ° (22'48) (22. We can understand what stability means in the present case. The equation is called stability equation or equation of small fluctuations. wk would be imaginary and so the factor exp(—iwkt) in rj(x. (22. which represents an expansion in terms of normal modes.50) This equation is identical with Eq. (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). 22. t) is the fluctuation (field).t). This number. and so would invalidate the procedure which assumes that rj(x. It is instructive to obtain this equation by yet another method.49) exp(iwkt)ipk(x). Hence we set ${x.7 linking the pendulum bobs).47) where n(x. (22. t) = ^ k ~ iL2)71^ + V "^c{x))'q[x.t): {w With the ansatz r)(x. is therefore also called "winding numbed. We are interested in studying (field) configurations $ in the neighbourhood of the classical solution cpc.22. (22.49) would imply an exponential growth in the future t > 0 or in the past t < 0. the topological quantum number. each finite energy configuration beginning and ending with a classical vacuum corresponds to an integral number of rotations about the iaxis (i. Eq.t) = <l>c(x) + r}(x. 22.6 The Small Fluctuation Equation 557 Thus if the classical vacuum corresponds to the case when all pendulums are pointing downwards.13a) and (22. If some w\ were negative. ' Equating coefficients of exp(—iw k t). t) of Eq.6 The Small Fluctuation Equation We now return to the Schrodingerlike equation (22. Inserting (22.t) is a small .13b) we have for small fluctuations rj(x.25) which we obtained as the condition of classical stability for eigenvalues w\ > 0.e. of the continuous curve in Fig.47) into (22.
20).51) (b) SineGordon theory: Here 1 — cos m m 9 V"(<f>) = m 2 c o s f ^ . t) have to be orthogonal to the zero mode (which is thereby circumvented) so that the Green's function of the expansion procedure exists. If that is necessary. we obtain V"(<t>c) . In fact. .A \ Inserting <f)c of Eq. Thus for stability w\ must not be negative. Classical Field Configurations fluctuation.50) has the general form of a timeindependent onedimensional Schrodinger equation.52) tan 6 := e±m^xo). This means that the stability is not universal but only local in a certain sense. approaches a constant value which is nonzero in both the <54theory and the sineGordon theory.x0) = —2m + 6m 2 (l — sech 2 m(x — XQ)) 6m Am' cosh m{x — XQ) = (22. V"(<b) = 2m2 + 6g2(f>2. 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. cos46» = 2cos2(2(9) .18). V"((/)c(x)).e.e. i. we obtain V'^c We have m cos ± 4tan" 3±m(:r—xo) (22. the zero mode. 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£^)*. Inserting (j)c of (22.558 CHAPTER 22.1. The equation of small fluctuations (22. (22. For x — ±oo the quan> tity acting as "potential". we shall see later that the fluctuations n(x.2 m 2 + 6m 2 tanh 2 m(x . i. and need 9 = tan" 1 j e±m^Xo) 1. In fact.
2 (22. ip+.53) = m(x — (22. The eigenfunctions of the discrete case vanish at ± o o and are squareintegrable.6 The Small Fluctuation Using the formula cos 20 we obtain cos 40 1 . .* For such a potential the spectrum can be b o t h discrete and continuous. (22. and as a consequence a limiting form of the small fluctuation equation of all basic potentials. A  *(* + !) cosh 2 z tp(z) = 0. which satisfy the boundary conditions V>(o) = o. Those of the other case are complex and periodic at infinity. We can therefore construct solutions of definite parity. — XQ)] cosh[±m(x — XQ)]: 1 . i.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 . Consider the equation fd2 dz2 .55) The differential operator is even in z.tan20 1 + tan 2 # Equation 559 1 — e x p { ± 2 m ( x — XQ)} I + exp{±2m(x — XQ)} sinh[=Fm(a.56) *We shall see later that the PoschlTeller equation is a limiting case of the Lame equation.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 PoschlTeller potential — and setting it is known as the A = 4*2 as the eigenvalue. ^+(0) d 0.22.e. We can investigate the spectrum as follows. we see t h a t the equation represents a Schrodinger equation with a potential which vanishes exponentially at ±oo. (22.ijj^ respectively. even and odd solutions.
Z ) + (A + Z2) "n+1 (n + l)(n + §) &n (22.e. when in the even case n ( n . 2n < Z. (22.( 2 n + l . (22. and in the odd case n+\)U+\l)+\(\ i. i.61) Discrete eigenvalues are obtained when the series (22. A * An = .57) Then the equation for x{z) 1S (22.Z ) 2 . i.59) Setting (22.60) n=0 n=0 oo we find the recursion formulas x+(o = E ^ " > n(nl) = x(o+=l £ £&»£n.4 n ( n . i.e.l2 = .58) = (1 + sinh2 z)~l/2x(z).Z ) + ^(A + Z2) = 0. 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) .e.560 CHAPTER 22. Since for £ — oo: * the function ^>+ is normalizable provided Z > n. + \{\ ) 2 ~Q"m 1/2 oo «n+l (n + i ) ( n + l ) ( n + ^ ) ( n + ± . + l2) = 0.e.Z)2. .( 2 n . A ^ An = .60) terminate after a finite number of terms.
w2 = Am2. From (22.11). (22.u2)—^V 2 dio and ip(z) = elkzV{to).is normalizable provided 2(n + . i.10) and (22.e. °2 i. 2 . (22.50) we obtain with z = m(x — xo) tp(z) = 0. m N<2. N = 0. We can summarize the results in the statement that the discrete eigenvalues are A = (ZAT) 2 .e. (a) ^theory: the equation Inserting (22.( 2 ..e.63) we see that the continuum starts at A= 2 2^ = 0.j < I .2Z = .22..1: w2 = 0. (22. m i. d w = tanhz. 3m 2 .54) by setting X = k2 Then 1 .62) where even (odd) iV are associated with even (odd) eigenfunctions. (22. We can see the continuous spectrum of Eq.6 The Small Fluctuation Equation 561 and the function ip. We can now derive the spectra of the stability equation for the potentials (22.2.62) we see that the equation possesses discrete eigenvalues w —2 .\=Kl2 From (22.Ny.. for N = 0.51) into (22.. the continuum is w > Am2.1.64) This is the equation of Jacobi polynomials V\a' (u/) with a = —ik of degree I in tanhz.63) d + K(2ik  2LO)—V ' dijj' + 1(1 + l)V = 0.65a) _dz2 cosh 2 z l=2. (22.N<1.
Thus it has no node and so represents the eigenfunction of the lowest eigenvalue. m m ipo{x) oc —<fic{x) = ——tanhm(x — XQ) = g cosh m(x — XQ) dx g dx (22. Thus either spectrum contains the expected eigenvalue zero. In the $ 4 theory we saw that (cf. (22.. . We know that 4>c is a monotonic function. . also be deduced from d(j)c/dx. Eq. so that its derivative is nowhere zero (except at ±oo). the zero modes.. . 22. and — as expected — this eigenfunction is even. i. Classical Field Configurations (b) SineGordon theory: Inserting (22.2 The discrete eigenvalues are now given by w ^l m^ = (lN)2. N<1.27)) .8 The zero mode as typical ground state. m d . . for N = 0: to2 = 0 and the continuum starts at w2 = m2. 22. It is now particularly interesting to look at their wave functions. This property can.e.53) into (22.50) we obtain with z = m(x — XQ) the equation (22. Thus 4>'c(x) has no node. d . .65b) dz2 cosh^J. The zero mode is depicted in Fig. of course.562 CHAPTER 22.8 and we see it has the shape of a typical ground state wave function.66) We see that this is a nonvanishing function for any finite value of x. %(x) Fig.=1A=^_.
. .z 0 )]oo = 8 . d . Finally we consider the normalization of the zero modes.o) ±4 m g l __ e ±2m(xx 0 ) 1 cosh m(a. .oc (22.= E(<f>c).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. (22.6 The Small Fluctuation Equation 563 In the sineGordon theory the zero mode is proportional to . — XQ) OC Hence the conclusion is similar to that in the previous case.t a n h x .67) J —( Mx)= and hence m w0i^ 1 2 4 m ±—tanhm(x — XQ).e. Imposing the condition F we have in the <I>4theory: /•oo dx[^o(x)}2 = 1. (22. (22.3 .68) d{mx) m(xx0) 4 m3 3 ^ g Ml J cosh / i. ipo{x) oc —<bc(x) ax oc 4 m d _i Q±rn(xXQ) —tan e g dx m D±m(za.e.70) .22. 1 o rrr tanh x — . without quantum corrections). 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 . i. .
73) n<(>a) = j ' d^V^)2 = a2~d / dd(ax = J' ddXl{V4>(ax)f d 4>{ax) 5ax . we consider E(<P) which we also write as (22. Under this transformation (the verification is given below) £(</>) ^ E(</>a) = T(cf>a) + P(</>a) = a^T^) since for instance + a" d P(0). Classical Field Configurations 22. ^For basic aspects of scale transformations see e. to underline the fundamentally different nature of topological and nontopological finite energy configurations.564 CHAPTER 22.e. We now wish to inquire whether static finiteenergy configurations could exist in more than one spatial dimension.7 Existence of FiniteEnergy Solutions So far we have been considering the case of one space and one time dimensions.72) with T{$) = I' ddx]{V(j))2 > 0. ddx {V4>f + v{4>) (22.g. P.* The general scaling argument used for this purpose was first introduced by Derrick t and is therefore referred to as Derrick's theorem. One can have different types of scale transformations — see e. for if such configurations exist in a realistic theory with three spatial dimensions. Olive [113]. Kastrup [145]. They may help. Affleck [5] who uses in his Eq. however.> 4>a{x) = 4>(ax). .41) for the energy of the classical configuration but now with respect to a ddimensional position space. H. J As a first example we consider the expression (22. H. This difference may not always be appreciated later in our very analogous treatment of (topological) instantons and (nontopological) bounces. it is important to understand their physical implications. f G . Goddard and D. I. (2.g.71) We now consider the scale transformation (with 'a' some number € R) 4>{x) . Derrick [68].4) <j>(x) —• a<f>(ax). i. (22. P(</>) = f ddxV{(f)) > 0.2d.T(4>).g. The arguments of Derrick were later rephrased by several authors whose line of reasoning we shall be using here. A. This question has important consequences. *See e. *The considerations of this section extend in part beyond quantum mechanics and into field theory. I.
xd). Next we consider more complicated cases. the second derivative is positive) only for d = 1.80) .76) we see that E(<f>) can be minimized (i.V(<P) (22. 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.73) we see that this implies (2 . (22.. to be finite the field <P ~^ 4>oo £ Mo for x — oo. It is also a virial argument in view of the virial theorem relation T{<j>) = P(<f>). (22.74) From (22. (22. a=l (22.79) In order to emphasize that (f> — (p^ in any direction.e. x = (x1. Can the energy also be minimized for more than one spatial dimension? To see this we stationarize E(<f>a) with respect to a. d = 1 minimizes the energy.e.75) Since both T and P are > 0 and d is integral.22.78) For the energy of a static configuration... (22.71). 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 ddimensional Euclidean space defined as the unit sphere S*1 = {xx 2 = 1}. Thus this scaling argument excludes the existence of static finiteenergy configurations of the given theory in more than one spatial dimension. i.d)T(<f>) = dP{4>).x2. Consider first the Lagrangian density £[&M = \{d^)(d^) .. since d2E(<pa) da2 2(2d)T(<f>).e. i. In fact.7 Existence of Finite Energy Solutions 565 We saw that the static configuration with a = 1. we set 0 dE(<pa) da J a = i (22.77) with V(4>) > 0 and x G Rd and the set of minimum configurations Mo := {4>\V{4>) = 0}. we write • = lim J r .\ eMo(22.
(ptx = <f>oo. 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 . in fact. —00. Thus in order to obtain a topologically nontrivial case we must have a different classical vacuum in association with each point at infinity.1 manifold Sd~1\d=i <• manifold Mo <—> 0 O = 27rn^ Q <—> 0_oo = 27rn'^ kink limits (nn'/ 0) const. i. i.and sineGordon theories we assumed $ to be real. is a connected region.38).81) and (since d = 1) S = ± 1 . 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).1. therefore. x = 00. Classical Field Configurations In the sineGordon theory (d = 1) the equivalent statement is (22. i. If we go to two spatial dimensions (d = 2). the easiest way to visualize points at infinity is to take a circle Sl of radius r and let r — 00. ( with — = lim <pc(x) 9J *>±oo TYl \ (22. that we have a set of disconnected classical vacua (</>oo><^oo. These observations are summarized in Table 22. = —00 : S = — 1 <—> ^oo = 27rn^ The topologically trivial case is given when the classical vacua associated with S1 = + 1 . 0oo 7^ <>< > in the case of the kink solutions) and /x associated with this a set of disconnected points in space (00.566 CHAPTER 22. — 1). Then each point at infinity is characterized by a different value of the polar angle 9. are the same. . Thus in this case S^1 is a disconnected set consisting of two points. —00 or equivalently S = + 1 . —1.e. Table 22. We see.e. of • course. one set can be mapped into the other. S1.e.1 (A) Topologically nontrivial mapping x = +00 : S = +1 x = 00 : 5 = . as is the case for the constant solutions. and. (n — n = 0) (B) Topologically trivial mapping x = +00 : 5 = +1 <—> <> o = 27rn^ /o a.
if we consider a theory in three spatial dimensions and so the unit sphere S2 = {x x{ + xl + xl = 1}. for instance. by allowing <fi to have three components in some "internal" (isospin) space. we would have only two points at infinity. i. This will be "nontrivial" in the above sense only if 0oo is independent of 9. if <p is real. This can be achieved e. i. i. map into. * n reach any other <^(oo) by simple phase shifts. the manifold of classical vacua would have to possess a similar geometry if the theory is to be topologically nontrivial. . phase transformations 6Q ^> 9Q + 59. therefore. i.83) . VMx)) with (0(oo) = linvxx) (p(r. the field must map position space into field space at infinity.e. assuming invariance of C under the transformations of the group SO(3) in internal space. if cj> were real and position space twodimensional. we have w e ca {Mx)}> and 1 3 1. Thus.3 (^real) v{4>) = / (22. 9) would depend only on r = x and hence the problem would be effectively onedimensional. If position space were onedimensional.7 Existence of Finite Energy Solutions Suppose.22.e. 9)) Mo = {</>(oo)</>*(oo)0(oo) = a2}. (22.e. to have topologically distinct classical configurations. We see. given any (00) = «Koo)e* = <?[<p(xMx)a2]2 567 say with 9 — 9$.e.1 . cj)(r.e. Then the entire set of configurations (f)^ = \4>oo\ exp(i#) would correspond to. 00. i. the one point S = + 1 . Similarly.82) Now Mo := \ 4>(oc) ^24>2 = a2\. —00 or S = 1 .2. On the other hand.• M0. that if the problem is to be topologically nontrivial. $(r) : Sl . i.e. Thus.g.
Glashow [108]. L. the winding number is n. $ — > > $ exp{i Xlo=i Pa. Summarizing we see that </>(x) is a mapping from S^T1 to <S^1. (b) The GeorgiGlashow model I! with nonabelian gauge field in 3 spatial dimensions with Lagrangian density (V as under (a)) £[AM. depending on d) and correspondingly (with n complete windings) the transformation of one classical vacuum into another. The symmetry of the potential (and Lagrangian) — i. a theory being described as 'gauged' if a gauge field like the electromagnetic field is involved. Classical Field Configurations The manifold A4Q is a sphere in a 3dimensional Euclidean space. it is effectively a surface (and so a continuous set) S2.Ta} in an 50(3) invariant theory and so on — determines the transformation from one point at infinity to another (on a line.V{4>a]. $ — $exp(ia) in a U(l) symmetric theory.e.2: Derrick's theorem applied to gauged theories Wellknown and important theories involving a gauge field (AM if abelian and A". Olesen [220].568 CHAPTER 22. A) + T{A) + P{<j>). % . Nielsen and P.3.e. In order to preserve lim y ^ i a\ such rotations must tend to the identity at r = oo. the subsequent discussion requires only the general form of the Lagrangians. A) = T(4>.. we cover <S^_1 n times.„ where G r = 9 M £ . $ — $ + 27 > • T in sineGordon theory.a M J . Georgi and S. 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. Example 22. + i ( D " f t W ) a .Av  dvAf i where 0 is a complex scalar field and the spatial dimensions are 2 or 3.e. 11 H. B. i. . a = 1.2. invariance under replacements $ — —$ in $ 4 theory. so that details of these models are irrelevant.</>] = ~G^Gai.e. going once around S%~1 (i. through each point at infinity). If. Any 4> £ MQ can be reached from any other cf> e Mo by application of an element of SO(3). the appropriate rotation. . i. sphere. if SO (3) nonabelian) are two models defined by the following Lagrangians which we cite here solely for the purpose of being explicit. In the following Example the use of Derrick's theorem is demonstrated by application to some gauged theories. circle..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) £(</>.
some contribution must be .A). Thus we set dE(<fix. (22.89) P{4>) are positive semidefinite and not all zero. Solution: Each contribution in (22.85) Here the difference in the transformations results from their difference as scalar and vector quantities which are defined by the transformations ^V) = 0(x'>. A). <0 >• <£A(X) = <KAx). J 4 ) + A 4 . ^ A ) = A 2 . = 0.87a) C?M„(x) — \2G^{\x). > Since T>IJ.. A.7 Existence of Finite Energy Solutions where T{<j>. 4>Y We have T«(4>A) = J ddx(d„4>x)2 = J dd . i.d)T(A) .22.87b) We consider explicitly the example of <j> — </>\(x) = <p(\x) for > T ( V(4>):= Jddx(d.T{A).d T ( A ) + AdP(</>). negative. (22.<^[x) transforms like a vector. (22. and so B ( ^ A .i T(AX) = X4~dT(A).84) and investigate the existence of finite energy classical field configurations. (22.e. (2 . (i)^ 2d T Considering the contributions of (22. 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 ). A) + (4 . (22.d)T(4>. T(A) = f ddxV(4>) > 0.d T ( ^ . P(tf>A) = A " d P ( 0 ) If £ is to allow a static finiteenergy solution.e.dP(4>) = 0. Mx') = E with a scale transformation x — x' = Ax. >• A„ x ( x ) = AAM(Ax). A A ) = A2dT(^. We now consider scale transformations with A £ R. 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>.86) (22.A) P{4>) 569 j / i F ( « ( P ^ ) t ( P ^ ) „ > 0.84) is positive semidefinite and so must be finite by itself. i.84) in this way one finds that the quantities involved transform T ( ^ . l j ddxGijGai] > 0.
** e. H. (c) For d = 3: All three terms must be present as in the GeorgiGlashow model.89) the contribution (4 — d)T(4>). We then have a situation of the potential like that described in the context **T. One can thus define a scalar field or wave function ty(x. In the following we consider a related theory. H. (22.90) and we would have to add to Eq. An explicit and solvable Skyrmion model in 2 + 1 dimensions is considered in D. (22. This case is.92) The expression ^ ( x ) ! is known as the "order parameter. A) or P{4>) for d = 3. (d) For d = 4: This is possible only if the sign of V is reversed — this is exemplified by a socalled pure gauge theory with socalled instanton solution (which we shall encounter in a simpler context later). W. There are other posssibilities if we permit higher powers of derivatives as in Skyrmion models.89). 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. J. (22. Thus such a contribution could counterbalance T{4>. tt The — now established — microscopic BCS theory of superconductivity was formulated some eight or so years after the macroscopic GinzburgLandau theory. ft V . The GinzburgLandau theory assumes that the static energy density is then given by £W0 = ^ W ( x )  2 + 7 + \aMx)\2 + \m*)t (2291) The parameter (22. if we allow a term T^ 4 ' {<f>) with fourth powers of derivatives.8 Ginzburg—Landau Vortices We have already referred to the NielsenOlesen theory. (e) For d > 5: In this case it is not possible to stabilize Eq.g. T^{<t>x) = \4dT(</>). (b) For d = 2: A theory with all terms would be a candidate — in fact the NielsenOlesen theory is an example.570 CHAPTER 22. D. We can therefore use our understanding of the BCS theory in interpreting the GinzburgLandau theory. (22.t) such that  ^  2 describes the density of Cooper pairs. 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. a > 0. timeindependent) field configurations is equivalent to the GinzburgLandau theory of superconductivity. Skyrme [253]. R. Ginzburg and L. Equilibrium states of the superconductor are assumed to be described by the timeindependent static wave functions ip(x). MiillerKirsten [268]. exemplified by soliton and sine—Gordon theories. L. .e. Tchrakian and H. 22. of course. is temperature dependent and changes sign at the critical temperature Tc. a = a(T) = a T —T c . We first demonstrate — as a matter of interest — that the NielsenOlesen theory for static equilibrium (i. Landau [111].89).
Clearly the length (see Eq. setting S£. (22. ^(00) = a (22. Sec.8 GinzburgLandau Vortices 571 of the complex Higgs model (cf.22.2): For T > Tc the overall potential has a single well (minimum).A) = iB 2 + i iXQ + 7 + ^ H 2 + /^4.93) J' (22. (22.e. from x < 0) means — as is familiar from electrodynamics — replacing V by V . assuming there is no superconducting material in this domain). 22. but for T < Tc it has a double well shape.e.95) This expression can be shown to be equivalent to the integrand in the NielsenOlesen theory. i. leads in a onedimensional case (considering this briefly) to the equation ^  = aV + /?^ 3 axz Imposing the boundary conditions (in the domain 0 < x < 00) ^(0) = 0.94) the solution is (in the domain 0 < x < 00) i>{x) = ^/^tanh(^x\ . is a measure of the distance from the surface of the superconductor to where . The parameter 7 serves to put the minima of the double well at V = 0.94) to check the dimension) 1 = I—a. Variation as before. Then the static energy density becomes (cf.(tp) = 0. 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. Example 22. Returning to the three dimensional case and switching on a magnetic field (applied from outside of the superconducting material. 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.2) £(^.
the order parameter \ip\ attains (approximately) its asymptotic value.e. The magnetic field on the other hand. exhibits a completely different behaviour — in fact. In the Maxwell equation See e.M.9 Behaviour of ip and B in the superconductor. i.572 no superconductor CHAPTER 22. V x M = j s . MiillerKirsten [215]. (22. (22.g. The density of the Cooper or super current is given by the typical expression of a current. h [^DV'V'DV]. J. 450.98) where M is the magnetization resulting from atomic currents. as indicated in Fig. 2mi Setting < / > \P\e vp B = ViA.96) ipH. . 22.e. H. inside the superconductor it falls off exponentially away from the surface (this is the MeissnerOchsenfeld effect generally described as the expulsion of lines of force from the superconductor below the critical temperature T c ).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 . p. i. We can see this as follows. Classical Field Configurations superconductor (x>0) Fig.9. 22. i. W.e. (22. yJ—a/[3.
5 to 10~6 cm). m Here A^ is known as the (London) penetration depth (or length) (in practice this is of the order of 1 0 . This is zero in the present case. 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. Superconductors of type II are characterized by the fact that when the magnetic field is again increased and beyond a first critical value. In the onedimensional case we have B{x) = B(0)ex/XL.e.8 GinzburgLandau Vortices 573 the quantity j is the density of a current applied from outside.M i.e. Using the relation "curl curl = grad div . m xi A B . Hence 0 = V x H = V x ( — B .99) Inserting (22. those of the socalled type I and those of type II..22. V x B = [i0V x M = /lois.97) this becomes (since curl grad = 0) AB = Mo—Vx m A = /x 0 —B. There are two types of superconductors. i.div grad" and V • B = 0. Approximating p by its equilibrium value.L B = O.100) we obtain 2 m ( 0\ The ratio XL/XQ is called the GinzburgLandau parameter. this becomes A B = no V x j 5 . Also dD/dt = 0 in a static situation like the one here. and V x (V x B) = / J 0 V x j s . (22. the normal state of the metal begins to reappear in thin vortices and not uniformly thoughout the metal as in the case of superconductors of type I. (22. K i ™*. These vortices therefore carry magnetic flux which is necessarily quantized .
J.102) In the onedimensional soliton case of Eqs. We can think of (e. See also P. The existence of these vortices has been confirmed experimentally.36). 'See the remark in H. In the NielsenOlesen model defined in Example 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.9) ^e^ '>\<poo\. Schaposnik [69] after their equation (3.. (22. Finiteness of E requires \<t>\ = \<fioo\ on this circle. (22. we must have that lim V ( H ) = 0. A. i. In the two models of Example 22.9). De Gennes [67]. 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). A.37). x—>oo Thus at spatial infinity </» has the value of a zero of the potential. (22. in other words. of two points on a line. i. in fact a periodic function of the polar angle with period 27r.101) (p(r. Classical Field Configurations (see below). Thus the asymptotic field </>(r. Abrikosov [2].e.574 CHAPTER 22.e. by the element exp[i%(0)] of the internal group A(s) = e * W : SI+SI.2 — identified as above with the GinzburgLandau theory — type II superconductivity occurs^ for A > 1/4 and type I superconductivity with the complete MeissnerOchsenfeld effect for A < 1/4. In all the models we considered so far we observed that the energy E contains a contribution P(V) = j ' afxVM). where exp(ix(0)) is an arbitrary phase factor. the space consists of two diametrically opposite points on the circle. 22.g) twodimensional space as being bounded by a circle at r = oo. G. 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.2 we saw that finiteness of P(V) puts no restriction on the phase of (f> so that r e (22. . 6')r_>00 represents a mapping from the circle at spatial infinity to the circle defined by x(#) (or. *A. de Vega and F.
In mathematical language S1 is therefore a compact.103) Later we demonstrate that this winding number n remains unchanged under continuous deformations (those of Eq.* 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. Naturally an integer cannot change continuously.X ( 0 = O)].22. The 1sphere is the unit circle in the space of all complex numbers.9 Introduction to Homotopy Classes 575 By a map or mapping / : X —> Y of a space X into a space y we mean a singlevalued continuous function from X to 1". Misner [96]. We may define this as « = ^[X(0 = 2 T T ) . abelian (i. Suppose <j>. The two theories we considered above in Example 22. Finkelstein and C. 22. 22.10. The map A : S1 > S1 defined on S1 C R2 by (22. The map satisfying X(x + y) = \{x)\{y) is said to be a homomorphism. W.e. Fig.102) is called the exponential map of S1 onto S1.118)). Then a continuous function of (j> on [0.10 States of winding number 0. Sl := {zeC\\z\ = 1}.1). (22. 2n] remains continuous if the function assumes the same value at the two endpoints. Thus smooth deformations *See D.1. We can illustrate the simple lowdimensional example under discussion by reference to a strip with Mobius structure. i. commutative).2 imply that such a map from a circle to a circle is associated with an integer which we called the winding number n.e. States of the strip corresponding to winding numbers 0 and 1 are illustrated in Fig. topological group with the usual multiplication as group operation. defines the orientation of this line element. 0 < 0 < 2TT. (22. . X is called the domain of / and y its range.
The gauge field required in the asymptotic limit (22. we can write 1 d h J ee rde w At h r^oo .. we required that lim r—>oo w o.104) so that d2x hJ Since A has direction e#.105) is described as "pure gauge" since it is determined entirely by the phase x(#).104) to be finite we must demand that g .e...e. elements in the same class have the same topological characteristic. The different kinds of connectivity of these topological spaces (e. Homeomorphism subdivides the possible spaces into disjoint equivalence classes (to be explained below). In our consideration of the NielsenOlesen theory we also required the gauge field A^ to behave such that the energy is finite. finite. r^oo 1 dx(9) n r o0 (22. it cannot be deformed continuously into the socalled "classical vacuum" (the latter is a constant configuration in our soliton example). A field configuration with n ^ 0 therefore cannot be deformed continuously into one with winding number zero.i'ldx(0) 9A Thus for (22. connectivity. Classical Field Configurations of the fields which preserve the fmiteness of the energy must preserve the winding number n. the winding number must be a constant of the motion. i. In other . The continuous deformation or distortion of a set of points does not alter the topological structure of the set (e. i.g. no hole. (22.576 CHAPTER 22. it retains this twist or hole under the continuous distortion) and is described as a homeomorphism.g.e. since otherwise the energy integral would behave like f d x f dr dr 2 J r J r lnr. if it has one twist as above or a hole.105) the corrections falling off faster than 1/r. one hole.) remain the same under homeomorphic transformations and are therefore called topological invariants. i. The winding numbers n are such topological invariants. For that reason n is called a "topological invariant?. Thus two homeomorphic spaces have equivalent topological structures. We observe that since time evolution is continuous.
22.<b rd9—4^ = n. with n = 1.e. the vector potential A. However. However. the gauge field. the number of flux quanta 2irh/g being the winding number n. Then we can construct a oneparameter family of maps or transformations which transform </>n (0) continuously into (j>h (6). cannot be pure gauge everywhere.110) . we can go n times around S^. into S i . We can calculate the magnetic flux $ through the region with $ = * rd9A0 = . Therefore. if A = V x .109) A W(0) = e^ W. 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". Ffiu. is also zero at r = oo. i. i.l].106) J g j r d d g Thus the magnetic flux is quantized. for B / 0 . i. (22. going once around S^.105) as «A(I)'AWV*<I> \(x) = J*V\ ( =• VxW)) (22. A field configuration with one loop (or twist). we can consider transformations which describe continuous transformations of a configuration with fixed n. Of course.e. cannot be deformed continuously into one with two loops (n = 2). In terms of the phase factor \{x) = eixie\ we can rewrite the relation (22.107) The fields <f>(r.9 Introduction to Homotopy Classes words.e. (22. the group space of (the phase transformations of) U(l) continuously. Suppose two such configurations are given by O) W(9) = ei&)V\ n fixed. it is determined by a gradient (l/r)d/dd. It follows that the field Fij. (22.105) insure that B ^ O . 9){^ and hence the functions (22. we have B = V x A = curlgradx = 0.9):=t(f>£\9) + (lt)(f)W(9) with te[0. the higher order contributions in (22.108) map the spatial circle S^.
Two curves which can be related to each other in this way. Still on the level of a mathematical text is (in spite of its title) the Argonne National Laboratory Report of G.111) In principle an expression like (22. i.e. (22. time cannot make a field jump from one homotopy class (see below) to another. i.112) are homotopic iff there exists a homotopy connecting (fin (9) and (fin (9).g. (22.113) . 9) is an equivalence relation. Example 22. Steenrod [261] and S. S%. Reuter [77]. to the internal group space of £7(1). 8) = cfinl\9). Hu [134].lQSl^S^ with H(0. Solution: As stated. Dittrich and M. In fact.9) = ^°\0). n' ^ n.9) is called a homotopy (emphasis on the second syllable). it is clear that this is also a homotopy. A brief introduction can be found in the very readable book of W. (b) symmetry and (c) transitivity.111) as a map from the space described by 9. are said to be homotopic (emphasis on the third syllable) to each other and one writes ^(0)^^(9). i. H. Classical Field Configurations which is continuous in the parameter t and is such that H(0. (b) symmetry and (c) transitivity. i. Thus by varying t from 0 to 1 we deform (fin (9) continuously into (fin '(8). We verify these in the following example. H(1. H. 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.3: Homotopy an equivalence relation Verify that the homotopy map H(t.e. A superposition of classical configurations with different winding numbers does not minimize the energy.T. n. these two maps (fin from S^.112) is an equivalence relation — and as such it is defined by the three properties of (a) identity. (fin (9) in the above example. we have to show that the map satisfies the three properties of (a) identity. and the space of t. The map H(t.0) = $\6).e.1]. i.110) could be written down for configurations with different winding numbers. the phase factors of (f)(9). Since the passage of time is continuous.8) itself must be a configuration with definite winding number.9) = <fiW(9) and H(l. e. like <pn (0).e. — S\ > (22.578 CHAPTER 22. But this type of combination is excluded sind H(t. We can consider (22. Thomas [270]. f T h e standard texts on homotopy theory are N.e. the unit interval I = [0. The relation (22.
* 22.0). 0). these groups are then called homotopy groups (cf. since and tf (1. \ M)\ H!(2t. Returning to the maps \(x) : Si ^ S1^ (i. ' .e) tf2(2il.0) for for 0<t<.9) H(0. is isomorphic to) the set of integers Z (positive. . so that H(t. <^(2).e. Two elements of 7ri [f7(l)] which belong to different classes are homotopically inequivalent.e.10 The Fundamental Homotopy Group 579 (a) Evidently the homotopy H(t. 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. which therefore demonstrates that the homotopy relationship is transitive. In some cases the set of homotopy classes possesses group structure (satisfying the group properties). Assume <p( ' ~ (j> and <p( ' ~ <> ' /( for continuous maps <^(0'. 9) = H(l. 0). negative or zero). (c) We can demonstrate the property of transitivity as follows. is such that H(0. Then H(t. 9) := H(l .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 onedimensional sphere S1. 0) = ^ (0) and 6(1.e.t. at t = 1/2 both i?i and H 2 give c/^1'. On the other hand H(0. 9) = H(0. then </>W ~ <t>{0)• We define an H(t. i? is a homotopy connecting </>(°) with <^>(2'. 0) = <£{2). Then there exist homotopies Hi and Hi connecting <j>(0> with (j>W and t^ 1 ' with </>(2) respectively. as they are called. We now define the function H ( f 0 ff . i. U(l)). Hence <f>(0^ ~ <p 2 ).0) = </>(0) is continuous there. Hence the set of all such maps breaks up into equivalence classes or homotopy classes.9) = <f>(0\6) and H(l. Moreover. so that H is a homotopy connecting <jyl>{0) with <f>^>(9). <f>W . 9) connects <pn (0) with itself continuously. (b) Next we check that if 0<°) ~ <p^.9) = . a class being a set of equivalent maps. below). . Thus 7Ti[C/(l)] = vrifS'1] corresponds to (i. (22 1U) At *=2!: But ^ ( ^ ) = ^ i ( M ) = #2(0. i<t<l. 0) = tf2(l. 0) = 4>W (0).22. being the continuous map connecting </>(°) with <j>(1'>. 9) = <f>W (0). we see that of the set of these maps some are equivalent and others are not. Homotopy classes are sometimes also called Chern—Pontryagin classes.ffi(O.
but TTI[S2] = 0: S 1 is a circle. and then introduce some of the related mathematics.) why 7i"i[S2] = 0. to make the result plausible. Appropriate maps. In order to really appreciate how topology comes into the picture here one should understand (e. Classical Field Configurations Thus there is a denumerable infinity of such homotopic classes or sectors. 52. 22. S 2 the surface of a sphere in a threedimensional Euclidean space.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. T (22.580 CHAPTER 22. Sketch of idea demonstrating that nilS1] = Z. 22. 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. T < 9 < 2TT.115) Varying t € [0. which is to be mapped into another circle Si indicated by dashed lines.11 for the special values of t = 1 and t = 1/2 with Xo(0) = to t(2vr 9) for for 0 < 9 < 7T. n=0: t=1: t=1/2: i.11 the circle drawn with the continuous line is the circle S%. p. In the diagrams^ of Fig.g. This trivial map into a single point is described as a "degenerate map" and is illustrated in Fig. Fig. 22. . How can we see this? We first sketch the idea. Rajaraman [235].e.11 The trivial n = 0 map allowing shrinkage to a point.
One can prove that the expression (22.117). 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. of (22.12 The n = 1 and n = 2 maps not allowing shrinkage to a point. i.h. is invariant under infinitesimal (continuous) deformations. 22. 2 the maps are wound around the dashed circle. to the case of winding number zero.118) is an infinitesimal (continuous) real function on the circle with (5f)o=o = (Sf)0=2n 1 See S.s. The maps for these cases are illustrated in Fig.e.e.117) 2^ d6X(x)\1(x). 0 < X(P) < 2vr. (22.s. Coleman [55].h.10 The Fundamental Homotopy Group 581 n=l: cutting of dashed circle n=2: Fig.116) In fact from this we obtain (see below) 2  n and this means r. In the cases n = 1.I** dOJn*W±e {8) X 2vr J0 — / 2Wo dO(iri) dO \ d6 l. d (22.117) in ±.e. We make another observation at this point. i. phase factor or element of U(l)): A(x) : Xn(x) = em*W = [Xl(x)]n. For winding number n we have the mapping (i. (22. 22. .103)) A(x) > \'{x) where 5f(x) = \{x) + i\(x)5f(x) = A(x) + 6\(x).12.22. the transformations^ (the result may be looked at as 8n of expression (22. and hence the winding number.
1 1 See N. the identity class (the group space S2 is simply connected. Hence i /"27r ^ i Thus we have shown explicitly that the winding number remains unchanged under continuous deformations. only one homotopy class.e. Steenrod [261].e. It is intuitively clear that any circle drawn on the surface of a sphere can be shrunk to a point. there is a theorem which says thatH TTg(Sn) = 0 for q < n. 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. .8. 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. Sec. In fact.582 CHAPTER 22. 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. il**( ^') r2Tr A ( 21 9 2 1) ' de\x2 *(«/W). Thus there is only the trivial map into a point. whereas 7ri(50(3)) = Z2 (i. the cyclic group with two elements (since 50(3) is doubly connected).e. i. an integer modulo 2) i. 15. Classical Field Configurations Under this transformation n changes by 5n Now.
and in particular the exponentially small level splittings of the double well potential. 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. finite Euclidean time) classical configurations (which are nontopological). 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. We consider such a theory now with action S((p) depending on the real scalar field 583 . In each of the cases considered we obtained the result by solving the appropriate Schrodinger equation. Together with the infinite Euclidean time limit this requires for the calculation of the tunneling contribution to the ground state energy the FaddeevPopov method for the elimination of the zero mode. 23.2 Instantons and AntiInstantons A onedimensional field theory is a quantum mechanical problem.Chapter 23 P a t h Integrals and Instantons 23.e.1 Introductory Remarks In Chapter 18 we considered anharmonic oscillators and calculated their eigenenergies. 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!). 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.
1.2 = 2m 2 (here m is the parameter in the potential V{4>) of Eq. Path Integrals and Instantons S(<f>) / dt \i>2 . Patrascioiu [110] and Chapter 18. Chapter 6) 1 2 . ! " ^ j = ^ 2 . (23. Fig.* /•tf CHAPTER 23.2) Fig. in particular E. Gildener and A. n = 0. 2 2 4 ™ = ~. 23. note that here the mass of the equivalent classical particle is taken to be mo = 1. 23.2. (23.1 — degenerate minima (positions of classical stability or "perturbation theory vacua") at positions 4>{t) = ±m = ± 0 O . and ^2(0) : m* .1 The double well'potential. "For comparison with literature. i.v{4>) <?VV m4 __ 1 .m ^ + §^24 ' ^2 > a (23J) The potential V(</>) has — cf. 2mo has as the normalized ground state wave function </>o(?) = <<?0) Thus for mo = 1 we have in our notation here a.584 <)>{t).1)) and hence E0 = mh/y/2. The Hamiltonian of the simple harmonic oscillator (cf.e.
2 — and we can deduce from this the boundary conditions._(0) = 0. but there is tunneling between the two minima if the central hump is not infinitely high. must have definite parity as in a physical > situation. These boundary conditions imply nonperturbative contributions to the eigenvalues which yield the splitting of the asymptotically degenerate oscillator . We know from the shape of the potential that the quantum mechanical energy spectrum is entirely discrete (no scattering).e. The states of the system here must be even or odd under x — ±x.23. The corresponding wave functions ip±(x) then must be peaked at the extrema of stability — as in the examples illustrated in Fig. </.x) = x2 V(x). we have the Lagrangian L(x. and so the question is: What sort of boundary conditions do we have to impose on the wave function other than its exponential falloff at 4> — ±oo? We can decide this if we observe that S = J dtL is invariant • under the exchanges 4> ~ * ±</>. We can develop a perturbation theory around either of the minima. 23. as we did in Chapter 18 — but as we saw. i. and in the above 4> corresponds to x. Fig. W(0) = 0. e.g.2 The wave functions of the lowest states. This tunneling affects the eigenvalues. this alone does not yield the level splitting. 23.In mechanics of a particle of mass 1 with ~ position coordinate x(t).2 Instantons and AntiInstantons 585 Classically the position <f> = 0 is a point of instability.
i. For small h we can attempt a stationary phase method. dr m+™ (23. we consider the amplitude for transitions between the vacua or minima at 0/^ — ±</>0 = ±rn/g over a large period of time. (23.9) . is finite if we go to imaginary.e. But quantum mechanically it can tunnel through the hump from one well to the other. (23. r . n) = f with the Euclidean action rTf SE(<i>) = / J TV f V{<f>}exp[SE(<P)/h} (23. that given by Eq.e.5) and iS = dLE d dr[d(d(f)/dT)} i. dr We observe that Eq. d(f> d ' dr2 dr It follows that the equation of motion is now given by 55.tf\ . Euclidean time T = it.0o.7. i.4) J <bi =—(An rTf (ITLE = I JT.2. We are therefore interested in the amplitude — called Feynman amplitude or kernel as we learned in Chapter 21 — (</>o. (23.6) d*l dr 2 .2 m 2 ( l m z (23.7) resembles a Newton equation of motion with reversed sign of the potential as a result of our passage to Euclidean time.7) so that (23. Then (0o. A classical particle starting from rest (0 = 0) at —m/g cannot overcome the central hump and move to +m/g. i.e.586 CHAPTER 23.£'(0) = 0. In the semiclassical path integral procedure that we want to use here.<h. The transition amplitude for this.e.£j — —oo (we assume the normalization factor or metric to > > be handled as explained in Chapter 21). dLE = 0.3) for tf — oo. SE.8) V(4>) + const. Equation (23. (23.U) = J'D{<f>}exp[iS(<l>)/h] (23. Path Integrals and Instantons approximations into an even ground state and an odd first excited state. classically forbidden domains then become "classically" accessible.2 V'(<l>) = .7) is to be solved for 0 with the boundary conditions m 0(r) = — for Tf — oo and 0(r) = > m for n oo.3).  .
is depicted in Fig. Fig. If the constant arising in the integration of the Newtonlike equation is chosen to be zero (see below). — (23. 23. we see that the equation is the same as the classical equation for a static soliton (i. In the present context of Euclidean time the configuration (f)c is generally described as an "instanton". Its trajectory.e.e. from fa — —rn/g. i. 23.3.2 Instantons and AntiInstantons 587 In fact. this means the total energy of the particle in Euclidean motion is zero. timeindependent) in the 1 + 1 dimensional <J>4theory which we considered previously. Thus we can picture the particle as starting from rest at one minimum of V(</>) or maximum of —V(<p). and hence kinetic energy gained is equal to potential energy lost.e. the classical path. and having just enough energy to reach the other extremum at 4>f = m/g where it will again be at rest. i. 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. . Thus we know the solution from there which is the kink solution (22.23.18).3 The instanton path.10) where TO is an integration constant. in the present case (T — To) = m tanh m(r — TO).
13).588 CHAPTER 23.5).r 0 ) ' (23.10) is given by (23. We also observe that the expression is singular for g2 —> 0. We also need to know how we can calculate the difference AE of the energies of the two lowest levels from this expression.TJ T fro. but with the same Euclidean action (23. For a nonzero constant we obtain the periodic instantons discussed in Chapters 25 and 26. This antiinstanton configuration rises monotonically from right to left as shown in Fig.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. (23.12) dimr) cosh m(T — TO) m tanh x tanh x 00 _ 4m*_ (23.14) But this is not yet enough.13) into Eq.4). Using Eq.10) which rises monotonically from left to right is the pseudoparticle configuration called kink in the context of soliton theory and instanton in onedimensional field theory or quantum mechanics with topological charge 1 and Euclidean action given by (23.^ .].4. Tf = oo. 23.* With this choice and inserting mo = j dr J = m we obtain* SE = m T lm4 2 g2 cosh4 m{r .8) we obtain T f SE dr[2V((j)c) + const. we obtain lim f+ OO.' .13). The configuration (23. Inserting the result (23. the constant must be zero.7) admits another configuration which is the negative of (23. (23. Also note that we can reexpress SE in the form 2 SE f0 c (oo) T Joo \ dr j J0 C ( where ^iw = .10) and is called the antikink or antiinstanton configuration and obviously (following the arguments of Chapter 22) carries topological charge — 1. (23.Tj) oc exp 4 m3 m^2 (23.11) For this quantity to be finite for TJ = —oo. Path Integrals and Instantons The Euclidean action of the instanton solution (23.7"/! ). The classical equation (23. 'Note that this integral may be evaluated for finite limits r = ± T and presents no problem in the limit T oo. thus indicating the nonperturbative nature of the expression.
(23.g.4 The antiinstanton.23.15b) has the shape shown in Fig.g. 0c(r) = &(T).TO) + ^ c ( r .5. Questions which arise at this point are: Is the nonmonotonic configuration (23.15a) we have a simple superposition. so correspondingly one defines the instanton or kink with topological charge + 1 and the antiinstanton or antikink with topological charge — 1. from the sketch or from the variation of SE. 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. in the case of (23. Clearly we are here interested in configurations of type (23.15b) In the case of (23. e. and an antiinstanton localized at (say) TQ. Both configurations comunicate between the two wells of the potential at < = —m/g and (ft = m/g.15a) or # ( r ) = 0(rro)^c(rro)0(rir)e(TTi)^c(r^)0(T2r). 23.2 Instantons and AntiInstantons •<l> 589 ^ ^ ^ 0 ^ ^ ^ _ _ ^ Fig.TQ are very far apart their sum differs from an exact solution only by an exponentially small quantity.15b) since we require configurations from vacuum to vacuum.10) we can ask ourselves: Does it make sense to consider the sum of an instanton localized at TQ as indicated in Fig. 23. b u t if TQ.15b) a mending together of an instanton and an antiinstanton at r = T\ with —oo = To < T\ < T2 — 00 for TQ. as we can see e. 23.5. (23.TO far apart and T'Q 3> TQ. / > Looking at the instanton (23.Writing ••• + L ••• and performing the variation in the usual way with 5<j) = 0 at r = ± 0 0 .15b) is not exactly a solution. 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. .15b) a solution of the classical equation? Is it a topological configuration — if so. #(r) = </>C(T . The sum (23.r0).
of course. Thus it belongs to the socalled vacuum sector of solutions and its topological charge (determined by the boundary conditions at plus and minus infinity) is zero.5 Superposition of an instanton and a widely separated antiinstanton. Ignoring exponentially small contributions the Euclidean action of the configuration (23. #(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 . 5^ In order to answer the second question we observe that (23.CHAPTER 23. Finally. at oo the variation is <<> = 0). g cosh m(r — TQ) (23.16) .15a) is not exactly a classical solution. since (23.15a) is a configuration which starts from 4> = —i^/g and ends there. vanishes as T\ —> ±oo (of course. Path Integrals and Instantons '. we see that in the domain [—oo.T 0 ) } ] .15a) can be argued to be zero. m/g / o / o ^ 0 1 \ c'\ o\ T2 ' ^^__ m/g Fig. In fact.r 0 ) At r = Ti the variation Sep is finite but nonzero so that we are left with an exponentially small contribution which. it also does not exactly stationarize the Euclidean action.e x p { . 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 . 23.2 m ( T o . T\] with LECIT = 0.
23.2 Instantons
and
AntiInstantons
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
fTi SE \Joo roo JTx \ / /
rTi
( [ * • • • + [°° •••)LEdr=
( f
\Joo
1
...
f
i— Ti 1
...)LE
J
= 0.
Joo
As a further example we consider a configuration consisting (approximately) of two widely separated instantons (or kinks) and one antiinstanton (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)
+^(rr2)^c(rr(S3))0(T3r) 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 antiinstanton. Again t h e configuration is nonmonotonic 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 antiinstanton 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 "quasistationary 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 timedependent 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(tlt)} EE&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 EE, dE 2^
(0)
^h(x/)^n0)(x)exV[iE(t't)]
n
J
iE(t't)
=
i6(xx')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 onedimensional $ 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 (/>4potential (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 (/)4potential (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>)eMiEn(tt')/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>)eME+(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 == EE+ 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 = e2*§ ± I Then
K{(J)Q,T'\(J)Q,T)
« =
{V(^o)}2exp[JE;o(r'r)]eA^')/2eAE(^)/2 2{V(^o)}2exp[JB0(r,r)/^]sinh^:AJB(T,r), (23.35a)
where we reinserted 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 antiinstanton. 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)+lJdTr]\T)L'E{4>)W
dr
/
00
\ it) ^^ ~ ^
jvi^eMSE^/h]
T){<p} eMSE(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 '•'•oneloop''' 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 timedependent — quantity is £n) V(r) = ^2^nrjn(T),
71=0
(23.45)
"The word "loop" refers to internal integrations which this nonclassical 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
\fJTn
'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+6sech22n2)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
JmT
, 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 Tdependent 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 socalled 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
/
JmT mT sinh 4z
32
dz cosh z
m°
+
sinh 2z
3z
mT mT
+—
ill
m 32
5
AmT
(
We observe how the largeT 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(xx'), 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(xx') 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 nowhere 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 FaddeevPopov 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) Jj partial integration of the right
r)JV(r) I" K ' ' JT
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(TT")+
(T
K(T,T')5(T'
T")CLT'
with
K(T,T')
.
JT
N{r'Y
(23.62)
where we allow for a general r dependence in the Volterra kernel as well. Discretizing the Volterra equation (23.57), i.e. Z(T) = r,(r) + fT K(T,r')r,(T')dT',
or more generally an integral equation of Fredholm type, by setting 5 = Ti+i — T;, i = 0 , 1 , . . . , n with TO = —T,rn — T, we can write the equation
z r
{ n)
= ^2
finiVin)
+
5
5Z
K T
( n>Ti)v(.Ti)
(23.63a)
We observe that K(rn,Ti) = 0 for % > n. The discretization here is a very delicate point. A different and convenient — but not unique — discretization (discussed after Eq. (23.66)) is to rewrite the equation as
n n ..
z T
( n)
=
^2 Snir}(Ti) + 5^2
n r n
K T T
( ^ i)~{v(n)
+
v(ni)}
cn—i
] T SnrtTi)
+  £
K(r,Ti)f,(Ti)
+ ~Y,
K{r,Ti+1)r]{n).
(23.63b)
Equation (23.63a) can be rewritten in the matrix form of a set of simultaneous linear equations: / z(n) \
2(T2)
/
i + i^Cn.Ti)
5K(T2,TI) 5K(T3,T1)
o
1 + <5K(T 2 , SK(T3,T2)
T2)
z( 3)
T
1 + SK(T3,T3)
0 1 +
0 0 0 0
6K(T„,T„)
V{T3)
V 2(r„) /
V
*See also I. M. Gel'fand and A. M. Yaglom [107], after their Eq. (3.4).
n) K(Tj. p.65) The quantity D is really the Fredholm determinant which is obtained as above by solving the Fredholm integral equation with constant integration limits.. 2! . Martin [43].TI) 0 K(T2. we obtain exactly the nonvanishing terms on the right of Eq. Felsager [91]. . (23. . This case has been considered in the literature^ where it is shown that if the Volterra transformation is considered as a Fredholm transformation whose kernel vanishes on one side of the discontinuity or diagonal. Thus the factor a inserted above has to be taken as 1/2. This determinant can be expanded in powers of K (more precisely in powers of a parameter A which we attach to K). n )d? (23. i. the kernels to the left and to the right would be associated with step functions 8(T — r') and 9(T' — T ) .4 Field Fluctuations 607 This equation represents a set of n linear equations which has a unique solution provided the determinant Dn of the coefficient matrix on the right does not vanish. Thus D„ = l + 5 ^ K ( r i .63a).T')dT' (23. X(TI. r ' ) is to be evaluated at r = T ' .66) •Dr) .T1)dT1 + J dn J dr2.. = 1 52 K(n.TJ) The factorials arise since every determinant of I rows and I columns appears l\ times when i.Ti) i=l : exp \ £ K(T.62) — and with insertion of a discretization dependent factor a still to be explained — we have in the present case D: . whereas the determinant arises only once in the original determinant Dn. 192.N{T) Finally we add the following remarks concerning the discretization (23.67) i=l v ) = lim exp V In ( 1 + ' L K(T.64). the kernel K(r. [44]. the value of the determinant is half (or arithmetic mean value) along the diagonal. The Fredholm kernel is in general — as in our illustration above with a Green's function — a quantity with a discontinuity between the two constant integration limits.exp • /_:T JV(T')" dr' = e x p [ . H.n) o K(TJ. as in Eq. Cameron and W. (23.64) D (23.T2) I exp I K(n. We observe that if we discretize Eq.n) z=l x (23. n — oo the expansion becomes > D lim Dn = 1 + n—*oo J K(r1. (23. T. are summed over all values from 1 to n.e. In the limit S —* 0 .n) i=l N lim exp £*(T.63b). Our final step is to relate the VolterraFredholm determinant to the required functional determinant.lnJV(T)}] N(T) N(T) (23.TI) K(T2. Thus in our case one of the kernels is zero and we have to decide what we do when.23. With this specific discretization the Jacobi matrix becomes diagonal and hence we obtain with the determinant coming exclusively from the diagonal* T>z ~Vri and then Vz Dr) = lim exp In TT ( 1 + K(T.64) as above and consider the determinant of the expression.68) D above. where a f 1 See R.7 Y.64) Inserting the explicit expression for the kernel given in Eq. (23..a { l n i V ( r ) . so that (with T *T) 1/2 •Dz N(T) (23. ^This is the method explained by B. r i ) + ..j.
J dz(T) J T>{z} J da•Dz hJ dz{T) v{z} I f_A\^+W)i{T)} J H § I"exp [ .72) The endpoint integration dz(T) = dz(rn+i) can be combined with this.L dT{ \ (i(r) exp T>z 1 N(T) + 1>n exp N(T)\2 ' N(T) J a2 N2(T) 2 JV 2 (r) We perform the functional integration as in Chapter 21.3: Implementation of the endpoint constraints Using a Lagrange multiplier a insert into the functional integral (23. Eq. (23. t h e range of T is subdivided into n + 1 equal elements of length e with (n + l)e = 2T and TO = —T. T + N(T) / T' = T JT dr' N(T') .^ . (23. N(T') .N(T) .57)) z(—T) —> 0) ^)/>'£^') = iv(r)/_T/T'A(__l_)2(T0 T JV(T') T . I2 and using induction one obtains (cf.2o) 2 rc^O.608 CHAPTER 23. Briefly. f(T) = N{T) J^ dr'^^z(r') (23. so that (with V^N(T)/N(n)) 7t = J V{z}^[JT_Tdr[\[z(r)+ia^)' — lim n—*oo. Evaluating Ii.0.(T) + /_ dr — .22)) In = (7T6)"/2 Vn + 1 exp e(n+l) ( Z n + l .l + £7t) «i (23. In = dzi dz2 • • • dzn e x p . Eq.Z ( T ). Solution: From Eq. Thus 0 = 2 (rf) + f(rf) or 0 = z{T) + f(T). represents a constant translation. Path Integrals and Instantons Example 23. (23. (23. (23.70) and this now requires also integration over the endpoint coordinate z(T).61) we obtain the constraint on the function Z(T) which results from the endpoint constraint JJ(T^) = rj(T) = 0 on the fluctuations T){T).59) the endpoint conditions »)(±T) = 0 and evaluate the integral. In every integration / dzi the contribution €7. rn+i = T. I0 = — f dz(T) f 1>{z} f da T'Z x exp T>r\ «/>5^> + ix(*<T> + '"< T >/• T *>m™ 7V2(T') (23. n J=.69) Incorporating this condition into the functional integral Jo with a delta function 1 = / dz(T)S[z(T) we have (cf.Z i . and hence may be ignored. .( . With partial integration (in the second step) we obtain (since (cf.59)) + f(T)] = — f dz(T) f°° dae^W+f™.71) Hence (with h = 1) l = j .e—*0 In.73) . (21.— .
3 T h e Faddeev—Popov constraint insertion The FaddeevPopov constraint insertion implies the complete removal of the zero mode. D. Eq.12).£Q) — aTo]dTod£o. / — / / We started off with V(T) = ^CnVniT).11). J co § dr Recall that / f ^ doce^ = yfVfp." The method consists essentially in replacing the single integration J d£o by a double integration involving a delta function. / d£o <9r0 5[F(TQ. (23.e. Faddeev and V. 11 L. This result verifies Eq. £n = ~2 dTT](T)7]n(l (23. we are left with§ Io 1 1 2TT Vz oo / Vz dee exp oo * r 2 rT 2 dT^n N*(T) L T [N^NiT)]1'2 /2TTh I.4 Field Fluctuations 609 In the required limit the argument of the exponential is proportional to 1/T and thus vanishes in the limit T = (n + l)e — oo. Gildener and A. (23.10) and (21.23.4. dxv 0(r) = (f>c(r) + V(r) and set 1 TOO a — const. (21. Popov [89].75) or more generally a threedimensional surface integration j dau by the volume integration f cftx through the relation dF(x) 6[F(x) — otTo]d x. j / dTT]n(T)r]m(l r . N. . ^E.76) Since the zero mode T]O(T) is proportional to d(f>c{r)/dT with — SE. i.74) where we reinserted h. (3.. a = const. 1/2 T W 2 (r)J (23. Patrascioiu [110]. Ignoring the remaining metric factor with the reasoning explained > between Eqs.59) and agrees with that in the literature.^ 23.
in the direction of the zero mode. / dTm{r)n{T).e. (23. this constraint implies that we go to the subspace which is orthogonal to the zero mode.610 this means that CHAPTER 23. in the case of the soliton of the static 1 + 1 dimensional theory this is the kink coordinate.e. (23. We now impose (i. Then cn = ~2 / dr(pc(T)rjn(T). (23. i.81) 2 / dr»7o(r)0c(r) = . The delta function 8[F(TQ. e.81) the rdependence is integrated out and hence the integral does not provide the function F(TO.76) we set 4>c{r) = ^CnVnir). in the direction T]O(T) at the point TQ of the parameter r which parametrizes the trajectory of the classical configuration </>c(r). demand validity of) the constraint £o = 0. i. (23.75) enforces the collective coordinate ro to be given by F(ro.77) The set {r/ n (r)} constitutes a basis of the fluctuation space about the classical configuration. The number £o is the component of the fluctuation rj(r).g. .82) This condition says that (j)(r) is not to possess any component in the direction of T)O(T). Path Integrals and Instantons _^ ^ w . at r = ro. Thus.75) requires.78) <t>ij) = Y.^n{r).£o)/a. In the integral of Eq. We consider the following integral / ( r 0 ) := .79) (23. where L = in + cn. We establish the function ^(To>£o) with the following reasoning. i /*oo (23. (23.83) **The collective coordinate in the present context is also called "instanton time". A position ro of this classical configuration is called the collective coordinate** of (the field) 4>{T). In analogy to (23. if the set {Vn(T~)} spans the Hilbert space of fluctuation states about 0C (at collective coordinate position r ) .J drm(r)Mr + T0) (23.£Q) — aro] contained in Eq.^O) with dependence on ro which the replacement (23.80) i r or •i /*oo dr7?o(r)0(r) = 0. (23.e.
p. (23.23.86) = Jdrrjoir^Mr F(TO.4 Field Fluctuations 611 for an infinitesimal translational shift TO of the position of 4>c(r). 158.y/SET0 = 0.<t>c{r)} .89) Here the factor App is called the FaddeevPopov determinant.87) l/M) (23.90) Now tt 0^C(T + TQ) ro=0 <9T0 dr (23. .91) " F o r more discussion of this point and related aspects see also B.£O) F(r0^0) we have defined as + ro) + T?(T)] = / ( T 0 ) + p£o.85) PJ W i t h Eq.88) ro0 Then in Eq.84) (23. (23..^) .y/S^ro] (23.Zo)y/SETQ We now define A FP '• =0 • (23. jdrrj0(T)4>c(r) + JdTm(r)(p'c(T) X/SET0.(23. We have AFP = /F(T0. We evaluate this quantity as follows in the leading approximation. .£O) OTQ = = £ " . Felsager [91].J dTrj0(T)[(t>c(T + TO) + ??(T)] i r /• I a rfT77o(T)—0C(T + TO) + sr — / dTrj(T)r)o(i # C ( T ) ./ drri0(T)<pc(T) + PJ (23. (23.82) we can replace here 7?O(T)0C(T) by 77O(T)T7(T). so t h a t with F(TQ. Expanding /(TO) about TQ = 0 we obtain (with (J>'C{T) = T)Q(T)\J13E/'p) /(TO) = = Hence / dTT]Q(T)[(j)c(T + To) .75) 1 = AFP f dT06[{F{rQ.
/(TO) ^/SE~T0} V~SET0 c d£. Thus the result is that the integration over the tangential or zero mode component £0 is replaced by an integration over the collective coordinate 7 o multiplied by the FaddeevPopov determinant. ~ There is a quick way to obtain the result (23. iytQ 0C(TO + dr0)  4>C{TQ) = dT0^'c(T0) = dr0 7 0 (r 0 ). using the relation (23.e. i.612 CHAPTER 23. 0 d r (23.e. (23. 770(7"). i.93)./ dT0AFP. =~ (23.0AFP8 4o H P .93) where AT = TT — 00.94) ' ' T h i s is an important result which will be required later in the explicit evaluation of path integrals.92) We find therefore that the value of the FaddeevPopov determinant is given by the square root of twice the classical kinetic energy or simply of SE^ The relation (23. if ^ + d£o = 0.90).75) now becomes d£0AFP8[f(T0) = = / dr0 / J&T J + p£0 . fs d(/)c{ro) = We have also drj(T0) = d£0r]o(To) + ^2d£ir)i{To). if the coefficient of 770(70) vanishes. .91) into Eq. Inserting (23. Consider. in d(j)c(To) + dr/fa).e. P J AT (23. the "static energy" invariant. P the relative sign being of no significance. The factor p is a normalization factor which we > retain here only for ease of comparison with literature and may otherwise simply be put equal to one. Path Integrals and Instantons Thus a shift TQ in the direction of the zero mode (i.e. i.77) between the zero mode d(^ c (r)/dr and the fluctuation vector in its direction. we obtain (remembering that % = pd(j>c/dT/y/S^) AFP = 4 = f dr ( ^ V \/!5E JT \ dr * = yfe. tangential to (J)C(T)) leaves this and hence the square of this quantity. 7 Thus there is no fluctuation along the direction of the zero mode in the sum.
This very important intermediate result is the contribution of a single instanton or kink configuration to the p a t h integral (ii has been defined by (23.4 Field 23.60a) in not involving the zero mode contributions.35a) since the dependence on Tf — Ti = 2T of both expressions differs significantly. (23.97) is a constant still to be evaluated. We determine I'Q by observing t h a t this differs from IQ of Eq.60a) for I0 and Eq.43)).98) . (23.52a) and replace in this J d£o by the expression (23. (23.55b). so t h a t h = det r)/(n\ppATexp 4 m3' ATAFPdet(^ \ a£ jexp (23. We have to take into account also multiinstanton configurations which almost stationarize the action integral and satisfy the same boundary conditions as the instanton.23. (23.4 Fluctuations T h e single instanton contribution 613 We now return to the multiple integral (23.96) does not yet allow a comparison with (23. we can set I 2TT ^o = l'o\ wlp2'' which with Eq.A T A F P i " o e x p T—>oo p where I'o n^O. We observe t h a t the result (23.95) P Here £o 2 in the exponential is multiplied by WQ which is very small for large T and approaches zero in the limit of T —>• oo.93).4.96) lim . Since w\ is for finite T given by WQ of Eq.55b) implies I'o w0p m \ ' LOQP '2K irh) m \ ' irh) 4:\/6mp mAT (23. We can now perform the Gaussian integrations and write the result lim T^oo p 4 m3 exp 3^J (23. The source of lack of similarity is attributed to the fact t h a t the contribution of a single instanton to the p a t h integral is not enough. (23.
. Gildener and A.* Finally we rewrite the expression (23.55b)) w0 = v0e VQ 23. the result becomes h lim 2 m r ^ ^ 1 e . in the present case we also have to include multiinstanton configurations in the form of one instanton plus any number of instantonantiinstanton pairs which describe back and forth tunneling between the two wells.5 I n s t a n t o n .101) 2T > oo (cf. (23.100) We observe that the dependence on the normalization constant p cancels out. Although we choose p2 = 1. (23. their Eq. ^See E.4. they have to introduce the inverse of that factor on the far right of their expression (3. *For the validity of the calculation the instantons or kinks have to be far apart. Patrascioiu [110].32). the approximation is then known as the "dilute gas approximation".20).2 m T e x p 2 irgh ' T»oo (23. In view of their normalization of the normal modes. (3.614 CHAPTER 23.102) *For comparison with the parameters of E.i n s t a n t o n contributions As already mentioned.100) in such a way that it exhibits the crucial factors of which it is composed.e. i.4>C exp h (23.TiMiy+'Mr «>. ~ rKo')e{Tx . 0c M. T0 < 4l) <C Ti « 42) < • • • «C T 2 n + 1 = T — oo (23. Thus — the subscript 1 indicating that it is the Feynman amplitude for the oneinstanton contribution — lim where with A T ATAFPI0 W0 SE{.99) limlATAFF/0f^^eA^exp 4 m3 3 / ^ Inserting the explicit expressions for IQ and App. Patrascioiu [110] make the replacements \QP <> 2g 2 and p?GP <» 2m 2 .a n t i .r). Gildener and A. we drag the factor p2 along for ease of comparison with literature. This means we are concerned with configurations like* 2n+l d(r . Path Integrals and Instantons It follows that the kernel becomes* h = Urn ATAFPI0( T^oo p W^TT ^L exp 4 m3"1 3^h 4 m6 .
r/l&. 2 .(l>i) (23. the total Feynman amplitude is given by (^/.17b) and there perform first the integrations over the intermediate configurations whose endpoints are not fixed.4>i.Tl\(j)i.fo) = Id2 St SE(<Pf. .23.4 Field Fluctuations 615 for n = 1.Tf\(p2. 23. . r0(3) = 2T/3. T2 = T/3. 7 i = .<j>i) .105) where we write 'proportional' since only the classical instanton action depends on the endpoint configurations.Tl}{4>l.2 T / 3 . here (f>2. Taking these configurations into account. 1 0 6 a ) and SE{(J>I.102) distinguishes between instanton (rising from left to right) and antiinstanton (rising from right to left).103) For instance for n = 1 one has the configuration <jyc (T) which has the form shown in Fig.H)2 + .<j>i) + 2 d<& 1 r)2 *? k? + (<f>2 .r 0 ( 1 ) = . (3) (23.(r) = ^ 2 n + 1 ) ( r ) + 77(r) with r ^ ) « 0.106b) SE{<t>l.Ti} OO J — OO /OO OO OC OO d(f>2 J — OO LeSE{<l>f.<t>ff + h . this means we have to consider the amplitude (23.4>l)eSE{4>l. (n = 0 implies the single instanton solution).T2\(f)l. (23. (23. . Since these endpoint configurations. r 0 ( 2 ) = 0.6. (23. The factor ( .l ) i + 1 in the sum of Eq.<t>2)eSE{<t>2.T2}((t)2. ( 2 3 .<Pi)2 + • • • .r / 3 .<t>f) + 2 d</>{ SE{4>fi4>i) + ld2SE 2 d(/){ ( h . Consider (with T 0 = T. Proceeding as before we now consider first in analogy with Eq.104) Considering the case n — 1 of a kink with one additional kinkantikink pair. we can expand the Euclidean actions here involved as follows (the first derivative vanishing at a solution): SE{<j>f. are varied.36) (/.4>I) ld2SE SE(c/>i.Ti) = X^/' T /I&> r i> ( 2 n+1) n=0 (23. T3 = T) d(f>l((pf.
106a).e. in the case of l2n+i therefore ry2n.107) with (the evaluation is given below) 7T 1/2 7 = IT 1/2 1/2 li&SE/dfi) {d^SE/dcft) 2m J (23. of course.616 CHAPTER 23.105) can in other respects be treated in analogy with the single instanton amplitude. each of which implies a multiplicative factor 7. The allowed positions of the individual kinks have to be such that their order is maintained. d2SE{4>) d(j)2 2m. the dilute gas approximation.2d2SE ~c 4>i d(f)i e x p SE^f.2J. Thus when in each of the three cases the zero mode elimination has been performed with the help of the FaddeevPopov method. each of the three intermediate kernels contained in Eq.4>f) d4>\ 4>s72 (23.4>i) / d(f>2 e x p {<h . It has to be remembered.3 9 L 3m 2 and 9SE((/)) ± .e. i.108) where the last expression is obtained as follows. that each applies only to a fraction of the overall Euclidean time interval of r = 2T. Path Integrals and Instantons where the first expression in Eq. (23. rn ^2 J2H 9 1m' ±2gcj).4>i) { 2 " 2d SE .12) we have JM~°°) i.(01 .. i.e. It follows that — always within our approximations — c (^TflfaTjW SE{<t>f. i. (23. On the assumption that the kinks of the instantons are widely separated. the . ±m/g (23. With our remark after Eq.106b) is to be substituted into the second relation of (23.109) The result of these integrations is that the amplitude l2n+i involves 2n intermediate integrations. dT 9 70(oo) V m J sE(4>) = ± m Hence d2SE{<t>) dej)2 .e. (23.
4 Field Fluctuations 617 replacement (23.V ^0 SE{4>C X exp mAT lim — e ~mAT Sinh T^oo 7 IjATAppIo vo exp h (23.(*) o ax = 2 T / 3 .6 the collective (i) coordinate TQ has to be to the right of TQ . we now have to integrate over three collective coordinates. Am3 2m J rW ( r ( 0 and this followed by setting aj. (23.e. Gildener and A.A 3 r>oo 7 ' 3! exp SE(<J>C) T6 VQ mAT FP10 3 h (23. (3) ° JrW: 2T/3 (2T)3 .r/^. 23.(2) (1) (2) (3) _ ( 2 T ) 3 JT JT JT 3! (23.35a) which contains AE in the argument of sinh and obtain for the level splitting A J resulting from consideration of the instanton plus any number of instantonantiinstanton pairs A*£ = 2n7AFP/0^=exp 2TT 1/2 4 m3 m V2 / m \ 1/2 4 v / g ^ exp ( 2TT irh _±rrv^ 2h ^With shifts r. Patrascioiu JT/3 ° ir< >2T/3 1 dr.r.)(3) 1 o A T 3 Ao lim .105) t h a t each of the three amplitudes contributes a factor e x p [ .2 T / 3 this product becomes that given by E.23. and TQ to the right of TQ This implies t h a t in this example we have the product of integrals^ rp (3) (2) . In the case of (<^f. a 2 0.7 3 ^ — .a 3 [110].Tj)( 3 ) these considerations imply the following result parallel to t h a t of the single instanton calculation (23.93) has been applied.T/0j. i.111) Finally we sum over any number of instantonantiinstanton pair configurations and obtain for the Feynman amplitude oo 1 oo 1 ( {JATAFPIO 2n+l I = lim V n=0 J 2 n + i = lim .101): (^/.110) Finally we have to recall from Eq.^ .m A . then sign reversals TQ i ) .112) We now compare this result with expression (23. T ] with ] T 3 = 1 A » T = A T = 2T. and in the case depicted in Fig.
. MiillerKirsten [30]. We see that the method — above we included a lot of highly nontrivial details which are passed over in other literature — is considerably more complicated than the method of the Schrodinger equation. We shall see that the path integral applied to excited states is even more complicated in necessitating the evaluation of elliptic integrals.618 and hence AE A. (23. we obtain AE = 2A* and agreement with the result (18. AE.' We see therefore that the classical Euclidean configurations 4>c can be considered to be responsible for the quantum mechanical tunneling which gives rise to the level splitting AE. 23.113) for qo = 1: A1E „ .113). MiillerKirsten [163]. Gildener and A. Patrascioiu [110]. Liang and H. W.W We also see that the level splitting is a nonperturbative expression (with an essential singularity at g2 = 0). (23. The difference between levels. We observe one more aspect of our calculation. J. W. K. J. by either replacing A T by 2AT or doubling A*. W. — the relation between the parameters is ^LMK = ^I^GP = 4 r " 2 a n < i C\MK ~ \^<3P = 92. Path Integrals and Instantons 8V2hm5/2 = exp ( 4m3\ . has been obtained without explicit use of canonical commutation relations (i. . 2 V TT [ i ( g „ _ i ) ] ! V 9 J V 3 9 HJ For an elementary discussion see S. To help the comparison we note that — with indices indicating the authors. explicit quantization). Finally we realize that the computation of (23. here as 1) is also taken into account.e. whereas the method of the Schrodinger equation in this case is of the same degree of complication as for the ground state. (18. Achuthan. J.With this correspondence A\E of Eq.182) (with mass = 1 and H = 1) becomes the following which implies twice the result (23.^r . l~2h = 2qo+2\ = m /4m3Vo/2 / —*exp 2 P 4 m3 \ — .113) This is (in slightly different notation) the result of Gildener and Patrascioiu [110] for consideration of the instanton plus associated pairs. MiillerKirsten and A.182) of the Schrodinger equation method provided the difference in the mass of the particle (there taken as 1/2. J. H. l CHAPTER 23. ^See P.113) is very complicated.Q. Taking similarly into account the antiinstanton (since the physical level splitting does not distinguish between an instanton and its antiinstanton). Bose and H. Wiedemann [4].5 Concluding Remarks In this chapter we used the path integral to obtain the level splitting of the ground state of a particle in the double well potential. even without explicit specification of the prefactor. GP for those of E. In fact the equivalent to writing down canonical commutation relations is the evaluation of the Gaussian integrals which enter the prefactor of the expression in Eq.
in a somewhat different formulation. In order to really appreciate the difference between topological and nontopological configurations (the solitons and instantons we considered previously being of the first kind) we begin with a recapitulation of some crucial aspects of the former. however. Previously we considered in particular the quartic or symmetric double well potential for a scalar cp.Chapter 24 P a t h Integrals and Bounces on a Line 24.e.1 Introductory Remarks In this chapter we are concerned with nontopological classical configurations on R1. ™=£('^° < 2 «> Here the two symmetrical minima of the potential are called vacua (classical or perturbation theory vacua as they are sometimes called because ordinary RayleighSchrodinger perturbation theory would be performed around these points). 619 (24. Potentials of this type play a significant role for instance in models of the early inflationary (expanding) universe (as a result of its cooling).a 2 ) 2 + c^T2.2) . i. Consider for instance the temperaturedependent potential yr(0) = i A ( 0 2 . Since problems of instability are ubiquitous and occur in all areas of microscopic physics their understanding is of paramount importance. and their interpretation and uses. We shall see that these classical configurations are associated with quantum mechanical instability and hence with complex eigenvalues.
Path Integrals and Bounces on R1 Here dVT{4>) vanishes for . J 4. A2 _ a — \„2\ u ) ~r u/\y> 2 (24. but implies instability for T < Tc.a2)4> + 2cTz 2 0. its symmetry under .e. 24. We see that <fi = 0 minimizes VT(</>) for temperatures T > TC. T = 2\(<f>2 . > / V[<H 'true'vacuum 'false'vacuum Fig.620 CHAPTER 24. a spontaneous symmetry breaking leads to the double well (i. We can destroy the reflection symmetry of V((f)) (i.1 The asymmetric double well.3) Since d2VT{4>) 2 d^ we see that ~ ^ ' > (d2VT(<f>)\ V # V # 2 2 A=o 2 2 = 2(cT 2 .2=a cT /\ Thus there is a critical temperature T c := (X/c)l'2a. = 4(ACT2 cT 2 ).e.A a 2 ) . Thus if we start with the system (like the early universe) at a high temperature and then allow the system to cool. at the minimum of the potential the case < = 0 changes to the case 0 ^ 0 ) .
On the other hand we shall also see that (a) bounces describe vacuumtoinfinity transitions. False vacua play an important role in models of the inflationary universe* and elsewhere.e. We recapitulate briefly a few points from our earlier detailed treatment of (1 + l)dimensional soliton theory. the quantum mechanical vacuum). Guth and S. .) .24.^J >o. L. M. see e. and (c) determine iQE of E which results from a nonvanishing probability of tunneling away from the classical vacuum only.g. Cornalba. the vacuum being understood as the classical. S. (24. tf(*) =^ 1 . <44 2) ' This is now an asymmetric double well potential as in Fig.1 with (24.g.t In the following we consider first (after some recapitulations) a simple model of a false vacuum. also in contexts of string theory. . The Lagrangian density of the theory is given by C = dp*d»*U(*).1 Introductory Remarks 621 exchanges 4> <* — <P) by adding e. the higher minimum is called a "false vacuum". We calculate the classical configuration — which in this case is called a "bounce" — which corresponds to the soliton or instanton in our previous chapters (but is not topological) and we use this in order to calculate the lifetime of the unstable state (or equivalently the imaginary part of the energy). A. perturbationtheory vacuum). Pi [126]. V(<j> = m/g) = 2e. Costa and J.5) We see that one minimum of V now lies higher than the other. (b) they are minimum configurations of the energy E (instantons of the Euclidean action SE). (24. and (c) determine the lowering of the energy to the true quantum mechanical ground state (i. T False vacua and bounces occur. Penedones [59] and references cited there.g. for instance. Since the true vacuum minimizes the potential. We shall see that: (a) solitons (or equivalently instantons in the onedimensional theory) describe vacuumtovacuum transitions. which (b) are saddle point configurations of E (or SE). a linear term as in w^^'A1^'*0V(<l> = m/g) = 0. 24.6) The EulerLagrange equation is given by • $ + */'($) = 0 (metric + .7) *See e. H.Y.
Thus dx J —c 1 k'\2 {4>'cy + c/(0 c (24. so that the EulerLagrange equation becomes the Newtonlike equation <t>"{x) = U'(</>).e. We have seen previously that the same equation is also obtained from the second variational derivative of the energy functional.13) For small w\ > 0.7) yields the following equation called stability or small fluctuation equation. We observe that this kink solution is odd in (x — XQ).t) dt 0 are written (f>(x).12) Since <fic(x) solves Eq. h " 52E(<j>) 8(j){x)5(j>{y) +u {Mx)) = ip {x) k wlipk(x).8) This equation possesses the kink solution <t>c{x) m tanhm(a.e. which is an important point as we shall see. dx2 + u"(<j>) 5(x y). i.622 CHAPTER 24. Path Integrals and Bounces on R1 Static fields given by d$(x.14) . — XQ) (24.9) 9 for the doublewell potential given above. small perturbations around </>c(x) do not grow exponentially with t (the zero eigenvalue requires special attention). The energy of this classical configuration is obtained by evaluating the Hamiltonian for (f) replaced by <j>c. (24. we see that (24.11) We investigate the classical stability of the configuration by setting in the full t—dependent equation for small ipk: iw $ ( s . (24. i. (24.10) Inserting (pc and evaluating the integral we obtain E&c) Am6 3ff 2" (24. t) = <f>c(x) + Y^ iPk{x)ekt (24. (24.8).
In the case of the double well potential (24. eigenvalues w\ > 0 (actually the stability is only local in the sense that E is not minimized in the direction (in Hilbert space) of the eigenfunction associated with the zero eigenvalue. i.1) the fluctuation equation becomes 6m 2 ipk(x) = wkipk(x) (24.2 The inverted potential.e. kink. . 3m 2 and w\ > Am2.oo \ x== + CX) Fig.17) .1 Introductory Remarks 623 For socalled "classical stability" the differential operator has to be semipositive definite.16) V[<H x= oo 0 d2 .15) cosh mx From our study of the solutions of this equation in Sec.8 we know that this equation possesses two discrete eigenvalues (one being the eigenvalue zero. (24. o 2 + 47TT dx / \ J x=+ oo \ d 4>c/dx 1 ' 0 x = . thus the zero mode here is given by d . i. d \m .e. i. underneath the particle's velocity. 22. . 1 —<pc = ~r~ —tanh mix — xo) oc dx dx [ g cosh m(x — XQ) (24. configuration (for our </>4theory). the other a positive quantity) and a continuum which are given respectively by w\ — 0. We have seen that general considerations tell us that the zero mode is given by the generator of translations applied to the classical. 24.24.e. the zero mode).
Then the energy satisfies f°° . Hence in this case there must be a lower state with a negative eigenvalue and an even (ground state) wave function.20) The number 1 on the right is the socalled topological charge which is obtained as follows. We see that this pseudoparticle starts from rest at one peak of — V at x = — oo and in losing potential energy gains just enough kinetic energy to reach the second peak at x = +00. 4 TTT.2.e. where it is then again at rest. Below we shall see that the configuration described as a bounce is given by an even function of x. and — clearly — we can resort to a classical interpretation of this function.e. In this case the usual Newton equation reappears with an opposite sign of the potential. The kink solution can now be looked at as the classical solution of a onedimensional field theory. Path Integrals and Bounces on R1 We observe that this zero mode is an even function of (x — XQ) in agreement with our expectations for the ground state wave function. which has the solution 4>c(x). Integrating the classical equation (24. i. 4>"{x) = U'(4>). Before we consider such a theory in more detail we recall the topological aspects of a soliton or instanton. 24. quantum mechanics. we obtain the BogomoVnyi inequality (<t>'{x) = V2U)2 > 0. (24.18) This inequality is saturated by the first order BogomoVnyi equation </>'(x) = ±V2U. and also represents a nodeless even wave function — in complete agreement with our expectations for a quantum mechanical ground state wave function. The classical configuration is then called a pseudoparticle configuration or instanton. i. F (24.8).21) The charge is defined as the space integral of the time component of the current. The associated zero mode represents the velocity of the pseudoparticle as indicated in Fig. so that its derivative representing the pseudoparticle's velocity i