•
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
Thus this first part . but also about complex energies that he encounters in a parallel course on nuclear physics. its mathematical foundations. like periodic potentials. Thus important level splitting formulas for periodic and anharmonic oscillator potentials (i.e. 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. one problem being that there are few nontrivial models which permit a deeper insight into their connection. in Feynman diagrams in quantum electrodynamics). However.XVI illustrative examples in any treatise on quantum mechanics and — excepting various kinds of square well and rectangular barrier potentials — leave the student sometimes puzzled about other potentials that he encounters soon thereafter. With various techniques and deeper studies. has not always been understood well. 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. With the growing importance of models in statistical mechanics and in field theory. basic postulates and standard applications. in fact. any problem more complicated is frequently dispensed with by referring to cumbersome perturbation methods. Excluding spin. Chapters 1 to 14. and it will be seen in the final chapter that potentials with degenerate vacua are not exclusively of general interest. again point the way: For scattering problems the path integral seems particularly convenient.g. we recapitulate the origin of quantum mechanics. These basic cases will be dealt with in detail by both methods in this text. screened Coulomb potentials and maybe singular potentials. numerous problems could. the aforementioned exactly solvable cases. with degenerate vacua) were first and more easily derived from the Schrodinger equation. that is the Coulomb potential and the harmonic oscillator. In the first part. but arise also in recently studied models of large spins. The introduction to quantum mechanics we attempt here could be subdivided into essentially four consecutive parts. whereas for the calculation of discrete eigenvalues the Schrodinger equation. 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. 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. be treated to a considerable degree of satisfaction perturbatively. even though the expansions are mostly asymptotic. To what extent the two methods are actually equivalent.
The most prominent examples here are the double well potential and its inverted form. i. 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. the method of matched asymptotic expansions with boundary conditions (the latter providing the socalled nonperturbative effects). and their eigenvalues. which is presumably the only such singular case permitting complete solution and was achieved only recently. 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). 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. We also consider inverted double wells and calculate with the path integral the imaginary part of the energy (or decay width). the chapter thereafter deals with Yukawa potentials. The potentials with degenerate minima will be seen to reappear throughout the text. we derive respectively the levelsplitting formula and the imaginary energy part for these cases for arbitrary states. we deal mostly with applications depending on perturbation theory. 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. 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. Using perturbation theory. and the elliptic or Lame potential — here introduced earlier as a generaliza . This is followed by the important case of the cosine or Mathieu potential for which the perturbation method was originally developed. Thereafter the concepts of instantons. Chapters 15 to 20.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. In the second part. periodic instantons. After a treatment of power potentials. 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. The earlier Chapter 17 also contains a brief description of a similar treatment of the elliptic or Lame potential. 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 Chapter then deals with Schrodinger potentials which represent essentially anharmonic oscillators.e.
the Mathieu equation. 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. The physical behaviour there (in the transition region between quantum and thermal physics) is no longer controlled by the Schrodinger equation. and whenever available also with the results of WKB calculations. This puts Schrodinger equations with e. however. We then illustrate the relevance of this in the method of collective coordinates. this comparison on a transparent level being one of the main aims of this text. The application of path integrals to the same problems with the same aims is seen to involve a number of subtle steps. for instance. it is. anharmonic oscillator potentials on a comparable level with. 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. With a fully systematic perturbation method and with applied boundary conditions. In fact. This method is therefore more complicated. 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. except that these are no longer of hypergeometric type. from our precise reference to unavoidable elliptic integrals taken from Tables).g. 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. such as limiting procedures. The introduction of collective coordinates of classical configurations and the fluctuations about these leads to constraints. the Schrodinger equation can be solved for practically any potential in complete analogy to wellknown differential equations of mathematical physics. The particular solutions and eigenvalues of interest in physics are — as a rule — those which are asymptotic expansions.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. Employing anharmonic oscillator and periodic potentials and reobtaining these in the context of a simple spin model. for instance. quartic. 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. cubic). All results are compared with those obtained by perturbation theory. we can make the following observations. an approximation whose higher . Comparing the Schrodinger equation method with that of the path integral as applied to identical or similar problems. Our fourth and final part therefore deals with elementary aspects of the quantization of systems with constraints as introduced by Dirac.
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. whose research into asymptotic expansions laid the ground for detailed explorations into perturbation theory and large order behaviour. other topics have been left out which are usually found in books on quantum mechanics (and can be looked up there). Watson [283]. D. W. D. Throughout the text some calculations which require special attention. formulas taken from Tables or elsewhere are referred to by number and/or page number in the source. are relegated to separate subsections which — lacking a better name — we refer to as Examples.XIX order contributions are difficult to obtain. Their deep involvement in the attempt described here is evident from the cited bibliography. Tchrakian (Dublin) and Jianzu Zhang (Shanghai). The author has to thank several of his colleagues for their highly devoted collaboration in this latter part of the work over many years. thereafter University of St. Park (Masan). T. we also consider at various points of the text comparisons with WKB approximations. Whittaker and G.g. 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. Since this comparison was the guideline in writing the text. This endeavour developed into an unforeseen task leading to periodic instantons and the exploration of quantumclassical transitions. with the source given in the bibliography at the end. which is particularly important in the case of elliptic integrals which require a relative ordering of integration limits and parameter domains. e. in particular Professors J. Dingle (then University of Western Australia. Whittaker and Watson [283]. Andrews). an additional motivation was that a sufficient understanding of the more complicated of these problems had been achieved only in recent years. For ease of reading. also for the verification of results.g. The author is deeply indebted to his onetime supervisor Professor R. at the end of a chapter (after troublesome turning of pages). Instead a glance at a nearby footnote provides the reader immediately the names of authors.* H. J. MiillerKirsten *In the running text references are cited like e. not the least for permitting a more detailed and hopefully comprehensible presentation here. Liang (Taiyuan). . B. Nonetheless. The line of thinking underlying this text grew out of the author's association with Professor R.Q. K. B. so that the reader is spared difficult and considerably timeconsuming searches in a source (and besides. like E. H.g. the references referred to are never cited by mere numbers which have to be identified e. N. shows him that each such formula here has been properly looked up). As a rule. as well as applications and illustrations. 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.
that the formula he had established for the energy distribution of electromagnetic black body radiation was in agreement with the experimentally confirmed Wien.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.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).1 Origin and Discovery of Quantum Mechanics The observation made by Planck towards the end of 1900.Chapter 1 Introduction 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. Kirchhoff's law in thermodynamics says that in the case of equilibrium. the amount of radiation absorbed by a body is equal to the amount the body 1 . We do not enter here into detailed considerations of Planck. which involved also thermodynamics and statistical mechanics (in the sense of Boltzmann's statistical interpretation of entropy). Although practically every book on quantum mechanics refers at the beginning to Planck's discovery. Thermal radiation (with wavelengths A ~ 10~ 5 to 10 . Instead. Planck had arrived at his formula with the assumption of a distribution of a countable number of infinitely many oscillators. very few explain in this context what he really did in view of involvement with statistical mechanics. we want to single out the vital aspect which can be considered as the discovery of quantum mechanics.
l ) where x = ^ = ^ . h = 6.e. (1..e. y J c 3 \ex .2 CHAPTER 1.626 x 10 .38 x 1(T 23 J K'1. T) were known and tested experimentally. kT kXT (1. The (equilibrium) radiation of the black body can be determined experimentally by sending radiation into a cavity surrounded by a heat bath at temperature T. of the photons or photon gas) in the cavity with both possible directions of polarization (hence the factor "2") in the frequency domain v. and then measuring the increase in temperature of the heat bath.2) and the . Introduction emits. v + dv in equilibrium with the black body at temperature T. Let us look at the final result of Planck. u(u. In Eq. i.3 4 J s. The parameters k and h are the constants of Boltzmann and Planck: k = 1. (1. and the other in the region of large A (or AT).T) = 2*?£(?)kT.T) = dv3eC2U/T.e.e. T)du is the mean energy density (i. A being the wavelength of the radiation.1) ' y Here u(v.1 Absorption in a cavity. energy per unit volume) of the radiation (i. These expressions are: (1) Wien's law. i. the formula (to be explained) u(u. How did Planck arrive at the expression (1. Fig.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. 1. radiators.1) c is the velocity of light with c = u\. Black bodies as good absorbers are therefore also good emitters. It was found that one expression agreed well with observations in the region of small A (or AT).
2) and (1.T) 2^^kT. but — here the discreteness appears properly — only in elements (quanta) e. and he had succeeded in finding such an expression of the form u(v. To this end he considered Boltzmann's formula S — klnW for the entropy S. so that W represents the number of possible ways of distributing the number P := NU/e of energyquanta ("photons". exp(x) ~ 1+x) and "large" (exp(—x) < 1).47TZ/ 2 {x small).1 Origin and Discovery of Quantum Mechanics (2) RayleighJeans law: u(i>. in the first place Planck had tried to find an expression linking both.T) = av e6"/Ti' where a and b are constants.1) in regions of a. C3 being constants. Planck now imagined a number T of oscillators V or iV oscillating degrees of freedom.T) = 2^C3T. that the formulas (1. Indeed. We see. where the Fig. he searched for a derivation. discretization begins to enter. When Planck had found this expression. (1.2 Distributing quanta (dots) among oscillators (boxes).e. which are indistinguishable) among the N indistinguishable oscillators at . T) u(i/. Moreover Planck assumed that these oscillators do not absorb or emit energy continuously. e xhv. (x large). and thus over a discrete number of admissible states. This is the point. every oscillator corresponding to an eigenmode or eigenvibration or standing wave in the cavity and with mean energy U.3) Ci. 1. we obtain: u(i/. C2.3) are contained in Eq. Here W is a number which determines the distribution of the energy among a discrete number of objects. (1. "small" (i. . 3 (1.1. Considering Eq.1) as approximations.
U{T) being the average energy emitted by one oscillator. 1. p.g. We visualize the iV oscillators as boxes separated by N — 1 walls. Agreement with Eq.3 Comparing the polarization modes with those of a 2dimensional oscillator. u(i>. as in most other Tables.1)!P! (1. and the second law of thermodynamics ((dS/dU)v Example 1. The Stirling formula or approximation will appear frequently in later chapters. i.6) Fig. v + dv. W. Magnus and F.T)dv. with riydu — 2 x —w—dv. N * oo.4 CHAPTER 1. e = his. S. h = const.343(2).7) *See e. by multiplying U with the number nvdv of modes or oscillators per unit volume with frequency v in the interval v.1) u = vmrri (L5) as the mean energy emitted or absorbed by an oscillator (corresponding to the classical expression of 2 x kT/2. . as for small values of e). e. We now obtain the energy density of the radiation. 1. I. M.g. (1.e. = 1/T). with the quanta represented schematically by dots as indicated in Fig.4) IniV! ~ JVlniViV + O(0). Then W is given by w = (N With the help of Stirling's formula* {N + piy. i.2) requires that e ex is. (1.3. 940. formula 8.2. p. one obtains (cf. there not called Stirling's formula.e. Oberhettinger [181]. Introduction temperature T. (1. Gradshteyn and I. . Ryzhik [122].
UJ2/C2 + K? = 0. dj\l dM . We observe that u(v. L for i = 1.3. rii = 1. (1.. '''From the equation I \ JW . where^ 2 [2KUY .3.1)) The solutions of this equation are ^max = 4. In terms of A we have u(X.7)..1. KT = I J .V 2 ) E = 0. as in electrodynamics. d ™4*±\IL> — —dv = nvdv dv dv _8 3 \ c / 14 8 2 4TTV2 = 3 dv 83 ^ ^ = ^ ^ ' as claimed in Eq. so that . 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 = 0.1 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. so that the derivative of u implies (x as in Eq.1.2. .2. We obtain therefore u^T) = Unv = 2^fJ^—i.8) This is Planck's formula (1..7) for instance..UJ = 2KV.965 and xmin = 0. i.3 (as for ideal conductors).i = 1.T) has a maximum which follows from du/dX = 0 (with c = vX). v + dv.2. so that (lvL\A 0 I I = rr. We obtain the expression (1.3. 2 2 2 r2 L K — 7T n .T)dX = ^ehc/*kT_idX. Then L^j = nrii. .. The number with frequency v in the interval v. (1.1).. is given by . nvdv per unit volume. The number of possible modes (states) is equal to the volume of the spherical octant (where n^ > 0) in the space of n^.e. (1.
i. x — oo) the mean energy vanishes (0 < U < * > oo). However.10) Thus here the socalled zero point energy appears.l. .e. Lorentz and Planck that Eq.1. and from which the constant h can be determined from the known value of k. One might suppose now. that it is easier to consider first the case of T = 0.e. which did not arise in Planck's consideration of 1900. which had also been known before Planck's discovery. This is Wien's displacement law. . Introduction he 4. n = 0. 2 . . which can not be eliminated without a different approach.2 (1. We shall see later that in the case of this linear harmonic oscillator the energies En are given by En= (n + jhu= U + I W h=—. Thus we have a rather complicated system here.6 The first value yields AmaxT = CHAPTER 1. which can assume the discrete energies en — nhv. Since temperature originates through contact with other oscillators. of course. If an oscillator with thermal weight or occupation probability exp(—nx) can assume only discrete energies en = nhu.10).965K = Const. that of an oscillation system at absolute temperature T ^ 0.8) could be derived much more easily in the context of statistical mechanics.—v <L9) We observe that for T — 0 (i. such a procedure leads to contradictions. . One expects. We therefore examine such contradictions next. but one can expect an analogy. 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 . We are not dealing with the linear harmonic oscillator familiar from mechanics here. that we arrive at quantum mechanics simply by discretizing the energy and thus by postulating — following Planck — for the harmonic oscillator the expression (1. we then have at T = 0 independent oscillators. . ra = 0. A. Later it was realized by H. the behaviour of the system at zero absolute temperature. (1.
e.ln(TV ..In . d . 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.2 Contradicting Discretization: Uncertainties The farreaching consequences of Planck's quantization hypothesis were recognized only later. so that 1 f ds\ T — \dUjy . {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) .kT This means U is then the classical expression resulting from the mean kinetic energy per degree of freedom.i n .k dU 1+U\ f ln(l T + U\  U U k ( e . 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".1.1 ' u~ e . we obtain T V S =fc[ln(TV+ P . with Heisenberg's discovery of the uncertainty relation. 1. kT/2.1)! . . In the following we attempt to incorporate the above discretizations into classical considerations* and consider for this reason socalled thought experiments (from German "Gedankenexperimente").1 Solution: Inserting W into Boltzmann's formula and using In TV! ~ A In T — TV. i. 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. around 1926.= .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 e \U e u = exp(e/fcT) which for e/kT — 0 becomes > .2 Contradicting Discretization: Uncertainties 7 Example 1.1)! .
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
2) If we have only one degree of freedom (i = 1).p% are therefore again observables. the equation of motion of the observable u. chapter VIII. Compared with an arbitrary function f(qi.t). which can be observed directly at time t. chapter VII. All functions u(qi. Dirac [75]. x^fdu. which all together describe the state of the system.1).P. 6t>0 5t Real quantities which are directly observable are called observables. It was only with the development of quantum mechanics by Heisenberg and Dirac that Poisson brackets gained widespread interest in modern physics. We can therefore consider Eq.g.^ ^ j .2) as This equation is.24 CHAPTER 2. One now defines as (nonrelativistic) Poisson bracket the expression^ With this definition we can rewrite Eq. ^As H.Pi. one obtains the Hamilton equations* • OH . A system consisting of several mass points is therefore described by a number of such variables. .. (2.pi) of qi. Goldstein [114] remarks at the end of his chapter VIII. (2. M. du \ x^/dudH du dH\ .1).1). One can verify readily that Eq. The following properties can be verified: *See e. dH In this Hamilton formalism it is wrong to consider the momentum pi as mqi. qi(t + 6t) .3). the standard reference for the application of Poisson brackets is the book of P. since . this expression is simply a functional determinant. . (2. (2.e. H. whereas the velocity requires observations of space coordinates at different times.qi(t) qi = hm f . It suggests itself therefore to consider more closely the properties of the symbols (2. Rather.Pi) is contained implicitly in the canonical variables q^ and pi. the entire timedependence of observables u(qi.1) as d .) = £ [wm + WiK) = £ [WiWi . (2. The total time derivative of u can therefore be rewritten with the help of Eqs.n n. Hamiltonian Mechanics is wellknown. in analogy with Eqs. Goldstein [114].4) contains as special cases the Hamilton Eqs.4) as the generalization of Eqs. (2. A. S »(«. i. (2. Pi has to be considered as an independent quantity. as mass times velocity. (2.
(2.5d) is useful in calculations. we expand A and B in powers of qi and pi and apply the above rules until only the fundamental brackets remain. Since Eqs. (2) linearity: {A.Pk} = 0. in other words without any reference to the original definition (2. it is irrelevant whether we write {A. Bx} + a2{A.B}C + B{A. B}.5a) (2. These are {Qi. B} is completely evaluated. If we evaluate the Poisson brackets for qi. {qi. but could be multiplied by a complex number): {A. A}} + {C. 25 (2. The original definition of the Poisson bracket will not .Pi. where A and B are arbitrary observables. that of the linear harmonic oscillator.6) We can now show. can be solved solely with the help of the properties of Poisson brackets and the fundamental Poisson brackets (2. i.Qk} = 0. {C.B}.5b) (3) complex conjugation (note: observables are real. (2. (2.2 The Hamilton Formalism (1) Antisymmetry: {A.5c) The first three properties are readily seen to hold.5d) = {A*. B}} = 0.5e) = {A.BC} (5) Jacobi identity: {A. the Poisson bracket {A. (2.6) give the values of these. we obtain the fundamental Poisson brackets. for example.A}. like here. which combines the Hamilton equations.4).B}C or C{A. {B.B*}. C}} + {B. (2. As an example we consider a case we shall encounter again and again. B2}. Later we shall consider noncommuting quantities.5d) above. we wish to evaluate {A. {A. then the ordering is taken as in (2.3) of the Poisson bracket.6). that the very general Eq.2.C}.B} = {B.e. {pi.Pk} = 5ik.B}* (4) product formation: {A. (2. As long as we are concerned with commuting quantities. a i S i + a2B2} = ax{A. If. Property (2.
\(p2 + q2)} = l({q.6). from which we infer that q(t) — qocost + posint. "4' = q.7) into (2.yPot3 + •••• (2.. (2.qot2 .10) we deduce q = p = q. or q(t) = qo+ Pot .10) In classical mechanics one studies also canonical transformations. q = {q.8b) p=q.. we consider as Hamiltonian the function H(q.p) is an ordinary function is also irrelevant. (2.p.11b) (2. Hamiltonian Mechanics be used at all. These are transformations qi—>Qi = Qi(q. (2.t).8a) and use the properties of the Poisson bracket and Eqs. Since constants are also irrelevant in this context. and so q = q.26 CHAPTER 2.H}.8a) (2.. Pi—> Pi = Pi(q. and Pi are ordinary real number variables and that H(q. and P={p.p.11a) 'q' = q. (2. q + q = o.p}p + p{q..H}. (2.8b) We insert (2. From Eqs.p) = ±(p2 + q2). (2.q2}) = = 2i{q. (2.P2} + {q.4) we have for u = q.t).p.7) q = [q. In the evaluation one should also note that the fact that g. According to Eq.9) and (2.p\) P. Then we obtain: (2. (2.12) .9) Similarly we obtain from Eq.
B}q. (2. i. Eq. Example 2. verify that the Poisson bracket of two observables A and B is invariant. {PhPk} = 0.3) we can now express the Poisson bracket {^4.1: Canonical invariance of Poisson bracket Assuming the invariance of the fundamental Poisson brackets under canonical transformations Qj = Qj(Q'P)>Pj = Pj{Q>P).B}Q>P.1 we verify only Eq.Qk} = 0. which means that a Hamilton function K(Q.qk} = 0.p <> Q. those at a time t — 0. this transformation is described precisely by the equations of motion but we shall not consider this in more detail here.14) With the help of the definition (2.P.13) We write the reversal of the transformation (2.B}q.15a) °raS {AiB}Q. Of course.Pk} = 0.e. i. B} of two observables A and B in terms of either set of canonical variables. Example 2.e.Pi back to their constant initial values. . *H.e.12) Qi—>qi = qi{Q.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. that (dropping the subscripts therefore) {Pi. {QhPk} = Sik.t).e. i. {Qi.pk} = 5ik. Goldstein [114]. Pi^Pi=Pi(Q. The proof requires the invariance of the fundamental Poisson brackets under canonical transformations.* Hence in Example 2. {qi.P. In classical mechanics we learned yet another important aspect of the Hamilton formalism: We can inquire about that particular canonical transformation.P is canonical in the sense defined above.t).2. _ax p__dK_ (2. (2. (2. (2. i.15a). .2 below contains a further illustration of the use of Poisson brackets.15a). i.P provided the transformation q. {qi. for which Hamilton's equations hold.P One can then show that {A. P) exists. as {A. (2.3 deals with the relativistic extension. and Example 2.2 The Hamilton Formalism 27 for which the new coordinates are also canonical. chapter VIII.e.p = {A. which transforms qi.
Thus we extend q and p to spacetime vectors (t. Mittelstaedt [197].P^pr.15b) dA {Qk. we have —Et. § E x a m p l e 2.B}q. (2. (2. The relativistic Poisson bracket (subscript r) therefore becomes du dF dq dp du dF dp dq du &F _ du dF ' {u.E(t). their product Et — qp being relativistically invariant. OQk Pk}q.p) .3: Relativistic Poisson brackets By extending qi. 1 „ H = p + m0gq and solve the canonical equations with Poisson brackets for initial conditions q(0) = qo. we obtain analogously dA dQk' Inserting both of these results into the first equation. Solution: Relativistically we have to treat space and time on an equal footing. we obtain as claimed by Eq. dA {Qk.p^— + {A.q/c) and (E.Qj}q.28 CHAPTER 2. V V °Qk dPk dPk dQk J E x a m p l e 2.A}q. 236. {A. 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 „ „•. § See P.pt to fourvectors (in a (1 + 3)dimensional space) define relativistic Poisson brackets.s F 5 7 r ={A.B}QIP.( 9A = {Pk.A}q.F}r ~di d~E ~ d~E~8t Consider F = H(q. p.+ dQ. Qk}q.15a) r„ ™ v . .p^ "Pk.p(0) = poSolution: The solution can be looked up in the literature.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.Pj}q.pc). we obtain dA {Qk. Replacing here A by Q and B by A. Thus whenever q and p are multiplied.p = dB dA dB \ rA 1 {A'B^ E b r s r .p^v dP.p = J2 d~Fk Replacing in the above A by P and B by A.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).
.Po. (1. 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. it is the equations of motion which lead from Go(q. we have A = (p apaq — J . Probabilities Single particle consideration We continue to consider classical mechanics in which the canonical coordinates qi. i.. numerically zero. du . . Let us assume that at some initial time to the system may be found in a domain Go(q. to ._.p) to G\{q. and E as a function of t).t). one obtains Gi with q = q((lo.y.to.p). since H is expressed as a function of q and p.3 Liouville Equation.1 Liouville Equation. i. and the space spanned by the entire set of canonical coordinates is described as its phase space. Probabilities 29 (This is.dt y/l  qZ/c2' 2. of course.p).po is a point on Go. But now we consider a system whose phase space coordinates are not known precisely. Then . du q\ P = —.po) = Gi(q.t).e.e. This probability is evidently proportional to AqAp. We consider first the a priori weighting or a priori probability. .)•'=(*)**£. and at time t > to in a domain G\(q. I . H(q. dudH s l u .14) and area A of the phase space ellipse given by Eq. in the case of the linear oscillator with energy E given by Eq.p) . c dt 2 2 w^ du du dr .3 2.16) For example..po.3. so that Go(qo. dE . p= p(qo. g. Instead we assume a case in which we know only that the system is located in some particular domain of phase space. (1. also boundary points of one domain are mapped into boundary domains of the other.E(t)}r = Hence at Relativistically we really should have clu/dr.Pi are the coordinates of some mass point m.t0.. where dr is the difference of proper time given by dudH 1 du dE(t) du — = 1 du . (*. if qo. (2.. We distinguish in the following between two kinds of probabilities.p). 2TTE UJ and hence g oc — . rrr .p) around some point qo. Of course.17a). but partial derivatives of F do not vanish. g oc AqAp. Since Hamilton's equations give a continuous map of one domain onto another.2.Po.' ±=j1V c \dtj' u^.
— = — Aq.1): d . Show that the total volume of phase space covered for constant E is 8n2IE. Solution: Integrating over the angles we have =2TT fe=7r 2frEsin.5 thereafter provides an illustration of the a priori weighting expressed in terms of energy E.14) t o (1. in view of this independence it can be expressed in terms of the conserved energy E. which means. ? « + •sin 2 6>/' 21E 2IEsin2e' in spherical polar coordinates.30 CHAPTER 2. dt dq dq dp dp d2H dqdp d2H = 0. this has the same value at a time to. Hence from the difference: — ./.15) — an ellipse of area § dpgdp.. (1. as is demonstrated by Liouville's theorem in Example 2. Thus g must be independent of t. Hamiltonian Mechanics If g depended on time t it would be dynamical and would involve known information about the particle.5: A priori weighting of a molecule If the rotational energy of a diatomic molecule with moment of inertia / is 1 / 2 . and hence g oc %ir2IdE. — dp A A — In(AqAp)s = — + — = Example 2. dq Aq dq 2 dq Aq dq 2 qH to the right and q to the left.4: Liouville's theorem Show that A q A p is independent of time i.eded<t> = 8TT2IE. Je=0 . dt dq and with with Hamilton's equations (2.p^. Pi 2 / \. the (pg. dpdq and similarly dt = — Ap.4. 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. = 2TrIEsm6.)curve for constant E and $ is — as may be seen by comparison with Eqs. Example 2. Example 2. as at a time t'0 ^ to • Solution: We consider dln(A9AP)rf(A<Z) ' .
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.20) .*). d d 1 P = JJdqidpi.3.e. i. 2.p. (2. Since W has the dimension of a reciprocal action.p. The total number of systems N is obtained by integrating over the whole of phase space. whose positions in phase space are characterized by points.q + dq.t)dqdp = N.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. it is suggestive to introduce a factor 2KK with every pair dpdq without. by jr = P(9»P.1 The system moving from domain Go to domain G\. i.t)dqdp= 1.t).p.3 Liouville Equation. Probabilities 2.19) Thus W is the probability to find the system at time t at q.Thus we assume a large number of identical sytems.p + dp.2. (2. W=p(q.p. Thus dn is that number of systems which at time t are contained in the domain q. however.18) With a suitable normalization we can write this / W(q.Po. (2.e. leaving the basis of classical mechanics! Hence we set / «w)^ =. (2. dn — p(q.p.17) F ^GT^^ 0 Fig.
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.p) denote the velocities in directions q and p — p{q.t)dp{ jt Q. i. 2.a t dt + q+dq.p. p + dp G P O q qndq Fig. * q Fig.3 and establish an equation for the change of the number of points or systems in G in the time interval dt. 2.3 The region G.p. We are now interested in how n ov W changes in time. we take into account.e. how the system moves about in phase space. i. The number of points at time t + dt in domain G.P dt p(q + dq. we consider the domain G in Fig. — if vq(q.p. The equation of motion for n or W is the socalled Liouville equation. Hamiltonian Mechanics We can consider 2irh as a unit of area in (here the (1 + l)dimensional) phase space. t + dt)dqdp.p / .2 The ensemble in phase space.32 CHAPTER 2.e. that in our consideration no additional points are created or destroyed. t)dqdp p(q. p(q.p.t)dp( — I . t + dt)dqdp — p(q. i. In order to derive this equation.p.p) and vp(q. In doing this. 2.
p. the probable motion of the system under consideration.t) dt {H(q.p(q. vp(q. so that dt ~~dq\P Hence dp) + dp\P ~dq)~ ~~dq~~dp~ + dp~~dq~ ~ { .P.p(q + dq.t)vq(q + dq.p(q.p.t)vp(q. t)dqdp = p(q. 33 Dividing both sides by dqdpdt this becomes p(q.21) we can also write dW(q.19).20) and (2.p)=p = 3H —.P *' % = iH>P} ( 2 . (2.t)^ = 1.t)vp(q.3 Liouville Equation.p) . p. (2. (2.t)vq(q.22) The generalization to n degrees of freedom is evident: The volume element of phase space is .p + dp)dtdq.t)} with JW(q.21) with Eq.P)].t)vp(q.p.p. .p.p) . (2. Comparison of Eq.p)dtdp .p + dp. dH Mq. Probabilities and thus p(q.p.p). .p) = q = g^.t + dt) .p(q + dq.t + dt)dqdp — p(q. (2. = K(9.p.21 ) This is the Liouville equation which describes the motion of the ensemble or. t)vq(q + dq.p(q.4) shows that p and u satisfy very similar equations. .p) dq p(q.p.2.t)vp(q.t)vq{q. put differently. .p)dtdp +p(q.p.t) dt p(q.p + dp) dp = or  However.p + dp.P)]K(Q. With Eqs.W(q.p)dtdq . p.p.
e. (2.e.4).24) implies that p is a constant in time. We shall see that we have two possibilities for this. We now inquire about the time variation of the expectation value (it) of u. Hamiltonian Mechanics where is the probability for the system to be at time t in the volume q.26) we described the time variation of the observable u(q.4). Example 2. and hence that equal phase space volumes contain the same number of systems. With Eq. The first and most immediate possibility is — as indicated . i.p.4 Expectation Values of Observables Let u = u(q. p+dp.that the density or probability W(q.p).po are the initial values of q. and this means — since these systems are contained in a finite part V of phase space — that dt We have in particular. Thus in Fig.t)(^J. p. (2. q+dq.24).1 the area Go is equal to the area G\. (2. Equation (2.24) since the total derivative is made up of precisely the partial derivatives contained in Eq.p) be an observable. i. We define as expectation value of u(q. t) depends explicitly on time t (if determined at a fixed . 2. (.p)W(q. (2. for i<«>!/«<*p>"w>(^)".34 CHAPTER 2. We deduce from the Liouville equation the important consequence that ^ M = 0.p.p (cf. since no systems are created or destroyed. that dtj\y v if qo.p) the following expression: (u)=Ju(q. 2.
Po.t).f(q0. W is the density in the neighbourhood of a given point in phase space and has an implicit dependence on time t.t).t). (2.p0. (2.Po.p) assumes certain values) that depends explicitly on time. (2.t)}(^y.34) S e e also H. Reversing Eq. Sec. Goldstein [114]. 1 (2.f(qo. Observables 35 point in phase space). .2. Thus we can write.t)W0(q0.Po.p) = u(q.0) = W0(q0.p. since W oc p is constant in time according to Eq.t).p.p). (2.p. where we used Eq.t).28) = Ju(q. we can also employ a more complicated consideration.22).24): W(q. ie.31) i.e.po.p) = uo{qo.p.t). We verify this claim as follows. Then Eq. (2.0) = u0{qo.^ Solving the equations of motion for q.p. and hence u(q.t).t). With these expressions we obtain for the expectation value (U)Q: (u)o = Ju0(qo. (2. (2.po. we can express these in terms of their initial values qo. we have Qo = g(q. We expect.p0)(^^J.p.0) =u(g(qo.p){H(q. at t = 0.Po. 8.Po) at time t = 0.27) becomes !<»>  /«(*P>!"W>(^)" (2.po.4 Expectation Values.p.t) W(q0. and the time variation d(u)/dt is attributed to the fact that it is this probability (that u(q. of course. (2.po.32) In this expression the time t is contained explicitly in the observable u(q. so that Q = g(qo.t). that (u) = (u)0.W(q.i).33) po = f(q. (2.Po.30) p = f(qo.29) The distribution of the canonical variables is given by W(q.8.t).29). However.t) = = W(g(q0.po.
t).39) Substituting this into Eq. d dt {U)0 = dt.t) dp M du\ ° q 'P=f(qo. (2.36) = (2.f(q.POit)i p=f(qo.t) v{g(<j(q.p0) = W(q.31)) W0(q0.t)J(q. (2.Po.37) and (2.35).p.0) u(q.p.37) as had to be shown. p = f(<j(q.t) d t _ (du\ \ d<i dq / P=f(qo.po)(~ .38) we obtain £<«)„ = .t). (cf.u(q.) f / u (q0.p).t).f(q.p.po.p).t)\—^\ = (u).t).p.30): duo(qQ. (240) .P0.t)W "(q0.=9(. (2.f({H(q.p)W{q.po.25) into Eq.Po.Po.t).f(q.29) q = g(g(q.p. (2.t) = (2 30) (2.t).p.P0. (2. we obtain («)o = / u{q.36) and (2. Hamiltonian Mechanics so that on the other hand with Eq. (2.(dq0dp0\n We deal with the partial derivative with the help of Eq. (2. (2.p)})q=g(QO.M 27Th i .) Wo(qo.t).p. (2.t) . Eq.p.t) M f 9u\ V °P / P=f(q0. Inserting Eqs.t) dB__ (du\ °P \°PS P=f(qo.30) Moreover. (2.p.(2.t)." V " " ™0' > • • » « » 0 .o.t).P.p. (2. .P)}). we obtain an expression which is different from that in Eq.Po)(^^)n.t) 0H_ 0 q =  {{H(q.t).t).\„.35) Inserting these expressions into no we obtain uo(qo.t).p.p).32).32).u{q.Po.f(g(q.30) u0(g(q.P.36 CHAPTER 2. Taking now the total time derivative of (it)o.po.f(q. i.t) (2.28). Eq.0)=u(q.t).e.PO. [ du0{q0.t). / ^fdq0dpc .
In the first case.28). x r „. and the time variation of (u(q.p).p.28). lim ——uW — 0. the observable u is treated as a function u(q. (2. we obtain {H.t)} .45) the Liouville equation. 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. from the properties of the Poisson bracket.p)}W(q. (2.42) the relation d .43) ——uW = 0.t) of u assuming certain values q and p. .2.t)}.p).41) This expression contains {H.u(q. W}..34) and use (2.W(q..31).45) in agreement with Eq. However.(2.p).44) iji^ioo dpi (which is reasonable since the density vanishes at infinity).28) and (2.. Alternatively we could deduce from Eqs.41) and (2.p.t) dqdp\n 2irhJ (2. (2.p.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. u}W + u{H. This timedependence is described by the Liouville equation dW(q. The phasespace integral of a Poisson bracket like I (2.p. Consider {H.4 Expectation Values.42) wxtf)" ~dHd(uW) dqi dpi vanishes under certain conditions.W} in Eq.25) and (2. uW} = {H.t) dt {H(q. ^ „. we obtain zero after partial integration of I and hence from Eqs.W(q. u}W instead of u{H.p){H(q.p)) is attributed to the probability W(q.p. dt <«)o v f u(q. Observables 37 Here we perform the transformation (2. (2. . / d q d p \ n 2irh J (2. described as "Schrodinger picture".p). so that ~d~t <«>0 = J{H(q.
F(«. Hamiltonian Mechanics In the other case.po are constant initial values — we have du0(q0.po. .5 Extension beyond Classical Mechanics With the above considerations we achieved a general formulation of classical mechanics.H(q. " This means: Only in the Heisenberg picture dW/dt = 0.. the probability of the initial values Wo(qo. " T h e author learned this approach from lectures of H.4) of an observable on the other. called u Heisenberg picture". which are described as "Schrodinger picture" and "Heisenberg picture"— all this on a purely classical level but with the use of canonical transformations." and the explicit timedependence is transferred into the correspondingly transformed observables Uo(qo. The time dependence of the expectation values can be dealt with in two different ways.p) It = {u(q. 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.rt. The equation of motion is then that of an observable. (cf.46) We thus also recognize the connection between the Liouville equation. Eq.t) dt du(q. and the equation of motion (2.p0.po) is assumed. and probabilities W.p)}. (b) As usual a muliplication by a complex number is defined. but does satisfy the usual associative and distributive laws. the reason being that — since qo.p).e. i.** 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. (2. which.38 CHAPTER 2. (c) A multiplication of the quantities among themselves is defined. Koppe [152] at the university of Munich around 1964.t) dt __ ~ (2J30) ( 4) du0(qo.t). as we observed. This formulation deals with observables u representing physical quantities.4)) ^ = {«(. 2. does not have to be commutative.p)}.po. as the equation of motion of an ensemble or of a probability distribution on the one hand. (2. which describe the state of a system. for which the axioms of a commutative group apply.
for a phasespace integral we define the word or symbol "trace". (b) TV(cm + j3u) = oTV u + /3Tr v.p} = 0 • — ^ ~[q.2.48): (a) TV u* = (Tr «)* = TV^. (d) Tr(ut. ox In view of our later considerations. it is helpful to introduce already at this stage some additional terminology.49) With these considerations we have reviewed aspects of classical particle mechanics in as close an approach to quantum mechanics as seems possible.p} = l {q.26)) (u) = Tr(wW). The quantity corresponding to W in quantum mechanics is the socalled "statistical operator'''.q}=0. also called "density matrix". We shall then interpret as "canonical quantization" the procedure which allocates to each of the fundamental Poisson brackets (2. Eq. Moreover.q} = 0 {p. That this is the case. (2. Introducing it here assumes.5 Extension beyond Classical Mechanics 39 Quantities satisfying these properties define a linear algebra. (c) TV (u*u) > 0.p] = l. Thus we can write the expectation value of an observable u (cf.48) In matrix theory the symbol "trace" has a welldefined meaning. which all apply to (2. so that we consider this next. can be deduced from the following characteristic properties of a trace. ~[q.6) a socalled "commutator''' [A.p]=0(247) One verifies readily that the commutator relations are satisfied by the differential operator representations qx > x. and to be able to correlate these with the above classical considerations. if u ^ 0.p)(^y. (2. (2. therefore.) = TV (vu). that its use here implies the essential properties it has in matrix theory. y:\p. In Chapter 9 we attempt a corresponding approach for classical systems with . also written "TV". px > in—. by Traceu:=Ju(q. B] := AB — BA in the following way: 'i {q.
a constraint) has to be taken into account.40 CHAPTER 2. . and. electrodynamics. Thus we can now proceed to prepare the ground for the extension of classical mechanics into an operator formulation. However. it will be shown that in electrodynamics.e. i. Hamiltonian Mechanics a wavelike nature. the Poisson brackets require modification to Dirac brackets. excepting the Poisson brackets. since gauge fixing (i. 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.
= {ipi} is called a linear vector space on the set of numbers I 6 {C}.Chapter 3 Mathematical Foundations of Q u a n t u m Mechanics 3. 3. A set M. which turn out to be those of the theory today known as quantum mechanics. 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. of measurable quantities. i. Building upon this. i.e. 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. we can define the Hilbert space as the space of states of a physical system.2 Hilbert Spaces We first recapitulate some fundamental concepts of linear algebra and begin with the axioms defining a linear vector space. 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. quantum mechanics: The Hilbert space as the space of state vectors representing the states of a physical system. In this chapter we therefore introduce important basic mathematical concepts of this noncnumber mechanics. if the elements ipi of M satisfy the usual axioms of addition and K 41 . and selfadjoint operators in this space as representatives of observables.e. We found that the Poisson algebra permits extensions to noncnumber formulations.
~ip2) '• M.ip2. 2 . if every vector ip E M can be associated with numbers Q . ^ i = 0 if ^ = 0. (0 : null element and with complex numbers a and j3: a(tpi + tpj) = a{Wi) lipi = = < M). (ipi + ipj) + ijjk = ipi + (ipj + ipk). . .i = 1. . In each case n linearly independent vectors are said to form a basis. with the properties (a* G IK): (^2. if numbers eti.  ..1) aipi + aipj. n. and n is the smallest such number.2) Vectors ip\. —»• IK. .ipn are said to be linearly independent.^) called inner product.4) The vector space . (V>. (a. 2 . E (3.e.^1) = (01. i=l (3. are linearly dependent.3) If all on = 0. fa.aiV>i + a 2 ^ 2 ) = ai(ip. (3 5c) + a2{ip. n is called the dimension of .tp2.5b) .ip2). ./?eK). .5a) (3. Mathematical Foundations of Quantum Mechanics multiplication by complex numbers.M is said to be a metric vector space or a preHilbert space. If n + 1 elements ^ € M. such that n ^ = YJCi^i.M. • • • . (ipi. so that n 5 > ^ = 0. x M. . f > 0 if V T ^ O .02)* (hermiticity). ( n (3. (3.ipn are said to be linearly dependent. i = 1.42 CHAPTER 3. (3. i. (aptyi. . A + tpj = tpj + i'i. the vectors il)i. . .^i) . if any two elements ip\. ip2 of this space can be associated with a complex number (V'l. n exist (not all zero).
9a) we start from if) = ipi + \if>2 6 M.^1) 2 h 11 0 <%«e + M .tp2 £ M. linearity in the second component. for arbitrary A and ip2 ^ 0: (V>i + AV>2.7) In addition the following relations hold in a metric space M.9b) = 0 (Pythagoras theorem).9e) 11^11=1 We restrict ourselves here to some remarks on the verification of these wellknown relations.rh) (3.9a) (3. The distance between two vectors if>i. (3. In order to verify Eq. WA+Ml2 = ll^il 2 + ll^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. meaning oneandahalffold linearity). ^ i ) + A * ( ^ .V>i + A</>2) > 0 .2 Hilbert Spaces 43 where the asterix * means complex conjugation. \\1p1 + ^211 < HV'ill + IIV^II (triangle inequality).^2)1 < H^ill • 11^211 (Schwarz inequality).i>2\\2 = 2'i/'i2 + 2'i/'22 (parallelogram equation). (310) (^2. tp2 £ M: 1(^1.9c) HV'i + tp2\\2 + \\1p1 .. (3.is defined by d(ipi.*h)=\\AH\(38) (3.^) = ai*(V'i.2.9d)   V i   = sup  ( ^ i .e.3.V') +"2*(V. (3. which we can write 0 < ( ^ i . (3.. ^ i ) + A ( ^ i . if Wi. The norm of the vector ip (preHilbert space norm) is defined as H:=(^)1/2.)> (36) i. antilinearity in the first component (also described as sesquilinearity. The first two properties imply ( a i ^ i + a2ip2. »2 2 . for ip\. ^ 2 )  .V.^ 2 ) =   ^   2 + 2 ^ i . ^ ) + A 2 (V2.
V 2 ) = 0 .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.* Two vectors i f i . .^)1 <IIV>il.12) We omit here the verification of the remaining relations (3. ^ e M are said to be orthogonal if (</>!.\\ih\\. Mathematical Foundations of Quantum Mechanics 1(^2.9a). HIHI (3. hence verification in each case is necessary. (3. Examples of metric vector spaces: (1) Let M.V2) IIV'l2 + V'22 + 21(^2.9e).9d).ii) thus verifying Eq. o ^ V i = o. (3. (3.^1)1 Vi 2 + ^2 2 + 2   ^   .44 or CHAPTER 3.11). We also omit the verification of the following properties of the norm of a vector tpi € M with ^ € X . ll^ill + llVdl. a £ K : ll^ill HVill 11^1 +^211 M If for a vector tp e M: > = < = 0. for which CO 2 := V^l^l 2 < 00.10): ll^+^H2 = < <   ^  2 + V22 + 2K(Vl.  ^ l l = (HlMI +   « 2 . (3.9c). beginning with A = 1 in the second line of Eq. (3. In verifying the triangle inequality we use this result (3. so that   ^ l + ^ 2   <   ^ l   + ^2. be the set of all column vectors v\ V = (Vi) = I 2 V with complex numbers Vi.
g)= [ /(x)^(x)d3x. for which the scalar product. are combined to an equivalence class [/] (with space L2). etc. (2) Let M.5b). Eq. . In order to avoid this difficulty. with inner product oo 45 (v. Elements of the classes are then called representatives of these classes.e.5c). and one defines addition and multiplication by complex numbers with respect to these classes. which differ solely on a set of measure zero. which satisfies relations (3. (3.e. But this applies also in the case of any function which is nonzero only on a set of measure zero. (3.[$]):= / JM /(x)* 5 (x)d 3 *.5c)) / ( x ) = 0 = > ( / . = 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. •'^ \ 0 otherwise. / ) = 0.3. is defined by ([/]. With the scalar product (f. 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).5a). all squareintegrable functions / which are "almost everywhere equal". i. (3. The Schwarz inequality is then oo oo \M\<* ^2\Vi\ ^2\Wj\2.w) :=J2v*Wi. i. Then L2 is the space of all these equivalence classes. */ x } = / /o for x = 0. 2 and so on.
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.
4b) (4. if (u\v) — 0.e.62 CHAPTER 4. which we write (u\.5c)). (2) (u\u) > 0.3 Linear Operators. that the vector \u) E V is associated with a vector in V. if x\u) = 0 V kets \u) E V. It follows that we can now write the Schwarzian inequality (3. Dirac's Ket. Hermitian Operators A linear operator A in the space of ketvectors V acts such that A\u) = \v) EV .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\. (4.(t. if \u) = 0 (see (3. the norm squared. \v) E V are said to be orthogonal.5) 4. V \u) (in the case of £) with infinite norm!). (3) (u\u) = 0.9a): \{u\v)\2 < (u\u)(v\v). Also: (xi = (X2I) if (xi\u) = (X2\u) for all \u) E V. As in our earlier considerations we demand for (v\u) the following properties: (1) (u\v) = (v\u)*. or (4.4a) \v) = J\(m<% and (4. i. Two ketvectors \u). 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. The dimension of V is the same as the dimension of V.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). This expression is linear with respect to \u) and antilinear with respect to \v).\u) = (u\0*.
3 Linear Operators. BA = 1. The product vectors l?^ 1 )^ 2 )) span a new vector space. Multiplication by unity.e. as we can see as follows. The inverse of A is written A .e.or Kronecker. The operator A.e. also acts on the dual space V.. Two operators A and B are said to be inverse to each other. the space Vi <S> V2 has the dimension . if AB = 1. let (771 6 V be the bravector defined by this functional. Furthermore. Then one says: (771 results from the action of A on (x. on bravectors are similar to those on ketvectors. i. Then the following rules apply: A + B = B + A. defines the unit or identity operator. 1 and \u) = B\v). This product is commutative. The inverse of a product is (AB)'1 =B1A~1. we have \u^u^) = X1\v^u^) + \2\w^u^) and so on. This does not always exist. if \v) — A\u) i. Hermitian Operators 63 with A = 0 if A\u) — 0 for every vector \u) £ V. In this way the operations of operators A.or direct product of vector spaces. Let the following bravector be given: (x\ £ V. The product of such vectors is written: \u^u^) = \u)^\u)^. Then (x(^4it)) is a linear functional of \u) £ V (since A is linear). It follows that (r..4. and one writes: (V\ = (X\A. Also A = B it for every vector \u) £ V : (u\A\u) = (u\B\u). ltt) = \u). denned as a linear operator on V. i. B. the space Vi <S> V2. with n)(1) = Aiw)( 1 )+A 2 k) ( 1 ) . Finally we define the tensor. If Vi has the dimension A/i. Let \u)(l> be vectors of the space Vi and similarly ?j)(2) the vectors of the space V2. AB^BA.. Furthermore linearity applies.\u) = ((X\A)\u) = (X\(A\u)) = (x\A\u).
we write \v) = A*\u).6) for all \u).7b) (4. in the product space. \t) G V.e. Every such operator is then also associated with an operator. then \v) is a linear function of \u).3).e. Every operator A^> commutes with every operator A(2\ i. i.and BraFormalism an operator in the Mi A/2. Examples of operator relations: (1) (AB\u)(v\Cy = C^\v)(u\B^Al (4. If (v\ = (u\A. Since we demanded the scalar products to have the property (t\v) = (v\ty. Dirac's Ket. i.e.7a) (4. by replacing in the above the bravector (t by <tflt) (AB)* = B*A\ (cA)* = c*A\ (A + 5) t = A* + B\ (4. since AWAW\UW)\UW) = \vWuW) = AWAW\UW)\UW).7d) . (v\t) = (u\A\t). AW\U)W = \V)W. also Sec. Next we construct the following scalar products: (t\v) = (t\Ai\u). [A^. Let A^> be an operator in the space Vi and A^ space V2. 3. Assuming now that \v) G V and (u\A G V are conjugate vectors as explained above.64 CHAPTER 4. we obtain the conjugate relation <tAt«> = (u\A\t)* (4.A^] = 0. i. AW\uW)\uW) = \vWUW). and (v\ = (u\A.7C) («)(u)t = \v){u\. the operator A^ is called hermitian conjugate or adjoint operator of A (cf.e.g. As a consequence we arrive at the following properties (e. here designated with the same symbol.
e. The commutator of two hermitian operators is antihermitian: [H. Hermitian Operators (2) The conjugate bravector of AB\u)(v\C\w) is {w\C^\v){u\B]A]. An hermitian operator A has the properties (a) that all its eigenvalues are real. It follows that a is real. i. but only if these commutators commute. Every operator A can be written as the sum of two such parts: A=\(A + A*) + \(Atf) anti—hermitian . The operator H is said to be positive definite. hermitian The product of two hermitian operators is not necessarily hermitian.e. . [H. or (u\A\u)* — {u\A^\u) = (u\A\u). if H = H' and K = K\ then (HK)* = HK only if tfrf = KH. H^K] The separation of HK into hermitian and antihermitian parts is therefore HK=±{HK + KH)+l[H.K}. i.K\. as is also (u\u). so that (n^4n) is real. so that (a)(a)t = a)(a.K]^ = = [HKKH]* KHHK = = K^H^ [H. Obviously the quantity \a)(a\ is an hermitian operator. for we had (\u){v\)i = \v){u\. K] = 0. if H is hermitian and (u\H\u) > 0 for all ketvectors \u). and (b) that the eigenvalues with respect to \u) are equal to those with respect to (u\ and vice versa. Analogous considerations apply in the case of bravectors (u. Verification: Since A — A^ and A\u) = a\u). if H — H' and antihermitian. 65 The linear operator H is called hermitian.4.3 Linear Operators. it follows that (u\A\u) = a(u\u). An important theorem is the following.
i. which includes vectors of infinite norm. Then every vector \u) £ "K can be written \u) — its) + \us*) with \us) e S.e. Ps2 = Ps(4. but the continuum vectors are normalized to a delta function. Ps is idempotent. Ps\ua. The above theorem remains unchanged.u').8) The projection operator has the following important properties: Ps is hermitian. If these form a complete system.66 CHAPTER 4. This follows from observing that (u\Ps\v) = (u\vs) = (us\v3) = (us\v). A continuous spectrum can immediately be included by passing to the extended Hilbert space. so that (u\Ps = (us\. Dirac's Ket. Next we introduce the very useful concept of projection operators. i. The operator Ps which projects onto S is defined by the properties: Ps\u) = \us) = Ps\us). Assuming A\u)=a\u).9b) (4.and BraFormalism Moreover.e. (4. {us\us*) = 0. ( n  z / ) = 0 . \u8*) E S*.e. we have 0 = (uArt) — (u. so that (v\u) = 0. i. (v\A = b(v\. Orthogonality: Two eigenvectors of some hermitian operator A belonging to different eigenvalues are orthogonal. since a ^ b. a ^ b. (n\n')=5nnl. then the hermitian operator defined on this space is an observable. as we can see as follows. (u\u') = 5{v . it follows from A\u) — a\u) that (u\A = a(u\ or the other way round. since a is real. (u\Ps = {us\.Au) = a(v\u) — b(v\u) — (a — b)(v\u). i.e. The entire set of ketvectors spans the extended Hilbert space.9a) . Let S be a subspace of the Hilbert space !K.) = 0. and let S* be its complement.
p = 0..e. iV) is a set of orthonormalized states. Obviously N PN = y]ln)(re n=l is the projection operator onto the A^dimensional subspace SN.. . Very similarly we can construct operators which project onto the subspace spanned by the continuum states. Ps is projector onto S. 67 (4. Let p be an eigenvalue of the projection operator P. If for these (by construction) the projector P2 onto a subspace S2 is P2= J*I f2\odm. c = (a\u). p 2 . (1 — P)\u) are orthogonal. i. so that (Ps2 ~ Ps)\u) = 0.e. (4.4.e.9c) This property can be seen as follows. i. The quantity \a)(a\ is called elementary projector. Example: The vector \a) normalized to 1 with (a\a) — 1 spans a 1dimensional subspace. where £ is a continuous index. Ps2 = PsThe only eigenvalues of Ps are : 0 and 1. \ua) = c\a). i. Hermitian Operators This property follows from the observation that Ps2\u) = Ps\us) = \us) = Ps\u). Set \u) = \ua) + \ua*) with (a\ua*) = 0. Then 0 = (P2P)\p) and hence p = 0..e.9d) Thus the vectors P\u). . \u) arbitrary. i.3 Linear Operators. so that \ua) = (a\u)\a) = \a)(a\u). as can be seen as follows. Let a continuum state be written £). 2). P\p) =p\p). Then (a\u) = (au a ) + (a\ua*) = (a\ua) and so {a\u) — {a\ua) and hence {a\u} = c(a\a) = c. (1 — Ps) is projector onto S*. = (p2p)\p).1. The projection \ua) of an arbitrary vector \u) onto this subspace is ita) = a)(aw). if the set of vectors 1).
be a system of basis vectors in this space. in this case the spectrum also has a continuous part. Degeneracy will be discussed at various points in this text. . Sec. The eigenvalues Aj then form a discrete sequence with associated eigenvectors \ui) € !K. i. which we introduced earlier.1 and 11.e. 4. that one and the same eigenvalue Aj is associated with several eigenfunctions.. 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. ' The projector onto the subspace with eigenvalue Aj can then be written Pi = ^2\ui. Coulomb potential. Observables are representatives of measurable quantities.e. Dime's Ket.114b).r\ "An example is provided by the case of the hydrogen atom. In general it is possible.68 CHAPTER 4.. are linearly independent).6.and BraFormalism and hence '6 Numerous properties of the projection operators Pi are selfevident. See e.114c). and Eq. the operator P is correspondingly P2\u)= f2\0m\u). u2). Let us assume that A is an hermitian operator with a completely discrete spectrum (as for instance in the case of the harmonic oscillator). 8. which are orthogonal to each other (i. However. Let \u\).. one says the degree of degeneracy is r . see Eq. Examples 8.r)(ui.g.4 Observables Operators which play a particular role in quantum mechanics are those called observables. In the case of the infinitedimensional Hilbert space with a countable number of orthonormalized basis vectors. P = £n)(n. If there are r such vectors.6. (11. (11.
r)(ui.4 Observables The dimension of this subspace is that of the degree of degeneracy.r \(ui. PA = y£pi i = Yt\ui.r (4.4.r\ = l. and the eigenvectors \ui. the projection of this operator is the entire space. in the case of degeneracy with (ui. Applied to an arbitrary vector \u) G "K Eq. ^ A i P i = A = ^Aiui. The uniqueness of the expression (4. Together with the orthogonality condition.e. i. 69 i If A is an observable and if the spectrum is purely discrete.r>(ui.r)(ui. i.r) (the convergence.\i)Pi = 0. which we do not enter into here.10) follows from the fact that i Applying the operator A to the projector (4.r\ui/.e.r =^ i. Let us set {A . then PiPi' = 0. If Aj ^ Aj'.r)(ui.e.10) expresses the completeness of the orthonormal system. (4. is always implied).r\u). we obtain i i i i i.11) i.10) This expression is known as completeness relation or closure relation. the operator A is completely determined by specification of the eigenvalues A. The operators Pi are linearly independent. It follows that the norm squared is given by (u\u) = {u\PA\u) = ^2(u\ui.r'} = Sii>6rr>.r.10) gives the linear combination it) = ^ i. (4.10). i.r \ui.r\u)\2. the relation (4.e. i i (4. and APi = XiPi.r\u) i.12) . or also as subdivision of unity or of the unit operator.
and BraFormalism This is the expression called Parseval equation which we encountered with Eq. The ketvectors \uv) are orthonormalized to a delta function. > rV2 + / dis\(uu\u)f \uj)(ui\Xi + / \(v)\uv){uv\dv. (3. Then f(A) = f(A)PA =^/(AOk)^! + / f(\{v))\uv)(uv\dv. Dime's Ket.70 CHAPTER 4.e. Let v be the continuous parameter which characterizes the continuum.e. for instance. i. (u„zv) = 8{v . the eigenvector \ui): f(A)\ui) = f{\i)\ui). In a similar way we can handle functions f(A) of an observable A. (4. as explained earlier. for instance exponential functions. 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. now however. If the spectrum of an observable A consists of discrete as well as continuous parts.v'). Then we denote by \uv) the eigenvector of A whose eigenvalue is \{v). (note: {uv\A\uvi) is the matrix representation of A) A\uv) = \(v)\uv). 1/2)) PA = 22 \ui)(u*\ + / dv\uv){uv\ = 1. i. written in Dirac's notation. (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). we have the corresponding generalizations.14) . One defines as action of f(A) on.17) previously.
(4. For many practical purposes the use of the Fourier transform is unavoidable. again later.. i. Therefore we want to reexpress the Fourier transform (3.17) The Fourier transform provides the transition from one representation and basis to the other. We shall return to this theorem. with energy E. form a complete set of commuting observables.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. as a formal orthonormality condition. for which the completeness relation is f dk\k)(k\=l. Extending this we can say: A sequence of observables A.C. First of all we can rewrite the integral representation of the delta function.B. (4.58) in terms of ket. Eq..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. here presented without proof. B].16) represent subdi . which all commute with one another. i.and bravectors.. which is discrete in the case of discrete energies and continuous in the case of scattering states (with continuous energies). and (b) if they possess a uniquely determined system of basis vectors. if (a) they all commute pairwise.5 Representation Spaces and Basis Vectors We began by considering ketvectors \u) in which u is a parameter of the (energy) spectrum. / dx\x){x\ = 1. where the symbol ip is already indicative of the Schrodinger wave function ij. (3. [A. In the differential operator representation the problem to establish such a relation is known as the SturmLiouville problem. 4. \k)€Fk.16) Correspondingly we also have a complete set of basis vectors {\k)} of an associated vector space F^.59). Finaly we recapitulate the following theorem from linear algebra: Two observables A and B commute. (4.15) and (4. In the onedimensional case we write {x\x') = S(xx'). their commutator vanishes.e. if and only if they possess at least one common system of basis vectors..4. x) € Fx. In the following we consider more generally ketvectors \ip) as representatives of the physical states.e. Since both expressions (4.
which are representatives of the states of our physical system.x') = (x\x') = (x\t\x') = (x\ f dk\k)(k\x'). (x\k) = ^=eikx. called wave function. (4.21) provides the ketvector \ij)) in the A. (4.space representation.18) According to Eq.1a).19) Comparison with the orthonormalized system of trigonometric functions (4. (4. Rather \x) and \k) serve as basis vectors in the representation spaces Fx.15): 5(x . ${k) := (k\if>). .e.L e " * * ' = {x'\k)\ V27T (4.22) All of these expressions can readily be transcribed into cases with a higher dimensional position space. Obviously we obtain this by inserting a complete system of basis vectors of Fk. (4. this expression has to be identified with — / dkeikxeikx\ i.59) or Eq. Dime's Ket. i.20) The representation of the corresponding bravector (^1 G IK in the position space Fx is correspondingly written (il>\x) = {x\ipy.2a) etc. V 27T (k\x') = . <a#) = [ dk{x\k)(k\ip). shows that these expressions are the corresponding continuum functions (the continuous parameter k replaces the discrete index n). 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). il>{x) := (x\i/}) : "K * C. we can rewrite Eq.72 CHAPTER 4. i.Fk. The Fourier representation ij){x) = )= V 27T J Ieikx^{k)dk (4.e.and BraFormalism visions of the unit operator.e. (3. The vectors \x) and \k) G F are not to be confused with the vectors u) or \ijj) G !K. (4.
i. if we do not take into account the complementary part of the universe. 5.t). We postulate the (timedependent) Schrodinger equation* (with the Hamiltonian as the timedevelopment operator) and obtain the quantum mechanical analogue of the Liouville equation.Chapter 5 Schrodinger Equation and Liouville Equation 5. This will then also clarify the role of p as an operator in the quantum mechanical analogy with the Liouville equation.p. 1.e. which is the quantum mechanical analogue of the classical probability density p(q. the calculation of which for specific cases is the subject of Chapter 7.1 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. 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). 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. Schrodinger [244]. We mentioned earlier (in Sec. 73 .2 The Density Matrix We shall establish a matrix p.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 states \i) £ "K. T . \a).j = where J2 (J\A\i)5baCatCbj* = X>'l4*>/0« a.b.e. i.b. as well as 2~] «)(* = 1 in the subspaceof pure states i (i\j) = 5ij. which in Dirac's notation we can write as = J]C a »i) a.. of an operator that acts only on the states of the system we are interested in.74 CHAPTER 5.i.i. The difficulty we have to face is the impossibility to specify precisely the stationary state of our limited system. \i). i.e. \a). 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). called the density matrix. we have (a\b) = 5ab. i eH ®^' (5J) We assume that the states are orthonormal and in their respective subspaces also complete. the matrix can be diagonalized by a transition to a new set of states ^The hermiticity can be demonstrated as follows..3a) and N J  a ) ( o  = l in the complementary subspace. p" ( C C * ) t = CTC* = p.. Schrodinger Equation and Liouville Equation the measuring instruments.j ij (54) Plj = (i\p\j)^YCiCJ*a (55) Here p is an hermitian matrix. is then given by (A) := {MA\^) = Yl U\{b\A\a)\i)CaiCbj* a. The actual and real state of a system is then a superposition of two types of states. a (5. that is. but also of the states of the complementary set of the universe.^ Since p is hermitian. where i represents collectively all quantum numbers of the state of the system under consideration. (52) (5.. The states \a) £"K'on the other hand are the corresponding states of the complementary part of the universe. These are the quantum mechanical states which are also referred to as microstates... we have to deal with a socalled "mixed state". (a\i)=0 = {j\b). are called "pure states"..3b) The quantum mechanical expectation value of an observable A.
Gibbs introduced the concept of ensemble. in the position state representation or xrepresentation.^'\i'^i'\> (58) since then (f\p\j") = ^Ui'ij'li^ii'lj") v = ^Ui'Si'fSi'j" v = UjiSj/j. In this sense we have p(x'. i..2 The Density Matrix 75 \i') 6 "K. e.4) we had the expression (A) = Y/Pij(j\A\i) i. . This ensemble is interpreted as a large set of identical systems. a (512) *In his reformulation of the statistical mechanics developed by Boltzmann.e. (see below) eM.3a). a projection operator.) = YJc t . which are all subjected to the same macroscopic boundary and subsidiary conditions. a complete orthonormal system with properties (5.7) p = ^2\ii)(i. i. in a specific representation. If the ensemble corresponding to the system under consideration consists of Z elements (i.6) (i'\p\i") = Wi'Si'vi = real.2) and (5. (5. can be achieved with the help of the completeness relation of the vectors \i): \i') = J2\i)(i\i') i. or (5. (5.* We can specify the state of our system..e. In the following we write simply \i) instead of \i'). then (cui/ '^Zi^i)Z is the number of ensemble elements in the pure state i). i i (511) In Eq. but occupy in general different microscopic.3 with Pi^YtCaiCaj*. as we discussed previously. The ensemble can be represented as a set of points in phase space..g. Thus the real system is in the pure state \i) with probability uii/ J2i Ui.9) is diagonal. P ^ ) = i)(#) = (#)i) (5.e.10) and is therefore. This projection operator represents a quantum mechanical probability as compared with the numbers ui{ which represent classical probabilities.ji(x / )i*(x). which today is a fundamental term in statistical mechanics.e. (5. where x represents collectively the entire set of position coordinates of the (particle) system.x) := (x'\p\x) = yju. as we saw earlier. seen macroscopically of Z copies of the system).5. Thus the pure states \i) form in the subspace of the space of states which is of interest to us. This transition to a new set of basis vectors in which the matrix becomes diagonal.j(x'i)(ia. quantum states.\p\i")(i"\ = Y. The operator Pi = \i)(i\ projects a state \<p) G "K onto \i).
(5. we have Tr(p) = l. Thus Tr(j>ii)<i)=l. Eq.15) 5^0'l5^w«l*X*b') j i = 1 » or J^wi(it> = ^ W i = l. Schrodinger Equation and Liouville Equation We now define the operator p by the relation (cf. (5. Eq.16) (5. We can also show that Ui > 0.18). (5.14) we obtain (Ai') = ^2(j\i')(i'\i)(i\p\j) ij = ^Sji'Su'Pij i. (5. In particular for A = 1.14) which we can write (A) = Tr(pA). (5. To this end we set A —>• Ai' := \i')(i'\ (no summation over i'). On the other hand.76 CHAPTER 5.4). We recall that the elements of p originate from the coefficients Cai in w = Y^cai\a)\i) = Yl ( E ^ M V ^519) .18) Substituting this into Eq. Then the expectation value of the observable A is (cf. according to Eq.4)) (5. (5. (A.3 = pvv = uv. as claimed by Eq.13) {A) = ^(j\A\i)(i\p\j) ij = 52{j\Ap\j).17) Hence the sum of the classical probabilities is 1.) := (mm = (#') W> = l(#')2 > 0. i t (5. (5. 3 (5. Hence Wj/ > 0.5) above) (i\p\j)=Pij.
8). (5.t) and instead of (x\Pi\x) we now write p(x.3 The Probability Density 77 We see therefore that p defines the mixed states. In the following we are mostly concerned with considerations of systems in pure states.t) we now write tp(x. thus enters these states.t)\2.21) p=\i){i\=Pi. Thus we have a system in a pure state.§ {\x)} is a complete system of basis vectors in the position representation space Fx with dx\x){x\=l. and in the other states with probability zero. not in a mixture of states. (5.5.24) s (x\x') = 5{x . since they do not take into account the (interaction with the) rest of the universe. i. i.e.20) expresses that the system can be found with probability 1 in the state \i). For the case n we have TJ=J (520) (5. so that p{x. which is contained in the coefficients Cai. We see this more clearly by going to the position space representation. the only remaining one is 1. that — different from u>i — P.x').t) = \^{x. i. We now consider the latter expression. Equation (5.3 The Probability Density p(x. We observed above.t). ^2ui\i){i\. .e. defines the statistical operator. t) p= i The expression (5. one says. in which the numbers Wj represent classical probabilities and the operator \i)(i\ quantum mechanical probabilities or weights. by representing the vector \i) by the wave function i(x. The effect of the interaction of the system under consideration with the surroundings. If all ui but one are nought.e. Then {x\Pi\x) = (x\i)(i\x) = \{x\i)\2. the system is in a pure state. but we see here. that these are actually not sufficiently general. represents a quantum mechanical probability. (Question: Is the state of the universe a pure state?) 5. in all other cases the state of the system is a mixed state.22) In the following (x\i) is always to be understood as (x\i(t)}.t) = {x\i). (5.23) Instead of i(x.
e.22) can also be written (xi) 2 = (i\x)(x\i).e.78 CHAPTER 5. (5. can be found with probability 1 in the space R1. which is chosen in such a way that the desired analogy is obtained. Hence we postulate: The time development of a state \j) E "K is given by the equation ih^\j)=H\j).t)\2. (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.t) = \rp(x.3. Equation (5.^ The equation which is the conjugate of Eq. or the wave function tp(x. (5. Schrodinger Equation and Liouville Equation The relation (5. t h e equation of motion with states represented in a system for which the q's or here x's are diagonal).28) Differentiating Eq.8).27) is iHJ\ jt= m . i. the Schrodinger equation is always postulated in some way. The generalization to a threedimensional position space is selfevident: JK 1 / p(x.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. We therefore postulate first an equation for the time dependence of a state vector. (5. it is not "derived"). the expression (5. and the timeindependent equation should correspondingly be called Schrodinger wave equation (i. (5. This differentiation requires the time derivative of a state vector \i) E "K. and hence the integral over all space dxp(x.27) is known as the Schrodinger equation. with respect to time.26) 5. We shall convince ourselves later that the postulated equation is sensible (in fact. To this end we have to differentiate the density matrix p.t) = / dx{i\x)(x\i) = (i\i) = 1. . but the linear operator states are moving — even if the system is observably stationary). t). We now want to obtain the quantum mechanical analogue. 2.8) and multiplying by ih.27) where H is the Hamilton operator of the system. (5.25) J Thus the particle whose state is described by \i).
p also contains "quantum mechanical probabilities" (i. With Eqs. (3. (5. The correspondence to the classical case is obtained with the substitution h J< > <5'33> and dp/dt = 0 (cf.p}.4 Schrodinger Equation and Liouville Equation 79 Inserting here Eqs.Xi)\j).p} + ihJ2^\J)(J\. We can now consider the Schrodinger equation (5. The Schrodinger picture and the alternative Heisenberg picture will later be considered separately.5. (5.32) Comparing Eq. (5. we observe that here on the left hand side we have the total time derivative.27) and (5. such a formulation is described as Schrodinger picture.37)) (5. (5.27) in the position space representation: ih(x\\j) = (x\H\j) = {x\H(pi.34) . (cf. Eq. The reason for this is that in addition to the "classical probabilities".31) below) 3 j 3 = y £u>J{HPjpjH)+thyE^m\ dt 3 = [H.28) we obtain (see also Eq.e. i.24)).31) The eigenvalues u>i of p determine the fraction of the total number of systems of the ensemble describing the actual system which occupy the state \i).28) the states are considered as timedependent. We thus obtain the operator form of the Liouville equation.27). (5. M\/n\ (530) where in statistical equilibrium dt £^">01=03 (5. not the partial derivative. Eq.32) with its classical counterpart (2. i.21).e. dp/dt ^ 0. both uji and i)(i). (2. (5.e. the relation ih^ = [H.
**R.** In view of the close connection between time.4. appearing in the Boltzmann distribution. with respect to time t. i.4. in Sec. (5. 5. 0) or \il))t = \il>)t=o exp (5. T meaning temperature. to) = exp[—iH(t — to)/H\. operator" the exponentiated oper .80 CHAPTER 5. (5.t) = exp ip(x. however.e.and temperaturedependent Green's functions.t) = Hld .0): where H(ih—. 10. we write V>(x. we shall describe as "time development ator given by U(t. since the timedependence cancels out (the exponential functions involve the same operator H but with signs reversed.t): > ih—</>(x. but with respect to the parameter /3 = 1/kT. we can replace these states by corresponding timeindependent states. (5. P. „ / ith—.35) can be written'! ip(x. which we shall need later.x 1 ^ ( x .8) for p.36) in in The initial wave function or timeindependent wave function ^(x.Xi){x\En) = En(x\En)..e. or H\En) = En\En). this is an application of the BakerCampbellHausdorff formula which we deal with in "Later. For a mixture of states \i) of the system under consideration caused by the rest of the universe we have Ui ^ 1. i.35) Here we can look at the Hamilton operator as a "time generator3''. Schrodinger Equation and Liouville Equation or with (x\j) — ip(x.. in p = ^2i^i\i}(i\. it is plausible to refer to this equation at this stage. Feynman [94]. In expression (5. t ) .. . With the help of Eq. the state \i) is still timedependent. namely that the density matrix satisfies an equation analogous to the timedependent Schrodinger equation — not. n (537) This equation is described as the timeindependent Schrodinger equation.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. a (5.38) (x\j)t=o = Yt(x\E^(En\J)t=o. since the solution of Eq. 0) can be expanded in terms of a complete set of eigenvectors \En) of the Hamilton operator H.36) however.
x'(3) = Si e ~^W*V). we obtain p(x. x') = ^ i LUiMxWix'). Hence we have (with (Ei\Ej) = 5^) in what may be called the energy representation p = YJ"i\Ei)(Ei\.5. we can rewrite p. (543) Since H(f>i(x) = Ei<f>i(x).40). i. (1.39) therefore (x\p\x') = J2"i(x\^)(Ei\x'). i (5.4 Schrodinger Equation and Liouville Equation 81 Example 5.1).41 ) or with 4>i(x) := (x\Ei) this is p(x.9) with En > nhu) L0icxePE\ so that Yliui becomes = f3 = l/kT. i ( 5 . the expression _Zie^\Ei)(E>\ also as (see below) e0H (5. (5. (5. or ^ = ^ e 0Ei _m.40) 1. Eq.e. (5.44) r (TrePH)' (5.^ n tne position space representation Eq.x')^p(x.42) Inserting here Eq.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. (5.45) since on the one hand •PEi i i i and on the other hand .
48) = 1. This solution is the Green's function or kernel K(x. (5. the following relation holds: exp(A) exp(S) = exp ( A + B + i [A. (5. PN(0) := (EilpNiP^Ej) = (Ei\ePH\Ej) =e"^^ (5.(3).51) where the subscript x indicates that H acts on x of PN(X.The quantity pjv is obtained from the solution of the Schrodingerlike equation (5.45) without normalization as PN(J3) := e~pH. Wilcox [284].L [B. . We now rewrite the factor exp(—(3H) in Eq. [B. whose calculation for specific cases is a topic of Chapter 7 (see Eq.55)).49) In the position or configuration space representation Eq. Schrodinger Equation and Liouville Equation so that p is given by Eq. (5.x'. M.(3) := (x\pN(P)\x') = (x\e~0H\x').1: Baker—Campbell—Hausdorff formula Verify that if A and B are operators. (5.X'.46) is pN(x./3).47) (5. With Eqs.45). (5.x'. (5. Tr(pN) = / dxpN(x. [A.4> = ^ g ^ .51) is seen to be very similar to the Schrodinger equation (5.46) In the energy representation this expression is pNij(P) with PNij(P) = 6ij. (5.tt tt For a more extensive discussion see R.50) Differentiating this equation with respect to j3. (5.x'.51).x. we obtain = HxPN{x.52) Hence this expectation value can now be evaluated with a knowledge of pN. B}} . (7.46) we can write the expectation value of an observable <^Tr<„.45) and (5.82 CHAPTER 5.35). (5. Example 5. we obtain 9pN gpP) = SijEie^ = EiPNij(J3). (3). Equation (5. (5. Differentiating Eq.47) with respect to (3. A}} + • Solution: The relation can be verified by expansion of the left hand side. B] + L [A./?).
.1 Introductory Remarks In the following we consider in detail the quantization of the linear harmonic oscillator. 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. which is often described as "second quantization" as distinct from the quantization of quantum mechanics which is correspondingly described as "first quantization" (thus one could visualize the free electromagnetic field as consisting of two mutually orthogonal oscillators at every point in space). can 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.51) to (8. (8. Since the spectrum of the harmonic oscillator is entirely discrete.53)). E. Noz [149].the present form of quantum mechanics is largely a physics of harmonic oscillators. that the oscillator. This is a fundamental topic since the harmonic oscillator. S. can — in fact — be quantized in quite a few different ways. * Comment of Y. 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." Quite apart from this basic significance of the harmonic oscillator.. Kim and M. its associated space of state vectors also provides the best illustration of a properly separable Hilbert space. However. 83 .Chapter 6 Q u a n t u m Mechanics of t h e Harmonic Oscillator 6. The importance of the harmonic oscillator can also be seen from comments in the literature such as:* ". we shall see on the way. like the Coulomb potential.
of which the only nontrivial one here is [q.p.84 CHAPTER 6. For instance we can go to the special configuration or position space representation given by q^x. h = l. 2mo dx \ 2 2 (6. linear harmonic oscillator the Hamilton operator is given by the expression XJ 1 2 .' ^mgLOh by setting and *~. we can also proceed in a different way. hence we do not use a distinguishing notation. Quantum Mechanics of the Harmonic Oscillator 6. which is called Weyl ordering.1) for the hermitian operators q. To this end we introduce first of all the dimensionless quantities rriQUi fl q.to = 1 reduces to the parabolic cylinder equation. in the present case + [EmooJ2x2)ip = 0. Also observe that H does not contain terms like pq or qp. p^ih—.2) which for mo = 1/2. This equation would then have to be solved as a second order differential equation. so that there is no ambiguity due to commutation.MvT« 7=r)^(VTvir>' + («> 'Note that we assume it is obvious from the context whether q a n d p are operators or cnumbers. ox and then to the timeindependent Schrodinger equation Hip(x) = Eip(x). .2 The OneDimensional Linear Oscillator In the case of the onedimensional.^ The first problem is to determine the eigenvalues and eigenfunctions of H. This can be done in a number of ways. the boundary conditions on the solutions ip(x) would be square integrability over the entire domain of x from —oo to oo. 1 2 2 The quantization of this oscillator is achieved with the three postulated canonical quantization conditions. However. which is — in the first place — representationindependent.p]=ih (6. When such a term arises one has to resort to some definition. like taking half and half.
and [AB.A^] = 1. i\/2 Inserting these expressions for p and q into H we obtain H = hiu{A^A + AA*) = hi* (A^A + ^ . i. (6. C] = A[B. i.14b) [A* A.6) and p = y/motoh AAi =^. q are hermitian).1 ) . With the help of Eq.34d): [A.14a) we deduce for an eigenvector \a) of A^A. Prom these relations we obtain ( ^ = ^ .1) we obtain immediately the commutation relation [A. C] + [A. for ^ A  a ) = a\a). if A*A\a) = a\a). + 1).5) Reexpressing q and p in terms of A. (6. A*} = A^A.11) Thus the eigenvalues are real and nonnegative.BC] = [A.8) The eigenstates of H are therefore essentially those of N := ^ f A (6.C].9) We observe first that if \a) is a normalized eigenvector of N with eigenvalue a. A^] = A\ (6.e. (6. (6. (6.10) then a = (a\A^A\a) = \\A\a)\\2>0. (6.6. A\ we obtain h A + A^ rriQLU y^ (6.12) in order to obtain the following expressions: [tfA.7) (6. We now use the relation (3.e.2 The OneDimensional Linear Oscillator 85 (p.B}C + B[A. A] = [A\ A]A = A. (6. C}B.13) ( ^ A ) i t = A\A^A From Eq. (6. .14a) (6.
or Ata) = v ^ T T .  a) is eigenvector of A^A with eigenvalue (a + 1).l)A\a) = A(A^AA .14b) : (AiA)A*\a) = A\A^A + l)a) = A\a + 1) a) = (a + l)A^\a). A 2 a) is eigenvector of A^A with eigenvalue (a — 2). or \\A\a)\\ = Va. since this equation implies that the eigenvalues cannot be negative. unless A^\a) = 0. Thus for a certain value of n > 0.A)\a) A{AAU . (6. but (b) An\n>\ llA An+1a)=0.17) (6.20) .l)\a) = A(a . If we continue like this and consider the vectors An\a) ^ 0 for all n.21) .2)A a).e.16) This means. Similarly we obtain from Eq. (6. unless A 2 a) = 0.2A)\a) 2 (a . Next we consider the vector A 2 a). However. we must have (a) Let \an):= Ana)^0. (6. (6. In this case we have (A*A)A2\a) (6 (6.l)A\a).. in view of Eq.2A)\a) = A(Aa . (6. This would mean that for sufficiently large values of n the eigenvalue would be negative.18) = '= a ) A(A^A .l)\a) = (a . Quantum Mechanics of the Harmonic Oscillator (A*A)A\a) = A{A*A .15) Thus A\a) is eigenvector of A^A with eigenvalue (a — 1). Similarly \\A^\a)\\2 = (aAAta) = (al + A^A\a) = (a + l)(aa) = a + 1. (6. unless A\a) = 0. we find that A n a) is eigenvector of A^A with eigenvalue (ct — n). The norm of A la) is Aa) 2 = {a\A^A\a) = a ( a  a ) = a.11) this is not possible. (6.19) i.86 the relation CHAPTER 6.
i. (6.n ) = " A»a>=» ' = A«a)F = 1 .28) .27) According to (b) of (6.. Moreover.e. that an = 0. or a = n > 0. (6. we obtain an = \\A\an)\\2(6=) An+l\a) \\An\a) (6.18) we obtain for a = n: p t  n )   2 = n + l. \ I 2 2 x 87 Replacing in Eq. HAU+IO)!!2 = = (O\AAAWI\O) = {O\A{A^A + I)A^\O) for all n.n  a . so that A^\n)^0 In particular we have A^\0) ^ 0 and II^IO))!2 = (OIAA+IO) = (0\l + AU\0) = (00) = 1.22) With relation (b) of Eq (6.20) we have for a = n = 1: A2\1) = 0. In the following manipulations we always use the commutator (6. v. (6. \ 9 \ I / II 4 T ) .25) (6. so that (a(A n )tA n \a) __ \\An\a)\\2 ( a . (6.5) to shift operators A to the right which then acting on 0) give zero.6.26) <0A4 t AA t + A4 f 0) = (Q\2(A]A + 1)0) = 2.23) Hence the eigenvalues of the operator iV := A^A are nonnegative integers. (6. I „ .17) a by (a — n). (6.24) This is a very important relation which we can use as definition of the state vector 0).20) we obtain from this that the right hand side vanishes. From relation (6..' " /. For a — n = 0 we deduce from (b) of (6. But A 2 A t 0) = A(AA^)\0) = A(A^A + l)0) = 0. The state 0) is called ground state or vacuum state.I n U .20) that A0) = 0.2 The OneDimensional Linear Oscillator be a normalized eigenvector of A^A with eigenvalue (a — n).
A^} = 2A\ and [A. (6. as we can see as follows.(A*)n] = n(Ai)n\ (6. Similarly we find 2)oc^Ut0).29). The states \n) thus defined are orthonormal.31) (6.26) the equality l) = A t 0).5) and again (6. (A^)3] = [A. In general we have \n) = ^(A^n\0) (6. According to Eq.32) we have (nm) = .33) . we have 2> = ^=dAi\0). i.11) we have [A.30) (arbitrary phase factors which are not excluded by the normalization have been put equal to 1). (6.Al] = 2A^A^ + (A^)2 x 1 = 3(A^)2. (6. and in general [A.29) Hence we obtain and in view of Eq. (6. (6. (0^ 2 (At) 2 0) = 2.e.^ ( O l A ^ n o ) .34) (6.32) (6.(6. (At)2]A+ + (A*)2[A. But using Eqs. Quantum Mechanics of the Harmonic Oscillator A2AtA*\0) = A(A^A + 1)4*10) = (AA^AA^ + AA*)\0) = = = {AA\A* A+1) + A* A+ 1}\0) (AA^+ 1)\0) = (A^ A+ 2)\0) 20) ^ 0.11).88 and CHAPTER 6. and in view of Eq. A^A* + A*[A. (A*)2} = [A. l>ocAt0).
.1 0).1 (A + ) m .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 . .36) + 1) (6. i.34) and (6. . In view of the properties (6. 2 . (6.39) A^A\n) = y/nA^\n . we do not have creation or annihilation of any real particles as in field theory (hence the word "quasiparticle").L ( A t ) n + 1  0 ) = VnTT\n Vn! and with Eq.1) = n\n). chosen in close analogy to the postulated "second quantization relations" of field theory.40) (6. In view of the same properties A is also called annihilation operator and A^ creation operator of socalled u quasiparticlesv.33). .34) A\n) = ±=A{Al)n\0) .41) The contribution hu/2 is called zero point energy. . in quantum mechanics.38) (6. however.35) Inserting this result into Eq.35) the operators A^ and A are called respectively raising and lowering operators or also shift operators. we obtain (n\m) = 5nm We also deduce from Eq. The terminology is.8) we obtain therefore H\n) = hu (A^A + ]. Here. n = 0 . (6.J \n) = hw (n + i J \n).6. (6. . (6. (6. 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 .32): rf\n) = .1 ) . l . the eigenvalues of the Hamiltonian H are En = hw\n+\+ i ) . With Eq. whose number is given by the integer eigenvalue of the number operator N = A^A.1) (6. •(» (6. ldnm = n\5nm.'n\ and = v n ![ ( A t ) M i= + (6.37) n^)711}^) = v n  n .e. .
is defined by projection of operators on eigenfunctions of energy.7) the energy representation of the operators q and p. and — in fact — we have used this already previously (see e.3 The Energy Representation of the Oscillator The energy representation. En = hu(n + ).43) = = y/n(n'\n — 1) = y/n5n^ni. A^ from the relations (6..45) Correspondingly we obtain with Eqs. (5. (6.e. n = 0. (6. also called Heisenberg representation.2. ( 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 / (6.46) v^ .6) and (6.. i.44) yfn. other elements zero. other elements zero.1.39)).90 CHAPTER 6. Quantum Mechanics of the Harmonic Oscillator 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 ... Vn + 1. Eq.39): {n\A.38) and (6.g.42) We can deduce the energy representation of the operators A. 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. The representation of the Hamiltonian in this energy representation is {n\H\n') = En5nn. (6.
48) Applying from the left the bravector (x\ and remembering that (x\p\<f>) = we obtain ih—(x\(j)}. i. ™ou(x 2h \ + J _ d ] m = 0 (649) mQUJ dx CemouJx2l2h. 2 .32). (6. (6. Correspondingly the position space representation is given by the wave function <t>n(x) := (x\n). 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.4 The Configuration Space Representation We saw that the eigenstates are given by Eq.1) and (6.5) are also satisfied as matrix equations. The ground state wave function (/>o(x) is defined by Eq. mow / (6.6.e. (6. Hence ( * Recall J™ dxe™ * ? 2 2 2 1/4 W \ i/4 \ em^x*/2h_ (g 5 Q ) = yfln/w . Eq.47) LP 0) = 0 . A\Q) = 0. 6. by (cf.24).4 The Position Space Representation 91 It is an instructive excercise to check by direct calculation that Eqs. (6.3)) ^ (q 2n \ + (6.
c/>n(x) := (x\n) = . they are called Hermite polynomials. i. Vn! Now {XlA (6.2 n n! The functions Hn(£) are obviously polynomials of the nth degree. ) e"^2.l ) n f f „ ( 0 .92 CHAPTER 6. (6.51) \l 2h {X q \ m0uP)\l 2h \X m0u. Quantum Mechanics of the Harmonic Oscillator This is therefore the ground state wave function of the onedimensional harmonic oscillator. In order to obtain the wave functions of higher states. These polynomials are real for real arguments and have the symmetry property # n ( .l ) > n ( x ) .32) to apply the appropriate number of creation operators A^ to the vacuum state 0).L ( x  ( A t ) »  0 ) . so that <t>n(x) = ( .e.56) < ^ = We have as an operator identity the relation .54) (6 55) i/4 rzri7r 1 / 4 vo. it suffices according to Eq.0 = ( . (6.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 . (6.
d£" (6. (202"2.g. Magnus and F.4 The Position Space Representation i. 2£.58b) by W. (6.58a) ' T h i s definition — apart from the usual one — is also cited e.12(20 . p. 93 .57b) • Generalizing this result and inserting it into Eq.e. e.It follows that 0en(Pe/2^Lez 2 n\P?i2 — c2 d d£ ? = 2f. (6. 81. .en_ dt.l ) n e ^ In particular we have #o(0 #i(0 #2(0 #3(0 tf4(0 = = = = = 1. gain as an operator relation we have de = een± = een da ee/2±_^ee/2 d£ 1 u ^ d? dt. — = n+ d ~^ i 2 + d 2 de) ' 1+e ~ (6.2)m = (±t)m for some function VKO.6. ( 2 0 4 .12.54) we find that (observe that there is no factor 2 in the arguments of the exponentials!)§ Hn{i) = ( . Oberhettinger [181]. (2036(20.
Oberhettinger [181]. 80.59a) The differential equation obeyed by the Hermite polynomials Hen(£) defined in this way is (^2 .3) are (as given in the cited literature and with a prime meaning derivative) Hen+1(0 Hen(0 He'n(0 = ZHenifi) .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.(e^/2) or Hn(0 = 2n/2Hen(V2ti).59d) The normalization of Hermite polynomials is given by the relations: oo / d£ Hn(OHm(Oe~e oo = 2 n n\^5nm. £33£. £2l. Quantum Mechanics of the Harmonic Oscillator The differential equation of the Hermite polynomials Hn(£) defined in this way is ( ^ 2 ^ + 2n )ff"(O=0. 81. £. pp.59b) = (l)^2/2^. (6.94 CHAPTER 6. Magnus and F.He'n{£). = nHe^iO.6 £ 2 + 3. . (6.59c) The recurrence relations satisfied by these Hermite polynomials (the second will be derived later in Example 6. (6.^ + n) He n(0 = 0 (6.60b) J—< "See for instance W. (6. Mathematicians often define Hermite polynomials Hen(^) as^ Hen(0 Then He0(O Hex{0 He2(0 He3(H) fle4(£) = = = = = 1. £ 4 . = ffen+itO+ntfeniCO.60a) and I d£ Hen{Z)Hem{Z)e?l2 = n! V & „ m  (6. (6.
62) Joo The following function is generally described as generating function of Hermite polynomials Hn{£): F(S.g. 93. F(S.s)=F(£. W. Magnus and F. . Oberhettinger [181]. (6.s) = eM?(Zs)2}(6.4 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£) = ( .64) n\ See e. and so on. s) as a Taylor series. Their normalization is therefore given by OO /"OO / OO J —OO z (6.61) e?l*2nl2Hn{Z).^ = e^/42~n/2Hn(aV2). p.63) The meaning is that if we expand F(£.s) = J2 n=0 (6.l ) ^ e^HeniO 2 / 4  .6.0) we obtain + s fdFs 1! V ds s=0 + OF ds 2 dF ds2 = s=0 2(f d S )^«)a s=0 = 2£ = ffi(0.. Thus the generating function can be written F(Z. 2(£  s)£2& s=0 s=0 ~ ds = m sf2]e?= •««) : S=0 2(2e2l) = (2e)22 = ^2(0..
has parity + 1 . 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 . i.96 CHAPTER 6. W i t h Apx = moAvx we have = moLoAx Ax and thus h/2 Apx h/2 mQuAx Ax I h/2 .1 The comparative behaviour of the lowest three wave functions of the harmonic oscillator.e.1. for x(t) = 0 and p(t) = mox(t) — 0. We also see from the eigenvalue t h a t the energy of the lowest (i. i. Furthermore we observe t h a t the wave function of the nth state has n zeros in finite domains of the variable x. 6. and we see t h a t even and odd states alternate.e.e. 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 classical mechanics the energy of the oscillator can be zero. Quantum Mechanics of the Harmonic Oscillator Fig. E0 ~ —{Apx)2 + IrriQ l mQuj2{Ax)2 moLoh/2 2TUQ 1 moU!2h/2 2 TTIQU = Ku). in order to estimate the lowest energy of the harmonic oscillator. We see t h a t the lowest state is symmetric. 6. In fact we can use the uncertainty relation AxApx ~ H/2. ground) state is fko/2. mow' Apx = ^m0Luh/2.
6.2 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". Example 6. Thus. 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)). /classical orbit Here q goes from 0 to a. to 0. in corrected version instead with n* = n + 1/2. But it has a pole at infinity. The analogous Example 6. In Example 6. back to 0. Sec. pdq.e.2 contains only the answers since the evaluation of the integrals proceeds as in Example 6. to —a.4) to obtain by contour integration the eigenvalues of the harmonic oscillator. E .1/a O l/a Rez all at infinity Fig.: 27 T mov2a2. classical orbit i.6..plane pole at z=0 ^ . This observability can be taken as direct proof of the uncertainty relation.1. Solution: Consider the simple harmonic oscillator of natural frequency of vibration u. According to "old quantum theory" with integer n.plane Imz z . setting q = 1/z. which is the complete orbit. original path q . Finally we point out that the Hamiltonian of the harmonic oscillator can be diagonalized in many different ways.4 The Position Space Representation 97 The zero point energy plays an important role in atomic oscillators. The kinetic energy is always positive or zero. we obtain the integral .1: Eigenvalues obtained by contour integration Use the corrected BohrSommerfeldWilson quantization condition (cf. 14. and is directly observable. like twoatomic molecules. n*h. 27rmoi/ <f> V'a 2 — q2dq = n*h. and with potential V(q) = 2K2mov2q2 and total energy E = p2/2mo + V(q). In the plane of complex q the integral has no poles for finite values of q. mass mo.
for instance — without deriving all properties of the solutions.3. With a2 = E. z = 0 Vi 2Kmov(Ka2) = n*h = E/v.2: Contour integration along a classical orbit Setting q = 1/z. Example 6. 6. .) hv.c2 = (I + 1/2) 2 (natural units. mo = 1/2) the integral I2 can be checked to give the eigenvalues for the Coulomb potential Ze2/q (2 turning points). 2 . « qdq + I2 = f dq . but here we leave this .It follows that 1 = s2 i ^V»a*ai ~2f z=0 2wi r m '„2 = 7ria 2 La*Va 2 2 2 .1. E = Z2e4/4(l + n+ l ) 2 (cf. since these two cases.114a)). 6.65) f ^Va2z2 . in q at infinity). Eq. We can evaluate the integral with the help of the Cauchy formula (the superscript meaning derivative) Jc Hence we obtain (z . In the following we investigate in more detail the important differential equation of the harmonic oscillator. .( 6 . 1 . hence we consider the following differential equation in which n = 0 . (11. harmonic oscillator and Coulomb interaction.e.5 The Harmonic Oscillator Equation In the above we considered mainly the energy representation of the harmonic oscillator and encountered the Hermite polynomials. After removal of a Gaussian or W K B exponential. verify that II = d> dq yi92 aq + ft 2TTI 9 /y/F^I 9z V a + bz z=o .5 .. i. are the most basic and exactly solvable cases in quantum mechanics. i.a) m = 2 .V62 a2). we start with a Laplace transform ansatz for the remaining part of the solution.1. (m .e. Quantum Mechanics of the Harmonic Oscillator This integral has a triple pole at z = 0 (i.1)! (6. .e.98 CHAPTER 6.5. a J \d Solution: The solution proceeds as in Example 6.2b = Ze2. and obtain with this all properties of the solution.2TTI a ^ = . £ = n*/w = ( n + . 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). In Chapter 11 we perform similar calculations for the radial equation of the Coulomb potential — as in Example 11.1 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. / " ^ .
67) the latter being the differential equation of Hermite polynomials Hen{t) for integral values of n.J ±(ps(p))e*dp. Oberhettinger [181]. 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 .g. Then (with partial integration) *dt = ~*/Ps(p)e. (this determines the limits). we obtain P2/2 ln(ps(p)) = — p — nlnp or s(p) <x Tl+l P (6. 80.e.i ^  (ps(p)) P n P Integrating this expression. (6.66) First we have to remove the £2term. (11.g. and / dpf(p)pn the Mellin transform of / ( p ) . Substituting these expressions into Eq. since this can give rise to exponentially increasing behaviour (at most a factor t can be tolerated). Hence we set (with the chosen sign in the exponential) m=y{t)et2l* with *0.** We make the Laplace transform ansatz^ (with limits to be decided later) y(t) = [ s(p)e~ptdp.6. so that the integrand of the remaining integral vanishes. . .111)) the Mellin transform of e~p is n!.t*ij& + ny{t) = O. i. E.p t dp = \ps(p)ePt} .67) we obtain \ps(p)e pt] + I P2s(p) + —(ps(p)) + ns{p) e~ptdp = 0. W. Magnus and F. This has to be true for all values of t. (6.69) **See e. (6. (6.68) § = J p2s(p)e*dp. Eq. p. / dpf(p)e~px the Laplace transform of f(p). fp{ps{p)) = {v2 + n)s{p). We demand that C(p) := — \ps(p)e~pt] = 0. (cf.
6. so that p cannot go to infinity. We now investigate possibilities to satisfy the boundary condition C(p) = 0. Hence a possible path of integration > is from — oo to oo. i. since otherwise (with p = \p\ exp(i8)) we will get part of the solution from 0 to infinity. This can be attained if p is allowed to be large (t can be of either sign). 6.100 and hence CHAPTER 6.. y(t) oc 2vri J pn x dp for n integral and ept p 2 = coefficient of pn in ~^ (6. we can choose a path from zero to infinity (the condition will vanish at p = 0).3. This exponential assumes a maximum value when pt \—p2 = minimal. Quantum Mechanics of the Harmonic Oscillator ptp2/2 » ( « ) ptp2/2 dp with C{p) = (6. (6. then for large t we have y(t) ~ exp(i 2 /2).70) = 0 P. We can show that if \p\ is allowed to be large. as indicated in Fig.71) . of course). when p = —t (which is possible. = / • Im p Rep n not an integer Fig. The quantum mechanical wave function is then (cf. Therefore we take ptp2/2 y(t) .e. (3) We can choose a contour once around p = 0 in the complex pplane.n+l between the two limits of integration. so that y(t) ~ exp(i 2 /2). For singlevaluedness. n must then be an integer. Thus the only solution left which does not involve large values of \p\ is that with § Q for n an integer. (1) This will be satisfied if \p\ — ±oo. To see this. The value of the exponential at this point is ~ exp(£ 2 /2). (2) If n < 0.3 The path when n is not an integer.67)) M) * e"* /4y(t) ~ et +2/ 2 /4 But this is not permissible. Eq. consider the exponential in the integrand.
(676) . . Therefore the coefficient of (pt)n in exp(—pt — p2/2) is (—l) n /n\. 2 . the term of highest degree in t in the exponential corresponds to taking all p's from exp(— pt).if n is odd.3. pt p2/2 (6. The Hermite polynomial Hen (t) is now defined such that the coefficient of tn is unity.72) We now obtain the polynomial form with Cauchy's residue theorem as Hen(t) = ( .5. . n—v even ^ 2 ' Hence we obtain the polynomial of degree n Hen(t) = nl £ see below ( _ 2 ) ( n . In y(t) above. The exponential function here is described as the generating function of the Hermite polynomials Hen(t).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.l ) n n ! x coefficient of pn~v in V LJ_L e p 2 /2 *^ v\ u=0. . We can also obtain the derivative form of this Hermite polynomial.5 The Harmonic Oscillator Equation 101 with Cauchy's residue theorem.. We have 27T! Setting q — p + t. 4 .„y2 v l { ' ^ r (^) \ 2 ' where v = 0. The Hermite polynomial is therefore given by H ^{t){^—f—^r~dp. this becomes jfj. . if n is even and v = 1...P 2 / 2 00 C—tY ( .l ) n n ! x coefficient of pn in = = (~l)nn\ e~ ~ x coefficient of pn in V l z ^ e .6.
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.4: Orthonormality of Hermite polynomials Establish the orthogonality and normalization of Hermite polynomials. we obtain tHe„{t) (l)"n! ~ i. Solution: We have to evaluate °° 2 .59a). Example 6. (6. one can construct the tower of polynomials Example 6.70). ( n .102 CHAPTER 6.75) as Hermite polynomials. Quantum Mechanics of the Harmonic Oscillator which we recognize as agreeing with Eq.5 is a further application of the method for the evaluation of integrals that occur frequently in practice (e. Example 6. 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. 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).)} 2 /2dt = ePde(P 2 W)/2^^} J —CO . (n. 8.3: Recurrence relation of Hermite polynomials Show that tHen(t) = (n.1 '2dt using Hen(t) = (l)n27 r .g. n + l) = l.n + l ) f f e n ^ i ( £ ) + (n. We leave the consideration of the recurrence relation of Hermite polynomials (of paramount importance in the perturbation method of Sec. 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+<. Pptp 2 / 00 /2 dp.77) = t. in radiation problems.4. n — l)Heni(t). Reinterpreting the remaining integrals with Eq. and elsewhere.n— 1) = n. Hei(t) Henj. as in the normalization of asymptotic expansions of Mathieu functions). Starting from Heo(t) = 1. (6.3 and 6.7 that we use throughout this text) and their orthogonality and normalization to Examples 6. (6. Hen(t).e.
6.q) = = e . Hence the normalized wave function of Eq. g) is n! Next. g) is the product V^Tre" ^oM!(n/i)!(n + 2rM)! (6.82) ^ . so that /2 Q2(n+rrf in / ( p .jQ) n n!I ^—' ^Q fi\(n . the coefficient of qn+2rn the coefficient of p"qn+2r m exp(.n)\ i s an+2rij.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.q). It follows therefore that the following integral (with an exponential exp(—at) from whose expansion we pick later the power of i) is given by °° 2 / where OO eatHen(t)Hen+2r(t)et l2dt = n\(n + 2r)\ X coefficient of pnqn+2r in f(p. (6.5 nm .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.80) f(p.^ + 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 .66) is \Zn\V2w Example 6. (6.(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)./(n __ 2r — /i)!. zero otherwise (6. and 0 otherwise \/2n(—l)n+mn! if n = TO.78) \/27rn!.
. n.l ) ( n .i / ) ! ( 2 r + i/)! > . n + 2){n + 2.e '' ^ o^M ! ( n .104 CHAPTER 6. n + 2) + (n.e. n ) ( n . the relation (n. Prom Example 6. i. . the coefficient of T 2s »{sr u)\ As we observed.l . n + l ) ( n + 1. n + 2)]ife„ + 2 (t) + ••• .77).79). n + 3)(n + 3.^ ) ! ( n + 2 r .78).ra + 3)(n + 3.^ / 2 * . 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.Y^ i[ ^0vKnv)\C2r + ») cc2s A (n + 2r)! a 2 ^ . K is the coefficient of a / ( 2 s ) ! in the expression J . n + l)(n + 1.— — ^ .(t)e«2/2dt Now. n + l ) ( n + 1.A o?(.r . n + 2)(n + 2.r .n+ l)(n + 1.' .u)\ 2s<—" v\(n . n + l ) ( n + 1.n + 2) + (n.3 we know that upgoing coefficients are 1 and downgoing coefficients (n. n .r+v) a2s coefficient of in (n + 2r)\ V (2s)! h . i / ! ( n .83) = (n + 2r)!e a / 2 V „ „.' '2dt= / Jo t4 Dn(t)Dn+2(t)dt = V2^(2n + 3)(n + 2)!. n — 1) are given by the first index. Quantum Mechanics of the Harmonic Oscillator It now follows t h a t — in later equations with v := (n — /i). 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 . n + 4)_ffe n +4(t) + [(ra. n + 2)(n + 2. i..v)\(2r + v)\ > Q (2s)!(n + 2r)! AT 2~ s r 2V v\(2r + u)\{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 + l ) ( n + 1.( n + Jr). 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.e.t • 2l/2dt _ J : Q2 » a2(n+r^) ~ f ° ° f f „ t2/2 „ !ZoHen(t)e2 Wdt .e. n + 2) + ( n . by n!(7r/2) 1 / 2 .2s _oo ( a 2 / 2 ) i . A « 2 ( r '+ I/ ) 2 coefficient of (2s)! in (n + 2r)\e ' ^ ." ) a2^) r coefficient of (2s)! in f^ (s .A i ) ! (6.v)\ ^ Multiplying both sides by the value of the normalization integral. The result then follows with half the normalization and orthogonality integrals (6. n)(n. i.° i i f e T 1 ( t ) i f e „ + 2 r ( t ) e .v)\(s . we obtain the result (6. n + l)(n + l . so that now with integration limits from 0 to oo K = coefficient of a2s (2s)! in J a = 2.
With reference to this case. in Chapter 21 (with a different calculation).x'). These are the Green's functions of the timeindependent Schrodinger equation or. We 105 .2 Timedependent and Timeindependent Cases In the case of Green's functions we distinguish between those which are timedependent. we consider the sojourn time of a quantum mechanical particle at a point. 7. Let H^ = E^> (7. which we write here K(x.1) be the timeindependent Schrodinger equation which has to be solved. This example also offers a convenient context to introduce the inverted oscillator potential. and those which are timeindependent. We consider first timeindependent Green's functions. which we write G{x. The first of these cases will reappear later. of a differential equation of second order.x'.1 Introductory Remarks In this chapter we consider Green's functions of Schrodinger equations in both timedependent and timeindependent forms.t). which is normally not discussed. in Feynman's path integral as the kernel or free particle propagator. and is thus an important point to be noted here. The other example of the oscillator enables us to evaluate corresponding expectation values of observables in the canonical distribution introduced in Chapter 5. more generally. which is a point of unstable equilibrium for a classical particle.Chapter 7 Green's Functions 7. 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.
Green's Functions H = H0 + Hi. and Hi is some other contribution.g.e. where £ ( 0 ) = lim E. We can readily see that G(x. the relation X') = 5{x .x') or ^2^(x)^*(x') i = S(x . appropriately decreasing behaviour at infinity).g. 'The case of £(°) equal to some Ei is considered later. since the generalization to higher dimensions is selfevident.EW)xG{x. (7.5) M = Z*W=EW i l for Ei0) E * (7 6)  This holds since *For simplicity we consider the onedimensional case. e. i. of the kinetic part p2/2mo and some part of the potential. for instance. cf. The boundary conditions which G(x.106 assume a Hamiltonian of the form CHAPTER 7. although the explicit calculations can be much more involved. square integrability.x').g.3) i which has the specific representation ^2{x\i)(i\x') i = 5(x .4) where the functions ipi(x) are solutions of HQij)i{x) = E^iix).x') by recalling the completeness relation. (7. The timeindependent Green's function G(x.x') can be written t G (7.x'). (7. e. Eq. . (7. x') has to satisfy correspond to those of * (to insure e. We can obtain an expression for G(x. a perturbation part like (3q4.x') is defined by the equation* (Ho . aq2. i.13).2) where Ho consists.e.
For the solution V we can immediately write down the homogeneous integral equation (again to be verified below) *(a.7. (7.t) since ih^K(x.0) = S(xx').E^ of the eigenvalue.E^ We can check this by considering (H0 . (7.8) is satisfied as a consequence of the completeness relation of the wave functions ij)i{x). which consists of the perturbation part Hi of the Hamiltonian and the corresponding correction E .11) = = 3) = fdx'(H0 . We see that the initial condition (7. (7.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.. . The timedependent Green's function K(x. We can see the significance of the timeindependent Green's function for the complete problem as follows.#.7) with the initial condition K(x.E® (EE{0) Hi{x))m{x).x.x')(E .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.) = f dx'G(x.E<®)xG(x.EM)X*(X) ( .Hi{x'))^(x').x'.x'. We rewrite the complete timeindependent Schrodinger equation H^ = E^ as (#0£(0))* = (££(0) #/)*.t) = ^Eie^ihM^i(x') i = Y^eE^ihUx)^(x'). x'\ t) (7.£(0) .(*'))*fr') Hi{x'))^(x') f dx'5{x .x'){E .t).t) (K for "kernel") is defined as solution of the equation ih—K(x. i (7.10) The contribution on the right hand side is the socalled perturbation contribution.9) = i H0YieE^i%(x)^(x') = H0K(x. . x'){E . (JL x'\ t) = H0K(x.
H. In this case the perturbation is restricted to that subspace of the Hilbert space which is orthogonal to ipj. 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!). Booss and D.11) are called (homogeneous or inhomogeneous) Fredholm integral equations. 17. Examples in various contexts have been investigated by L. that (E .t) for the complete problem as follows.e. (7.11) 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). The timedependent Schrodinger equation ih\><l>)t = H\il>)t (7.15) ''This problem and its circumvention can be formulated as a theorem. then at best such a contribution would be something like cipi(x). Wiedemann [183].11) with a system of linear equations of the form Vi = Mijyj. .6) we would have (7.6) is not defined. MiillerKirsten and A. But if we assume that E^ ^ Ei for all i. This difficulty* can be circumvented by demanding from the beginning that 0= f dx'tf(x'){EEjHi(x?))V{x'). i. (7. Merzbacher [194]. In general. (7. it is possible to solve the integral equation by an iterative perturbation procedure (see later: Born approximation).x'. W. integral equations are more difficult to solve than the corresponding differential equations. Maharana. see E. But then the Green's function (7. D. (7. however. The theorem is also known as Fredholm alternative. Sec. see B.12) Equations of the form of Eq. (7. In cases where an inhomogeneous contribution is given. Instead of Eq. Bleecker [33]. This will be different.14) We can now see the significance of the timedependent Green's function K(x.108 CHAPTER 7. if we assume that E^ = Ej. 2nd ed.. the function ipi(x) would not be a solution of and it is not possible to add an inhomogeneous contribution. Green's Functions Can we add on the right hand side of Eq. J.
21) dx'K(x. (7. 109 (7.t) = J2 fdx/^n(x)^*n(x')^(x'.g.x'.x'. (7. (7. We use this in Sec.t) = where if(x^'. 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.15) which contains (7.0). H\En) = En\En).t)il>(x'.2 for the computation of the sojourn time.51) shows that we obtain a very analogous expression for the density matrix PN(P) a s for the Green's function K(x.. Merzbacher [194]. p. (5. <a#) t = ]T n J [dx'(x\En)(En\x'){x'\^)t=0eE^\ (7. (7.e.19) i/>(x. x'.0)eEnt/in.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. e. 2nd ed. (7. t)\2. Eq. *>0.The solution of Eq.16) As usual.g.7.Q) = 6(xx').20) the Green's function K(x.2 Green's Functions: Timedependent and Timeindependent permits stationary states \En) defined by \1>)t = \En)eE^ih. (7.20) is the timedependent Green's function. x'. this relation provides the probability density \ip(x. 1 See e. t > 0.t) = ^ e £ " f / « V n W < ( ^ ) . in different formulations. 7. (7.t) is known.17) is the initial condition of \4>)t.17) as initial condition is obviously W)t = Y. for the oscillator potential.' which obviously satisfies the initial condition K(x.22) According to Eq. i.18) or.x'. E. Comparison of Eq. .158.5. \En)(En\ip)t=oeE^\ n (7. t) with the Note that when K(x.t) describes the evolution of the wave function from its initial value ^(a^O). (6.7) with Eq. We can write this expression also as§ ip(x.
ie (7 . In the following we shall derive the Green's function for the case of the harmonic oscillator.23) along the contour C in the plane of complex E as shown in Fig.x'.i /f e «/. 7. With Cauchy's residue theorem we obtain I(t) := J2eEnt/ihMxWn(x'Mt) n = K(x. t) = J2 e £ " f M i ( x ) < ( x ' ) . we obtain / W : = .1 The contour of integration.110 CHAPTER 7.e.x') and K(x. Jc 27T „ hn . 24 )  ReE Fig.x'. As a consequence of the above considerations one wants to know the connection between the timedependent and the timeindependent Green's functions. we obtain then also the corresponding density matrix. t > 0.«^«2 W . x')6(t) (7.t). We therefore consider the following integral with e > 0: I(t) := i J ^eEt/*hGE^e(x. 7.t) K(x. Green's Functions difference that (5 = 1/kT plays the role of it. = J2 ^„(s)C(*') En — E We see that G = GE depends on E.x') = GE(x.25) in agreement with the timedependent Green's function . x'. between (with E^0' = E) G(x. (7. i.E . Inserting for GE+ie the expression above.1.
we obtain *I£I = h2 —(2AB).29) tJJ 2 2B(x .x'.e.0) = 5(xx').29) and (7. We therefore consider first the simplest case with Hn = P 2mo h2 a2 2mo dx2 (7. / dxK(x.x'.27). (7. and identifying coefficients of the same powers of t on both sides. (7. (7.31) The constant A has to be chosen such that K(x.e.30) Inserting Eqs.— d2 —2K(x. .t) t h2 = .7. B m0 n _ 2ih 2iK (7. which is moving in one space dimension.x'.3 The Green's Function of a Free Particle 111 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.X>.0) = 1.7). v ' 2m 0 v h i.26) This is the case of a free particle with mass mo. It is clear that it is nontrivial to solve an equation like Eq. K(x. i. Thus we try the ansatz. A and B being constants.30) into Eq.t) = ^eB^X In this case we have dK ~dt and dK dx 2 8K dx2 A A 2t3/2 A tV2 '^'K (7. 2mQK h h2 ih(AB) = —{4B2A). In this case the Green's function is the solution of the equation d hK(x. (7.x1) t B(xx')2/t B(xx')2/t 2AB 4AB2(xx')21 ~¥l 2+ W 2 (7.28) B(xx')2 t2 DB(xx') 2 /t (7.27) An equation of this type — called of the type of a diffusion equation — can be solved with an ansatz.x>.t).
e. 158.t) into Eq.(7. Merzbacher [194]. i.34) K(x. "See the excercise in E.20) and can then obtain ip(x.36) 2mo Then — provided that the parameters assume values such that the intergral exists — K{x. (7.36) ^ W Q ^ J emo(ii') 2 = —eP2lia 27T dke~^k^2^2 J„00 /2«tS . 32) It follows that x'. *>0.x't) = Y. t) = /jJo_ e mo(««') 2 /««.H We can insert the expression for K(x. p.oo 27T (7.t)= f^em2t/2m0iheik(x~X')_ J 27T (735b) We set a = i—.. j3 = i(xx'). (7. Mx)^M^) „ n so that J =^ .21). we have This is 1 provided A=JB= ^ (7 . v27r En^^.ihMxWn{x'). from K{x. we > have to make the replacements V ^ fdk.x'. (7. n For a free particle moving in the onedimensional domain \x\ < L — oo.35a) K(xy. (7. (7. 2m0 (7. (7. 1st ed."737) \27rii/i in agreement with Eq.112 CHAPTER 7. Green's Functions For parameter values such that the following integral exists.33) V Lirvnt Can we demonstrate that this expression can also be obtained from Eq.0)oce"Qx2+ifcox.zEnt.x'.t) — for instance for a wave packet given at time t = 0 of the form ^(x.38) .t) = — / dkeak2+Pk 2vr J.33).
7.t).e./ ) + fK(x. (7.25) — in the context of Feynman's path integral method.t) + ~x K(x. I (7.t) We now set h2 d2 = .42) .e. .t).39) In this case the timedependent Green's function K of Eq.8). i. (7. (21.x usn ot i. We consider the onedimensional harmonic oscillator with Hamilton operator HQ. .33).——K(x. K(x. .x'.x'. i.37) will later be obtained by a different method — see Eq. (7. HQ = ^—p2 ZrriQ + \m0uj2q2.0) = 6(xx') at / = 0.. We rewrite this initial condition in terms of £ and use for a = const.x'. the relation* 6(x) = o<y(os). x'.x'. 5{ax) = ± J eikaxdk = ArS(x). 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 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. / ) = ~^K(x.3 The Green's Function of the Harmonic Oscillator 113 The result (7. —KK(X. mQLu . §jK(x. x'. in this domain *Recall that S(x) = ± f eikxdk. —ih—K(x. ..t) = h2 2 d2 TW .40) becomes 2 .7) is the solution of d ih—K(x. z ..7.e. / ) (7.x'. (7. 1 2mo u>n ox n with the initial condition (7.t) 1 + m0u2x2K(x. (7.e. i.X . . .d r.40) Then Eq.
K { ^ f ) ^ ^ e ~ { ^ ' ) 2 / i f f r ° ^ ' (7 45> The same approximation is also valid for large energies E and for t or f small (near zero) in view of the relation AEAt ~ h.b2.45) in this sense.4a&£ + 2a .47) We insert this ansatz for K into Eq. P. Identifying coefficients on both sides. 50.t) become similar to expression (7.47) becomes (7. we must have /o = 0.c o t h 2 / .48) ocexp[{a(/)C 2 + 6(/)e + c(/)}] (7. f i. c(0)^.50c) (7. . that is. so that for / — 0: > a^—. (7. (7.e.42) and obtain the equation (with a1 = da/df etc.) a'£2 + b'i + c' = (1 .50b) (7. we obttain a' = l .50a) we obtain (7. Feynman [94].51) See R.4a 2 )£ 2 .e.48). Integrating Eq.50a) (7. Green's Functions to the particle behaves almost like a free particle. W ) .f) with a(0)»y. as the limiting case / — 0.x'.46) If we interpret Eq. (7. (7. (7.. i. it • is suggestive to attempt for K the following ansatz^ m.^ .45) in accordance with (7.b2.114 CHAPTER 7. p. a = .49) a=icoth2(//J0 ) w 2 "' 2tanh2(//0) * To ensure that the expression (7. b' = Aab.33). and we expect K(x. (7.4 a 2 .Z'. (/ = 2a.
p.7. as one can verify.x'.27). with A. x .e.33) for the Green's > function of a free particle.0). t. and to ensure that we obtain the prefactor of Eq.45). we can use K(x. x') of the density matrix p^ (with respect to the canonical distribution with (5 = 1/kT): pN(x. (7.55) 'For an alternative derivation and further discussion see also B.48). we must have (besides A = —£') = /m 0 u.53) or.In B.40) of the timedependent Green's function with Eq.* With this result we have another important quantity at our disposal. B independent of / ./3) = J IJIQUI 2irhsmh(hLu/kT) 2hSinh(hw/kT) \{X +X j C x exp ° S n kT lXX J (7.54) For t —• 0 this expression goes over into the expression (7.50b) Kf) = —r^77) smh 2 / A independent of / . (7.£) _  {(x2 + x /z ) cos tot — 2xx'} 2msm(o. (7.e. Finally Eq.47) we obtain ^ c o t h 2 / + ^ + ^ c ° t h 2 / (7.2.5.3 The Green's Function of the Harmonic Oscillator Correspondingly we obtain from integration of Eq. (7. Felsager [91]. K(x. as we shall see in the following.X'. i. / —————rexp y 27rmsin(u. (7. c(/) = i ln(sinh 2/) + ^ 2 c o t h 2 / .b(f).x'. we must have A = ('.i) (7. i.50c) yields for c(/).48) 6(0) = — £'/2f.t) = . 174. (7. . (5.t) to obtain this element (x. 7. in particular for the derivation of the sojourn time in Sec. In order to satisfy Eq. (7.51) for the density matrix PN(X. (7. if we return to x.c(f) K = 5 =exp Vsinh 2 / into Eq. 2TT^ Inserting a(f). Comparing the Eqs. 115 To ensure that in accordance with Eq. x'. c(0) = £' / 4 / .
^—• Irpiv J pN(x. we skip the algebra here. For instance we have with Eq. we obtain § < Z> = ^ c o t h ^ ^ ° — . (5. [ [ 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. <2 (7.» l.116 CHAPTER 7.56b) w v ' 2mQu} 2kT 2m 0 u ' What is the meaning of this expression? At temperature T the fraction (cf.e.x. eP E i = i + y.40)) 1 epEi 1 Ui = y. Eq. Eq. (5.x.. at temperature T): {A) = Tl{pA) = ^ > .55). 52. Thus the system is in a mixed state and the expectation value "Cf.52): Tr (PNq2) Jx2pN(x. p. (5. Green's Functions With this expression we can evaluate (cf.56a) the expression (7.P) 2.e. = jdxp(x. (7. 11 For T ~+ 0: u>0 . 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"}.uij>0 >• 0. P. . )x Inserting into Eq..p)dx Thus for (q2) = Tr(pq2) = ^2i(i\pq2\i) we obtain: 2 (7.x.52)) the expectation value of an observable A in the canonical distribution (i. R.eft >E*) E (7 57) ' of the number of systems of the ensemble occupies the quantum mechanical state i.56a) {q2) = = Y.p)dx W ) = —^r = —p . Feynman [94].
1 mo^ exp lmow / 2 2 ft X (7.J (w — ie)< e (n+l/2)Et e i(n+l/2)u.t) w—>UJ—ie to—n t—>oo moco/irh ^ ' piuit g—iwt exp m ^\(x 0 ^ . If we consider the system in the pure state \i).59a) E 0 = ^fiw. In this case the Green's function K(x.50) we obtain. t) of Eq. <*2>o dxx a •K j e IT I da r dxeax *\1/2 da a J _ 1 _ _ h 2a 2mou j —( . setting a = mooj/h.3 The Green's Function of the Harmonic Oscillator 117 (7. x'.e. and we replace to by LO — ie. 2 + x'2)le^2xx' to—>w—it t—>co 2 .59b) n+ l \ ftwt here 2hf exp _ i ( n + . (6.t "With the normalized ground state wave function of the harmonic oscillator given by Eq. 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. of K(x. /2N 2~7T: = eE°/iht(t>Q(x)(t)0(x') for £ > 0.7.54). We return to the Green's function (7.58) Next we explore the connection between the explicit form of K and the latter's expansion in terms of a complete set of states. We assume t > 0 and t —> oo.x'. x'. i.56b) is that with respect to this mixed state (whose cause is the finite temperature T). This is the first and hence dominant term of the expression (7.QUJ (7. (7. which means in the oscillator state \i) with eigenenergy fku(i + 1/2). e > 0. 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.21).54) is K(x.
In Example 7. i.e.59a). Considered classically.x'. The wavelength A is given by The phase velocity v<p.61) The word "plane" implies that the points of constant phase <p := k • r — uit at t = const. of dpN as pN(x. k = k. is defined as vv = . (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. which are planes. i. as in Eq.5 The Inverted Harmonic Oscillator We encountered the inverted harmonic oscillator already in some examples. 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.(3) = J2^EnMx)K(x') n „ ^ ° e^MxWo^') (760) 7. the velocity of planes of equal phase. Green's Functions For t large (i. However quantum mechanically in view of the uncertainties in position and momentum. the particle will stay there only for a finite length of time T. a particle placed at the maximum of the inverted oscillator potential (which is classically a position of unstable equilibrium) will stay there indefinitely.5.1 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)]..e. lie on surfaces.1 we estimate T semiclassically.62) . 7. to infinity) the contribution with n — 0 dominates.e. and with this we estimate the sojourn time T. (7. In a very analogous manner we obtain the solution of the equation for the density matrix. (7.118 CHAPTER 7.
/OO / / / oo /*oo \f{k')\\f(k")\ei^W^k"»dk!dk" l/(fe.7. xt— + — = 0. i. One defines as centre of mass of the wave packet that value of x for which d(p — = 0. (7. i ) = /*/(k')e i(k '" r '' .)ll/(fc")l[cos(¥'(fe. then ^(r.e.63) The relation u = w(k) is known as dispersion or dispersion law. ^ ( r .<p(k"))}dk'dk".k ') 2 . or dco' da The centre of mass determines the particular phase. u = w(k). for which \ip\ assumes its largest value: OO . (7. to the function ip given by oo / fik'y^'x^dk'. t) is the spatial Fourier transform of this Gauss distribution — as we know. oo (7. e.e. A wave packet is defined as a superposition of plane waves with almost equal wave vectors k.64) where / ( k ' ) differs substantially from zero only near k' = k. essentially again a Gauss curve. / oo J—oo . If we assume for / ( k ' ) a Gauss distribution.4 The Inverted Harmonic Oscillator 119 Every frequency u belongs to a definite (particle) energy E (cf. i. duo' da .Jt + a.65) Let f(k') = \f(k')\eia and <p := k'x .Q ( k . i. t) describe the motion of a classical particle? For reasons of simplicity we restrict ourselves here to the onedimensional case.e. the fundamental postulate on matter waves in Chapter 2): E = hu. a > 0 .)¥'(*:")) +ism(<p(k') . We now ask: How and under what conditions does the time variation of the function ^(r. e . ' t) dk'. The wave packet describes a wave of limited extent.
1: Fourier transform of a Gauss function Calculate the Fourier transform of the spatial Gauss function e . we obtain the deBroglie relation p — hk. 9 cku dk dhw dhk dE dp dE dp' We can also argue the other way round and say: By identifying vg = v. For E = hio.. Thus the expression g{k) is solution of the following first order differential equation g'(k) + Aff(fc) = 0. In the present case.120 CHAPTER 7. (7. we can use a simpler method. dE v 9 = .67) The centre of mass moves with uniform velocity vg called group velocity of the waves exp[i(kx — cut)]. = — (768) It is instructive to consider at this point the following examples.69) In general one uses the theory of functions for the evaluation of this integral.fc w> vv = ~a^ = S r a d P E «i7 Sd u .^ g=r a rad. however. a > 0. p = hk the group velocity is equal to the particle velocity v. The threedimensional generalization is evidently du . Green's Functions This expression assumes its maximal real value when <p(k') = ip(k") = const. as claimed for =0. . Example 7.— g(k). Solution: The function to be calculated is the integral dxeax OO ezkx. . i.e. 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 = . (7.
71) (7.69) and (7.. 2 f + . (7.43) one can verify the following important relation* 5[{x a)(xb)] = T—!— [S(x .73) 27T 7 .a) + 6{x . W.73) we obtain the requested representation of the delta distribution with 1 f°° 1 € Six) = lim — / dke~e^eikx = lim . or see H. (7.2: Representations of the delta distribution Use Eqs.7 °) With the help of this example we can obtain some useful representations of the delta function or distribution as in the next example.4 The Inverted Harmonic Oscillator Simple integration yields g(k) = c e " f c 2 / 4 a .74) Example 7. XZ (e > 0).. Example 7.69) and (7.72) Solution: From Eqs. (7.0 0 7T 7 0 7T e From Eq. . Since /•c 121 9(0) = J—00 dxe J_ we obtain S(fc) = ^ e " f c 2 / 4 a ( 7 .3: The uncertainty relation for Gaussian wave packets For the specific Gauss wave packet of Example 7. We have — 1 / * ° ° dke~eWeikx =  I / " 0 0 dke~tk cos kx =  1 ..1 verify the uncertainty relation Aa.70) to verify the following representations of the delta distribution: X2/€2 5(x) = lim ?—=e ^ o eV7r and 5{x) = \\m^r^.. J.75) ' E . Appendix A. (7. (7. where c = 9(0) is a constant. \a — b\ (7.&)] for a + b.. with partial fraction decomposition.70) we obtain 6(h) = lim — f°° dxeax2eikx = .r „• With the help of Eq. (7. (7. —±=ek*/ia2 / 4 " = lim 1 — — lim ^e* = The second important example can be verified by immediate integration. MiillerKirsten [215]. g .Afc = 81n2.7.
70) the Fourier transform of f(x) is (7. i. when Ax is very small. In quantum mechanics the square of the modulus of the wave function ip(x. i.« 2 ( * . Solution: In Fig.) leads to a very broad maximum of the Fourier transform \g(k)\.g.77b) (7.t) is a measure of the probability to find the particle at time t at the position x. requires according to Eq.77a) g(k) = e . The breadth A/c of the curve g(k) around k = kg.78) It follows that the product of the uncertainties is AxAk = 8 In 2. (7. A simple calculation yields A i = 2v/2~In2a.e.r)1/2a Re f(x) Ax Fig.2 The Gaussian curve. where  / ( x )  = m a x  / ( x )  / 2 .* o ) 2 / 2 . the width Ax is a measure of the uncertainty of the probability. at time t = 0). Thus a slim maximum of the curve of /(a.CHAPTER 7. The wave function corresponds to the function f(x) in the above considerations (e.76) 2ira The uncertainty Ax is defined to be the width of the curve at half the height of the maximum. Physically . Green's Functions 1/(27i)1/2a — 1/2(2. Thus a sharp maximum of the function f{x). and hence implies large values of Ak. 7. 7. where g(k) = max<?(fc)/2. (7.2 we sketch the behaviour of the Gaussian function p—x 11a ~—ikox (7.77a) small values of a.e. According to Eq. is Afe = 2 V 2 1 n 2  (7.
As G. i. In the same limit the momentum uncertainty Afc grows beyond all bounds. The function g(k) is therefore described as momentum space representation of the wave function f(x). nr = S(x).".5. In the limit a — 0 we obtain from Eqs.79) According to Eq.7. Barton recalls that W. Barton [15]. (7. The result (7. many supposedly elementary problems request the calculation of the sojourn time of a quantum system near a classically unstable equilibrium configuration. For reasons > of simplicity in the following we set in addition t mo = l.. In the reverse case a sharp localization of the particle in momentum space implies a correspondingly large uncertainty of its spatial coordinate. (7.. <7*» We assume that initially. Lamb at Oxford set the following problem in an examination in 1957: "A pencil is to be balanced so as to stand upright on its point on a horizontal surface. (7.u> = l. Barton mentions at the beginning of his paper.54) with the substitution ui — iu.81) *G.76) ex 2 /2a2 ^2na a>0 um 1/0*01 = lim a^O . 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. the less precise is the determination of its associated momentum. (7. This means the particle is localized at x = 0. which means all values of the momentum are equally probable. 7.2 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.75) implies therefore: The more precise the coordinate of a microscopic particle is determined.54) these parameters appear in the combinations moui/h (dimension: l e n g t h .t) = V 27Ti s i n h t exp 2 sinh t {(x2 + x'2)cosht2xx'} .h = 1. so that Ki0(x.20) the wave packet at time t > 0 is obtained from that at time t = 0 with the relation «. E. (7. Thus G.* We obtain the Green's function K{0 for the inverted harmonic oscillator from Eq. at time t = 0. Estimate the maximum length of time compatible with quantum limitations before the pencil falls over.4 The Inverted Harmonic Oscillator 123 p = hk is the canonical momentum associated with x.2 ) and U)t. Drastic idealizations are then required in order to reformulate the classical situation into that of a quantum system. O ) . .«> = / ^ C ' W .x'. ^In Eq.71) > and (7.e.
2 (7.83) ip(x.83) sinh 2 t + (i/2b2) sinh 2i 2 sinh £ 2A 2 sinh 2 1 (7.81) and (7..(Ax' + Bxf \/27risinh + (B2 + ^cothi)a.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 (7.82) i/f(x.t) = J oo dx' Setting A2:=±we can rewrite ip as ex b2 and AB sinh t' (7.124 CHAPTER 7.84) dx' oo tvirl/2b With some algebra one can show that (B2 2y and A2 sinh 2 t f7 s ^ sinn / + icotht) = . + cosh 2 1.88) .t) = J oo P .86) Evaluating the Gaussian integral of Eq.cosh t sinh t' Wisll2b ^ (7. 9 (7.(i/2) sinh 2t 2 2 2C 2 (i) + *c^.84) we obtain* ip(x.80) we obtain „/2 oo ex P 2 s i n h t { ( : C + x')^ht2xx'} \/27usmh . . ' 2\Aswht % cosh t (7. (7.. Green's Functions Inserting (7. = .85) = —r= o2 7 2 sinh 2i. A . (7.87) <9 2 „9 2 sinh21 i? = A + B f / ^ d x e " ^ / 2 " 2 = v^F<*.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 .
•&(t)=//c(t) = / or.£ 2 (7. since dy Jo the derivative dT_ ~dT For b = 1 we have f°° dt= / (7.94) / dxf(x)=g'(y)f(g(y)). t) the phase ip drops out and one has (x.89) = exp[x2/C2(t)} 7rV2C(t) (7.t)= where Co(t) / dx\^(x. (7.t)\' 2 exp[x2/b2] rrVaft In the following we choose 6 = 1 .t)\' 2 125 (7.95) .t)\2 Ji == 71 / V " Jo dO ? 6' (7. In taking the modulus of ip(x. C2(t) (7.86) has C2(t) := b2 cosh 2 t + 72 sinh 2 1.7.4 The Inverted Harmonic Oscillator The real part of Eq.. _2 1 exp C 2 (i) = l + 2sinh 2 i. 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.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. We have therefore .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).90) with the expected limiting behaviour \im \ilj(x.91) C(t)' (7.
B. 136.96) °° l Jr.From this we obtain C(t) C'(t) Z2 ^(r?2_.2)(p_7?2)(7. (7.=0 V^ C'(t) I Jt=0 V^ (7. V and (cosh t = y 1 + sinh2 t) 1 r CHAPTER 7 Green's Functions dr} dt C'lt) ' C(t)' (7. Dwight [81].(4 2 . p..97) sinh 2 1 = . + 0(e'2) (7. £ 1 dr\ 2Z %_ rl 2 2 yfx Jr.126 We set I C(t)' so that dT ~d~l Now. C(t) = .1 ) 2 \n J ^"•""ff^^V^K?. we obtain dT _ 1 ~dJ~l 'See for instance H.97) and obtain dT ~dl or with £ = ?y/7: 2 2 d^e.98) We insert this into Eq.99) .= \ / l + 2sinh 2 t..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.=o V(i v W + v2) oV "dT J L 1 f A x 2 __H_I r'=s/21 Adx'.
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.7. U) (7.4 The Inverted Harmonic Oscillator 127 for I sufficiently large. this is T ~ ]n(ly/mou>/h). Explain the parameters entering the calculation. (7. p. The equation there describes the classical "rolling down" of this tachyon.4: The sojourn time calculated semiclassically A tiny ball of mass mo is placed at the apex of an upright egg.101) A more detailed evaluation of the constant can be found in the paper of Barton cited above.100) Reintroducing the dimensional parameters we had set equal to 1.102) 'As a matter of interest we add that in the theory of a free scalar field x. is instructive in revealing the basic quantum mechanics involved in the finiteness of the sojourn time. T h e following Example 7. Calculate with the help of a semiclassical consideration the (quantum mechanically limited) maximal length of time which passes before the ball is observed to roll down.g.3 How stable is the particle on top of the eggshell in quantum mechanics? E x a m p l e 7. also motivated by this paper. 2 2 2mo mow x The classical equation of motion is therefore' it — uj x = 0 (7. Fig. It follows t h a t T~lnZ. 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. See e. . 7. B. Zwiebach [294].4. 238.
2 x(0) ^ where A i . Thus we set I = A cosh LOT + B sinh coT ~ (A + B)e"T.102)) p(0) = ^/(p 2 >. / 7 ^ 2 =2J(x }. and for simplicity we take (cf. i. (10. . but instead are subject to an uncertainty relation of the form AxAp > h.p(0) cannot be determined with arbitrary precision. ^(x2)(p2). the smallest macroscopic length.e. Ap= v /(p 2 )<p)2. so that raoto _ We assume now that at time t = 0 the ball is placed at the point x(fi) with momentum p(0). and solution x = A cosh Lot + B sinh cot. Green's Functions x = Aco sinh cot + Bco cosh cot. A p are defined by the mean square deviations given by (see Eq. (7. (p2):=mW(x2). (7.g.128 with x2 = LO2X2 + const.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. 2V ' Let A and Bio be the values of x and x at time t = 0. i. (x 2 ) h 2mou 2mou> and so l^J^ULl).e. CHAPTER 7.5)) Ax = <J(x2){x)2. v A x A p = \ (x2)(p2) v = m0u(x2). Thus quantum mechanically x(0). so that e. x(0) = A. For h —> 0 the time T —• oo in agreement with our classical expectation. Quantum mechanically x(0). so that x(0) + il^L mow It follows that For a minimal uncertainty (hence the factor of 2 in the following) we then have h= 2 Hence uT P ( ° ) _ O . 2 10 x(0) = B u . so that AxAp= In our semiclassical consideration we set therefore x(0) = y/{x*). We ask: At what time T > 0 does the ball reach the point with horizontal coordinate x = I. Eq. For a symmetric state like the usual ground state of the harmonic oscillator we have (since the wave function is an even function. the expectation value of x is the integral with an odd integrand) <x) = 0 = <p). pifi) have to be replaced by the positions of the spatial and momentum maxima of a wave packet.
For mathematical purists the question of convergence of the series is an even bigger challenge. EW+(3EW+(32EW +  ((3 can also be thought of as a kind of "bookkeeping" parameter in retaining corresponding powers of some kind of expansion). An example permitting convergent perturbation series is provided by the trigonometric potential cos 2x with onedimensional Schrodinger equation given by ip" + [E. in closed form) only for very few potentials. but this is no longer the case for an anharmonic oscillator potential like ax2+(3x4.(i)+/^(2) + ". one would set tt E = = ^(°)+/ty.e. In general perturbation series do not converge. In the case of the harmonic oscillator potential ax2 the Schrodinger equation can be solved exactly with ease. 129 .E^ is already a bigg problem.2h2 cos 2 a # = 0.e. It follows that in general one depends on some approximation procedure which is usually described as a perturbation method.Chapter 8 TimeIndependent Perturbation Theory 8. i.1 Introductory Remarks The Schrodinger equation can be solved exactly (i. 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. Frequently even the calculation of the next to leading contributions ip^.
3) *See e. the expansions in ascending powers of the parameter h2 can indeed be shown to have a definite radius of convergence. T. the expansions in descending powers of h2.2 However. which is known as the Mathieu equation.. E.1) and its terms with the behaviour of the following series and its terms: f(x) = 1 .* i. 581..130 CHAPTER 8. E = E^ + h2E^+hAE^ 71 72 7 + .g.g.• + (ir~ X X* 1' 2' r?l Xn + Rn(x). . In the following we explain the difference between convergent and asymptotic series.. the expansions i> = v (0) + /*V (1) + /*V 2 ) + • • •. p. W. (8.+ .= 0 < 1. J. N. 0. e. This convergence can be shown with D'Alembert's ratio test since''' lim . Whittaker and G. explore the latter in somewhat more detail and finally consider methods for deriving perturbation solutions of the Schrodinger equation. 8.g. The series n=0 of the exponential function is well known to converge absolutely for all real and complex values of x. Meixner and F . see e.g. Cll E = a_ 2 n + a_i/i + a0 + — + —r \ . It will be seen later (e.2) n—»oo n It is interesting to compare the behaviour of the series (8.e. perturbation theory and path integral methods — that the lightminded way in which perturbation theory is sometimes discarded is not justified. TimeIndependent Perturbation Theory In the case of this equation. h hl are in the strict mathematical sense divergent. 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. . ^For further details. at the end of Chapter 26) — and by comparing the results of WKB. Schafke [193]. (8. Watson [283].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. concerning also the uniform convergence of the exponential series..
However.9876. .4) for every value of x. p. . + 0. for larger values of x. We observe in the first place that since lim n—>oo co (8..0... and we see that the moduli of successive terms first decrease and then.3) diverges for every value of x. for x = 3..000 000 06 + • • • . the better the approximation obtained.2. From the Weierstrass product which defines this function* one obtains the series _ M*l)! = 7(*l) + E [ ^ . N. The theory of asymptotic expansions claims that if the expansion is truncated at the least term. . 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.000 002 00 .e. . + 0. / ( I ) = 1 — 1! + 2! — 3! H • However. + •••. It is evident that the larger the value of \x\. Normally a divergent series is characterized by an ever increasing behaviour of its terms as in the case of x = 1. Comparing the asymptotic series (8. As a second example we consider two series expansions of the gamma function or factorial T(z) — {z — 1)!.0. /(1000) = 1 .29629.l n ( l + ^ ) n=l L ^ ' (8. i.5) Cf.0.g.0. 235. .8..0..22222 . we have ^ ' ~ = 3 + 9 ~ 2 7 + 8 1 ~ 2 4 3 + 7 2 9 ~ 2187 + " ' ' 1 . e.33333 . Thus e.001 + 0. and in fact increase indefinitely..222. in spite of this the series (8..2 Comparison of Asymptotic Series with Convergent Series 131 where Rn(x) is the remainder sum. after reaching 6/27 = 0.3) can still be used to obtain almost correct values of f(x) for x sufficiently large. Whittaker and G. 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. We can see this as follows.1).22222 . the series (8.3) with the convergent series (8. Watson [283].. This type of behaviour is characteristic of the terms of an asymptotic expansion as we shall see in more detail in Chapter 20. T. begin to increase again.376 . E..4938.g... . +0.
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. although this has not been investigated except for a few simple problem^. 11 L. that E and \ijj) (i. p. Sec. ^E. It is interesting to observe the comments of various authors on this point. the dominant approximation is. Vol. p. E. 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. Messiah [195]. N. **A. Schiff [243].1. p. 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. I. the leading term of an asymptotic expansion. In many respects Messiah aims at more rigour in his arguments. In fact. 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. 371.J In2. whereas applications of the convergent series (8. . at a later point in his treatise Messiah admits that the expansions are mostly § Cf. 236.e. TimeIndependent Perturbation Theory where 7 = 0." This is a very clear statement which does not try to pretend that a perturbation expansion would have to be convergent. § However.5) are practically unknown. T.132 CHAPTER 8. it is sensible to make the assumption.5). 16. in fact. (16.41) and (16. This latter observation hints already at the importance of asymptotic series in applications. Whittaker and G. ip) can be represented by rapidly converging power series in e. Watson [283]. meaning convergent. Applications of Stirling's series can be found in all areas of the physical sciences (particularly in statistical problems).5772157— Here the series on the right is an absolutely and uniformly convergent series of the analytic function. This is not appreciated by mathematical purists. It seems that Schiff tries to cling to the idea that a proper series has to be analytic. It is therefore not surprising that he** says: "7/ the perturbation eV is sufficiently small. Thus one can say that the vast majority of expansions of this type in physics is asymptotic and not convergent. Merzbacher [194].1 between Eqs. for ln(z — 1)! one can also obtain the Stirling series ln(zl)! = ln(2vr)5 z+ (z." These "rapidly converging power series" in physical contexts are most likely very rare cases.
" Of these three authors t h e first. which have been studied in great detail.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. possess asymptotic expansions which have for a long time been important a n d accepted standard results of mathematics. We know from books on Special Functions t h a t all of these functions. 8. (8. solutions of second order differential equations. ft Jx du = 2tdt.^=e~x2 / e^^dt. there is no need for such bias.ft somewhat afraid t o say so himself a n d therefore with reference t o investigations of T . . Kato: "Indeed the perturbation expansion is in most cases an asymptotic expansion . *R.8. Messiah [195].11) A .= / e~t2dt. (8.7) This function has been considered in detail in t h e book of Dingle. Dingle [70].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. In fact. (8. 2 (8. (8.2 Comparison of Asymptotic Series with Convergent Series 133 asymptotic. 198 (German edition).. 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. II.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. Vol.10) Vn Changing t h e variable of integration to u = t2x2.2.. He says. Replacing t h e exponential in Eq. Merzbacher.. B. we can rewrite Eq.7) as / e'* dte* dt = 1 . p. (8.* and we follow some of t h e considerations given there. It is therefore suggestive to write iy poo <t>(x) = 1 .
(8. Using the reflection formula (*!)!(*)! = for z = n + i .12) VK Jo n V oo 7 n=o x) 2 2*Qn\(±n)\\x2) (8.16) . 7T (ni)!sin{7r(n+^)} ( „ . du.12).13) presupposes t h a t < 1. (8. Then e (n —\)\( 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.15) < 1 we can insert this expansion into Eq.134 we obtain CHAPTER 8.' we have 2 nH 7T 7T (8.^e~x \Ar i.i)[(_!)«' Then.e. « \~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 (8.14) sin7T2.e. we obtain 1+ i.^ X e U ~ ( 1 + * \ 2 . 2 9 f00 / Jo du e" 2(u + x2)^ u <£(*) = 1 T h e binomial expansion . TimeIndependent Perturbation Theory <P(x) = 1 .
19) where "~" means asymptotically equal to which in turn means that the right hand side of Eq. ' arg. In the foregoing discussion of the error function we assumed that x was real and positive.l)! = / Jo Then 2 cHnxdt.18) or that we insert the binomial expansion (8. It is fairly clear that the above considerations remain unaffected for argx < 7T. Frequently "=" is written instead of "~". (8. whereas in a convergent expansion the terms first increase in magnitude. in general. i.19) approximates <f){x) the better the larger x is. The expansion (8. (8.e.19) implies either that we ignore the remainder or the correction term given by the integral in Eq.e. In an asymptotic expansion the first few terms successively decrease in magnitude. Whichever way we look at the result. i. convergent) only in the restricted domain u < x2. 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.12) ignoring the fact that the latter is valid (i. (8.r<7r.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 (8.17) 0(X) = i. As in the case of the gamma function. . much more useful than the convergent expansion. that a large number of terms has to be summed in order to obtain a reasonable approximation of the quantity concerned. (8.—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^. /•oo T(n) = (n .15) into (8.19) originates. We can actually understand the deeper reason for this in practice. argx = 0. and in fact so slowly.e.8. the asymptotic expansion of the error function is.
136 CHAPTER 8.2 y V . (8. The situation becomes critical. V7T Jo Changing the variable of integration to v (8. =dv.^ = . TimeIndependent Perturbation Theory since for this range of phases the decreasing nature of the exponential is maintained.^ f V e ^ . (8. i. We can therefore rewrite the integral as (j)(iy) i 2 fy2 y —=e / e~v 1 . (8. however. VK Jo V^ Jo (8. we set x = iy in Eq.15).20) where we set t = is. 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). (8. Proceeding along lines similar to those above we write <f>{iy) = ^=ey2 I* e^^ds. We therefore consider this case in detail.26) or else return . we can try to proceed with Eq.24) We can read off the binomial expansion of the factor in the integrand from Eq.22) we obtain 2 4>{iy) = .7) and obtain 4>(iy) = —= j e~l dt = = / es2ds.21) = y2s2. dv = 2sds = . when argx = 7r/2. thus with \v/y2\ < 1.23) Throughout the range of integration 0 < v < y2 the integrand is real.e.vds. (8. (8.
28) ignores the integral contribution of Eq.19). (8.e. 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.28) we observe that the expansion (8. For v > y2 the second integral is seen to be f imaginary and thus drops out in taking the real part. Eq. (8. (8. i. (8. However..e. „2 \v J v^(ni)! y(nj)l ^—' y2n ie"2 /00 ^ ieir /•" W7r /„2 y ^—' _v^{n\)\(v n! .^ ^ p . for  arg x\ < 7r/2 and for arg x = IT/2 the error function 4>(x) possesses different asymptotic expansions. which will be studied in more detail below. (8.2 Comparison of Asymptotic Series with Convergent Series to Eq. i (8.28) n=0 We observe that this expansion differs significantly from (8.26) we can write* 137 v* Jv2 ie « „2 OO / 1 n=0 M ~k n! . for large values of y. (8. we obtain 4>(iy) = ^ey2TZ f°° e~v (1 . Suppose now we consider Eq.27).28) because then the correction term oc f°£ becomes smaller and smaller.27) and (8. ey is much larger than something of order 1/y. We have therefore <j>(iy) = ^e^TL fV e~v ( l . — (8.2 yir (8.^] * dv. and so the integral is real. . We know that for v < y2 the integrand is real. If we proceed with Eq. (8. Returning to Eqs. and the larger y.30) yvn f Jo V y J Cf.24). the better the approximation expressed by Eq.24).8. y y 2 ( n _i)! y arg(* = iy) = i * .27) Then .^ ) * dv JO V y /»00 /»' yV^ 1 i Vsfv 2 ey n oo J0 Jy' 1%) 'dv.29) where 5t means "real parf. i. (8.17): J£° e~vvndv = n\. e ^ ( i y ) ^ ^ .
.e.31) and (8. (8.2 OO rz=0 V . R. . (8. we use the expansion outside its circle of convergence and hence obtain a divergent expansion.8)) 4>(x) = <f>(x) we see that _x2 OO / ] \  n=0 for 7r/2 < I arg x\ < 3TT/2. since (8. Looking at Eqs. e xZ ^ (n — £)! „ ^ ) ..28). Chapter 1.25) and integrate from 0 to oo. (8. and • higher order terms of the associated series all have the same sign. is obtained for <f>(x) with arg x = —ir/2.31). Dingle [70]. Since (cf. B.33) we see that the Stokes discontinuities occur at those phases for which these expansions (i. This result is identical with that obtained previously.3!) which is (8.* We observe that this Stokes discontinuity is a property of the asymptotic expansion but not of the function itself. (8.e. With a similar type of reasoning one can show that the same expansion.30) can be written </>(iy) = ^ey2 fV ev(l^) yVn Jo V y J "dv = ^!rV"E^(4)V (8. (8.138 CHAPTER 8.26). We have therefore found that XTT n=0 ^ ' and . For the sake of completeness of the above example we continue the phase to  arg x\ > IT/2. TimeIndependent Perturbation Theory If we insert here the expansion (8. 7r = . *Cf.33)) would have • a maximally increasing exponential e+x . (8." — E ^ ^ forarg. The corresponding phase lines are called Stokes rays. i.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. Eq. y.
for arg x = 7r/2. i. though. Thus if we set x = XR + ixj.35) 8.g. —1 cancel out. we have e z2/2 _ 2 e(x Rxj+2ixRxI)/2_ The exponential is maximally increasing for XR = 0.8.2 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. '"Cf.37) Cf.e. p. 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.2. which can be written (for v not restricted to integral values) D^(x) DM(x) f xu(f>u(x). 5. pp. Oberhettinger [181]. Chapter 1. (8.34) the contributions 1. W. Eq.13.36) has an exponentially decreasing solution and an exponentially increasing solution for zero phase of x. These rules.rgx<n/2 argx>7r/2 _ X1T •<—' v — ( n=0 £ (ni)! X2)n ' oo (8. e. B. Dingle [70]. Magnus and F.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. apply predominantly to asymptotic series of the solutions of second order differential equations.32) and (8. 1/1 ipv{x) v1 vl (±ix). 91. See also Tables of Special Functions.e.2). Later we shall make extensive use of parabolic cylinder functions Du(x) which are solutions of the equation dx2 + v+ :X y =o . . (6. R. Equation (8. and so may not be universally applicable. and 4>(x) argx=7r/2 = 2 r2 ^ + <f>(x) a. 9 . in the sum of the right hand sides of Eqs.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. i.
D^(x) = = xv4>v{x)Jr\^l'1^'Kl'1\A~v~X^^) xv4>v{x) + \a. ' S u c h symmetries are extensively exploited in the perturbation method of Sec. R. ti asymptotic series (8.x 2 oo «*)  — E^fi=0 v .38)) has a Stokes ray at Dl '(X) the arg x = 7r/2. (14).e. Hence Dil){x) = xv(j>v(x). as can most easily be seen in the case of the Mathieu potential. At arg x = 7r/2 the exponential factor becomes an increasing exponential and late terms of the series have the same sign.36) is invariant under these replacements. = 7r/2. The continuation of (8. 0 < argz <  (8. Chapter 17. (8.e.e. 9) and XXI. Dingle [70]. R. with a real proportionality factor. Chapter 1. We consider Dv (x). 1 1 Cf. i. (838) We observe that one solution follows from the other by making the replacements^ x — ±ix. Thus D{J\x) (i. v — — f — 1.40) § Cf. 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. We also observe that the late (large i) terms of the series in (8. i. B. (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. x~l/~1ipu(x). > > The equation (8. . Chapters I (in particular p.140 where^ CHAPTER 8.ei<v+VxvHv{x\ argx = £ . Eq. Dingle [70].7. TimeIndependent Perturbation Theory i.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.39) (as in the example of the error function). B.37). 8.
41) the Stokes multiplier a is. (8. 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). = ir. = ir. we must have at arg x = 7 in the dominant factor on the right of Eq. 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. < 37r/2. The value of Du (x) on the other side of the Stokes ray. where in the first line on the right hand side the second term contains the real proportionality constant a/2. beyond arg x — 7r/2. Applying rule (2) we obtain D$\x) = U™ + ^iap) (xyMx) + a{x)vlMx) (843) for IT < arg a. i. i. Since Dv (x) is real when x is real. (8. with \x\ = (—x) for arg a. as yet an unknown real constant.e.42) T ^(eiwu i.2 Comparison of Asymptotic Series with Convergent Series 141 for argz = 7r/2. In (8.e. another half of the discontinuity appears on leaving the Stokes ray on the other side.41) to arg a.8.44) + ^iap) = 0 . We therefore proceed to continue (8. is given by • Dingle's rule (2): One half of the discontinuity appears on reaching the Stokes ray. < 7T.42) el™ + ±iafi\ {x)v<t>v{x) + a{x)vl^u{x).40).e. (8. sm(iri/) = —a/3/2 or a/? = 2sm(irv). Thus for arg x > ir/2 we have D$\x) = xp(t>u{x) + = 2]iaei<v+^x1/l^v{x) ^ < arga. e +fx2 _ 2 2 e±(x Rx I+2ixRxI) is maximally exponentially increasing (i. 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. (8.41) xuct)v{x) + a{x)~u~lipu{x). On reaching ir the part containing ^ ( x ) .e.
38) = cos7ru(x)u(f)1/(x) + (—x)ve~*x COS7T^— —r a(u)(x)l'1^u(x) (2i .1)! TT. i.45) Thus the right hand side of this equation remains unchanged if v is replaced by —v — 1.42)) for argx = ir (in the second line of Eq.e. a{y)P{y) = a(u .1) = ±a(y).1). i. (z)\(ziy. (8. We can see that the equation is satisfied by (3{u) = ±a(u . Eq. (8.1) (8. Hence the left hand side must have the same property.1) (8.45) is therefore given by a{v)a(v .e.—^ ON • — ^ > + ^ ^ L .50a ) (aMb) »W = f "M=(=^)i In order to decide which case is relevant we observe from Eq.45) that af3 = 2sin(7r^)) D^\x) = (8. or sin7r(2: + 1) sin irz' W ( = 7 ^ ) ! = 2sil"r2Comparing this with Eq.42) that a multiplies ipv(x). Dl \X) is (cf.14). (8.47) Equation (8. (8.51) . (8.e. a(u)P(v) = 2sin(7r^) = 2sin7r(i/ . TimeIndependent Perturbation Theory Now a and (3 can still be functions of v.42) we insert from Eq. . argx = vr. i.1) = 2sin7ri/.l)0(v .46) For a given function a(y) this equation determines /3(u). (8. V i=0 \U0 v 2w . 8. (8.48) We compare this with the reflection formula (8.142 CHAPTER 8. (3{v .v . i.e.48) we can set <8 49) ' ^ ) = {_V\)V PM = ^f> or ( 8 .
8.2.53) the factor (—n — 1)! is to be understood in association with (2i — n — 1)! in the numerator.A / ± sin^Cx)"M* 2 V ^ .1 ) ! 4~L i\{2x2Y i=0 V2^(x).50a) because this enables us to define normalizable parabolic cylinder functions. (2x2)1 = COSTTU(X) v . the second contribution vanishes and we obtain £>«(*) . i. Proceeding along similar lines we can continue the expansion into the domain (2) ir < arg x < 2ir.3 Asymptotic Series from Differential Equations In the case of the error function we obtained the asymptotic expansion from the integral representation of the function.) (8.36)) in the domain 0 < arg x < IT.. the error . (2inl)! (n1)! n! {n2i)\ So far we have been considering the asymptotic series expansion of Di.1. In a similar way we can examine Dv (x) and its Stokes discontinuities (at argx = 0.2^1 v y 7T i=0 ' (2i + v)\ z!(2x ) v (8.e.*"ei°> f ( ^ _  _ L _ = (!)««(.7r). (8. (8.36). which vanish at x = ±oo.. i. Eq..1 ) ( . 8. (8. ' ^ (i/l)!z! ^ ..3 Derivation of Asymptotic Series from Differential Equations 143 We choose (8.. We observe that for v — n = 0.„ _ix2^(2ivl)l e 2 > —. Then (in the second step using (—u — l)\v\ = — ir/sm(Trv)) DW(x) ~ c o s ^ ^ * f ^ . However.e.. (8. ' (X) (cf.e^ .7). Chapter I.e.53) We have therefore obtained in a natural way the quantization of the harmonic oscillator.I / .35). It should be noted that in Eq. solutions of Eq. u 1 y> {2i + v)\ 1 n 0. i. —f.52) (argx = 7r).)(. For details we refer again to the book of Dingle. (8.
S2. 0.e. (8.38) can be obtained.^ + 1 ) .e.37). x—>±oo (8. ^ + 2 . how series of the type (8.55) The large x asymptotic expansion can also be obtained from the differential equation. S\. (8. i.36)) 0 where + x2Q(x)y = 0. % = {. cf.56) ~l^(8 57) Q{x) =  We then set y = es^ so that the equation becomes (8. Eq. • • • can be determined.. Instead of dealing with Eq (8.58) Since x2 is the highest power of x appearing in x2Q(x) we set S(x) = ]S0x2 + SlX + S2 In a. ^ = 0 dxA dx with the boundary conditions lim 4>(x) = ± 1 .54) (8. (8..60) 2 x x l Inserting (8. x°. + — + ^ + • • • . \ + U (8.144 CHAPTER 8.. (8.36) in the form (coefficient of — x2 here chosen to be ~. and demonstrate. 5„=4 Sl=o. TimeIndependent Perturbation Theory function can also be obtained as the solution of a second order differential equation. (8. i. we obtain equations from which the coefficients So.54) we consider again the important equation of the harmonic oscillator.36). x. We write Eq. Eq. Thus 2S 0 Si 5o + 25 0 5 2 + ^ + ^ and so on. previously we had ^.59) and equating to zero the coefficients of powers of x2.60) into (8. Solving these equations we obtain = = 0.
36) have correspondingly e±x / 2 ) . 4~l~2 ' i i _i.19). 2 X 2Wi. » 1.91.z) is a confluent hypergeometric T.38). / i N . 2 ) + V2(l)! lt1! 2 1 — v 3x\ 1 2 2' 2 where iF\(a. a z a a ( + 1) z2 which has the unit circle as its circle of convergence. In particular we shall need the asymptotic expansions of parabolic cylinder functions in various domains. i. Nonetheless we shall need asymptotic expansions like (8. . \x\ ^> \v\ and (a) for  a r g x  < D„{x) 3/4TT: e 4 x x i/(i/l) 2x2  v(y\){V2){yS) 2Ax2  .b. etc.e.62) 4 \ 2 XiFi 2'2 . (8.+ X M W = Q.8.37). Whittaker and Watson [283]) is now somewhat elaborate (although in principle the same as before). Oberhettinger [181]. n 22e"T JziL (8. therefore. 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+ .37) with (8. do not enter here into their derivation and simply quote the result: W i t h \x\ . and it is clear t h a t higher order terms follow accordingly.61) (the dominant terms of (8.63) **W. function. We. p.3 Derivation Then of Asymptotic Series from Differential Equations 145 eS(x) = 2 e±^ xS2 1+ 0 (8. T h e expansions (8. Magnus and F.35) and (8.38) in a variable for the solutions of approximated differential equations. (8. In the following we are mostly concerned with asymptotic series in some parameter. The derivation of the asymptotic series from integral representations (cf. are asymptotic series in a variable x.
E. Whittaker and G. top of p.1 a2 an a0 + — + ^ + . Thus there is no contradiction^ between (a) and (b).0.4x2 (8.67) Cf. If Sn(x) is the sum of the first n + 1 terms of the divergent series expansion . Watson [283]. . whereas e~x '4 increases exponentially. 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.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. 349. 4 (8. T.66) 7 =e(^) 2 i(3x)2ei(^)(?Jx) decreases exponentially.4 Formal Definition of Asymptotic Expansions Finally we introduce the formal definition of an asymptotic expansion.146 CHAPTER 8. 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. 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. 8.+ — + ••• of a function f(x) for a given range of a r g x .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. N.4x 2 27F (*/!)!  e^e^x""1 1 + (v + l)(i/ + 2) 2x 2  (i/ + !)(»/ + 2)(i/ + 3)(»/ + 4) 2.
Sn{x))  = oo n—>oo (8.72) For the eigenvalue E near En and the eigenfunction \l/ near ipn we assume power expansions in terms of the parameter e. thereby (unknowingly) discarding the vast majority of expansions which have been used in applications to physics.69) and the consequent nonunique definition of the expansion. In fact.e. 8. (8. The equation to be solved is the equation H<S> = EI/J with H = H0 + eV. (8. It may be noted that Whittaker and Watson's internationally aclaimed text on "Modern Analysis". we set E^En * > ^ n = En°1+eEnV = + e2EnV + . 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 definition is due to Poincare (1866).68) The definition(8. in particular with regard to possible exponentially small contributions to the expansion for which e. . lim xneW x—>oo = 0 (8.g. first published in 1902.70) 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).8.5 RayleighSchrodinger Perturbation although Theory 147 lim \xn [f(x) . A critical discussion of the definition can be found in the book by Dingle [70]. which is assumed to be small.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. Chapter I. i. possesses a whole chapter on asymptotic expansions.73) Vn + q/#> + e2T/i2) + • • • .68) is (effectively) simply a statement of the observations made at the beginning. (8.
TimeIndependent Perturbation Theory We insert these expansions into Eq. i. The second equation can be rewritten as {Ho . . We insert the ansatz (8.77) The states \tpn) of the unperturbed problem span the Hilbert space"K.4 0 ) M = / ^C(4 1} . (8.i4 0) )V4 1} = {EM . i.n.e.e.72) we obtain oo roo dxifc{H0 OO / ~4 0 ) ) E • / °i ^ = / J— OO d x < i E n ] ~ V )^n.e. We obtain then the following set of equations: Hrj. 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. ^ ) = 0.V)i.As a consequence the state vl/) is a vector in this space. °° i J or 0 = EWJdx{rnVil>n). (879) We multiply this equation from the left by ip^ and integrate over x. With the help of the orthonormality condition (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. Ho^+V^n = EW^+E^n.75) HoW + vW = E^+E^+E^.77) and obtain {Ho ~ 4 0 ) ) E "i^i = (Ein] ~ W n .76) The first equation is that of the unperturbed problem.w » . (8.70) and equate on both sides the coefficients of the same powers of e.74) (8. J f°° dxrn E(^ 0) .n = 40)V>n. (8.148 CHAPTER 8. i. (8. ( ^ .78) into Eq. so that we choose / • dxrpM^O.
79). i.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. (883) Thus the procedure forbids the equality of any E\ '.76) which we rewrite as (H0 . We multiply the equation by ipm. we return to Eq. i.m ^ n and integrate: oo /•oo / oo dxrmY. S ' ti^+Q^)• We observe that Eq. We therefore consider Eq. (8. It is clear that we can proceed similarly in the calculation of the higher order contributions to the expansions of En and \I/n.e.^(EW ~ V ) ^ + E& / J—oo dx^miln .82) *> = ^ + E . the coefficients a\ .V)^ Setting + 42tyn.Vil>n) (880) In order to obtain ipn . the procedure forbids degeneracy and we can consider here only nondegenerate eigenstates. (8.4 0 ) M 2 ) = ( 4 1 } . (8. ^ n ) + 0(e 2 ).83) makes sense only if Ef] ^ EnV for all i + n.84) (885) ^2) = £ a S 2 ) ^ (excluding i = n for reasons discussed above).e. multplying Eq.8. with En .i ^ 0. (8. In order to appreciate the structure of these expansions it is instructive to go one step further and to derive the next order contribution. and hence It follows that to the first order in the parameter e: En = 4 ° ) + e ( ^ . (il)i. (8.Viprt e (8.5 RayleighSchrodinger Perturbation and so Theory 149 E& = {n\V\n) = Wn.
TimeIndependent Perturbation Theory Setting m = n we obtain with (^„.^(8.Vlpn)E.79) •/—oo oo / Using (8.^j)(V .86) E^ iy^n JJn J J i ' i^n For m ^ n w e obtain from Eq.90) + £ ( 4 0) 4 0) )(4 0) ^ 0) ) ^ +' . ipn ) = 0 i. En0) (8. j. Inserting the expression (8.81) we obtain (2) _ _ {lt>m. (8.150 CHAPTER 8.^w . n) £f) (8.V4>n){lpn.V^n) .85) together with (8. ^ n ) j&iift + ^n .VV .88) (Q) Inserting Eq.^ ) + $& w £ (^fc. (8.89) • • • where we guessed the term of 0(e 3 ). FVn) + e 2 ^ ( V > n .Vlpn)(lpn.87) (i>k. (8.Vlpn) E^ .86) into the expression for En we obtain En = 4 ° ) + e ( ^ n .88) into xjjT we obtain similarly * r i + f ^ i^n (lpi.2 (V> m . V ^ i ) X (Q) (Vi. ~Ej E E.V^j)(^j.B i 0 ) ) ( £ f .£i°») 2 Hence +^£o (£«? .4 0) ^+ e 2 E (lpk.V'>pTl (E^Eny V'fc (8.(1) (i&0) .e.Vlpn] M _ P (0K 2 ^r .
89). (8. + l^t /„(0) V~^ c kjCjn &(Efi»Ei )(EfE&»)\ 0) „(0)w. We can therefore rewrite tpn ': The perturbation theory described in Sec.(0 4>k The sum J2j^n includes a term with j = k which is exactly cancelled by the other contribution.8.. These functions generally obey one or more recurrence relations which are effectively equivalent to the differential equation in the sense of difference equations. Merzbacher[194]).90) can be found in just about any book on quantum mechanics. in the form i Adopting this procedure for the perturbing potential V(x) we can set V(x)i. i.91) The coefficients Cj follow from the recurrence relations.e.90) is the term with only one summation J2 m the contribution of 0(e 2 ). One would prefer to lump it together with the other contribution.lpn+i) = ^ Cj5k. 8.n(x) = J2crtn+ii (8. . An ugly feature of the expansion (8. We then obtain (lpk. The functions ipn(x) are eigenfunctions of some secondorder differential equation. These recurrence relations allow one to reexpress expressions like Xm^n as linear contributions of ipi with constant coefficients. sometimes expressed in terms of projection operators which one can construct from the states \ipm) (see e.n+i = Ckn(2) We can now rewrite ipn ' as / (2) _ V"^ c ^ n knc0 ~ 2j ~ TJfi)0 ) k^n (4 ^) 2 n(0).Vlpn) = y^Ci(lPk.7 is based on this procedure..5 RayleighSchrodinger Perturbation Theory 151 The expressions (8.g.
ipm) = 0 = (^m. m = I. all n. . Em can now be linear combinations of ipni'ipm.95b) H0(c2l^n and + c22ipm) = En°\c21^n + c22^m) (8. . (8.ipn).94a) * m = c 2 i ^ n + c22tpm + e ^ + e ^ + ••• (8.97b) are . are allowed to be equal since otherwise the expansion becomes undefined.Hence we write * n = CllV'n + CisV'm + € ^ + d 2 ^ 2 +••• . . + c22^m). i.93a) (8.152 CHAPTER 8.V)(c2l^n + ci 2 ^ TO ). that of double degeneracy En = Em . Equations (8.95a) (8. degeneracy. (8. . i. —I. Equality of eigenvalues.96) are effectively the known equations of ipnitpm and so yield no new information. I — 1 .e.e. (H0 + eV)^n and (HQ + eV)Vm = Emym. (8.94a). TimeIndependent Perturbation Theory 8.93b).97a) (8.93a).4 0) )</4 1} = (En1] ~ V)(cu^n (Ho . (8.93b) The lowest order eigenfunctions belonging to En.We now insert Eqs.70).97b) with ipn y£ ipm. is not uncommon. For example the magnetic states enumerated by the magnetic quantum number m. so that En = 4 ° ) + eE™ + e2EP + ••• . (8. however.6 Degenerate Perturbation Theory It is clear from the expressions obtained above for En and tyn that no two eigenvalues En '.97a) and (8. in a central force problem all have the same energy unless the system is placed in a magnetic field. (8..94a) into Eq. (8. Equations (8. (8. . (8. Proceeding as before we obtain Ho(cniln + Ci21pm) = En0) (cU1pn + C121pm) = EnVn. i. Em = 4 ° ) +eE$ + e2E$ + .4 0 ) ) ^ } = (Em] .96) (Ho ..94b) with (ipn.e. In order to deal with the problem of degeneracy we consider the simplest case.
and ( ^ m . Vlf>m)] . H0lP$) = C22[E$ ~ (Vm.H0^) . we obtain with (ipn.(lPn.m). (^n.98b) + Culpm) + cuV^m).EJ®(lPn. ^ ) ) = CU[EW .8. (8. Vi/>n)] . c2iV^n + c22V7pm).99) and ( ^ ^ O ^ ) " ^ ^ .Vil>m)] V^m). Then (^. . (^m. But (8.(Vn.V^n). Similarly Wn.ci 2 (V„.J#»a&> = 0j since ii4i = En (degeneracy).£ < « = ag)^) . V^n)\ . £ aW^°Vi) .99) can now be rewritten as (frvvA \1"'Vr\ )(cn)=w(cn) (siooa) .^ ( ^ .^ ) ( ^ m . ^ o ^ ) ) = 0.C22(Vn. Vl/./>£>) = C2l[E£> . cn(ipm.H0il>W) = cn[EW . (8.6 Degenerate Perturbation Theory 153 inhomogeneous equations.m. ^ ) = ( V . Wm.98a) Setting I = n. V^n).ipn n) = 0 = (ipm.ipm ) the equations (iPn. Equations (8. We now multiply these equations from the left by ip* and integrate. H0lpV) .C 21 (^ m .(^04°))41)) = EM^CuTpn (ipl. t f ) = a k 1 } 4 0 ) " Eg>aU = 0.cnVipn and tyi.y>m.H0xl>$) .
. Nonetheless we shall see that the Schrodinger equation of the hydrogenlike problem can also be separated in some other orthogonal coordinates.100a) determine the coefficients Cn. Let ip and < be two eigenfunctions belonging to the same eigenvalue E. It follows that ip and <j> are linearly dependent. if)' (/>' — = —. i. d>" E=^. the secular determinant. A onedimensional system.6. (8. as we saw.e. 9} ip. i> 4> Hence there is no degeneracy.101) This secular equation determines the first order correction En • The equations (8.E)tp = 0. dx i.^Vm) 0. TimeIndependent Perturbation Theory (l/>m. (8.3).154 and V (^m.C22.E™ (Vn. like that providing the Stark effect (proportional to rcosO). just as there is no degeneracy in the case of the onedimensional harmonic oscillator (Chapter 6). ip = c<p.This completes the derivation of the first order perturbation corrections if an unperturbed eigenvalue is doubly degenerate. It is then possible that a special perturbation.100b) determine C2i.Vlpn) CHAPTER 8.Vl/>m) J \ C22 J m \ C22 J V ' These matrix equations have nontrivial solutions if and only if (lf>n. (V>V</>V>) = const.e.ci2. the degenerate perturbation theory must be used. there is degeneracy in the case of the hydrogen atom with Coulomb potential (Chapter 11) which is based on threedimensional spherical polar coordinates r. 4. Then / > i>" 2— = V ip Hence 0 = V ' V . However. i. Solution: We consider the Schrodinger equation with potential V in the simplified form i>" + (V . If one wants to perform a perturbation calculation in such a case.Similarly Eqs. In this particular case therefore nondegenerate perturbation theory can be used (cf.Vrl>n) .^ ' V = — (•>!>'<t><t>'i>).1.e. Example 8. At x = ±00 the constant is zero. in particular in parabolic coordinates (see Sec.5). \nip = ln(j> + c o n s t . Example 11. 11.1: Do discrete spectra permit degeneracy? Show that in the case of a onedimensioanl Schrodinger equation with discrete point spectrum there is no degeneracy. . An example which emphasizes the significance of the spatial dimensionality in this context is treated in Example 8. cannot be degenerate because there is only one quantum number corresponding to one definite energy and hence cannot be degenerate. splits the problem into two independent onedimensional systems. .
102) can be written D<f> + (EQ . W.7 DingleMuller Perturbation Method 155 8. In the following we describe such a procedure as first applied to the strong coupling case of the cosine potential or Mathieu equation.102) where D is a second order differential operator and E a parameter.7 DingleMuller Perturbation Method We have seen above that although the procedure of calculating higher order contributions of perturbation expansions is in principle straightforward.V0)<f> = 0 (8.V0)<t> = (v. See also R. V = V0 + v.104) are chosen such that the equation Dcf) + (E0 . B.105) (8.103) such that if E is written correspondingly E = E0 + e. of course. Eq. Dingle and H. W. (8. .e. with some restrictions on the function V.106) (8. one quickly arrives at clumsy expressions which do not reveal much about the general structure of a higher order perturbation term. anharmonic oscillator potentials and screened Coulomb potentials. We assume that V can be expressed as the sum of two terms. Dingle [72] and H.e)<t>. H. B.104) "This method was developed in R. i. and the solutions then have to be matched in their regions of overlap. Muller [73]. and such that one can even obtain a recurrence relation for the coefficients of the expansion. Muller [216]. J.103). (8. J. The subdivisions (8. Muller [210]. the method has to be applied in each domain separately. cf. the considerations given below would be valid even if D contained only the derivative of the first order. also possible in areas not of immediate relevance here. It is therefore natural to inquire about a procedure which permits one to generate successive orders of a perturbation expansion in a systematic way. (8. It is immaterial here whether D does or does not contain a first order derivative — in fact. T For instance spheroidal functions can be studied with this method. W. in particular in applications to periodic potentials. J. Thus the method is a systematic method of matched asymptotic expansions. (8.t Since in general a solution is valid only in a limited region of the variable.* In various versions this method is used throughout this book.V)(p = 0. Applications of the method are.8. a potential. We consider an equation of the form Dcj) + (E .
m + . (8. Consider (D + Em= (D + E m V0)(4>W + ^(1)) = (D + Em .102) with E ~ Em.E0\ (the latter in a restricted domain). since D<f>m + (Em .) This result is obtained as follows.V0)<f>^ .Em+s)(/)m+s. Considering (frm+s we have.+s ' (8. v = {D+EmVo)<t>m.113) . m + s) which are determined by the recursion formulae. 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.110) If we write the next contribution <f>W to <p(°> in the form m+st (8.m + s)" in order to relate them to "steps" from m to m + s. TimeIndependent Perturbation Theory is exactly solvable and has the eigenvalues Em and eigenfunctions (f>m. (8. the function 4>(0) = <t>m (8.111) s then (see below) 0(1 = E (m. s an integer: (ve)(pm = 'Y^{fn. and so v (8.156 CHAPTER 8.Vo)(j)m+s = (Em .108) with coefficients (m.109) D(j)m+s + (Em .V0)4>m = 0.VQ) ^ s c s4>m+s = 22 s C s(Em ~ Em+s)<fim+s = ( u . We write the coefficients "(m. where m is an additional integral or nearintegral parameter (depending on boundary conditions).V0)<f>m+a = 0.e ) ^ = 2 ( m .107) represents a first approximation to the solution (f> of Eq.m + s)4>m+s s (8. Then if ]e < \E\ and \v . the equation Dcf>m+S + (Em+S .e\ < \V0 . m + s)0m+s.
e.7 DingleMiiller Perturbation Method and so (m. (8.110) and are therefore related to the inverse of a . (8.Q (m.113) must vanish.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 .m)4>m in (8. Eq.m) E J^ ( m ~ Em+S) ' • fa.119) follow a definite pattern and can therefore be constructed in a systematic way.m + s)(m + s.117) „ . The factors (Em — Em+S) in the denominators result from the right hand side of Eq. (8.m) — 0 (first approximation).118).m + s)(m + s.m) 1^1^ IW .m + s)(m + s. V ^ V ^ {rn.102) to order (1) of the perturbation (v — e).114) To insure that 0 = ^(°) + f^ ) H is a solution of Eq.8. (8. Thus (m. (8. /01_. We then find altogether ^  l^m  Em+S) + y^y^ SJ. .119) is effectively the secular equation). £ ^ T^o (^m ~ Em+s){Em — Em+r) ( m ' m + s ) ( m + g ' m + r ) $m+r (8.116) m ^ v (m.T? S The first expansion determines the solution 4> (apart from an overall normalization constant). We observe that the coeffcients of expansions (8. . .115) Repeating this procedure with cf)^1' instead of 4>^ we obtain the next contribution 0(2) =YY together with (m. .m) + > ^ . (8. .m + r)(rn + r. the coefficient of the term (m..m) — —.^ (Em .TP \(T? . and the second expansion is an equation from which e and hence the eigenvalue E is determined (i.Em+S) = 0 (second approximation).„N This procedure can be repeated indefinitely.n\ ^119) + .m + s)(m + s. (8.
423) + Z(2720 9 3 .158 CHAPTER 8. The eigenvalue expansion obtained is an asymptotic expansion. A further conspicuous difference between this method and the usual RayleighSchrodinger perturbation method is that the first approximation of the expansion.e. for instance. W. 2 .* Example 8. Imposing normalization later one obtains the normalization constant as an asymptotic expansion. . (8. for instance. Other applications can be found in the literature. § H.6q + 1)1 + 24(5? . (f>m in the above. The following Example 8. has been treated by R.38).2).[q(llq2 + 1) + 2(33g 2 .[ 4 ( 8 5 g 4 + 2q2 .1)( 2 + 64J3] — . as will be seen in later chapters. and is typical of a large number of cases. An explicit application of the method (not including matching) seems suitable at this point. this is — so to speak — the price paid for obtaining a clear systematics which is essential. does not occur in any of the later contributions. as will be seen later. Miiller [211].9) + 4096J4] + 0 ( l / g 3 ) . TimeIndependent Perturbation Theory Green's function. ^The potential A r 2 / ( 1 + gr2). Kaushal [146]. One should note that in this perturbation theory every order in the perturbation parameter also contains contributions of higher orders. An additional aspect of the method is the full exploitation of the symmetries of the differential equation. 1 . in which q = in + 3.120) n4 . as already mentioned after Eq. . (8.2: 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. This means the solution is not normalized.2 therefore illustrates the use of this perturbation method in the derivation of the eigenvalues of the Gauss potential. i. 2 k2 + g2 = ga(2l + q)^[3(q2 „3 + l) + 4(3q~l)l + 8l2] 32ag r .e. .I2q + 64) + 256Z 3 (41 g . 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. J. n = 0 . i. for the investigation of the behaviour of the late terms of the perturbation series. S. . This can in fact be done and will be dealt with later.7 1 9 2 + 32g + 2976) 3•215g2 +32Z 2 (252g 2 . since it will be used extensively in later chapters. The result is the following expansion. The systematics of the coefficients suggests that one can construct the coefficients of an arbitrary order of this perturbation theory.
<«»> In the limit g —• oo we can neglect the right hand side and write the solution ijj —* I/IQ. (8. . The solution i. Then a?ip dz2 fk2+g2 2ga l(l + l) 1 z*)i> ±(T'(^*and S=z2.^z2) = i>(a) leaves unaccounted for on the right hand side of Eq. 1 . „2 J„2JL.b. 3\ 2 V ' 2. ._ „!2 ^ ( " « 2 r r 2 > . . b.2 '(l + l) .• 0 ( ° ) = ipq(z) = zl+1ez2/4<S>(a. Setting q = in + 3 implies k2 + g2 ga(2l + q). \2 . Therefore in the general case we may set k2 + g2 = ga(2l + q) . (8.7 DingleMuller Perturbation Method Solution: We expand the exponential and rewrite the Schrodinger equation in the form rfV (ir 2 . 2 . (fc2 + <?2) 4ga and b = I \—. (. i/> with Dq = d2 . 11. multiply the equation by onehalf and set h = —a/g.z2 is a normalizable function if a = —n for n = 0 . S) is a confluent hypergeometric function.2) where 1/ .1). 1 1(1 + 1) 1 .V ! 4 I z S *! V2 ipq(z). .. a)ip(a) + (a.0(z) = zl+1ez2^fa. Sees. The recurrence relation for the functions ip(a) follows from that for confluent hypergeometric functions: z2f(a) = (a.123) the contribution fl<°> Ah + .122) where A is of order 1/g.123) The first approximation (note the last expression as convenient notation) ^ .b. 2 The solution XQ(S) = $ ( a .6 and 16. (8.l)ip(a . Then the equation is Dqi> = Ah+y 2 4 ^ 1 1 » fc i\ . We now insert this equation into Eq.121). We then set V>o(2)=2i+1ez2/4X0(2) Now the function XQ{Z) satisfies the confluent hypergeometric equation (cf. a . .8. .2a2 A. 159 •g'g'a'V U E Here we change the independent variable to z = y/2gar.a + l)if>(a + 1) + (a.
a]1=0. a + r) = i 2 \ m ?n z J 0(a) = J^ j= — m S m ( a . TimeIndependent Perturbation Theory (g . a + 1) = a = In general CHAPTER 8. As a check on the usefulness of the perturbation method employed here. of course. Now we make the following important observation that Dqip(a + j) = jip{a + j). when j = 0. (8. In fact. a + r) = 0 for r > m. The coefficients S m ( a . If we now evaluate the various coefficients in the last expansion we obtain the result (8. a) = 1. (a. Hence the next order contribution to tp^ becomes oo i+2 r [a a .i ^(i)=£V+i i=0 J2 j = (i+2). one could now use the RayleighSchrodinger method and rederive the result (8. where [a.160 where (a.S m (a. 0 = h[a. a . one can write down recursion relations for the coeffcients of powers of hl as will be shown later in a simpler context. a + r) satisfy the following recurrence relation: Smi(a.l . a ] i + L ' ^ J 1 [ a + l .^ 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.1) = a .Z. „ . So (a.123) and so in R(°> can be taken care of by adding to the previous approximation the contribution fitp{a + j)/j except.e. . a + r + l)(a + r + l . a + 111 r [a.a]i It is clear that the various contributions can be obtained from a consideration of "allowed steps".a + j}i+1^(a + j).b = (g + 3) . a + r) with the boundary conditions Sn(a. a + i)(a + r. fa. a) = b . the latter determining to that order the eigenvalue. _ J [a . 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. o + r) + S m _ i ( a .120). (a.2a = I + q. ( 5 m ( a . fa. i.3). a]i + h fa.2. a + r) + S m _ i ( a . .j^0 ' + jU+1 J ^(a + j) with h[a. 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.120).a]i = .a]2+L ^ J1[a + 2 .a — 211 . a + i) = 0 for i ^ 0. a + j)i/>(a + m). a ] x + L _ 2 J 1 [ « . a + r — l)(a + r — l .a + 2]i P . This equation shows that a term fiip(a + j) on the right hand side of Eq.a—lli. a ] i + 0(/i3) [a.
although not exactly straightforward. We therefore begin with the appropriate classical considerations with a view to generalization.e. except for the canonical algebra which has to be formulated with Dirac brackets replacing Poisson brackets.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. as we shall see in Chapter 27. 161 . i. 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.2 Reconsideration of Electrodynamics In this chapter we consider some aspects of classical electrodynamics with a view to their generalization in the direction of quantum mechanics. We 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. Regrettably this has to be left out here in view of lack of space. light. 9. and this means electrodynamics.Chapter 9 The Density Matrix and Polarization Phenomena 9. and we shall then see that these lead to analogous results as in Chapter 2. the generalization to many particles. proceeds along similar lines.
4) (9.3) (9.1. In general the constant Ao is complex. 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) Fig. 9.5) which is described as right polarization as compared with left polarization for which (A0R x A 0 /) • k < 0. (9.e. t) given by A(r. The Density Matrix and Polarization Phenomena potential A(r. This is the equation of an ellipse as indicated in Fig. 9. i. t) = A(XR cos(k • r — cut) — Ao/ sin(k • r — cot) and at r = 0: A(0.2) (9.oei(krwt> which satisfies the transversality or Lorentz condition k • A = 0. (9. t) = A 0J R cos(cji) + A 0 / sin(u.t) = 3ftA. A 0 = A0R + iA0I.6) . In the latter case one has the possibility of {A0R x A 0 /) • k > 0.i). It follows therefore that A(r. (9.162 CHAPTER 9. When Ao/? is perpendicular to Ao/ and Ao# = Ao/ the ellipse becomes a circle and the field is said to be circularly polarized.
the connection must be linear. / = 4 0 ) * 4 0 ) + 4 0 ) * 4 0 ) = (A(°\ A<°>) = A(°)2. the intensity of a wave. the vector potential A. AW = £ fc=l. (AW. i. Then their connection is given by Eq.8). We set therefore A 0 = Ai + A 2 . represented by a filter "F".k (910) We give the initial wave the label "0" and subject this to an experiment. (9.A« we can construct only the following scalar products with this invariance: (A(°\A(°)). The intensity 7 of a light wave with amplitude coefficients A\' .A i = 0 = k . however. i.g.7) We consider now measurements with polarized light. p. the fields E and B are observables. as e. A2.2 *i*4 0) (98) with coefficients F^. (9. we recall. the direction of polarization.9) The intensity is characterized by the fact that it can be looked at as the "expectation value" (1) of the unit matrix 12x2. .e.11) 'This means we consider the polarization. (9.A2 is the sum of the moduli. ie. 17.e. for instance.A«).9.A(°)). not. 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. (A(°).e. I=<l2x2>:=E40)*l*40)i.AW). (AW. determined with the help of a polarizer.2) the vector potential A is completely determined by its two (say (x. a calcite crystal or a film of nitrocellulose.t in which k and ui remain unchanged. which leads to an outgoing wave which we give the label "1". From the vectors A(°). i. (9.A 2 . Rae [234]. See for instance A. Here we are interested in observable properties of light waves — like.2 Reconsideration of Electrodynamics 163 As a consequence of the Lorentz condition (9.y)) components A i . Observable quantities can only be those which are invariant under translations and rotations. with k . In classical electrodynamics.
it must be real and hence (F) = <F>*.12) i. But we can convince ourselves that we achieve exactly the same as with our earlier consideration of observables. Then according to Eq. i.164 CHAPTER 9. (F)=Tr(Fp).e. (9. We write the observable quantity as (F):=Y.e.15) in order to rewrite Eq.15) This matrix is hermitian since (Pife)t = (AiAtf = (A*Akf = (AkA*f = {pkif = (Pik). To this end we observe that as a consequence of Eq. The Density Matrix and Polarization Phenomena Only the second last of these combinations describes the experiment. (A(°). (9. i. Thus (F) = J2FikPki = ^(Fp)a.k i (9. Thus. pki = AkA*.14) We now introduce a matrix p — called density matrix — which is defined by the relation pik := AiA%.13).13) the quantity F must be an hermitian matrix and thus representative of an observable. (9.A( 1 )) = ^ ^ F i. (9.16) i. so that (F) can be looked at as the expectation value of an observable F. if (F) is an observable quantity.A*kF£A* = E and hence F = Fl A p i tkAk. (9. the connection between the initial polarization and the final polarization.17) .A*FikAk. We use definition (9.13): j2^FikAk = Y^AiF*kAi = Y.k l f c 4 ° \ (9.13) This consideration of translation and rotation invariant quantities is somewhat outside the framework of our earlier arguments.k (9. (9. i.e.
(AAt) = ( £ £ ) .( J o ) . 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. This admixture is described by introducing matrices p.9. that of a pure state. i. A2 = b as </*) . 1 The classical polarization vector or wave vector A corresponds in the quantum mechanical case to the wave function ip.d "»=(o ! ) " In the case of an admixture with equal portions of 50%.21b) This state is still a pure state. we have (9 19) ' In this case there is no preferential direction. The density matrix for the pure state (9. mutually orthogonal polarization directions x and y described by the matrices (see also below) * .e.* For a lightwave travelling in the direction of z. "incoherently.19). The socalled pure case. i. is in general unpolarized. <»^> (9. i. (9. like that of a bulb. it is a statistical admixture of light rays whose polarization vectors are uniformly distributed over all directions. is given by a single polarized wave.2 Reconsideration of Electrodynamics In particular we have / = Tr(p).e.22) For a — 1. In the present case of a polarized beam of light we have two possible.18) The vector A describes a particular ray of light. But light. For a 45° polarized wave we have 1 . b = 1 we obtain the expressions (9. 165 (9.21a) follows from Eq. b = 0 and o = 0.e.15) with A\ = a. . (9.
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. We now inquire about the the way p changes in the process of the transmission of the lightwave through an apparatus. (9. b = —=.15) Pik — A i A k . in which the intensity I = Tr(p) is not conserved (e. (9. Equation (9. as a consequence of absorption).g.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. The Density Matrix and Polarization Phenomena P45° = I 1 For a 135° polarized wave we have a = and l M . (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". We set in analogy to Eq.8) describes the relation between the initial and final polarization vectors A^°) and A^1) respectively in a measurement of the observable F.™ V 0 n kmAm / ^^jPim^mk' . Pi35° = \ \ 2 ? )• (9.8) A(1)=E^40)> k R = R(<*).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. 1 = . Then according to Eq.166 and CHAPTER 9.Z^ 3. (9. l l\ " i °\ ) ' ^925^ 9.
9. that Io = Tr( / 0 (°))=Tr(pW) = / 1 . The Liouville equation describes the motion of an equal number of systems in phase space elements of the same size.t) in Chapter 2 in what we described as the Schrodinger picture there. and p(0' remains unchanged.28) (9. In the case of Eq. in other words without creation or annihilation of systems.q.30) Ti{FRp{®I$) = Tr(F.e. (9.33) i.e.34) .30) the dependence on a is contained in pW(a) = R{a)p^{a0)RJ'(a) (9. The two cases are completely equivalent and can therefore be described as Schrodinger picture and Heisenberg picture representations. on the other hand.29) where (i. the dependence on a is contained in F' = R^(a)FR{a).4 The Liouville Equation 167 pW = RpMRl (9. that the initial intensity Jo is equal to the final state intensity I\. The corresponding condition here is that no absorption of the wave takes place.27) The measurement of an observable F in the new state with superscript " 1 " then implies according to Eq. in the state functional) in analogy to the time dependence of W(p. i.9.pW). the observable F remaining unchanged. (9. i. here this contains the dependence on a. (9.e. We can interpret the result either as (F) ( 1 ) = Tr(F{Rp(0)R}}) or as (F){1)=Tr({RtFR}p<0)).31) (9. (9. the observable is transformed. (9.4 The Liouville Equation It is natural to go one step further and to derive the equation analogous to the Liouville equation.17) (F) (1 ) = Ti{FpM) = = Tr(tiiFRpW) F' = RtFR.31).32) (9. (9. In the case of Eq.e.
(9.46) suggests to define Poisson brackets of matrices by the correspondence {A. In Chapter 27 we shall see that the Poisson brackets here are actually Dirac brackets because the gauge fixing condition (9. Comparison of these equations with Eqs.37) (9.p(°\ao)]da.21) and (2.2) has to be taken into account.B].35) we have 1 = R*R = (1 + M W ) ( 1 + Mda) ~ 1 + (M* + M)da. In view of Eq.40) Equations (9. (2. (9. Eq.e. . (9.38) H = Hl R(a)p(°\a0)R\a) (liHda)p(0)(a0)(± + iHda) pM(ao)i[H.168 i.33)) we obtain correspondingly in the Heisenberg picture the equation ^ = +i[H.e.F}.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. For an infinitesimal variation of the state of polarization we have R{a) = 1 + Mda. For p ( 1 ) ( a ) = p ( ° ) ( a 0 + da) follows from Eq. The Density Matrix and Polarization Phenomena Tr(p(°)) = Tr(p^) so that = Tr(Rp^R^) RlR = 1. We set therefore M^:=iH.B}:=i[A.39) If the dependence on a is not contained in p^1' but in F' (cf.36) (9. CHAPTER 9.p]. the matrix M is antiHermitian: Mf = .M . (9. (9. (9.35) Thus the matrix R has to be unitary. i.32) in the Schrodinger picture that p^iao + da) = = = or ( w i t h p(°> —> p) (9. j£ = i[H. = Tr(R^Rp^).39) and (9.40) can be considered as quasiequations of motion. (9.
and consider timedependent perturbation theory. 10.1 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. the Schrodinger. (10.2 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. To insure that the 169 . 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).Chapter 10 Quantum Theory: The General Formalism 10. Thus the expectation value with respect to the vector describing the state in the Hilbert space does not depend on this phase factor. Heisenberg and interaction pictures.1) This expectation value remains unchanged if \u) is replaced by e%a\u) with a real. we establish their Heisenberg equations of motion.
3) These fundamental commutators determine the commutators of arbitrary observables A = A(p. and assume that A = A(p.Qj] = 0.Pj] = 0.Pi are called fundamental observables.q). B We first establish the following theorem: If A and B are observables and hence hermitian operators obeying the commutation relation [A. The phase space coordinates qi. q).2.3 represent the three Cartesian coordinate directions of x. however. The first step in the investigation of a quantum mechanical system is to specify its dynamical variables A. Such a correspondence with classical mechanics is not always possible. which clearly does not hold in the case of operators. Naturally one can also write (10. [qi. we assume this now.* a conspicuous exception being for instance a spin system. such as for instance of the angular momentum operators Li. In order to avoid multivaluedness one always starts from Cartesian coordinates in position or configuration space.2.g. where i = 1.Pj] = ihSlj. 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.1 U n c e r t a i n t y relation for observables A. (10.y. Quantum Theory: The General Formalism expectation value of the unit vector or identity operator is 1. 10.4) *Even in those cases where this is possible the operator correspondence may not be unique. \pi. y.B] = ih. For these the following rules of canonical quantization are postulated: Postulate: The operators q~i. Here. . we must have (u\u) = 1.2) (F(A)) = {u\u) <"lf.Pi are postulated to obey the following fundamental commutation relations (again leaving out "hats"): [Qi. classically qiPi = PiQi. E. (10.170 CHAPTER 10. z and for definiteness here — which will not be maintained throughout — the "hat" denotes that the quantity is an operator. Pi • Pi.
{A)2)1'2 {{A2) . (10. Hence (AA)2{AB)2 = (A2)(B2) = (u\A2\u)(u\B2\u). (10.7a) (B) = 0.1) (AA)2{AB)2 = > (u\A2\u)(u\B2\u} \{u\AB\u)\2. Since A = A— (A). AB defined by the mean square deviation AC=((C2)(C)2)1/2. (10.2A(A)}  {A)2)1'2 (10..0)1/2 = AA.5) are subject to the following inequality called uncertainty relation: AAAB In proving the relation we first set A:=A(A). B}= ih.5)) (for A\u).. (10.7b).10) . Thus AA = AA = ( i 2 ) V 2 . so that (A) = (A) .B. we have A2 = A2 . (A{A))(B{B)){B{B))(A(A)) (10.8) and. (10.(A)2)1'2 (1 °= a) ((A2 + (A)2 .6) B:=B(B). Then [A. (10..2(A)2 .2 States and Observables then their uncertainties AA.12) where \u) 6 "K is a ketvector representing the state of the system.2A(A) + (A)2. C = A. = \\A\u)\\2\\B\u)\\2 (10.9a).13) AB = AB = (B2)1/2.(A) = 0.(A)2 = (AA)2. (3. 111 (10.9) ((A2) + (A)2 .B\u) G IK as vectors) we obtain (for an application see Example 10.B] = ABBA > ^h. Applying to the expression (10. (10. using Eq.12) the Schwarz inequality (cf..7b) = = [A. Ai = = = ((A2) . (4.11) (A2) = (A2) . Eqs.10.
13) (AA)2{AB)2 i.19) (u\AB + BA\u) = (u\AB\u) + (u\BA\u) c*(u\BB\u) + c(u\BB\u) = (c* + c)(u\B2\u). and with Eq. (10.3.18) >R2 + I2. for a complex constant c.4). . Quantum Theory: The General Formalism Next we separate the product AB into its hermitian and antihermitian components^ and use Eq. (10.172 CHAPTER 10. 4. R= (AB + BA). The condition (a) is satisfied when. since A and B are hermitian. (10. and hence for condition (b) 0 = = tSee Sec. A\u)=cB\u). (AB + BA) This means (AB + BA) is real.e.17) as had to be shown. (10.15) as (u\AB\u) = R + iI.15) = = (u\AB + BA\u) = (u\(AB + BA)^\u)* (u\(AB + BA)\u)*. The equaltosign applies in the case when (a) it applies in the Schwarz inequality. (10. Zi (10. From this we obtain (uiJBu) 2 = i22 + / 2 . (10. and (b) when (AB + BA) = 0. We can therefore rewrite Eq. Then AB = \(AB + BA) + \(ABZ Z BA) (1 = 8) \{AB + BA) + \ih. Zi + BA)\ + ^ih.16) I = h. (10. AAAB>I=h. Zi Zi (10.14) Hence (u\AB\u) = /^(AB where.
z. in view of Eq. For these we have in the onedimensional case the relation [p. for which one component of L is "sharp"? Solution: Applying Eq. = 5R2 + G 2 > 9 2 = hi2(Lk)2. we have LiLj = (LiLj and ^ = y(u\LiLj so that \(u\LiLj\u)\2 Hence {ALx)2{ALy)2 > l + LjLi) + [Li.e. Ly are not "sharp". Thus in this case Lx. (ALZ)2(ALX)2 > \h2 = (Ly)2.3 OneDimensional Systems We now consider onedimensional systems with a classical analogue as described earlier.e. Example 10. we obtain {ALi)2{ALi)2 > \{u\LiLj\u)\2.i relation holds: (ALi) 2 (AL. . For states \u). The observables like angular momentum are functions of p and q. {u\LiLj\u) = R + iQ. 9 = h{u\eiikLk\u).7a). (Ly)=0. (10. i.10. (Lx±iLy) ^ 0.e. m\lm)). 9ftc = 0. ALZ = 0 (these are states \lm) with Lz\lm) it follows that (Lx)2 = 0. (Lz)2^0. h2(Lz)2. implying that the vector it) has to satisfy the following equation: A\u) = i($Sc)B\u) (10. (ALy)2{ALz)2 > ~h2(Lx)2. (A{A))\u)=i(Zc)(B(B))\u).e. What does the relation imply for states u) = \lm).) 2 > = x. the following uncertainty h2{Lk)2.3 OneDimensional Systems 173 Thus c = — c*.Lj] = (LiLj + LjLi) + ih£ijkLk. i. ALX 10. = 0. Separating the product into symmetric and antisymmetric parts. implying {Lx)=0. i.13) to operators Li — Li.20) or. ALy = (u\L±\u) ^ 0.1: Angular momentum and uncertainties Show that in the case of angular momentum operators Li. (Ly)2=0.y. such that Lz is "sharp". i. (10. q) = ih. + LjLi\u).
21b) now follow with the assumption that A(q. as remarked earlier. Next we have a closer look at the operator q. (10. Eq.e. Similarly we obtain \p. We have (cf.C\.p] = ih we obtain therefore [q. q]q + # > v] = ~2ihq and \p.A(q.i h ^ ^ .q).21a) and (10. Then (see below) we can derive for A the following commutator equations [q. (10. . and (c) the spectrum is not degenerate.21b) (10.21a) Equations (10. This is why the operator p requires a representation in the space of eigenvectors of q. i.p)] = .pn]=ih^(pn). (b) its spectrum is necessarily continuous. We obtain these equations as follows.p}p + p[q. for q\x) = x\x) we have p\X)=\x)(th^y Now let A be an observable. Q2] = [p. (3. Using [q. they do not have a common system of eigenvectors.B]C + B[A.174 CHAPTER 10.A(q.p)]=ih^^and \p.p2] = [q. A — A(p.p) can be expanded as a power series in q and p.e. Quantum Theory: The General Formalism Since this says that p and q do not commute (as operators).p] = ^ihp ~ Proceeding in this way we obtain [q. i. We demonstrate that (a) its eigenvectors have infinite norm (in the sense of a delta function).22) d ih—(p2).qn] = ih^(qn).BC] = [A.34d)) [A.23) = ihg(q2) (10.
i. an operator.24) lim SK(x — x') — S(x — x').21a) A by U. . 6K(xx') = — fK dkeik^x'^ 2TT JK sin K(X K{X — x') — x') (10. (10. (10. Replacing in Eq. and a a cnumber.e.10.f^ ^ J—oo dke ikx —ikx' (10. U] = ih^= aU.25) It is shown in Fig. We have (x\x') = = where f^ J—oo dk(x\k)(k\x'} (4 = 9 ) ±. we obtain [q.1 The delta function for K — oo. U is unitary. We thus see explicitly t h a t the vectors \x) have infinite norm (in the sense of delta function normalization).28) i.26) it Here p is an observable.e.3 OneDimensional Systems 6K(x) 175 Fig. 10.1 how the delta function arises in the limit K —> oo. For the cases (b) and (c) we consider the operator U(a — pipa/h (10. > We begin with (a). = eipa/h = Uia)'1 = U{a). Since p = p\ follows t h a t U\a) so t h a t U(a)U\a) = 1.27) (10. 10.
WU = 1. Evidently every eigenvalue has only one eigenvector. Since this holds for any arbitrary value of a in (—00. 10. CHAPTER 10. q) can be constructed representing a dynamical variable. that with U(a) = e~ipa/h = U. Quantum Theory: The General Formalism qU = Uq + aU = U(q + a).3.30) (10.1 T h e translation operator U(a) We saw above. i. In an analogous way by defining the unitary operator U(b) = eiqblh (q operator.q)\i>) = (i>\U^UA(p. we obtain qU\x) = U(q + a)\x) = {x + a)U\x). or qU(a)\x') = (x' + a)U(a)\x'). The eigenvectors have the same norm as \x): (x\U*U\x') = (x\x') = S(xx').29). cf. (10. Eq. q any other observable F(p. ^Observe that {iji\A{p.00). which means that the spectrum of this operator is a continuum. Thus all these values belong to the spectrum of the operator q. the spectrum is not degenerate.176 i.e. or qU\x) = U{q + a)\x) = (x + a)U\x).31) . 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 (—00.00).29) This means: U\x) is eigenvector of q with eigenvalue x + a. (10. we can see that also p possesses only a continuous spectrum (from —00 to 00). (10. b a cnumber).q)U^U\ilj).e. 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.
(10.37) we set (with e infinitesimal) U(e) ~ 1 Then (x'\U(e)\x") = 6(x' . i. (10. for instance.10.35) (10. It follows that U(a)\x') = = U(a)U(x')\0) = U(a + x')\0) \x' + a).x") U{x'\p\x" % pz.36) i. \x') = U(x')\0). Comparing Eqs.x" .x')\ = l.x')5(x" = (x" + a\c*c\x' + a) x'). offdiagonal elelemts of the {x}representation of the momentum operator.x') = = i. (10. (10.38) h 6(x' .a).e) ~ 5{x' . i.hm or „'i„i™" s> i ™' ~/ (x'\p\x")\ = 8'{x' x").x")c(a.x" . In the expression of Eq. We choose the phase such that U(a)\0) = \a). It also follows that (x'\U(a)\x") = (x'\x" + a) = S(x' .e./ (10. \c{a. i. (10. (10.31) and (10. it follows that 5{x" . U(a)\x') = c\x' + a).e.3 OneDimensional Systems 177 Moreover.32) we see that \x' + a)ocU(a)\x'). since q\x') — x'\x').x") .e.33) (10.e.5(x> .37) With the help of the latter expression we can calculate.e) (x \p\x ) = . the operator U(a) acts to shift the value x' by the amount a.x" .32) Since U is unitary. (10.39) .e. (x'\p\x").34) {x"\U\a)U{a)\x') c*(a. The operator is therefore called translation operator. we have q\x' + a) = (x' + a)\x' + a).
Differentiating U(t.44) (10. 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)m0)). (10. 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.35). 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.40) 10. In this case the classical Hamilton function does not depend explicitly on t.§ It follows that (x"\p\x') = d'(x' This expresses that p is hermitian.tQ) = eiH(tt<Mh.L fdpeipx.178 CHAPTER 10. (5. Let \UE) be an eigenvector of the Hamilton operator H with eigenvalue E. i. (10.4.t0) = HU(t.e.t0).° ° d(p) (ip) e"« = S'(x).e.42) ^Observe that 6'(x) = £ f ^ dpipe** = ± / . This is in conformity with our earlier postulate in Sec. .to) is called the time development operator. .45) (10.43) (10. (io. 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). a conservative system.4i) We assume first of all. 6'{x) = !.f dpipeipx. 5. to) with respect to t we obtain ihjtU(t. See Eqs. The operator U(t.to) exists which is such that^ m)) = u(t.x") = (x'\p\x")*. H\uE(to))=E\uE(to)). Quantum Theory: The General Formalism We mention in passing that.27) and (5.e. i. so that U{t. since 5(x) = .
*l)^(*l)> = I7(*2.48): [/(Mo) = 1 .45) has very general validity. (10. Eq.t0) or ih—U{t.47) e.to).44)) we have U(t.t0) = U(t + we have U(t t) = lim ^ + Mo)^(Mo) = n (1046) = U(t. (10. Evidently U(t. Ufa. t0) (10.4 Equations of Motion 179 We now demonstrate that Eq.48) with U(to.e. (10.t)U(t.t).to) (differentiating we obtain immediately Eq. (10.t) = l^eH{t). (10.*l)^(*l.to) can be expressed as an integral as the solution of Eq.tiM*i. From Eq. (10. In the immediate considerations it is not necessary to restrict ourselves to systems with classical analogues.t) and therefore [U(t. Hence U(t. i.t0)]1 Let us set for small values of e: U(t + e. t0) = H(t)U{t.t) = l.to) = 1.49) . Then with U(t + e.z / dt'H(t')U(t'.t0) [U(t + e.t) = 1.10.48)). = U(t0.41) we obtain ^(t 2 )> = = J7(t2.t0)\^(t0)).t)l]U(t.t 0 )^(*0)> U(t2. t0) = E/(t 2 . The operator H is d efined by this relation.t0)U(t0. (10. But obviously (cf.
it follows that U(t + dt. Another such description is the Heisenberg picture. Since the only measurable quantities are moduli of scalar products.180 CHAPTER 10. and these remain unchanged under unitary transformations.e. the Schrodinger equation ju(t.(£ + dt)) = U(t + dt. to) can therefore be interpreted as a product of a sequence of infinitesimal unitary transformations.e. 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^). the predictions of different descriptions are identical. (10. requires the identification H(t) = H= Hamilton operator. Then \Mt)) = u(t. (io. Now let \tp(t)) be the state of the sytem at time t. 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. The operator U(t.41). Comparison with the equation postulated earlier. 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. >(*)> = i. whereas the physical quantities are represented by observables in "K which do not depend explicitly on t. The eigenvectors of these observables are also constant in time. i.t0)\ii>s{to)) (10. (10. t)\t/>(t)) =U~ and since H is hermitian.53) .t) = l % ^Hdt) \il>(t)) Hdt (10. 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 instant at which a measurement is performed on the system. Eq.30).51) Since </.5o) ihjtm)) = Hm)).t0) m0)). (5.
HH]. Differentiating Eq. i. if ^ = 0.48).e.t0)\ips(t)) = \ips(to)) = constant in time. = so that ihjtAH or ihjtAH = = [AH. t0)AsU(t. (10.HH].i).With explicit time dependence.56) U^[As.10.HH]+ihU^U. but here. (10. even if As is not timedependent.e. as a rule. 181 (10. t0). i. As(q.58) HHAH (U*ASU)(U^HU) [AH. in order to obtain the Heisenberg picture (description). However the associated observables are timedependent. we have to take into account the contribution dAs/dt. subject to a continuous change. so that a matrix element (i.p) in the Schrodinger picture dAs/dt = 0. {AH.4 Equations of Motion and \^H) = U\t.54) Observables transform correspondingly with a similarity transformation. For As(q.59) This equation is called Heisenberg equation of motion. we obtain ih^AH at = = U^HAsU + ihU^^U at + U^AsHU (10.(tf HU)(rfASU) (10. It does not always have to be the case that As does not contain an explicit time dependence.57) and rf[As. (10.e. for instance in cases of nonequilibrium or if part of a system disappears through absorption. Thus. (10. 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.55) and using Eq. (10. In the Heisenberg picture it is HH = U]HU.H]U + iW]^U.55) Thus AH is explicitly timedependent.p. we consider only cases without explicit time dependence of As. a scalar product) remains unchanged: AH(t) = U\t. Here H is the Hamilton operator in the Schrodinger picture. .H]U = = = U]ASHU AHHH — U^HASU .
in the case of the fundamental dynamical variables qi. Solution: We have for an observable A (A) = (i>H\AH\TpH). 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. For an extensive discussion see B.H] = 0. Quantum Theory: The General Formalism Wave mechanics is now seen to be the formulation of quantum mechanics in the Schrodinger picture.21b). (10. In our treatment of representations. [CH. 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. A. Finally we add a comment concerning conserved quantities. This holds also in the case of the Schrodinger picture. we used in essence unitary transformations of matrices. provided the Poisson brackets in the latter are replaced by commutators. i. if ffCH = 0.61) Thus conserved quantities commute with the Hamilton operator. 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. since in the transition from one to the other the commutator remains unchanged.2: Energy and time uncertainty relation" Use the Heisenberg equation of motion and the general relation (10.21a) and (10.Pi): and where the expressions on the right follow from Eqs.Pi we have in the Heisenberg picture (replacing in the above AH by qi.182 CHAPTER 10. .13) to derive the uncertainty relation for energy and time. In the treatment of the pictures we used unitary transformations of operators (and vectors). We see that formally one obtains Hamilton's equations of classical mechanics. An observable CH is called a conserved quantity. We also observe that the momentum pi is conserved when \pi. In particular. Example 10. Orfanopoulos [224].HH}=0.e. (10. In many cases the Schrodinger picture is more amenable for explicit calculations.
4 Equations of Motion Then. Let \ip) = \"4>H) represent the state of the system. (A).2 > h. we can replace ([AH.65) and rA AA d{A)/dt' . H\tl>) = {H(H))\f). is displaced by A A in the interval A T = TA.13) and (10. at ^±AH>h.\HAM\2) \(nA. a position or momentum observation is independent of t. this is practically the same as at time t for times t with \t — t'\ < T. since \IPH) is timeindependent. AW = {A(A»V>. > (10.e. (10. i. . 2 d{A)/dt " and.66) This means. H] = 0.62) and dAs/dt > \WAH\i.67) 1 d(A)/dt is oo and AE = 0.64) = 0. We construct the following vectors using Eq. and TA O C (10.68) To put it The eigenvalues a of A (i.p) we then have 0=^M) = U[A. (10.7a).Such a characteristic time interval can be defined for any dynamical variable.e. of A\<t>) = a<£)) are then called "good quantum numbers". (10. then if a measurement is made at time t'. in a different form: The eigenvalues of an observable are called "good" if [A.H]) = 0.62) The Hamiltonian H does not depend on time.)\2 ( = \{i. The probability density P ( r ) =  ^ ( r )  2 = (^r>(i#> is independent of t. 183 dV ' \ dt >"•"»» + ( £ ) • (10.HH]) Hence we obtain by ih(A).15): (AAf(AH)2 Using Eq.10.e. If r is the smallest such characteristic interval of time. i. the "centre of mass" of the statistical distribution of A. i. dt in. (10.63) so that with Eqs. (10. We defined earlier stationary states as states with a definite energy E and wave function * = V( r ) e x P ( — i E t / h ) . This means that for a physical variable A(q.e. (10. if we set AH = AE we obtain TAAE (10.H}\f)\2.
First we obtain from Eq. (x)/ E o (E')5(E .72) = V'EoW 7/2h + 0i(EE )/h  oo . Quantum Theory: The General Formalism 10. i. the particle density diminishes in the course of time (t > 0).£ ^ + n/2)^ o ( x ) e{^1/2+i(EE )}t/h 0 o =0 . Recall for instance radioactive decay.69) and (10. (10. then fE{E') = S(EE') and integration with respect to E' yields again t()E(x. The wave function of such a state with energy EQ and lifetime r = h/'y > 0 could.t) — ipE(x.t) of such a state must be the superposition of states with different energies about EQ and a corresponding weight function / so that ^ ( x . (10..(x)fEo(E')e«EE'Wh 2irh f dE'^E.69) (Observe that if ip consists of only one state with energy E in the domain of integration of E'. as is observed. the wave function ^ ( x . (10. (10. This means.71) From Eq. i ) = J dE'^El^)eiEltlhfEo{E').70) by From Eqs. But the number of radioactive (thus excited) nuclei decreases exponentially. t) = ^(x) exp(—iEt/h) describes states called bound states of unlimited lifetime.e.69): oo poo p / oo *PEo(x.Eo(x)eiEot/he'^2h9(t). This means that in the case of decaying particles. Energy and decay length of the radiated aparticles are characteristic properties of naturally radioactive nuclei. for instance. /oo (10.H ) ( £ .5 States of Finite Lifetime When E is real and P(x) is independent of t and J dxP(x) = 1.E1) 2irhfEo{E)4>E(x).tyEt/hdt = = = / J—oo dt J dEJrl>E. (10. On the basis of the uncertainty relation AEAt > h these must have an uncertainty AE in their energy spectrum. t)e iEt h ' dt = V>£0 (x) / JO <fte{7/2+i(i5£b)}t/ft (10.184 CHAPTER 10. have around EQ the form ipEofct) = i.)exp(iEt/h)). States with finite lifetime (At = r ^ oo) are those whose probability density falls off after a certain length of time called "lifetime". the wave function ipEo(x.70) we can determine the function fEo{E) integration with respect to t. states with sharp energy (AE = 0).70) we obtain: oo / oo TPEO (x.
Considered as a function of E. £/ = £/W(Mo)^'(Mo). VE(X) ^ ^ (£ > = dbfift+T/* (la73) This result is known as the BreitWigner formula. (C/(0) unitary).e. where U is the solution of Eq. ihU(t. (10. A state is discrete if its immediate neighbourhood (in energy) does not contain some other state.6 The Interaction Picture One more motion picture of quantum mechanical systems is in use. The state with lifetime r — h/'y is not a discrete state. (10. W)) = U(t. (14. 7 > 0. the formula says that JEQ{E) possesses a simple pole at E = Eo — 27/2. This description contains effectively parts of the other two pictures and is particularly useful when Eq.10.73) precisely by assuming with fEo{E)^8{EEQ) a "smearing" of states around the energy EQ (with uncertainty AE).e.g. to) be an approximate but unitary solution of this equation. Let U^°\t. whose real part EQ specifies the energy of the state and and whose imaginary part 7/2 specifies the lifetime r = h/j.e.45). We see therefore that the states with lifetime r < 00 and TAE > h belong to the continuum of the spectrum.e.76) **For specific applications see e. i. every unitary transformation of states and operators defines a possible "picture"). (10. 10.11). . (10. They are called "resonance states".41). of course. We obtained the result (10. and is called "interaction picture" or "Dime picture". Eqs. which means that a nucleous with discrete energy EQ does not possesses some other admissable level close to EQ. that** EQ.75) (10. 185 i.6 The Interaction Picture Prom the last two equations we deduce for E close to ipE0(x). ^(*o))> then \xp(t)) for t > t0 follows from Eq. We saw previously that if the state of a system is known at time to. i.105) and (20.t0).45) can be solved only approximately (in actual fact.e.74) Thus the solution of this equation is the main problem. i. i.to) = HU(t. (10.toM{to)).
83) ih^U^ ishermitian.t0) 0 = l. i. H(°Ht) = ih^uW.78) at with the initial condition c/(0)t f ^ ( 0 ) _ ^ ^ 1 V \ at J U\t0.186 Then CHAPTER 10. dt so that #(0)t ( t ) It follows that H{0\t) = dt = H<®{t). (10. For a "good" approximation U « [ / ' ' we have HU^ihjtU^^0. with Eq. with Hamiltonian H = H^ + H'. £/(°)£/(°)t = 1. (10.77) C/(o)t itiuWfLu^HUMu'ih^U' at or i. = HU<®U'.e.e. i.79) ^'~°i.e. Now let H^(t). i. (10.h^U> = at ( io.81) But [/(°) is unitary. be defined by the relation H(°)rj(0) _ i h ^ l = 0 (exact)> n (0) = eiH(o)t/aj (1080) i.e. [/' has only a weak ^dependence (i. varies only slowly with t).e. In addition we have (10.e.e.78) (10.82) (10. Quantum Theory: The General Formalism ih^(uWu') i. .
91) Since the Schrodinger equation is given by ihjt\^s{t)) = H\1>s{t)) = ( # ( 0 ) + H'Ms(t)). (10. (10. (10.78) at (10 85) (10.90) = U^H'\^s(t)). U<®\Mt)) = \Mt)).88) and as interaction picture version of an observable A the quantity Ar(t) = tf(°>t AgU^. H = H{®{t) + H'.10. (10.87) = We define correspondingly as interaction picture state iM*)> = t/ (0). We now multiply this equation from the right by U' and insert the first expression on the right (of Eq.78). (10. (10. (10. (10.6 The Interaction Picture 187 We use now the subdivision of H into unperturbed part and interaction. lfe(*)}.80).85) = where we used Eq. We also set H^U^H'UW. and obtain from Eq.86) at = U^HU^U'.89) where ^s(i)) and As are state and observable in the Schrodinger picture. U^H'U^U' (10.e.92) .84) U^ where H' is also hermitian. (10. Then ihjt\Mt)) = ih~u^\Mt)) = u(0)ihi^W and H'^it)) = U^H'U^U^\^s(t)) + ih(^) \Mt)) (10.U^HU^U' H'fi'. 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. (10. i.
comments after Eq. (10.e. we have i. In order to obtain the equation of motion of an interaction picture observable Ai. (10. A^t) = U^AsUi0). cf.93) Thus the vector ipi(t)) satisfies an equation like the Schrodinger equation.60b) with iff0) instead of H.95) Thus in the interaction picture the physical quantities A are represented by timedependent observables which satisfy a type of Heisenberg equation (10. (10. Since H'j is assumed to be small (i. (10. ot i.60a) or (10. H'j = UWH'UW. we differentiate with respect to time the operator defined by Eq. (note the difference between H' and H'j) ihjt\Mt)) = H'AMt)).e. (with ihjtAI(t) = [AI.80) (and multiplying by t/(°)t) ih~\Mt)) = u<MH'uM\Mt)). (10.e. i.e. dAs/dt=0) .90) and replacing the left hand side of this equation by the right hand side of Eq.H\0)}. ^ (0) l^(*)>. the vector varies with time (different from the Heisenberg picture).92) ^°)^V/(t)) + ^ ( ^ )  V ' / W > = ^ (0) ^ (0) l^(*)> + ^ .o)AU(o) at +uWAsHWuW UWHWU^UWASUW + iww^uw dt +U^ASU^U^H^U^ (10. i.e. a perturbation so that H ~ H(°)).188 CHAPTER 10.e.89). so that with Eq. Quantum Theory: The General Formalism we obtain by starting from the right hand side of Eq. i.55)) at + imm<Ms_u(o) at dt _umH(. near zero and \tpi(t)) is almost constant in time.94) Then we have (normally with dAs/dt dt (io=8o) — 0. (10. (10.94) _Hf)Al 1 + lhdAl+AlHf). (10.89)^0.
(10. We assume again that the spectrum and the eigenvectors of Ho are known. \Mt)) = +' \Mto))^fdt'Hi(t')\Mto)) ft J dt' f dt"Hi{t')Hi{t")\^I{tQ)) + (10.e. where according to (10. (10.t0).e.87) and (10.96) in which the perturbation part H'(t) depends explicitly on time and HQ replaces H^0' in the previous discussion. H0\ipn) = En\(fn).94) Hi = uW(t.99) (1098) Instead of the original Schrodinger equation we now solve Eq.101) Iteration of this inhomogeneous integral equation yields the socalled Neumann series.10.7 TimeDependent Perturbation Theory We consider timedependent perturbation theory as an application of the interaction picture. {ipn\<Pm) = Snm. ^ \<Pn){Vn\ = 1n (10. the system is in the Schrodinger eigenstate \ips ) of iJ 0 with energy Em. ih±\Mt))=Hi{t)\Mt)). We obtain the operator U^ from Eq. (10.t0)H'uW(t. Since . i. (10.100) (10.e.97) The equation to solve is Eq.98).102) We assume that before the perturbation H' is switched on at time to. (10. i. (10.i0) = e<Ho(tto)/ft. Integration with respect to t yields \Mt)) = \Mto)) ~ \ \ dt'Hi{t')\^j{t')).93).80) as E/(°)(t.7 TimeDependent Perturbation Theory 189 10. We have the Hamiltonian H = H0 + H'. i. H' = H'(t).
i. assuming that this supplies a sufficiently good approximation: IMQ) = \fm) lz\ n dt'HW)]^). we can write l 4 0 ) ) m = elHot/hWm) =e . 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.102) and truncate the series after the first perturbation contribution.105) In the interaction picture the iJostate m at time t is \Mt)) = eiHot/hWP)m = \<pm). 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'. (10.105) rt n Jto h ' Jt00 t rt  x  l h ' Jto t0 (10. (10.88) and obtain > n ^ V s ) = {<Pn\eiHot/h\ll>s) = {<Pn\eiHot/hUM\Mt)) = <¥>*(*)>• (10.108) . (10.106) We insert this expression into Eq.103) for m — n and use Eq. Hence we set ^> = £°«(*)l4 0 ) >n.103) The actual state \ipg) of the system is solution of ih\^s) = H\^s). (10.e. the time dependence of \ips } can be separated as for stationary states. (10.^ V ™ ) . (10.190 CHAPTER 10. Quantum Theory: The General Formalism HQ is independent of time. The corresponding amplitude or appropriate matrix element is n ( 4 ° V s ) = 5> m (i) n (v4 0) V4 0) >™ = m fl "W ( 10 .104 ) Here we insert Eq. (10. This probability is the modulus squared of the projection of the Schrodinger state \ips) o n t o \ips )„.107) Jto With this we obtain from Eq.
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^.Em)t' (<Pn\H'(t')\pm) = Wnm(t).111) Different from Eq. (10. (10.109) 10. (10. (10. A further example is provided by adecay. but that otherwise the perturbation is timeindependent. < KO?t . » — < (10. where the momenta of the aparticles form a continuum.112) for a^0. We assume that the perturbation is switched on at some time to. One reason why we begin with these simplifying assumptions is also to obtain reasonably simple expressions.Em)t] _ sin2 at aH {±{EnEm)}H Consider the function St{a) :1 sin2 at ir a2t t 7T for a —• 0.110) With this we obtain from Eq.Em)t} \{tfn\H'\ipv h 2h(En .109) the transition probability Wmn(t) = jp Jo . if in the scattering of a particle off some target its momentum p changes to p'. or if by emission of a photon with continuously variable momentum an atom passes from one state into another.En (10. This is the case. for instance.Jr(En—Em)t _ ^ (ipn\H'\(pm) {En — E„ sm{±(En . Thus we set H'(t) = H'0(t).25) we here have the positive quantity (note t in the denominator) s i n 2 { ^ ( £ n .8 Transitions into the Continuum In many practical applications one is interested in transitions into a continuum. t o = 0.
10.2. Quantum Theory: The General Formalism T h e behaviour of this function or distribution is illustrated in Fig.2 The delta function for £ — oo.Em). t—>oo Hence also (shifting t to the other side) sin2{^(ffn. Hence lim St(a) = 5(a). so t h a t Wnm would be zero. 10.io.114) It follows t h a t for large but finite times t Wmn(t) 2vr ~ — t5(En Em)\(vn\H'\tpm)\2. We assume t h a t the matrix elements for transitions into this infinitesimal element can be taken . Fig. Otherwise.113) {±JEnEm)¥ or n m irtSt En — ^ 2h 1 s i n 2 { ^ ( £ n . though sufficiently small.192 CHAPTER 10. (10. if the energies belonged to the discrete spectrum.— = o (En 2h En 2h5{En . In the case of transitions into the continuum we have to consider the transition probability into the interval dEn at En. and no other state would be nearby.Em)t} r i 7 TTT. .E7m)t} = (10. > 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.ii5) This expression has a well defined value unequal to zero only if the energies En belong to the continuum. we would have En — Em ^ 0.
Fermi [92].t The result makes sense as long as the uncertainty AE in the energy satisfies for finite but large times t the relation . Solution: For test functions <p{x) € S(R) the delta distribution is defined by the functional I &(x)4>(x)dx = <j>{0). _ 2irh AE > > Se. f See E./ t>oc TZ J_00 dx .W .118) where 8e is the separation of states around En which in the case of the continuum goes to zero. 266.* In view of the usefulness of Eq. pp.10. F Jo . for instance from a potential. We evaluate the integral 0(0) = 1. (10.117) in many applications.117) were derived by Pauli in 1928. t We choose a test function which falls off at infinity faster than any inverse power of x: <Kx) = e .113) this is* hm .3: Application of test functions to formula (10. Let p(En)dEn be the number of states in the interval dEn. r = J E ^ W =p(£m)y(</>n#W2n (10117) The formulas (10. Schwabl [246]. If we are concerned with pure scattering. i. '"Recall also the integral J ^ ° dec sin 2 x/x2 = TT. (10.115) and (10. (10. the transition probability into this set of states is J2Wmn(t) = fdEnP(En)Wmn(t) n •* = p(Em)2lTl{iPnlfliPm)l2t. Then. so that initial and final states belong to the continuum.113) Use the method of test functions to verify in the sense of distribution theory the result (10. then Eq.8 Transitions into the Continuum 193 as equal.113). Pauli [227].4). In the case of the functional (10. p. .117) provides the outgoing particle current (see Example 10.116) The transition rate V is the transition probability per unit time. (10. Fermi gave this formula the name "golden rule" . 75. 142. Example 10.r Z x 1 ^ ) = 4>{Q). sin 2 (tx) ax i—( * According to F. if p(En) varies only weakly with En. in the article of W.e.
Gradshteyn and I. iknx . S.194 CHAPTER 10.tan p 2 + a 2 . E x a m p l e 10./ cfc a n p 2 +fc 2 . .ln(l + 4t 2 ) + — t a n _ 1 ( 2 t ) = — tan 4t7T tTV T J< (oo) = 1.117) for H' = V(~x) the differential cross section da _ / n V dH ~~ \2Trh) f dxV(x)e where K = k m — k n . 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). 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. Solution: We set ¥>n(x) : so that {<Pn\H'\<pm) = ^3 / d x e i K ' x V ( x ) . . Ryzhik..( 1 / 4 ) ln(l + 4t 2 ) + t tan _ 1 (2<). dEk = h2kdk//j. 2 T / ^ M. Then p(Ek)dEk Since Ek — h2k2/2[i.^ V v 4 ) = ^ T 3 e * k ° *^ko)e*° * = — v . . p = 1 we obtain I = . .a 2 ' i that for a = b = t .e " 1 * 1 = 2 lim — . § We have r Jo \p\x dx sin ax sin bx  V.<p).. . 77^ + . = 0 . L3/2 ¥>m(x) : L 3 /2 The periodic boundary conditions ipn(xi) = fn(xi ± L/2) imply kiL/2 = ±nj7r. Jin § m = See I. p. K = k 0 . * ^ = . V +(ab)2 a _x 2p6 7 m —^ '.b2 4 p 2 + (a + b)2 2 b _j 2pa V 1 fc Urn . Quantum Theory: The General Formalism with the help of Tables of Integrals. The ingoing particle flux is therefore v / L 3 .„I.. Hence the result is 0(0) as expected.( ^ V V • < ih 2 _ i k n . . 491. with v the velocity of the particles.947. . x .. 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. Then ^ . 1 . . E = h2k2/2fi. M. formula 3.k. and hence . and is obtained as ih. with n .? 2TT / L \3k2dkdUn 1 /V(x). . = ( — ) dk= 2ir (—  2irJ k2dkdfl.4: Differential cross section from the "golden rule" Obtain from the "golden rule" (10. . . L \3k2dkdUn It follows that T. ^o n .
^ We start from the Hamiltonian H = HQ + H'. Thus we set ^(x. Schiff [243]. t) = (xl^) therefore in terms of the eigenfunctions of Ho with timedependent coefficients.122a) (10.122b) S e e also L. Thus we assume that the eigenfunctions (fn = (x</?n) of Ho are known. 193.t)+ / n a{E.10. For a possible application of this result. 10. We expand the wave function V( x . (10./continuum = <San(£)</4°)(x. see the discussion after Eq.x)^(E'. 148. (21. (10. by which we mean without leaving the Schrodinger picture. .t) = ^On(t)^(x.120) In the case of Eq.t).119) where H' = H'(t) is a timedependent perturbation term and Ho\ipn) = En\ipn).121) .9 General TimeDependent Method For a better understanding of the "golden rule" derived above.69). we present another derivation using the usual method of timedependent perturbation theory.x) = S(EE'). (10. pp.119) only the time dependence of HQ can be separated from the state \ip) in the form of the exponential factor of a stationary state.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. Here S stands for summation over the discrete part of the spectrum and integration over the continuum with the orthogonality conditions J and fdx. The problem is to solve the timedependent Schrodinger equation ih^\i(>) = H\r/>). 1 rfx^°)*(x)^)(x) = 8mn.t)^°\E^t)dE (10. (10.^°>(E. specifically in the case of the Coulomb potential.
/ *. .126) + H')<pn^)eiE^h}. (10. .196 CHAPTER 10. since f rJ>(°>(E. dnptMrnW = { ^ _ En) (10. The second term on the left of this equation and the first term on the right cancel out. Observe that these conditions do not contradict Eqs. .x.129) .E') = 5{E .122a) and (10.. .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.127) i n c a g e Qf c o n t i n u o u g £.122b). (10.126) ihak{t)eiEktlh where = San{t)eiEntlhH'kw (10. V>(x. f 6kn in case of discrete tp's. into dip (10. Next we use the orthogonality relations .iJ5 »*/Vn(x).x.120). t) = San(t)<pn(x)eiE^h. H'kn = J d x ^ ( x ) # V n ( x ) = (<pk\H'\<pn).125) where an(t) — dan(t)/dt.x). (10.127) we obtain from Eq. ^0)(E.t) = eiEtlhy(E..i) = e. Then J d^*k(X)[ihSdn(t)cpn(X)eiEt/h = J d^*k{x)[San{t){En + San(t)En<pn{*)eiEntlh] (10.x..„„.t)rJ><®(E'.. Quantum Theory: The General Formalism The discrete and continuous eigenfunctions of HQ are Vi 0) (x.123) We now insert these expressions into Eq. (10. (10. (10.t)dx. We multiply this equation from the left by </?£(x) and integrate over all space.. = e^EE'^h5(E Thus with Eq.„.e.128) .E'). (10. i..
and we set ak°}=5km or ak0) = 8(EkEm).130) H'kn by \H'kn and at the end we allow A to tend to 1. Now we set ak(t) = ak0){t) + \a£\t) + X2a[2\t) + ••• (10. SH'kn(ty^E^l^\t) (10. (10.134) . (10. We obtain ih{ak°\t) + \ak1\t) + \2ak2\t) = SXH'kn(t)e^. we obtain the equations: iha[°\t) iha£\t) iha[s+1\t) = = = ••• . 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. (10.^ [a^(t) E h + } + \a£\t) + A2a£>(t) + • • • ] . Here we assume (our assumption above): At time t = 0 all coefficients ak are zero except one and only one. let's say am / 0.132) These equations can be integrated successively. (10. since it is equivalent to the latter. SH'kn(t)e^~E^ha^(t).e. (10. In order to solve Eq.130) This result should be compared with the Schrodinger equation (10. to perturbation theory in the lowest order. determine the state of the system at time t — 0.9 General TimeDependent Perturbation We can rewrite the equation as Theory 197 iMk{t) = SH'knan(t)e^EkE^h.131) and insert this into Eq. i.130) we develop a perturbation theory. (10.120). they are fixed by the initial conditions. i. We observe that the amplitude ak of a definite eigenfunction tpk has effectively replaced the wave function ip.133) The numbers ak = const. Thus first we have ak0) = const.e. Comparing coefficients of the same powers of A on both sides. In the first place we replace in Eq.10. We restrict ourselves here to the two lowest order equations.130) along with the parameter A in front of H'kn. 0.
i. Integrating the second of the equations (10.137) . (10.136) Comparing Eqs.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. (10. ) ' X V(x')dx'.132).ml2sin2{^frn (En . (10.136) we see that both "amplitudes" are Fourier transforms and look similar.e. Quantum Theory: The General Formalism depending on whether Em belongs to the discrete part of the spectrum of Ho or to its continuum. (10. H'^y^"^^/hd£. (10.Emy in agreement with Eq. whereas ak is a Fourier transform in time (a distinction which is easy to remember!). i. What is the physical significance of Eq. ipm) into the state k.k .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.e.135) and (10. ifm{kl) = f J —oo (10. In this case we obtain from Eq. We shall see that in this approximation the scattering amplitude is effectively the Fourier transform of the potential. an expression of the following type AS?) = " 2 ^ 2 / e j ( k .111).135) We put the additional constant of integration equal to zero. we obtain it*™ = f J—oo SH'kn{t')e^E^'lh5mndt'.198 CHAPTER 10. so that ihakl\t = oo) = 0. ip) in the context of timeindependent perturbation theory.e.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. 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. The amplitude / B o r n is a Fourier transform in spatial coordinates. Equation (10. (10.
i be respectively the momentum.y.m.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. Z<i = Z) y(r) = _ M ^ . The Coulomb interaction on the other hand is of specific significance for its relevance in atomic and nuclear physics. Both cases therefore also serve in many applications as the basic unperturbed problem of a perturbation theory. 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. Angular Momentum In considering the Coulomb potential in a realistic context we have to consider three space dimensions. The electrostatic interaction of the particles is the Coulomb potential (later we take Z\ = 1.2 Separation of Variables.Ti.Chapter 11 The Coulomb Interaction 11. Let r : = (x.2. 11. r=  r . ( 1L1 ) We let pi.z) be the vector separating two particles which are otherwise characterized by indices 1 and 2 with electric charges Z\e and Z^e. such as the electromagnetic field (the quantization in these contexts is often called "second quantization". 199 (1L2) . meaning the inclusion of the creation and annihilation of field quanta). the position coordinate and the mass of particle i with i — 1.
4) P = p i + p 2 .3) for the individual position coordinates we obtain ri=R+ •^ —r. mi + m 2 va\ (11. z) one verifies that j9_ = dxi_ _d_ 8X ~~ OX dXl dx2.U7&. . But now P2 2mo = "ip/pi 2 \mi m p2\2 m2J  = m0/pi2 2 \m\ p22 m2.200 CHAPTER 11. P = pi + p 2 . r = rir2 mi + m 2 (11. so that as expected according to Eq. y.3) and centre of mass momentum P and the relative velocity p/mo by setting r» . (11. /'Pi p = m0 P2 A v. Z) and r = (x. (H4) \mi m._d_ _ _d_ dX dx2 ~ dXl d dx2 nifi^i [ ' etc. Solving Eqs. p are such that as operators in the position space representation or {R. Y. 2Pip2 m\m2 pip2 mi + m 2 . (11. The Coulomb Interaction We introduce centre of mass and relative coordinates R and r by setting R = m i r i + m2r2 • . (11.4). mi + m 2 iTi'2 r2 = R ^ —r. 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 ' .2/ mi+m2 The mass mo is usually described as the reduced mass. Similarly d_ _dx^_d_ dx dx dx\ dx2__d_ _m^_d_ dx dx2 mi dx\ mo_9_ m 2 dx2 etc. r} representation they have the differential operator forms Setting R = (X. so that as expected we obtain the expression for p in Eq. where m 0 = mim 2 .5) The momenta P .
9) ft = ftcm + ^ = centre of mass energy + energy of relative motion. Canonical quantization of the classical theory implies that we pass from the Cartesian variables rj.y. we mean n = rx s x.10) One can convince oneself that the transformation from (ri. r) is a canonical transformation. p = W. 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).2 Separation of Variables. (11. (1114) m0 We observe that the equations of motion in R. pj.10) we set H = Hcm + H.13) * and = M ' 1V1 P = 0.e. [Ri.= ^.j = x. (i. with M := mi + vn^R = ^ P.11.12) (11. The canonical quantization must always be performed in Cartesian coordinates. Angular so that Momentum 201 ^ ^ . etc. .Pj} = ih6ij. r = —.Ri. Ri.z).pj. (11. 2(mi + m<i) 2mo 2m\ 2m.pj. P on the one hand and those in r.il) ""Op' i.+ P.+ ^2.16) Although we write ri.+ ^(r) 2(mi + m 2 ) 2m 0 (H. (11.8) and so W =777 v + 7 T . p on the other are completely separated. The motion of the centre of mass is that of a particle of mass M = m\ + 1712 in uniform motion along a straight line. (11.Ri.2 It follows that p2 2 P2 n2 n2 n2 (H.r2) to (R.pj]=iMij. for which we postulate the commutator relations* [ri. (11. ' d and 'AW OH'ATJ P =  ^ dR <9# <9r ' (ii. Hence Hamilton's equations apply and we obtain. Pj over to operators ?i.Pj). Pj (which in the following we write again ri.15) In accordance with Eq.
r) = *(R)V(r).24) . (11.19) For stationary states with energy E the equation possesses a complete system of eigensolutions * which can be written tt(R.17) (11. In these coordinates we have /\rip 1 d r2 dr \ tdil> dr + r2 sin 9 09 \ S m 1 d_ . V(r) = V(r). (11.e. ip: x z = r sin 9 cos ip. r ) = J E t o t a i*(R.18) In the position space representation the time independent Schrodinger equation is therefore 2M' A fl ) + h2 2mo A r + V(r) * ( R . = rcos9.21a) L*+v{r)] ip(r) = Ei/j(r) (11. (11. 9. df\ 39 J + 1 d2i/j r2 sin2 9 dip2" (11. to that of the Schrodinger wave equation h2 Ar + V(r) tp(r) = Eip(r). —n < tp < ir.21b) + E.Q 2M' + V(r).23) y — r sin 9 sin tp. where Hcm$(R) = and Hip(r) = with ^total — ER (11.20) r2MAR_ $(R) = ER*(R).22) If.202 CHAPTER 11. i. 0 < r < oo. (11. Thus the twoparticle problem is practically reduced to an effective oneparticle problem. it is reasonable to go to spherical coordinates r. 0 < 9 < ir. 2mo (11. as in the case of the Coulomb potential.r). The Coulomb Interaction nc H = 2m.
)In Example 11.3.32) (11.2 Separation of Variables.1 it is shown that pr is hermitian provided lim n/>(r) = 0.11.j.2.3.e. if i. Angular Momentum 203 We can rewrite this in a different form by defining the operator * : =«. = (p2 + ^ V (r + 0).i ' . (11. We have I2 = (r x p) • (r x p).31) . i—>0 (1L25) (11.A fc2! d ( d^ ^ = _tfU2dA r\dr and (compare with Eq. n ( . 2 1 9 / 1 9 .12 — I2. (11.2. (11. that the above expression follows from l:=rxp=ift(rxV).33) i. (11. i( i.28) .j. = —1. the operator —ihd/dr is not hermitian! The operator pr obviously satisfies the relation [r. .k a clockwise permutation of 1. with tijk — + 1 . = 0.27) We can now write (see verification below) p V = h2Ar^ where 2. (l + . k = 1.24)) 2 + M\ d dr2 J = tfL^(r^A\ r2 dr\ dr J d2ip (11.5r = . j . _ h2 \ . otherwise.2._0di})\ x We now demonstrate that 1. k an anticlockwise permutation of 1. (11. is the operator of the relative angular momentum.29) . i.3.pr]=ih.26) However. where (r x p)j = ^eijkrjpk.
(11.15) we have fjPkTjPk = fj(rjPk and TjPkTkPj = = rjPkipjrk rjPj(rkPk + ihSjk) = TjPkPjTk + ~ ihSkk) + ihrjPj.25)) r • p = —ihr • V = rpr + ih and hence 8 = (r • p)(r • p .ih5jk)pk (11.ih)pr and x 2 2 9 9 2 2 Q O (11. Eq. (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.ihr • p ) . * ' S Since r =xz we have (cf. The Coulomb Interaction = &jj'&kk' — Sjk'Skj'.38) + y* + z .42) (11. r— = r • — or or (11.43) r2(p2p2). This means that l2 = so that 9 9 1 (11. (11.39) + ih).27) we obtain r{prr)pr = r(rpr .(r • p ) 2 + 3ih(r • p) .44) .ih) — rpr(rpr With Eq. (11. ih6jkrjpk (11.k Using Eq.35) j.k ~ rjpkrkPj) (11. (11.41) o = r pr.37) .j.36) It follows t h a t (note the part defined as —5) l2 = = ( r 2 p 2 . (11.40) (11.ihr • p r2p2(rp)2 < + ih(rp).204 It follows t h a t 2_^£ijktij'k' i CHAPTER 11.
 h f l d —(r<l>*(r.<p))ril>(r. ¥ p)r 2 drdn s = (p r 0.(r.— (riP(r. 9.2. H 1 2mo pi + K) +V(r).1: Proof that pr is hermitian Show that if Vv lipe H = C2(R3).45) oe Thus in Eq. Pri/Oi V0. 9 € [0.26).e'iP^} .1 Separation of variables We deduce from the expressions of H and l 2 that [H. prip) = = I <j>{r.e. or J .d_ r dr the operator pr is hermitian. i.e. Angular Momentum 205 Comparison of Eqs. 6. 8.11. r := r : lim rip{r) = o i .6l.[ 4>(r. (11.2TT].28) and (11.30) I2 is to be identified with the square of the angular momentum operator. 6. (11. r—>0 il>(r.6. S o l u t i o n : Set d£2s = sin.ip) = (<£.Q ' 2mor2 with the condition (11. <p)r2dfls = .30) yields l2 = I2 = sin sine— BinO^r 09 + d^2 (11. Then {<j>.<p G [0.e. ip 6 X>Pr C 7i. (pr4>.<P) = Etl>(r.e..2 Separation of Variables. <p) d °° . 11.<p)drdna " r r=0 J \ll§r~(r4'(r.e lim rip(r) = 0.[ i J With partial integration with respect to r this becomes (<l>. /'(r. <p))r2drdns % J r or r<t>(r. <p)*rip(r} 0. <p)*Pri>(r. 0.48) .47) Example 11. The Schrodinger equation of the relative motion can therefore be written V1 V 2m.6d0dip.l2} = 0. pr = .46) (11.e. (11.<p))drdns. i.<p)*?(ri. <p)* .! /').d. reR3.Pril>) J / r= dQs rcj>(r.<p) (11.7r]..
lz]=ihlx. zpx .e. (11.lx] = ihly. and this implies in the terminology we used earlier that their determination cannot be "sharp" simultaneously. y.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. xpz] = y\pz. [lyA = 0. Since 1 = r x p .e. Example 11.l2] = 0. In order to determine the eigenvectors of the operators l 2 and lz one defines l± = lx±ily. z we have altogether [lx. We can now proceed as in the case of the harmonic oscillator and determine their eigenfunctions.ly]=ihlz. [ly. i.53) Either of these operators is the hermitian conjugate of the other.49) By cyclic permutation of x. so that [lz.lz (or l .e. (11.206 CHAPTER 11.51) (11.pz]x = ih(xpy . g i = x. for any two of the components there is no common complete system of eigenfunctions. The Coulomb Interaction i. However. H and l 2 possess a common system of basis eigenvectors. zpx] + [zpy.q2 = V. we have [h.52) This means that l . they are incompatible variables. 2 2 2 = 0. (11.ly) are simultaneously "sharp" determinable variables and hence have a common system of basis functions. These operators here play a role analogous to that of the quasiparticle operators . i. ly\ — —tfi(lylx + Ixlyji [hi h\ = 0. 11.2: Verify that [h.xpz] = [ypz. We therefore first search for a complete system of eigenfunctions of l 2 .50) Thus the components of 1 do not commute pairwise.l2} and correspondingly [lx.qj]=iheijkqk S o l u t i o n : This can be verified in analogy to the preceding equations (i = l .lx or l . z]px + py[z. 2 . 3 . ?3 = z). ly] = [yPz ~ zpy. [lz. /z> *xJ = l <i\}ylx + tx'j/Jj [h.ypx) = ihlz. (11.
i + ] = [l2.59) = 9) h(l0 r)\lr). . l\l) is eigenvector of lz with eigenvalue (IQ — l)h. In the present context the operators are called shift operators for reasons which will be seen below.60) Similarly for l+ with the help of Eq. (11. [ig.54) On the other hand we obtain from Eqs. (11.+ ll.61) But lo is by definition the largest eigenvalue of lz.).* Thus we have (1 being a bounded operator) lz\l) = l0h\l) Then from Eq. (n. assuming a finite dimensional representation space.54) l 2 = \{l+l~ + 11+) + ll = l+l.11.62) 'At this stage it is not yet decided whether IQ is an integer or not! But we know that IQ is real. A^ in the case of the linear harmonic oscillator.r) := (Z_)rZ) so that lz\l .54): lzl\l) = {llz .57) i.51).r) (11 = {l02)h{l)2\l) (11. (11.i+] = m+.3 Representation of Rotation Group 207 A.h~lz(1156) (11. Similarly we have lg{l)2\l) and more generally lz(l.l)hl\l). (11.55) One now defines certain ketvectors as eigenvectors of lz which span the space "K\ of the appropriate states.52): 0 = [ l 2 . (11. Hence we must have Z+I0=0.lh)\l) = {l0 .i] = m. With the help of Eq.e. in particular one defines \l) € CKj as eigenvector of lz with the largest eigenvalue lo. [i+M = 2%iz.54): izi+\i) = (i+iz + i+h)\i) = (lo + i)ta+\i). (11. Therefore (lo + l)h cannot be an eigenvalue of lz. (11.Z_] = [ l 2 . (11. (11.58) {l0 : max. since lz is hermitian. (11. We write or define: / . y .Y\l) = (lQr)h(lY\l). (11.49) we obtain the following relations: [iz. and with Eq.
l\l .55)) we have 12Z_Z> = l. Since l 2 commutes Hence Z) is eigenvector of l 2 with eigenvalue with l_. (11..65) We assumed that "Ki is finite dimensional so that the sequence of eigenvectors of lz.From Eq. i..+ 11+) + l2z\l) = i+i.56) that l2/n) = (11 6) (l+l+l2lzh)\ln) (lll.54) (n = 57) (11.n). The Coulomb Interaction (l+l. [12.Z_] = 0. (11.56) CHAPTER 11.e.e.66) It then follows from the second of relations (11. (11.64) i.e. (cf.e. \l).+ i2z (11+ + 2izh) + 1 \ \i) = (izh + ti)\i) i0(i0 + i)h2\i). IQ(IQ 1 0 (11.60) and (11. (11. i.65) for r = n: l2\l n) = l2(l)n\l) = lo(lo + l)h2\l .n)}h2\ln). l\l) is also eigenvector of l 2 with eigenvalue Zo^o+l)^ 2 .\lr) terminates at some number (let us say) r = n (n a positive integer).64) we obtain on multiplication by Z_: lVl\l) and more generally l2(f)PJ> (11.67) i (11 60) = But according to Eqs.63) + l ) ^ 2 .68) . (11. Eq.208 Thus for l 2 we obtain: V\l) ' = 11.h)\ln) {(lQnf(lQ.n) (1 = 60) {l)n+1\l) = 0.. (11.\ll).l2\l) = l0(l0 + l)h2l\l).60) = Vli\l) = l0(lQ + l)h2li\l) i0(io + l0(l0 + i)h2(i.y\i) l)h2\lr). (11. i. (11..
m) = = = = y/lo(lo + 1) — m(m +l)\l.m) = = mh.. A. lz) are then: (l..Z 0 ).m) : Mo).e. \l — n) are eigenvectors of lz with eigenvalues loh. As in the case of the harmonic oscillator we introduce normalized basis vectors. 17. We write these* for integral values of IQ: \l. \l — n) into one other. (11.l)h \/(lo + m + l)(l0 ... Brink and G. we skip this here.m) with — lo<m<lo2 (11. (11.m) = IQ(IO + l)h2\l. The dimension of the representation space is therefore n + 1 = 21Q + 1. (11.{lo . 24. II.m y/lo(lo + 1) — (m — l)m\l. (11.m).. R.11.m). y (lo .g.1).63) and (11. In the following we consider primarily the case of integral values of IQ. p. p.3 Representation of Rotation Group so that with Eq. y/(l0±m + l)(l0^m)h. According to Eqs. *D. We therefore write the 2/o + 1 orthonormal vectors: \l. . lo = l0n. Satchler [38].n)} = (l0 . The basis vectors \l). i.67) we obtain (observe IQ(IQ 209 + 1) = —lo(—lo — 1)): *o(*o + 1) = Wo . § See e..m) (l. (IQ — l)h. The phases are chosen such that§ l+\l.n){l0 ..n .m)\l.m / + l)h .65) these are eigenvectors of l with eigenvalues Zo(^o + l)h2...l_.m\lz\l. Vol..m + 1)(Z0 + m)\l.. (H71) These expressions should be compared with the corresponding ones in the case of the harmonic oscillator.70) they satisfy the eigenvalue equations (with m = l0Jo l.  U o ..n) 2 .m) — mh\l..l0): l2Z. The normalization factor in Eq..m . M. Messiah [195].l)h.1)..70) According to Eq.69) Prom this it follows that IQ must be either half integral or integral. lz\l.. The nonvanishing matrix elements of 1 = (l+.m) = mh\l. —loh.. lQ = .2.m).71) can be established in analogy to the method used in the case of the harmonic oscillator. The operators l+.l.lz transform the vectors \l).m±l\l±\l. m + l)h. where lz\l.m) l\l. (11.
<p) = i0{io + lzY£(0.. /o.. CHAPTER 11. .72) In view of the spherical symmetry it is reasonable to use spherical polar coordinates r.210 Then.74) We choose the eigenfunctions of l 2 and lz as basis vectors of the ("irreducible") representation. We now perform the analogous procedure for the angular momentum or. ip) = e(9)$((p). for which the eigenfunctions \ip) are given by l2 l0(l0 + i)h2\ip). as remarked earlier. as above. (11.77) $((p) = m2<b(y).e..76) d2 (11.With the ansatz Y(9.1. mhY£(9. where IQ = 0. i. $ oc eim*.l0)=0. .75) the variables contained in the expression (11. (11. In these coordinates we obtain: h l± so that h2 = xpv .73) 1 d d sm0 06 \Sm9dd) + (11. ih d_ dip' lx ± ily = ±he^ d I ^ ± zcotfl^ .. We . (11. . 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.74) can be separated: 1 d sin 9 36 2 1 sm6— 89 j — sin 0(9) = h(l0 + 1)0(9).<p) = i)n2Y^{e^). —lo + 1 .yp. and m = —IQ.2. 11. d2 sin2 9 dp2 _' (11. with i2Y£{e. The Coulomb Interaction Z+Mo> = 0.4 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. l_\l. as this is sometimes described. ip. for the simple "rotor". . 9.<p).
11.79) .78) and the completeness relation oo I ' YrVMYTVrf) = °^~' ± J2 Y7n*(f}. The functions Y™ satisfy the orthonormality condition dfilimT' = / Jo < W sm8d9Yr(9. (11.V>)Yir'(8.<p) = (iy 47r(Z + m)! In particular for m = 0: 1/2 PI11 (cos 6)eimi?.0) is real and positive. Uniqueness implies immediately that m has to be integral. and that their phases are such that Yt0(0.76) and (11.r o ) ! YT{e.80) y? Furthermore Y°l = V^fcosO).n>).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. m > 0.(5cos 03cos0). We mention in passing that in spherical coordinates d lz = xpy . (11.=o m=l sin 6 The connection of these hyperspherical functions with the associated Legendre functions P™.\ = W . This is the case if one demands that l 2 and lz possess a common complete.^ = *(n . (11.<p) Jo = 6mm'5w.82) . so that im(ip±2ir) The functions Ylm(8. These functions are defined in such a way that they are normalized to unity on the unit sphere. y2° = \ (3cos 2 01) 2 V V 4vr V 16?r 3 W—. (11. is given by (2Z + l ) ( Z . / 3 / 5 J — casO.* ? ^ .t»'\Y7n((l'.i». (11. orthonormalized system of eigenfunctions.77) and o these are determinable by the requirement of uniqueness of the original wave function and its square integrability.4 Angular Representation of Angular Momentum 211 observe that Z and m define the eigenvalues of Eqs.(p) are called spherical harmonics.ypx = tfrK(11.
e.d. The associated Legendre functions or spherical functions PJn(x) satisfy the differential equation (1 .. i.86) y = 0.90) . the designation s..g.2xy' + 1(1 + 1) 1x The number m must be an integer already for reasons of uniqueness of the wave function.h. i..m = 0. oo (11. in the replacement ip —> if + 2K..85) (11. If we did not know yet that I = IQ = an integer > 0. exp(imip).1(1 + l)]xK+i = 0. (11.m(m + l ) ] i f = 0.x2)y" .. (1 .x2)P[' . 22 (11.2xy' + 1(1 + \)y = 0. we would now have to search for those values of I for which the solutions are square integrable.2. Setting y = (lx2)m>2Pr(x) m . is also in use. (11.x2)Pf .e. It is instructive to pursue the appropriate arguments.2xP[ + 1(1 + l)Pi = 0. i.2(m + l ) x i f + [1(1 + 1) . of Ytm.p.83) Their connection with the Legendre functions Pi (x) is given by the relation pr{x) with the differential equation =£^ Pi{x) (iL84) (1 .l)xK+i~2 i . respectively.88) we obtain for every value of x ^2 ai(K + 0( K + * ..87) we obtain the equation (1 .e.Y^O»[(K + i)(/e + i ~ 1) + 2(re + i) . i.89) Inserting this expansion into Eq. f. Instead of the values I = 0. (11.x2)y" . The Coulomb Interaction The number I is referred to as the "orbital quantum number" and m as "magnetic quantum numbed.88) y = ^aiXi+K. For the solution we use the ansatz of a power series. i=0 (re lowest power > 0).e. (11.1.212 CHAPTER 11. (11. Consider the equation for Pi(x).
M.95) Vl = rxiThe hermiticity condition for pr then requires that (cf.11.94) from the left by r) h2 d2 h2 2 + H! + \)IL. Eq.93) where Xl(r) is the solution of the radial equation (cf. . Margenau and G.E Xl(r) =0 (11. ai(re + !)« = 0. when the series terminates after a finite number of terms. (11.92) This is a recurrence relation. p.1 we obtain by comparing coefficients QOK{K 213 — 1) = 0.26)) yi(r = 0) = 0.2 + 2mo dr 2mor V{T)E Viir) = 0 (11. (11.96) + V(r) . We have therefore (multiplying Eq. Comparing the coefficients of xK+3 we obtain aj+2(K + j + 2){K + j + 1) = OJ[(K + J)(K + j + 1) .*" 11.29)) 2 fe2l 92 Pr = & r 7Tororz We set (11. (11. Eq. these are of little interest to us here.46)) p.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. l(l + l)h2 _2mo + 2m. but also those for a different series for a. We see that this happens precisely when I is a positive or negative integer or zero.94) ^For further details see H.1.5 The Radial Schrodinger Equation for Hydrogenlike Atoms For i = 0. Murphy [190]. 93. they have to be polynomials (recall the orthogonalization procedure of E.1(1 + 1)].78) to be square integrable. . For the solution in accordance with (11.Qr2 with (cf. l 2 and lz are the solutions of the Schrodinger equation of the form (11. first line of Eq. Schmidt). (11. This is precisely the case. (11.91) from which we deduce that K = 0. > 1. For \x\ < 1 we can obtain from this the coefficients of a convergent series. However. (11.
96) can be shown to possess a regular solution there which behaves like rl+1{\ + 0(r)) in approaching r = 0. This applies to the Coulomb potential but also to Coulomblike potentials like screened Coulomb potentials or the Yukawa potential.1.101) 2 r ^ We set  6 = ( _ e ) 1 / a = ( = (11. Equating the coefficient of 1/r2 to zero.e. Z\ = Z2 = 1 in Eq. (11. which behaves like 1/r1 in approaching r = 0. Jo Jo We thus have a problem similar to that of a particle of mass m in the domain (0.e.214 with CHAPTER 11.99) e ip(r) = Eip(r).98) 2mo e2 _ 1(1 + 1) yi = o e+ 2 y'l + h r r2 2m0E e = (11. and others. (11. oo) of a onedimensional space.96). s = 1 + 1. He + . 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. In order to verify this behaviour we set <VflV>r> yi(r) = rs(l + air + a2r2 \ ). we obtain the socalled "indicial equation!'' s(s . and substitute this into Eq.1)+1(1+ 1) = 0. r m0 = meMp me + Mv (11. The Coulomb Interaction (11. (11. In the first place we investigate the behaviour of yi in the neighbourhood of the origin. nor the condition of normalizability for l ^ 0.100) with (11. The equation also has an irregular solution. L i + + . i. We now consider the hydrogen atom with V(r) — —e2/r (i. for instance. We assume that V(r) has at r = 0 at most a singularity of the form of 1/r.102) . Obviously the irregular solution has to be rejected. since it satisfies neither the condition yi(0) = 0. In these cases the behaviour of yi near the origin is dominated by the centrifugal term and thus Eq. an analogous consideration applies to hydrogenlike atoms like.97) / r2dr\Xi(r)\2 = / \yi(r)\2dr.1)).
Magnus and F.6 The Discrete Spectrum of the Coulomb Potential 215 For E > 0 the solutions oscillate in the region r — oo.103) °y moe2 ( = 0. 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. . We set x — We set as Bohr radius* a = and 1 v = rea rriQe ft V (47Fe 2KT. It remains to determine the function vi(x) for a complete determination of the solution.105) 11. (11. We set therefore Vi X J. (11.88. (11.104) W i t h these substitutions we can rewrite Eq.11. for E < 0 they > decrease exponentially.105) t h a t for x —• oo there is a solution behaving like exp(—x/2). We can see from Eq.e. Oberhettinger [181].a * = 0.6 11. (11. (11. *The factor 47reo appears in SI units. behaves there like xl+1) extended to an exponentially decreasing branch at infinity.e. W. hypergeometric (11. p. T This equation is known as Whittaker's equation.100) as^ d2 dx2 1(1 + 1) x v___\ 4 2/1 = 0. Inserting the ansatz into Eq.108) This equation has the form of a confluent x$" + (6 . i. (11. (11.105) we obtain the equation for V[(x).+if>x/2 vi(x).x ) $ ' .e.g. d2 x—j 1 + (2l + 2x) ax d j dx (/ + 1 .v) vi(x) = 0. see e.107) equation.6.106) W i t h this ansatz we separate from the solution the behaviour around the origin as well as t h a t at infinity.529 x 1 0 " 1 0 meters) 1 V/2 e2f hc\ m0c^1/2 2E 2moE J (11.
z) = ^f^.216 CHAPTER 11.112) The gamma function is a nonvanishing analytic. are very strict in reserving the factorial exclusively for positive integers and the gamma function for all other cases. and for both cases.e.106). more precisely.1/2)!. . since the exponential function (generated by the infinite series) would destroy the asymptotic behaviour of the wave function which we separated off with Eq. with many of these around in some calculation.111) Important properties of the function T(z) are: (1) zF(z) = T(z + l). (11. and hence the *Some writers. The function T(z) is defined by the integral /co T(z) := / eHzldt. (11. (z)!(z . it is convenient to use only the factorial notation.(0_i)!(6 + n l)!n!' (11.z(zl)\ = z\. particularly pure mathematicians. 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. The Coulomb Interaction The regular solution of this equation is the confluent hypergeometric series (also called Kummer series) 1 . (2) r ( z ) r ( l . 02zl / i\ (3) r(2z) = — ^ r ( z ) r ( z + . i. meromorphic function.! ) ! = imlFI' c a l led inversion or reflection formula. + ' ' ~ 6 1 ! + 6 ( 6 + l ) 2! + Z. (11. called dupZication formula. (5)r(n + l ) = n ! .J. the factorial for an arbitrary argument. (4) r ( i ) = V5F. However. 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.109) In our case we can write the series The expression* r(x + l) = x\ is the gamma function. ^ ( 2 ^ ) ! = 2 2z z!(z .
i.113) This is a quantization condition. i.110) as follows: T(l + 1 + p . and so to the number of different values of m. • • • • <»•»*> It should be noted that the magnetic quantum number m does not appear in this expression. This number is equal to the number of finite zeros of the radial part of the wave function (excepting the origin).I) r(vl p)sm{n(i/ I . . The quantum number n' is called radial quantum number.101) to (11.1. the orbital or azimuthal quantum number I can assume the values 0..e. if the series terminates for certain values of u. . for z. 2 . = n = Z + l + n'. in our case: a = 0. .2.1) + n = n2.114c) (Recall that for an arithmetic series a + (a + d) + • • • + (a + (n — l)d) = na + n{n — l)<i/2.e.V))T(v . One defines as the principal quantum number the number n.. This is achieved.. which is thus seen to be a consequence of demanding the square integrability of the wave function. must be a polynomial. (11.a " \ 2moE we obtain the energy spectrum of the discrete states with angular momentum I ie t m0e2 / 1 N1/2 * =  ( s ) ' ^ . (11.6 The Discrete Spectrum of the Coulomb Potential 217 latter would not be square integrable. . which are also described as its nodes. Eqs. every n.11.. .v) T(l + lv) _ ~ K sin{7rQ/ . i. n' = 0 .e. 1=0 (11.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.104)) 1 n =v = K. With (cf. 1 .1) + n = n(n . Thus the infinite series must break off somewhere.» • » • » .) 'Observe that with property (2) of the gamma function we can reexpress the ratio of gamma functions in Eq. 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 . d = 1. (11. For every value En.
The following spectroscopic description of states is customary: nl —> ns. . from n. .l = 2 to n = 2.. and from n.. (c) The nucleus was treated as pointlike and not as something with structure. ne.4.1 = 1 (called diffuse series.§ One also writes n2l+1l.1 = 0 (called sharp series.\ I = 0 or 1. 2 . hence d). stand for I = where n denotes the principal quantum number and s. ' O n e describes as Lyman series the final state with n = 1.. where the superscript 21 + 1 at the top left of I gives the degree of degeneracy. . relativistic corrections are of the order of v2/c2 ~ 0(En/moc2) as we illustrate in Example 11. 11. from an initial state n > 2. hence p).'. One describes as Balmer series the final state n = 2. (b) The spin of the electron was not taken into account. I = 0..1. hence s). nd.d.. 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. . np..1 Hydrogen atom energy levels. 0. . nf . I = 0 to n = 2. 11..1 = 1 to n = 2. The spectrum of the hydrogen atom thus has the form shown schematically in Fig. / = 1.'. The above treatment of the hydrogen atom has obvious shortcomings: (a) The hydrogen atom was treated nonrelativistically.p. from an initial state n > 2.1.. and hence the fine structure of the energy levels is excluded. respectively.1 = 0 (called principal series.218 CHAPTER 11.
844. • (11. . We mention here three different definitions used in the literature. . Apart from factors. so that one always has to check the definition.414(3). Vol. l . W.2 (11.r f .115) ' =0 (ri . 11.116) 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. 3 .2l { + 2. Gradshteyn and I. 2 . Thus Messiah [195] defines the original or stem Laguerre polynomial as : >Lr(z) = *L°r(z) = e'^(e'zr). however. The mass of the hydrogen atom is given by massif = Mp + m e + E < Mp + me.x). a + 2:*) = t v . • fc. (11.110) is usually reexpressed in terms of orthogonal polynomials known as associated Laguerre polynomials. in fact. and I. M. • ( . (11. formula 7. Messiah [195]. p. see. the polynomial contained in y\ of Eq. With the help of Eq. several different definitions are in use. Appendix B. S. it is given by^ *Lkr(z) = ^ ^ f 1 *ir.r = 0 .p)\(2l + 1 + p)\ The spectrum (11.= t .114b) yields the binding energy of the states.z) = (r + k)\ Lkr(z). Oberhettinger [181].6.115) 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.117) The function *Lk(z) is a polynomial of degree r.11. "1'f„+11" P] „.118a) where Lk. Ryzhik [122].6 The Discrete Spectrum of the Coulomb Potential The associated radial wave function is yi(x) 219 = xl+1ex/2<f>{n'. Unfortunately the definition mostly used in mathematical literature differs. 84. Unfortunately. . .115) is effectively the associated Laguerre polynomial normally written Lk(z). (11. p. Each can be recognized by reference to the generating function. k + l. I.2. Magnus and F. for instance.{z) is the associated Laguerre polynomial as normally defined and mostly written Lr (z).1. We distinguish between different definitions in use. (u. . For clarity and later use we denote the function used there by its normal form L^.
Vol.d z—^z + {k + l~z)—+r Lkr{z) = 0.z . r r Z r=0 v tr k k (11. is defined by^ dk (—'\\k(r^\2 (11. 163. ^ F o r this definition and the following equations we refer to I. Example 11. N.. L\(z) = 2z. For our purposes here it is convenient to refer to yet another definition of the associated Laguerre polynomials. There at some points n + 1 is misprinted n + 1. L1(z) = l . pp. and Ll(z) = l. Messiah [195].220 CHAPTER 11.120) ' r=0 It will be seen that the orthonormality condition is" z ezJc*Tk.1.„\*Kk. The Coulomb Interaction The differential equation of the associated Laguerre polynomials is (same for *Lk{z)) d2 .122) "See A. (11.119) dz dz In particular one has (we cite these for later comparison) LQ(Z) = 1.3 we show in a typical calculation how this normalization condition can be obtained with the help of the generating function. Sneddon [255]. Ll(z) = 33z + ±z2. It is said that this is the definition preferred in applied mathematics.** The associated Laguerre polynomial here.2.3): e2t/(lt) (1 . 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. **I.„M„ z"*L. 162 . which we write ^Lk. Sneddon [255].. L2(z) = l2z + z2. I. .(z)*L TJz)dz 3 = KP + kV}Opqit Jo (11.164.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. The reader is there warned that '•'•care must be taken in reading the literature to ensure that the particular convention being followed is understood'. p. Appendix B. .121) In Example 11. . N.
Solution: We consider the following equation of the form of Eq.% ' ( * ) + by{t) = 0 with ansatz y{t) = I s(p)e~ptdp.11.4 + 2z.118a) and (11.122) we see that tLkr+k(z) = (l)k*Lkr(z) = (l)k(k + r)\Lkr(z). (11. '"''Comparison with Eq.4z + . (11.124a) (i*) pi The orthonormality condition is (cf.120)) it)k& zt/{lt) ~*P p=k fc (11. ^L\(z) = . and partial integration. Jfpd>2s{p))eptdp. ^L2(z) = 2 .124b) Comparing Eqs.3: Generating function of Laguerre polynomials With a Laplace transform ansatz for the solution of a second order differential equation like Eq..123) For the purpose of clarity and comparison with the polynomials of the other definition it may help to see that here in particular % ( « ) = !. (11. The generating function of the associated Laguerre polynomials so defined is (observe (—t)k as compared with (11.124c) Example 11. . *L\(z) = .119) derive the generating function of associated Laguerre polynomials (11. (11... Example 11. N d rk 0.119): ty"(t) + (a + 1 . Then.8): (11. with y'{i) = — / ps(p)e~ptdp ty'(t) = t f ps(p)eptdpp2s(p)ept s(p)e Pt J ^(ps(p))eptdp.z .121). (11.1 . and *L\(z) = .1 8 + 18z 3z2. The limits will be determined later.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 .. (11.107) implies k = 11 + 1 and r = n + I. and obtain with this the normalization condition (11.120). i L1(z) = l .
/ ( l — <?) —— = G(t).] from the above equation: C(p) := p(p + l)s(p)ept = 0.p + 1 = 1/(1 .q) and P~ t ( .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. p and one has (as polynomial of degree 6. Case (1) with p —• oo implies y(t) ~ / dppa~1e~pt ~ t~a. Then the preceding two equations (inserted into the original differential equation) leave us with I eptdp d {p(p + l)s(p)} . 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. Thus in the case of Eq.. s bK> (lq)°+i (11.p ) o + i 7 t = 0 .119) with a and b as there. 2TT_ Jq=0qb+l(lq)a+l 1 . Hence the expression in square brackets vanishes. j p = (P 0 "V?. Then dp = dq/(lq)2. and (4) around p = 0 if b > 0 and integral. (3) around p = — 1 if a + b is a negative integer. agreeing with Eq.^ . since for n > b : (b — n)\ —• oo.p ) « + i q c + i ( i . .124d) In order to obtain the orthogonality and normalization of the associated Laguerre polynomials.. Case (2) implies the exponentially increasing solution ~ e + t .118b)) Ll{t) = = coefficient of pb in ^ (p + l ) a + t > n=o £ ( ~Pf" "" coefficient of pb~n in (p + ! ) « + ' ^ = /g <*>" (a + 6)! n! —(. case (3) similarly.a = 21 + 2. p° Possible paths satisfying this condition are: (1) If b < 0 from 0 to oo. (11. t > 0. (2) from —1 to oo. the only possibility is the last.222 CHAPTER 11. It follows that 1 d i I _i. n! (6 — n)!(a + n)! An alternative integral representation is obtained by setting p = q/(l — q) or q = p/(p + 1). with requirement of a finite solution. (11.( a + l ) p s ( p ) + 6s(p) dp 0. The Coulomb Interaction since d(p2s)/dp = ps + pd(ps)/dp. which removes [. this is the rejected irregular solution. We demand (to be considered later for integration contours) the following condition.) 2 J J p 6 + i ( l . we insert this contour integral representation of the polynomial in the following integral and obtain __?(_) = P da <* ." b + 1 e~ptdp = coefficient of pb in g(p) := (p + \)a+be~pt. 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.
m) is 3 2 &nlm = a~ / NnlFnl (—^Y^O. (11. (11.127) . 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. 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. where Fnl(x) = yi(x) (11. /•OO / Lg(t)£?(t)taefctt: (a + c ) ! . so that I = (c + a)! 1 f .125a) and the factor a~ 3 ' 2 with a = Bohr radius. is introduced for dimensional reasons. Eq. . Clearly only the second form is sensible.(x).—i.3 T h e eigenfunctions According to the previous discussion we obtain n2 orthogonal eigenfunctions in association with the eigenvalue En. I.111).11. Jo (11. 11. f a p)(i 1) r + 1 l (1P9) J The coefficient here required is obtained from the expansion as .?).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.c (al)\ d(col)! c(c + a)\ c!a! with use of the inversion formula (11.—: <f h_nJ_[ dp = (a + c)\ T^ if c = o. and obtain r taet[P/(iP)+q/(ig)+i]dt Jt=0 With this the integral I becomes = .6.112).
11. 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 + /„.2. This behaviour is indicated in Fig.v (6n2  6nk + A:2 + 6n .n+l na l2 ° ^ 2 na n+L+lV) (2n + 2/ . + . For the ground state (n = 1.129) 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 .130) We see that the mean value of r is the larger the larger n is.131) . thus on the average the electron is the farther away from the proton or nucleus the larger n is. a (r)is = ~a.(l) n+l.. (11. 2 (11. The Coulomb Interaction One finds in particular (see also Example 11.7.3fc + 2).8):* OD  <n!>3 (nk)\ (nfc) (n\)3 i£ n (3) = _ .3(2Z + 1) + 2} 2 {2(n + 0 .(2Z + 1) + 1} = ^[Zn21(1 + 1)]. = 0) we have ) r /u =.6(n + Z)[(2Z + ! ) . and for the explicit derivation of the second case below Example 11.2Z .224 CHAPTER 11.128) These expressions permit us to obtain the mean or expectation values (1/r) and (r).1 + 1) (11. (11.! ] + (21 + l ) 2 .
In the case in which I assumes its maximal value n — 1.6 The Discrete Spectrum of the Coulomb Potential r. 11. In this case (r) = = For (r 2 ) one obtains < r 2 ) = n 2 ( n + ^ ) ( n + l)a 2 .2 I 3/2 Kl r 225 \2 1 R„.e.2 1 A r = V (^2) .134) *These are taken from I. Problem 5.2 Hydrogen atom wave functions.J n 1s 2s 3s *r Fig.j . S.20. This corresponds to E ~ — e2/2n2a like classically with circular radius n 2 o.133) for large r. Sneddon [255]. with n +1 replaced by n and 21 + 1 by k. pp. we have and tp becomes particularly simple.132) (11.11. = l a nl n N'nll rFnl.( n . i.(r) = x W 2 n + 1 =. v ^ V2n + 1 (11. / ! J — .. \f{r ) ~ n a. In the Kepler problem of a particle of mass mo in Classical Mechanics with the Newton . 173/174.^ We obtain therefore for the quadratic deviation (r) .l ) n ] =  [ 2 n 2 + n] an[ n + (11. or (r ) ~ n a 2 2 ^ [ 3 n 2 .
= ngh.6. i = 0. In order to obtain an impression of relativistic corrections in a simple context. Solution: We have H = yjm%c4 + pIn cylindrical coordinates. Thus we have to evaluate the following gravitational potential written V(r) = —mofi/r.135) This expression becomes very small for large values of n.e.4: The relativistic eigenenergy using cylindrical coordinates Using the formula for the relativistic total energy E of a particle of mass mo and charge e moving in the Coulomb potential V = ~Ze/r (with E numerically equal to the corresponding Hamiltonian H for a conservative system). pJ = —— and j> Pidqi = nth.4 the relativistic equation in cylindrical coordinates.226 CHAPTER 11. we have pg = const. for these values of I it is not distributed with uniform probability over the spherical surface. (c) The nucleus was treated as pointlike and not as something with structure. However. relativistic corrections are of the order of v2/c2 ~ 0(En/moc2).3. i.e. . determine the energy E by evaluating the BohrSommerfeldWilson integral of "old quantum theory". the electron remains practically localized within a spherical surface of radius (r). p2 = p2 + — pg • eV. The above treatment of the hydrogen atom has obvious shortcomings: (a) The hydrogen atom was treated nonrelativistically. The Coulomb Interaction or TT = ^ = (11.— > where ampler is the length of the semimajor axis of the elliptic orbit. we consider in Example 11. Example 11. identifying yj (r2) with aKeplen w e see that T K ep l e r = ^(r2)^ = ^(n2af/2 = *^a3'*n*. the states Ytl are predominantly concentrated in the xyplane). r. 6.r. (b) The spin of the electron was not taken into account and hence the spin fine structure of the energy levels is excluded. „ _ 'Kepler — the Kepler period is given by o Z7ra 3/2 Kepler . \/v vT 7 v^ See also Example 14. and separating this in cylindrical coordinates r. but rather like classically in a plane (i. Since 6 is a cyclic coordinate. This means. Thus. in which periods are obtained from the BohrSommerfeldWilson quantization condition.
136a) (11.2. ( v c + i ) . z=^rj).d r Kmlc2 + T U^\±AnlK2^Ur 2 2 \ c / r * + * ! + $*—».136b) £ = r ( l + cos#) = r + z.11.9. which correspond to a new choice of polar axis. <p = <p. where r = ^/x1 + y2 + z2. y — 0: V ^ c o s tp. ip are the spherical coordinates. (and subtract moc2 to We then obtain with a = e2/hc ~ 1/137. Thus E2 2EeV(r) e2V2(r) nrh = a> prdqr = ilc2 . (p). which are related to Cartesian coordinates in the following way. (11.. where ip = 0. In the case of degeneracy with respect to the magnetic quantum number m (as in the present case) one can find linear combinations of hyperspherical functions Yj^(9. The deeper reason for this are symmetry properties of the problem. 2 2 Therefore the energy is no longer a function of n = nr + ng. p.V .4 Hydrogenlike a t o m s in parabolic c o o r d i n a t e s The Schrodinger equation of a hydrogenlike atom can also be separated in parabolic coordinates £.* See L. .T). as also indicated in Fig.(p . r\ = r ( l — cos#) = r — z. Schiff [243]. but depends very slightly on ng alone which gives rise to the "fine structure" of hydrogen lines. Whenever there is degeneracy.6 The Discrete Spectrum of the Coulomb Potential 227 integral which we achieve with the "method of poles at infinity" explained in Example 16. the "fine structure obtain the ordinary nonrelativistic energy) Z2ei c2h2[nr + ne^jl 1/2 E — mac + 1 Z2ei/c2h2n2] 1/2 VTIQC 1+ •+ ng^la Z /r. 11. it is possible to separate the Schrodinger equation in other coordinates. 11.3.6. 86. It follows that (with h = 2nK) Z2e c h?ni 2 2ixnrfr = — 27r nehtjl  Ze2E >^m2c2E2/c constant'. where the constants are identified by comparison. y=y/%qsanp.
11.22) becomes — as we show in Example 11.3 — h2 ( 4 r d (rd^\ d_f df\~ + 1 d2xl>\ £?7 dip2 2Z1Z2e< Z+V •ip = Eip.3 Parabolas £ = const. 2 . (11. a separation constant. = 0. . . In terms of the coordinates (11. and r\ = const. . The Coulomb Interaction Fig. We rewrite the first of these equations with the help of the relation M (w ^  m + e) .228 CHAPTER 11. Then 1 d2$ izimip $ oc e $cV = —m where the separation constant m2 must be such t h a t m = 0 .137) One now sets the total wave function ip = £(£)!N"(r7)$(</?) and divides the equation by ip. d_ dirt) + TUQE fm0Z±Z2e2 2h2 V ^2 A V m (ri'N) Arj1 2n Since these equations are alike. so t h a t the variables can be separated. one obtains the following two equations: £ ) + TJIQE A m2n  « ' m = o. with zaxis as axis of rotation. there is really only one basic equation to solve here.136b) the Schrodinger equation of relative motion (11. for uniqueness of the wave function. Multiplying the remaining equation by (£ + r /)/4 and setting the £part equal to —A. 1 .
11. E x a m p l e 11. . .e. T h e n the first of the two equations becomes 1 £) n' m2 — 1 4 p + 4p" ( P 1 / 2 £ ) . as well as the Laplacian A = V and the Schrodinger equation.± ( m + 1) = 0 . In the spherical polar case we had (cf. Eq. 2 . 2 . Adding now / . " . i. 1 . . n 2 = n" . We leave the calculation of the degree of degeneracy to Example 11. m0Z1Z2e2 ^ 2 = n (say). .6 The Discrete Spectrum Then of the Coulomb Potential 229 A 22 (£ 1 / 2 £) + moE 2h2 A m2 — 1 +£  4£ 2 (cT1/2£) = 0. T h e n by comparison with the spherical polar separation the answer is as in Eqs. . we set (since E is negative for the discrete spectrum we consider) ITIQE =*"• H J?P ^( <9p: 1 / 2 i m0ZiZ2e2 A = n". 1 . .' / 2 p m / 2 L 2 Z + i = m ( / 9 ) ) Analogously we have N oc eCT/2<7m/2L™ (a). .i n +n = n i + n 2 + m + l = we obtain the energy eigenvalues given by E = n i = n' \(m + 1) = 0 .114b).105)) Hence we write I = (m — l ) / 2 . from Cartesian to parabolic coordinates. ph h2p2 _ h2 m2Z2Z2eA hn 4 2 _ ~~ m0Z2Z2e4 2ft2 n2 ~ 2m 0 ~ 2m 0 This agrees with our earlier formulas like (11.o. To obtain a closer analogy with the separation in the spherical polar case. . (11.106) and (11.5: The Schrodinger equation in parabolic coordinates Perform the transformation of the metric. . (11. . ds2 = dx2 + dy2 + dz2. .6.125b) (dividing by p 1 / 2 ) £ oc e . p = /?£ and cr = /^r?.
Eq.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.2d? dr)}. Similarly we proceed with dx and dy.[df + dr]2 . n 2 (1UJ)2 The coefficient of a .?e 9£ J 4 t d \ g^gv 9 \ dri \ gv dr] J t d \ g^gr.l)a + (n . are elements of length. 0 there are two. (n _ !) n = n2 §See e.l ) ( n .2)d/2. Beware! For m = 0 there is only one mstate. Thus we want to obtain the coefficient of wn~1 in £2 . i. and so dz = (d£ — drj)/2. . as t h e two signs of t h e exponential indicate.e. 2 + ^dV .230 CHAPTER 11. \— < / — cos ipdr] — \/£r] sin ipdtp.g<p 9 \ ..2) and summing over m. </. last chapter. and one obtains 2 = _ 1 4{ £ ) * (t + 7 )\ j. Magnus and F. (11.e. We use the method of selector variables. for instance.136a). In the case m ^ O w e have n\ + 77.2 . when m = 0 = 77. n the double sum ]T]2 is TI — m. + l 4 {^^)drl f i + v \ . Thus dz2 = . Observing that the highest possible value of m is n — 1 (i.* 71 — 1 n—1 n— 1 n n+ 2 E ( n ~ m ) = n +2 E n _ 2 E m= n + 2n(n . / J„^2 . for m / . 2 _ = (9id02 + (gvdvy + /„ . 9 \ 9<p\ gv 9f\ _ 1 a2 a / a\ a / ay £77 dip2 E x a m p l e 11.1) ~ . (11. 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 = 1.2 = n — 1.136a). Here we have (see preceding text) 711+77. (11. > .e. = e±im^e("+^'2{par'2L^{p)L^ (a).2 = n — m — 1 and therefore the number of the coefficient of u}n'm1 . W. Solution: For the wave functions ip = £(£)J\f(r])&(ip). N2 where g^dS. The Coulomb Interaction Solution: We begin with z of Eq.. We consider first the case m = 0. .g. i. Thus. dx = d{ y £77 cos ip) = . {g^Y. 2 2V? V^ One now performs the square of this and obtains similarly dy2. etc.1 in this expansion is seen to be n. we have to find the number of values of the quantum numbers n\. Oberhettinger [181]. and attach a factor to to each term introduced in the wave function. We then sum all possibilities and select the coefficient of ojn~1 in the result. / — cos ipdl.tx2 . 'Recall that the sum of an arithmetic progression is given by a + (a + d) + (a + 2d) + • • • + {a + (n . i. . The general formula for the Laplacian is§ ^ follows that 1 [9£\ A = ' 9 \ gr. " . 712 and m which give n. Since the lines of constant £ are orthogonal to lines of constant 7 (they cut perpendicularly).E E n l „ . all cross terms cancel in the sum of 7 the squares.2)d} = (n. z = (£ — rf)/2.114c).e. see Eq.
eF{z). if we expressed the perturbation in spherical polar coordinates. .6). i. .128) if one makes there the substitutions n —> n\ (or 712) + m.9) one finds that _ = %2 jm+2 {n2+m)\ (ni + rre)! +1 : > in. we can relate the expressions Iq to those of Eq.•* n (l) = / Jo It follows that . L ni\ ny.™(p)] 2 By contour integration (see Examples 11. Solution: Using formulas and results of above we have to determine (v) = eF{z) = leF(e.11.1 (cf.) = ^eF(p . In fact.124c). Ikr+k.drid<fi oc (p + o)dpda. (11.g. 2(ni +712 + TO + 1) Therefore the change in energy of the state to first order is given by.—r~n = 3(mn 2 ). I.(p)L™ (<7)]2(p + a)dpda rev where JJ = / Jo dpe~ "p"[I.r+k(l) = / Jo e~*xk[(k + r)\Lk(x)f = [(/= + r)\flk. {p — a) — = dxe~*xkvLkn (z)] 2 . . with Eq. 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£. Thus e. { 6 ni2 + 6 n i ( m + l) + (m + l ) ( m + 2 ) } .6 The Discrete Spectrum of the Coulomb Potential Example 11.e.127) and setting n = r + k. . with z = r cos 9. (11.7: Stark effect in hydrogenlike atoms 231 Consider a perturbation v = eFz. with RayleighSchrodinger perturbation theory). s 0/ .a).128). 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. with the connection » (11. With the help of the wave functions we derived above.r.. 8. Sec. fc — m. Z2 —> e.e.eF / nh2 \ {v)=3(n1n2)^—lr2j Thus in this case nondegenerate perturbation theory could be used as discussed in Example 8. Apart from the power of the leading factorial (which results from the definition of the associated Laguerre polynomial in the generating function). However.8 and 11. which will then give the change in energy of that state to the first order (i. = j (2ni+m+l). ( ! ^ ± ^ ) l { 6 n 2 + 6 n i ( m + 1 ) + ( m + 1 ) ( ? n + 2)} ny. with Z\ —> Ze. where F is the electric field.{ n i {2ni + TO + 1} + {ni j=± n 2 } 6(nin2)(ni+n2+TO+1) ^±n2} ^7—. we would have to use degenerate perturbation theory and obtain the same result. (11. .e. these expressions agree with those of Eq. i.
124d) of the alternative integral representation obtained in Example 11.9 we consider integrals with an arbitrary power of t.g )(l. (11.232 CHAPTER 11.9 ) ] m +h 1 yJo 0 ' 1 —pP 1 q 119 lpq _ (l9)(lp) = a With (see Eq. ) (l_pg)(m+2) Now the coefficient of qn in (1 — g)(l — p<j) .128). In Example 11.a.9 )« Solution: Using the generating function for both associated Laguerre polynomials. 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 .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 .p ) r + 2 [(lp)(lg)]"l+1(lpg). The Coulomb Interaction E x a m p l e 11..7 and (remembering the definition of the associated Laguerre polynomila there!) the second of relations (11. the integral arising here becomes with Eq. the integral / is the coefficient of pnqn in the expression _ / . L™(t) = coefficient of qn in G(t) e tq/(lq) (1 . E x a m p l e 11.1 ) ! (n .0 ° e *[l+P/(lP)+9/(l<j)] t m+l di) t  P  9 l [ ( l .b.3: /"OO I := Jo et[L^{t)]2tm+1dt. (11.l)!r! X We now evaluate the integral / as the coefficient of pnqn _ ( m + l)! [ ( l. _ „ .p ) ( l .1)! n\ This verifies the expression I™t in Example 11. s = at.p ) ( l .n+2 in the expansion of i n c o e f f i d e n t o W (m + 1 ) 1 ( 1 . c positive integers.p n _ 1 ) in (n — 1)! J p"(n + m + l)! n! p n . tt„:_ (1 .9) with the help of the generating function G(t) of associated Laguerre polynomials obtained in Example 11.^~T^——^ ) = coefficient of (pn .x ( n + m)! ( n . and is a special case of Example 11.3 by one power of t. dt = ds/a.3).l ) ! ( m + l)! in the expansion of p " ( n + m + 1)! p " " 1 ^ + m)! 1 _t . .8: Laguerre linear expectation value integral Evaluate the following integral (which is different from the normalization integral of Example 11.1 ( n + m)!"[ _ (n + m + 1)! (niy J n! (ra + m)! _ (n + m)! (n .pK — — —L .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. (n + r .( 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)!.
q)(l . — (cr)!r!(6r)! Kv.f ].L. i.?(t)t a e.1 + y w)\(a)y (r — w)\w\(j3 — r — l)!(j/ — w)\ Special cases: (1) In the case a = 0.(i) = i ! ( .p) .e. we can reexpress 1(a) in the form: /(a) = (2^)2 J J p t + l ( l . In case (a) we have u \ 7 V^ ir fb + c\ 7(a) = Jo / Tli u.(*) from the coefficient of a1: />00 233 1(a) = / Jo dte.^ ( l . ° ^ + a ( l . so that = (q+fc±£) .e.Pq + Q(1 " 9)(1 " P)1_a_1 = ti f p^ X COeffident f ap)]"01. we have 2r = 6 + c = even. .111) of the gamma or factorial function. . = c (q + fe)! W ° (^)!(^)!(^)! ~" (2) In the case of one power of a.6 The Discrete Spectrum of the Coulomb Potential Solution: One considers the following integral and obtains /£.w(r) = and / a ) = r\(P ~ r .p ) ] . we have to have y + (c + b — 2r) = 1. (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 . 7£.p) _lpq It follows that (using Cauchy's residue theorem in integrating around q = 0) 1{a) = (2^ / / ^ ^ [lpq c [ 1 . i(a) = / Jo dttaet{p/(lp)+q/(lq) + l+a} = Q .e. c + b — 2r = l (i. c + b even) and/or j / = 0. i.l ) * x coefficient of a* in the expansion of 7 ( a ) . where (a+**£)! c j n Jo = .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. c + 6 — 2r = 0 (i.124d) of an associated Laguerre polynomial. Using the contour integral representation (11.™( r > (1 + a)P r £^0y^w Jc+6_2r = . (a) j / = 1.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 «.11. c + b odd).a _ 1 = [1 + a ( l .e.r ) ! a ' .a t [Lg(i).
cannot really — and this means in a rigorous sense — be applied in this case. . In view of the slow falloff of the potential at infinity. we consider the traditional treatment here and refer the interested reader to literature in which it is demonstrated that the derivation given below finds its rigorous justification on a basis of distribution theory. 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. Zie given by h2 _ A _ ZiZ 2 e 2 V>(r) = Ei/>(r). is the work of J.138) " 2 ^ =2 ° ' 2 m Uo 7 = ZtZ2e2 ^T (1L139) * Screened Coulomb potentials are considered in Chapter 16.1%^ 11.~* 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 .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. 2mo Here we set (this defines the velocity VQ) j? E ^ 1 (11. + (!±<)«.^)(. These results are seen to agree with the expressions 1^. Since such a rigorous treatment is beyond the scope of this text. ^A very readable source to consult. The Coulomb Interaction b + c\ 2 J (b¥)\(Pb¥w)\{a) ($±£ w)\w\(J3 ^±£1)1(111. however parallel to the Coulomb case. t We have the Schrodinger equation of relative motion of particles with charges Z\e. like the asymptotic condition and definitions of the scattering amplitude and the phase shift. R. b\ V cited in Example 11. also with regard to earlier literature. .7.)!' f^)(. the standard principles of nonrelativistic scattering theory. Taylor [267].234 and (only w = 0 and 1 arise in the sum) CHAPTER 11.
(11.141) Inserting this expression into Eq. v = ik(r . (11. z) for \z\ —> oo? We can find the appropriate formula in books on Special Functions:* $(a. (11.e.z) = Wi(a. (11. i. a(a +1) z2 ] (11.136b).M) = l + . for r — oo) • ipr = Aeikzf(r .. Magnus and F. what is the > behaviour of 3>(a.108) and (11. W.b.109) we can therefore write the solution ipr = Aeikz$(i>y. .e. (11. with the asymptotic expansions (11. 86 . for instance.143) and v = ikrj. (11. where ^. (11.11. A = const.87 under Kummer functions.b.^ W ) = 0.146) See. pp. b. In particular the equation possesses a regular solution of the form (also ip ~ exp(ikr)..144) *(a.145) U ' ' j " T(6a)1 J ^ 71=0 r(a)r(a6+l) n! ' (7T < axg(z) < vr). i..140) The simple Coulomb potential permits particularly simple solutions.z). „ az 1.140) and using (11. . (11.+ ^  + . Oberhettinger [181]. that we encountered previously.138) becomes 235 ^A + k2 . According to Eqs. Our question is now: How does ipr behave for r — oo. writing Vv = Aeik^v)/2f(ri) we obtain the equation d2 x d f(v) = 0.142) This equation is of the form of a hypergeometric equation. ..b. z = rcos<9.z)).7 The Continuous Spectrum of the Coulomb Potential Then Eq.z).z) + W2(a.
if tbr were „ikr ibr ex el"z + f{6) r (11.^l*7 l + O (11. J\") ci(kr~/\n2kr) (11.z) + la)T(n + b.ik(rz)) where ^ ~ Aeikz——^—f't^ e ife(rz) + W2(n.a) (z) na)"~ r(la) T(ba) n\ ' (11. The Coulomb Interaction W2(a.e.l. (11.) of a particle with incident momentum hk in t h e direction of z for large distances away from t h e scattering centre.154) .151) exp[—ryln fc(r — z)] — exp[—ijIn = kr(I — cos 6)) exp[—ryln 2kr — ry In sin 2 (0/2)].ik(rz))} =^ + ^d.152) Since z = r c o s # .236 n\Zzab CHAPTER ^T(n '£ fo 11.147) (—7r < arg(z) < 7r) It follows t h a t for r — oo.z)T(l . = r cos 0.153) The asymptotic solution represents t h e stationary scattering state (E = const.b. > A = Aeikz[W1(i1.kz+l^ktrz)) C .7 ln*(rz)) k(r .150) ^(rz)]1"^ 1 + 0 T(«7) i. We now define a quantity f(9) by t h e asymptotic relation lb O < ^(. i.148) .151) with Eq. with 2. Without t h e logarithmic phase factors. a comparison of Eq. (11.152) 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. A(j)11 A where \ " r(i + \i) 1 j(kz+"flnk('rz)) 7 f ( l + Z7) _ e i(fcr.149) (11. (11.e.l.i 7 ) (11.
(11. the current density is * h . This cross section is defined by the ratio of the outgoing particle current (in direction 9) to the current of the ingoing particles.n . the Coulomb potential has an influence on the incoming wave even as far as the asymptotic domain. also called phase anomaly. 11.11.4 Variation of the observed differential cross section with 0. we see that this expression is the exact analogue of a current density in classical considerations.1 The Rutherford formula The above considerations permit us to calculate the differential cross section which is a measurable quantity. The problem of the logarithmic phase. We define as current density the quantity h J := 2imo [^(v^)v(v^n. do dQ K 12 K Fig.156) . 11.155) Since the gradient is proportional to the configuration space representation of the momentum..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. The quantity /(#) 2 would then be a measure for the scattering in the direction 9. z — — oo. 2im. that we encounter here is a characteristic of the Coulomb potential.7. In other words. the effective range of the Coulomb potential is infinite. (11. For the incoming wave ipi = exp(ikz). As a consequence of the masslessness of the photon.Q hk mo v0.
154) we chose the prefactor of the incoming wave in (11. 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. (11. which yields the scattering cross section.e. In reality this divergence does not occur. In Eq.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. 'screened ZlZ<2e2 cr/r0 ° . thereby screening itself off from the surroundings.e. the square 7 2 ). Correspondingly the other part of (11. We make a few more important observations.159) as the Rutherford formula which can also be derived purely classically. i.152) yields as density of the outgoing current Jr = ^\f(0)\2vo(11. We recognize the result (11. in the current this would in any case contribute only something of order 1/r. This implies effectively that the Coulomb potential becomes (virtually) a screened potential or Yukawatype potential of the form V . (11. is called scattering amplitude. That this formula retains its validity here in quantum mechanics is something of a coincidence.238 CHAPTER 11. This implies that scattering takes place for the attractive as well as the repulsive potential. since in actual fact the Coulomb potential occurs only in a screened form. for 0 —> 0. 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. The Coulomb Interaction In the case of the Coulomb potential we now ignore the logarithmic part of the phase (see discussion later). We see that a depends only on the modulus of the potential (i.152) as 1.153): 4fc2 s i r  The expression /(#).
i i f (''°>/Vk(x. (11163) (11.e. the wave packet in the absence of a scattering potential.e.k(x) is a solution of the continuum with energy h2k2 E = Ek = — >0.2 fr2 . and if H is timeindependent one has the stationary wave function ^(x.^ ^ % ( x ) . 11. i.8 Scattering of a Wave Packet We saw earlier that in quantum mechanics a particle is described by a wave packet.e. t) = U(t. of the equation r fi.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. 2mo V + F(x) k (11. 2 (11162) where Vk(x) = ^ ( x .*'^ e1 piEk{tto)/h (x x i)= (2^3^ ( k ) roT ^ k o l * . t 0 ). (k).165) ^ ( x ) = £ fe Vk(x).11. / dk 7^340(k)4(xxo. m0 m0 and V. (11. (11. t 0 )^k(x.yJ ^ j A k o W 727)3/2 e 0.4. which in the case of free motion (zero potential) are simply plane waves.164) Ek~Eko + {kk0)v0h. t0) = e. _ x o _ v o ( t _ toh io)_ em0^VoKtt0)^o)ix (1L167) 0 l/> () ) fxXn t)~ gik(xxo) .166) Let the incoming free wave packet at time t > t$. 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. i.t). whose maximum moves with the particle velocity.x° ' r j . be given by ^ f dk AL. The wave packet is a superposition of waves ip(x.t) = e .161) where Ak0(k) ~ A(k — ko) and ^ k (x.t). i. The particle velocity v is v = — . t 0 ) . v0 = . 11. of continuum wave functions. and which provides the uncertainties arising in quantum mechanics.
10 we show that the solution V'k(x) of Eq. Eq. (11.x °Vk(x).ip) is the scattering amplitude with respect to the incoming plane wave.k0) ~ k0 + and the expansion of E^ in Eq. k = kez.170) A ko (k) e .x + _ / k ( e j ¥ ) ) ikxo (2TT)3/2 zkxo = <>(* r y (2vr)3(27r)3/2e M ^ ) e *o)/^] e ikx 0 (11.e.169) and (11.172) h h k0 {rv0(tt0)} (11. From Eqs. (11.ik .xo.<p) + Ol Akr l (11. (2TT)3/2 eik* + —fv(P.163) we obtain </>ko(x .171) The phase of fk(0.t) = J ^ A k o ( k ) e .173) .169) The wave packet in the presence of the potential is obtained from the free form (11.i) dk (2TT (11.240 CHAPTER 11. ^k(xxo) ^k(x) ~ = e.166). i.^ ( * .t o ) / V k ( x .ko) k0 (11. (11. With Eqs. Then with k = ^(ko + k .x 0 ). (11. can be written A (X) = ( 2 ^ + W J ^ ^ ^ ( ^ ( x ' )  eikx 2mn f P»fexx' (11168) We saw earlier (cf.154)) that for x — oo.167) by replacing the plane wave exp(zkx) by this scattered wave.165) ko • (k . The Coulomb Interaction In Example 11.167) we have ^k0(xx0.iB *(**°)/ R °'tj + 1 pikr e * .k 0 ) 2 ~ Jk20 + 2k0 • (k . this solution > has the following asymptotic behaviour in which f(6. i.161) and (11. the stationary scattering wave. (11. ip) could be carried along but we ignore it.e.
r ' ) = 47r<5(r . (A + fc2)t/.z Fig.x 0 .11.(«)/ ipko(x .x 0 . A particular solution of the Schrodinger equation is — J G(r.y)eiko'Xoe'[fcor£. (11. and determine the Green's function G(r.174) r /k0(^.(r) 2rrao V(r)V(r) 7/>(r) (2TT)3/2 ^dr'G(r. t) f. '.r'). (11. U(T) = ^ V ( r ) . the solution can be written iMr) : (2TT)3/2 ^dr'G(r. Solution: We define the Green's function G ( r .154).r')l/(r>fc(r'). 11.r')V(rW). T h e scattering amplitude is t h e same as t h a t given by Eq.5 Scattering of a wave packet with scattering centre O. r 241 (0 LO) f ^ Afc o(k)rikxnri[fcor£fcoato)/ftl i(kko)x' = /k0^.171). with the energy of the maximum momentum of t h e wave packet.fco(tto)/ft]^0o)(x/xo. (11.r ' ) .r')U(r')ip{r')dr' /4T7 (as we saw earlier) . .*0ko (x .^y /k0^. however.to) has t h e form of t h e wave packet in radial direction as indicated in Fig.167).^e (2^3(27r)3/2e y (27r)3(27r)3/2e (11. Adding a suitably normalized solution of the free equation. *. t) .8 Scattering of a Wave Packet so that with Eqs. r ' ) by the equation (A + fc2)G(r.e.to). 11. E x a m p l e 11.10: Schrodinger equation as integral equation Demonstrate the conversion of the Schrodinger equation into an integral equation. Here ^ (x' — xo.5. i.
displaced slightly away from the real axis as indicated in Fig.O O ™ fc2 (e*fc'r _ e ifc / r 1 fc ) =  / fc'dfc' i(fc' 2 fc 2 ) Since this integral does not exist.fc2 ' 2 ^ i fc72^"fc2 Integrating out the angles we obtain 1 G roo />7r 2n fc' rZn j. e > 0 small. « = 52 / J _ Z100 7rr 7o / A fc'dfc' 2 / 2 • fc2 (fc' *(fc' . 9{k') = 1 1 S(T) = j G(r) 1 ^ A j dk'e i k ' r . .Rek' kfc/2K Fig. we relate it to a properly defined contour integral with the two simple poles at fc' = ±Vfc 2 + ie = ±(fc + ie/2k). which is such that the contribution along the infinite semicircle vanishes.6. 11. dk' ik'*r 2TT2 fc'2 . 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.fc ) oo dfc' 1 d nik r 7rr dr / . one obtains another Green's function given by G_(r)= r eikr. 11. The Coulomb Interaction We find a set of Green's functions G(r. Imk . Hence we write G(r)G+(r)./2 dk'd(—cos 6)dip iki.e \ r dr\ k ikr 1 ikr 0 With the choice of an analogous contour C— in the lower half plane.242 CHAPTER 11.=k+ie/2k 1 d f. r ' ) = G ( r — r ' ) with the Fourier transforms G(r) = J dk' 3 (k')e ik ' r . Thus we consider the contour integral taken around the contour C+. These are the socalled outgoing and ingoing Green's functions.6 The contour C+ in the upper half of the k'p\ane.
Eq. whose continuation to infinity is again given by (cf. e.g. . 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. Thus we proceed as above.2l + 2.9 Scattering Phase and Partial Waves 243 11. (11. b. Before we introduce the general form of the socalled partial wave expansion. Then £^<fc + (2l + 2 £ = ~2ikr. Then yW (11.£).140) (previously we solved this only for the bound state problem) § .z2 27fc 1(1 + 1)' yi = 0.178) (11. under the stated conditions 2/07 = Z\Zie2 = —Mo. 21 + 2.176) " 0—A ~ (I + 1 + iy)<t>i = 0. i.0 + W2(l + 1 + iy. z) + W2(a. z) = W^a.2Z + 2. Eq. we define first the scattering phase for the case of the Coulomb problem proceeding from our previous considerations. (11. the logarithmic phase as a consequence of the slow decrease at infinity and a special symmetry which permits separation in other coordinate systems. i. z).179) = = elkr{kr)l+1F(l e (kr) ikr l+1 + l + ij. b. (11.145)) F(a.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. (r) (11177) Similar to above we obtain a regular solution y) ' with 0j = F(Z + l + i7.11.175) yl' + k Here we set (as in the case of the hydrogen atom) yi = eikr{kr)l+lUi). (11.e.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.21 + 2. 2ikr)] [Wi{l + 1+11. b. If we proceed from the radial Schrodinger equation.e. where Mo is a parameter in Chapter 16. (11.
147) this becomes asymptotically (r) Vi lkr eikr(u„\l+l (kr T(2Z + 2) '—(2ikr) \T{1 + 1 .i f c r + i 7 In 2/cr+i7r(i+l)/2 + We set T(l + 1 + 17) = r(z + i + i 7 ) ' r(z + I .i 7 ) „ikr—ij\n2kr T(2l + 2) e e* jTv(lliy)/2 2l+i —ikr r(z +1 . . Then (r) r—»oo Vi 7T7/2 T(2Z + 2)e*7'*e i8i 2'+1I\Z + l + i 7 ) i5. (11. +ikr—iry In 2fcr—i7r(Z+l)/2 _i_ — iSi— ikr+i^ln2kr+iTr(l+l)/2 ! We set I^TiT^rsin(J:r"7ta*"^+4)'<1L182) 6 = arg T(Z + 1 + ij) (11.l e*"" + T(l + 1 + H ) T(l + 1 .180) ^ ' (11.244 CHAPTER 11. T(2l + 2) (2fcr)*T« (Alliy ikr . (2kryr{i" V 7 . The Coulomb Interaction Using Eqs.146) and (11.j syyll iy il( _ .' . Then e2i5< = Reie Re~ J2i9 9 = Si.i 7 ) I T 2l l—X—i^j ( + 2) 2ikr( ?.ii)' (11.181) so that eldlT{l + 1 .ry) = e"^r(Z + 1 + ry).«7) „ .183) r(Z + 1 + 17) := Reie.n) • e „—ifer+i7ln2fcr _eMr(i7Jl)/2 r(z + I +17) T(2Z + 2) /2 • ikriy In 2fcri7r(J+l)/2 r(z + 1 .
. indicates that — looked at analytically — the scattering amplitude f(6) possesses at E — 0 a branch point of the square root type... g = e«* = rSi + 1+ *V*ri (11.188) in the form y/E. The expression e 2i5j _ 2i5i—iirl is called Smatrix element or scattering matrix element.e. with Eq. 245 (11. However. considering I —• an(E).184) The phase 6i = 8i — lir/2 is called Coulomb scattering phase. i._W2)_ ( 1 L l g 6 ) The factor e2lSl between the Jost solutions (in the regular solution continued to infinity).11. i.n ' . (11. (11. a positive integer. One defines as regular Coulomb wave Fi(j. That the energy E appears in Eq. . as we saw. kr) the regular solution with the following asymptotic behaviour: r 0 (11.1.139) n' = 0.187) contains almost the entire physical information with regard to the Coulomb potential.185) J F} ^° sm(AT7ln2£T + ^ ) (also called regular spherical Coulomb function). (11. where the continuum of the spectrum starts. Squaring this expression we obtain the wellknown formula for the binding energy of bound states (as a comparison with Eq.e. For instance the poles of S are given by Z + l + z7 = .114a) reveals). i.. i.2.u\ respectively with the following asymptotic behaviour: u(±) r^o e±i(fcr_7ln2fcr. In the present quantum mechanical considerations I is.9 Scattering Phase and Partial Waves Hence 5j = argr(Z + l + i7).e. Analogously one defines as nonregular spherical Coulomb function or Jost solution the outgoing or incoming wave u\ .e. considering I as a function of E extended into the complex plane.
Singh [252]." These Regge trajectories can be shown to determine the asymptotic high energy behaviour of scattering amplitudes and hence cross sections.7 A Regge trajectory of the Coulomb potential. The Coulomb Interaction and plotting these for different values of n. (11. 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. We mentioned above that the logarithmic phase does not arise in the case of shortrange potentials. we obtain the partial wave expansion of the scattering amplitude as follows. (11. lman(E) jump from E<0 toE>0 o A LU infinity  E<0 Rea n (E) Fig.246 CHAPTER 11. and thus in general. 11. For the Coulomb potential a typical such trajectory is indicated in Fig.191) . 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. In these cases.189) where as before the incoming wave in the direction of z is ikz ipin = e (11.7.190) and the scattering solution „ikr Vwt = / ( 0 ) — + 0 V. 11. we obtain trajectories which are known as Regge trajectories.
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. W.~ J > Z + l)Lil*e~ikr . Magnus and F. 21 .22. The mathematical expression can be found in Tables of Special Functions (there cf.195) with the scattering matrix element S(l.^(cosfl) (1L194) This therefore yields the following partial wave expansion of the scattering amplitude = ^lY.192) Thus the incoming and outgoing spherical waves have a definite relative phase. pp. .^2l + l^S^k) lik (11.k)=me2iS^k\ in which r\i^\ indicates absorption. Oberhettinger [181]. (11.196) "See e.i f c r . When we solve the Schrodinger equation and calculate the asymptotic behaviour of the regular solution we obtain correspondingly V'reg .r ^ V ^ Wostf). (11.eikr\P^cos9). 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„ ' .9 Scattering Phase and Partial Waves 247 For a spherically symmetric potential the amplitude / depends only on 9. r e. (11.193) where r\x e2l5t determines the change in amplitude and phase of the outgoing spherical wave.11. 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 ' .g.i * r 1 v^ f ne2ibl — 1 \ = k^2l W +1 \^r^)Pl{cosd)—. 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.
This page is intentionally left blank .
It is very instructive. It is not surprising. the computation of finite lifetimes of a quantum mechanical state. Razavy [236]. Thus in this chapter we consider traditional "square well" potentials in onedimensional contexts. In fact. states of a finite lifetime. Such a vital quantum mechanical phenomenon is socalled tunneling. however. that quantum mechanically a particle may have a small probability of being in a spatial domain. which in one form or another. as — for instance — for trigonometric or double well potentials.* *See e.e. that the explicit derivation of such quantities in nontrivial contexts. M. to explore first much simpler models in order to acquire an impression of the type of results to be expected. i. in general. which a particle in classical mechanics would never be able to intrude. therefore. Similarly difficult is. in texts on quantum mechanics is rare. namely the instanton (or more generally pseudoparticle) method. 249 .e. For various general aspects and applications of tunneling we refer to other literature. one of the main and now wellknown methods to perform such calculations. was developed only in the last two to three decades. These newer methods will be presented in detail in later chapters.g.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. In particular we consider cases which illustrate the occurrence of tunneling and that of resonances. can be found in any traditional text on quantum mechanics. i.Chapter 12 Q u a n t u m Mechanical Tunneling 12.
2 Continuity Equation and Conditions The statistical interpretation of the Schrodinger wave function ^(x. .t)\2 =1 (12.t) (12. i.250 CHAPTER 12.4) (12.2) we have = o + ~v[(vv>*)V'V>*(W)] = vj. Quantum Mechanical Tunneling 12. ihr(x. M^ Then dt + Vj 1=0. Correspondingly one interprets as probability current density the expression j x ( < ' >=iH v ^7 v <4 = [W{x. (12. We assumed this already in the preceding since we interpreted the normalization integral / Jail space dx\ip(x.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.e.t)\* (i2 2)  With the help of the Schrodinger equation ihiP(x. According to Eq. t) was developed by Born in 1926.t) that the one particle concerned is somewhere in space at time t. r (12.5) L j • dF IFo is the probability that per unit time a particle passes through the surface FQ. so that p = T/>2 is the probability density for a spatial measurement at time t.1) as the probability described by ijj(x..t).t) = HiP(x.
It is therefore sensible to demand that V( x ) a r i d V'0(x) be continuous.2 . even for discontinuous potentials. as shown by P. However. Delta potentials in more than one dimension do not permit bound states and scattering. 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. the equation can be integrated at every point.12. This case plays a special role. t For the singular delta function potential a modification of these demands is necessary. since this potential implies a discontinuity of the derivative at its singularity (as will be seen below). From this follows that both p and j have to be finite and continuous everywhere.2a6(x)]il> = 0. 1 0 ~x o a= f See L. Nonetheless. Schiff [243]. higher dimensional delta potentials do not have properties like that in one dimension. 29 . where „ mpVo V(x) =+V0S(x). 12. 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.6) . we can do this also here. ~hr> k = n^ (12 7)  . 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. 'Maybe the reader dislikes being confronted again with the onedimensional case.3 The ShortRange Delta Potential 251 The timeindependent Schrodinger equation is a second order differential equation in x. with regularization their study is very instructive for illustrating basic concepts of modern quantum field theory. pp.33. which is the case we consider here. Tarrach [119]. although in general small. If there is no absorption. This effect is known as "tunnelincp. as we mentioned above. It is therefore quantum mechanically possible for a "particle" to pass through a classically forbidden region. Thus it is possible that the probability for the particle to be in such a domain is nonzero. Gosdzinsky and R. In much the same way that we define there reflection and transmission coefficients. Such a region is for instance the domain where V is larger than the total energy E. the problem is analogous to that of the scattering process from a potential in a onedimensional case. If ^>(x) and V^>(x) are known at every point x. The Schrodinger equation for this case is* ip" + [A. (12.3 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.
(12. i.ikB) = 2aC or 2a(l + B).e. Consider the case of a purely outgoing wave in the region x > 0 (no reflection back from infinity).2 a J 5{x)4>(x)dx = 0.: (i2iQ) Ce for x > 0. Quantum Mechanical Tunneling Integrating the equation over a small interval of x around 0 we see.' : It follows that ik . e Q (12.e. (12.12b) a=—^— C = a + ik and hence B = C1 Thus for i/> we obtain = ?—.11a) ikC . (12. a — ik £Qr < { eikx a eikx •t" . i. [<//]0+ . (12.= 2a^(0).[^']o.13) ~A R= for x > °(12. In the region x < 0 we then have both types of waves: the incoming or incident wave to the right as well as the wave reflected from the potential. that the derivative of i/> is discontinuous there: W\U + k2 J i>dx .8) (12. The continuity conditions at x = 0 are therefore for V : and according to Eq.1) .e.12a) (12. i.11b) 2aC.9) States with k > 0 correspond to those of free particles that move with constant velocity v = hk/mo.14) One defines respectively as reflection and transmission coefficients R and T the squares of the respective amplitudes. except in domains of the potential V.252 CHAPTER 12. a 2 — a 2 a — ik a2 + k2 ' .e 1 + B = C.(ik . 2 (12. . _ fQr x < Q^ eikx _ geikx ^=\ ~ Blkx *x .ik(C .ikC = i.9) for ij.
R. f Ce~KX c. 'Note that the physical pole has negative imaginary part. = .g. Newton [219]. See e.3 The ShortRange Delta Potential and T= ^rr a — ik 2 253 =1R.(K)C = . at k2 = —a2.2 o ^ ( 0 ) . the quantity T.K 2 < 0.) and V>'(0+) .2 = .e.ip'(0) = .16) The corresponding wave function proportional to etkx for x < 0 associated with the pole for k = — ia is^ rl){x) = V^e~aW.17) da#(s) 2 = 1. Omnes and M. and C follows from the normalization: OO A. when normalized to 1. Thus ^ = Ce. 2m 0 (12. i. We write the solution ^ so that ^(0+) = ^ ( 0 .12. K = a.2 a C . (12.e «x = J \ CeKX for for x > 0. as coefficient of the outgoing wave. / oo In the bound state problem the Schrodinger equation is ip" + [k2 + 2a6(x)]ip = 0. z<0.a l s l. poo nc s~t2 / oo dxe dxC2e~2aW = 2a\x\ _ u 2C2 / JO a Hence C = y/a and therefore ^ = VaeQK (12. i. or h2 E = —a2.19) Thus the delta function potential supports exactly one bound state. ( 1 2 . .e. possesses simple poles at a ± ik = 0. .i 8 ) .K C . oo (12.15) Furthermore we see that similar to the Smatrix which we encountered in the case of the Coulomb potential. Proissart [223] or other books on scattering theory like R.
. The latter contains also specific applications in nuclear physics. 12. (12. or in the lectures of E. 12. (12. V(x) V=E a v=y Fig. x€[a.20) for for \x\ > 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. Schwabl [246]. \x\ < a. x G [a. 58.1 The square well potential.a]. Similar treatments of square well potentials or barriers can be found in most books on quantum mechanics. We set 2m0(E + VQ) h2 so that ip" +fc2V>= 0 ip" + K il) = 0 2 (12. Quantum Mechanical Tunneling 12.21) The solution of the Schrodinger equation then consists of the following parts: Aeikx + Be tp{x) = { Ce~ iKX lkx iKX with with with xe[—oo. 55. We want to consider the case of stationary states for E > 0. p. The Schrodinger equation is iP"(x) + 2 ^[EV(x)}iP(x)=0. oo]. —a].254 CHAPTER 12. p.22) + De Feikx + Ge~ikx *We follow here largely F.1. Fermi [92].
I (12. .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. (12. (12.24) yields F ^ G. fc Af(a) = .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 \ . ).27) v ' Similarly Eq. M(a) .12. ( V C n vD ).f )e" t\ K\0—tKa—ika .ikGe~ika..iKa ? ikAe~ika and . Fikeika .26) K^gifoa—i/ca fc.ikBeika = inCeiKa + DiKe~iKa. ) A l ( pika p—ika pina K ina kC p—ina \ _p—ika J \ Kp—ina J ( c\ U/ (12. 2 l ^i i K\ iKaika O \ (1 + %)e ina—ika (1 . (12.24) The first set of equations can be rewritten as / ( p—ika / p—ika pika pika U pika A \ \ / \ / 1 pina Kpina p—ina Kp—ina \ \f ° ){D ) \ B ) ° or A N \ Bt i. (12.25) ^\=M{a)lCD where 1 / f l — « \ irea+ifca /i . (12.23) Ce~iKa + DeiKa iKCe~iKa + DiKeiKa = = Fe ifca + Ge"*fca.e.
(12. (12. \R\2 = T(E)^sm2Ka (12.e sin 2na IJlika e B = r)F sm2na. Quantum Mechanical Tunneling Hence with the complex quantities IK (12. i.34) and as transmission coefficient (note e is complex!) \T(E)\2 = 1 .a].31) we have A AT(a) = )MFo B ^r)sin2na . x € [a.4T(£)e ifci. the quantity T{E) := £ = A lika cos 2/ta — je sin 2/«x (12.a . (12. we obtain A = F( cos 2na — .37) .34) into 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.36) One defines as reflection coefficient \R\2 the modulus squared of the reflected amplitude divided by the amplitude of the incoming wave.e. (12.22) we obtain ' Aeikx + AT(E) I sin 2Kae~ikx T/>(X) (12. .a ] . i.e.(1 + e 2 ) sin2 2na 1 .35) with with with x G [oo. xe[a. = < Ce~lKX + DeiKX .256 CHAPTER 12.33) One defines as transmission amplitude the ratio of outgoing to incoming amplitudes.oo].i^ 2 sin2 2KO 1 + ry?7* sin2 2/ta Inserting Eq.32) [cos2«.—2ika [cos 2/ta + \e sin 2«a]e~ (12.
e. in the domain of A imaginary and K real. (12. p. and k real.35) and using Eq. (12. (12.34) that the amplitude F = AT{E) of the outgoing wave has a simple pole where cos 2KCL e sin 2KCL = 0. We see in particular from Eq. B.e. formula 406.4 Scattering from a Potential Well On the other hand according to Eq.44) .e. they have solutions. 2\k K. . where cos 2/ca With the relation§ cot(2«a) = [cot(«a) — tan(/ta)].43) Thus we set k — —i ' H . 83. Li ( — H— I sin 2na = 0. nor when both parameters are purely imaginary.12. i. which is satisfied if either / = g or / = — g'1.e.39) 2 i. However. (12.38) as expected. an equation of the type f~1f = g19. cot(fca) = i— or tan(fta) = K K .31): 1\T(E)\2 257 1 . for VQ < E < 0. (12. (12.42) These equations have no solution for K. (12.12. 2m0\E\ h? ' (12.41) i. (12. i. Dwight [81].40) we obtain cot(na) — tan(/«j) — i K IK .\ sin2 2KQ •n 4 sin2KaT(E) — \K\ >  IRI2 (12.
2Ka = nn.e.+ \tan(2Ka) where T is given by" 2 1 (K 2\k 1 (K k\ /n .fl = ( . t&n(2Ka)ER = 0.' We observe that T(E) has poles at those values of E which supply binding energies.l .34).47) These states in the continuum are called resonances.49) Taylor expansion of the denominator in powers of (E — ER) yields . A = 0.g.42) which then leads to bound states with associated even and odd wave functions. t2 2 (12. 2m0\E\ and Kcot(Ka) = y—^—L. . The vanishing of A implies that because k is imaginary the factor exp(ikx) does not diverge exponentially for x — . i. We now return to the scattering states. . (12.65. Then according to Eq. Schwabl [246].50) See e. around the values given by Eq. (12. . Also. i. (12.258 and obtain the equations CHAPTER 12. T(E)e2ika = — r . (12.e. from Eq. . „ .35)) for E > 0 sin2Ka = 0.20). At their positions cos(2/«z)£. ± 2 . (12.46) This condition implies the eigenvalues E^ER = n2^^ V0>0.c o . 2m0\E\ Ktan(«a) = — y——^— . F. 64 . The infinity of the amplitude of the outgoing wave corresponds to a zero of the amplitude of the incoming wave. k\ d(2Ka) K) dE \ E R 1v/2^a 2 h 2ER + V0 ^ER~(ER + V0) r 1 (12. (12.46).45) These are the equations that one obtains from the connection relations for the case of Eq. This can be understood.l ) n . Eq.e. pp. (12. The remaining wave part becomes exponentially > decreasing. ± l . n = 0 .48) We expand T{E) around these energy values. (12. Quantum Mechanical Tunneling . The amplitude of the transmitted wave assumes its maximum value where (cf. = 2/rn (EER). (12. i. and is itself infinite there. "Note that dtan(2K<z)/'d(2na) evaluated at ER is 1.
12. We consider a potential consisting of a chain of rectangular barriers as depicted in Fig.5 Degenerate Potentials and Tunneling We obtain therefore T(E)e2ika = = (^ 259 iT/2 EER + iT/2' (12.12.e.^ l / . 2a<x<2a. V(x) x=2a x= =2a .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. V(x) = 9(x + a)VQ6(ax). T(E) has poles at E= ERi: (12. A potential of this type is known as a PenneyKronig potential.' ''' 5a 3a \ a ~~~. s \ V0 /• ~\ 'X . 12. = 0.51) i.52) Here ER is the energy of the resonance and 2/F the lifetime of the state.' ' 5a \V / Fig.2 The PenneyKronig potential. These are precisely the resonance states. Since we have reflection as well as transmission there are states with repeated reflection and finally transmission. We restrict ourselves to the period — la < x < 2a. 12. The Schrodinger equation to be solved is ^"{x) + ^(EV(x))i. a 3 a.2. . This result is physically very plausible. .
(12. and hence for = Eip(x). and we choose (12..260 CHAPTER 12.2 = W h 2 _„2 = 2mo(E . x + 3a <> ™ since the particle cannot penetrate into the infinitely high wall. . i. This boundary condition implies the quantization of the energy with ^ ( 2 a ) = n7r. with V as parity operator.a ) = 0 = ^(a). (12.e.Vo) n. = EVip(x).e There is the additional boundary condition (for Vo = oo) ^ ( .53) *w«w^( « j)'*^{i i ^ .. .. i. 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. Hip(x) so that with Hi/)(x) — Eip(x): H[^{x) ± t/>(x)] = E[^{x) ± xl>(x)\.2.. Quantum Mechanical Tunneling (a) We consider first the case of VQ = oo. VHV1 Hip(x) = Eil){x): VHV~lV^){x) = EVtl>{x). E^En = ^ ^ t n = 0.1.55) The Hamilton operator of the problem. In view of the periodicity we have ^(x) = ^(x + 4a). HV^(x) = H..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.2 < a . the differential operator is even in x. We set .e.
12. that these are such that for VQ — oo they approach those of the previous case. This wave function is no longer restricted to the interior of a box. (12. We > obtain even and odd functions ip± by setting ^ = ±1.a]. VV(°) = °. At x ~ — a the connecting relations are Asin(k2a) kAcos{k2a) and at x = a: BeKa±Be~Ka K.2).63) = KBe~ F = . we shall not exploit here.60) The boundary conditions that ip± have to satisfy. We expect.62) = nBe~Ka T nBeKa = ^n[BeKa q= B e " M ] .12. — a].2a]. which. of course..2a. ^kAcos{k2a).Be Ka (12. We can write the solution as consisting of the following pieces: { Asink(x + 3a) BeKX + Ce~KX with with i £ [ . This means that i>±(x) = <  = ±1. V+(0) = const.5 Degenerate Potentials and Tunneling 261 We restrict ourselves again to the period — 2a < x < 2a.58) D sin k(3a — x) with xE[a. (12. however. BeKX±Be~KX with xe[a.a]. We therefore have different boundary conditions. At x = ± a we now have to connect ip and ip' of neighbouring domains. ±Asink(3a — x) with x€[a. (12. (12.il>]?0. xe[a. V'(O) = const. The central region of the solution is the classically forbidden domain. — a].61) = Be~Ka ± BeKa = ±[BeKa ± Be'Ka].2a].59) Asink(x + 3a) with x & [~2a. (12. = Ka ±Asin(k2a). so that (W meaning Wronskian) W[tl>+.. We demand these for even and odd wave functions ip±(x) which we can clearly construct from ip(x) (as examples see the dotted lines in Fig. are ^_(0) = 0.
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. (12.66) Inserting these expansions into Eq. + 0 ( e . (12.67) The value of k belonging to the upper sign. say A. we set tanhx = l . i.e.4 x ) and cothx = 1 + 2e~2x + 0 ( e " 4 x ) .2 a . say. » we expect the eigenvalues to approach asymptotically those of the case Vo = oo. where Akoce'2™.nir + 0[{2ak .68) . the odd solution with. in the present case it is easier to continue with the above equations. It is plausible therefore to use appropriate expansions.2 e . (with reinsertion of nir/2a for k) 2ak ~ nir .mr)2}. i.+ • • • 1! cos^(n7r) 2ak .even. F ^odd (12. i. is the one belonging to the even solution. (12.262 CHAPTER 12.3 Level degeneracy removed by tunneling. In general one now rewrites the equations in matrix form. (12. Quantum Mechanical Tunneling Fig.65) and we expand tan(2afc) about nir. the limiting value for K = oo: tan(2a/c) = = tan(n7r) \ (2ak — nir) 1 — ^—.e.nir)2}} = 1 ± 2e~2Ka.— (1 ± 2e~2A + O (—). = AA.e.Thus we observe a splitting of the asymptotically degenerate eigenvalues: kesympt. For large values of Vb. k = fc0dd. However..e.nir) + OU2ak . for VQ — oo. 12.64) we obtain ~[(2ak k . and is — as we see — less than that belonging to the lower sign case. Taking the ratio of the equations in both cases. i.
Compute this probability. exp . i. The following example is an attempt to transcribe the quantum mechanical effect into a macroscopic situation. y(x) = sin and y(xo) = 1. . b > a. Fermi [92]..v = 0. 12. ^v + ^[E. V ^ 0 are given by the equations ^o + " J .5 Degenerate Potentials and Tunneling 263 Thus the effect of tunneling — here an exponentially small contribution — removes the degeneracy of the asymptotically degenerate states. Solution: The wave functions i/>o>Vv in regions V = 0. (12. exp r rb rb / •J a 2mo (V(x) .4 A car meeting a bump in the road.V{x)\i. 12. .** " W ~^7~ Fig. . .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. a sinusoidal bump in the road with a height of one meter at its peak as indicated in Fig. Classically the car is unable to overcome the bump. 1pV 2ro 0 4>o '.2 dxy/V(x) (12. 2mn. 0. but quantum mechanically there is a finite probability for this to succeed. 57.69) where the potential V(x) is (with g the acceleration due to gravity) V(x) y'(xo) Setting w(x) = 7r**E.12. w(x0) •• = = mogy(x). 2m0E „ 2mnE ..e.3. as indicated in Fig. Example 12.4.1: 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. 12. '. and is a reformulated version of a problem set in lectures of Fermi. p.70) xo = (a + b) 2' .^ 0 = °' Hence in dominant order •00 — exp .
Emde [143]. 950. one evaluates an approximate probability of exp[6.71) The integral can be looked up in Tables of Integrals. we obtain ( .055 X 1 0 _ 3 4 J s .9191 ~ 1.e. ^ ) = 2 ^ .4 x 10 3 9 ].807 m s . n I .1.y) is the beta function (see Eq. formulas 3.621.2 [fr/2 ~ I)'] (Ml)! 2 ' where B(x. M. Quantum Mechanical Tunneling = y/mog 2(1) — a) /""'' / dw sin1/2 n Jo w. We have 8m 0 : exp ^(ba)/4 x 2_1 ^^W(s x °' 9191 (12. We can now evaluate the probability. S. : 2 ^ B ( ^ . p.1. 14. **For instance E.1 / 2 ) ! = v^r) TT/2 /. dx sin ' x •• Looking up the value of the factorial in Tables. p.**. 369 and 8. Ryzhik [122].72) Assuming a mass of 1 ton of the car (i.264 we have ro / dxyfV(x) Ja CHAPTER 12.2 . (12. h = 1. We find™ (with ( . which give /•TT/2 dx sin'M . . Jahnke and F. p. and assuming a length of b — a = 100 m of the base of the bump.384. 1000 kg) and using the following values of the natural constants: g = 9.17)).l . (15.1 / 4 ) ! = (4/3)(3/4)! = (4/3) x 0. Gradshteyn and I.2.
we consider the linear potential in three space dimensions that forbids the particle to escape. p. Finally. * A highly abridged treatment of this problem is given. in Chapter 15. This is the problem of Galilei in quantum mechanics. 13. the last case allows only a discrete spectrum. we consider the corresponding case of a particle above the flat surface of the Earth. Thus.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. 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.* In Chapter 14.1 Introductory Remarks We distinguish here between three different types of linear potentials.3. 265 . 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. Then considering the probability distribution determined by the squared modulus of the wave function.Chapter 13 Linear Potentials 13. Siissmann [266]. 144. each of which propagates with the classical momentum and energy of a Galileian particle. Example 14. in the book of G. whereas the first case permits only a continuous spectrum. We shall see that in this particular case the wave function can be written as a superposition of de Broglie waves. This is the quantum mechanical version of the problem of Galilei in one space dimension. 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. for instance.
Therefore we use the Fourier transforms il){x. V(x) = —mogx.2) which is the case with a discrete spectrum.1 S u p e r p o s i t i o n of de Broglie waves We consider a particle of mass mo falling freely in the timeindependent homogeneous field of the gravitational force F(x. (13. (13.1).t) = mogx + V0. and yields rriQX = mog. c> 0. A different > "linear potential' is the socalled "confinement potentiaF V(x. (13. x > 0. The equation of motion follows for instance from the derivative of the constant energy. (13.2.t).= l V 27T J_oo 1 dkeikxijj{k.t). We consider here the case of Eq.4a) f°° ${k.4b) . (13. i.t). 13.1) where g is the acceleration due to gravity and we can put Vo = 0.= l V 27T Joo dxelkxip(x.t) = m0g with potential V(x. Recall the classical treatment of the freely falling particle.t) = . g > 0. The present case is an exception and a good example of its applicability. (13.3) In general the momentum representation is of little importance.e.t) = .266 CHAPTER 13.t) = c\x\ + V0. from d_ m x2 0 dx + V(x) 0. We see " that the case g < 0 can be related to the reflection x — —x. 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. In the present case of quantum mechanics we have h2 d2 2mo dx2 m0gx ip(x.
< (13. multiplication by a test function and subsequent integration).4a) d ~ f I dk ikx dk 2vr (im0g) iP(k.6) the latter equation becomes d ~( m0gt ^(^M<=+=?.e.t) dk nikx k=oo '2K (im0g)il)(k. The vanishing of the boundary contributions is to be understood in the sense of distribution theory (i. We rewrite this equation in the following form th (13.13.t) dketkxiP(k.6) and wish to solve it.5) m+tm9Wcr{ktt)=2^^ktt)t (13. Applying the total derivative d/dt to this function.t). 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.8) With this result and the substitution (13.t)im0g^k.6) the argument k is then replaced by this.7) applied to every quantity in Eq.2 The Freely Falling Particle: Quantization 267 and consider the following partial integration in order to reexpress the last term in Eq.t). • fd ~(' m0gt %h . (13.6). i. To this end we consider the following substitution in Eq. (13. (13. we obtain the differential operator on the left of Eq.t) m0gx /2TT m0gxip(x.6) (and to avoid confusion we do not introduce new functions and variables): k > k + mogt (13.9) . d \ ~f1 dkr[ k + m0gt ^r't (13.t) the expression (13.7) In the wave function on the left hand side of Eq.3) yields us now by inserting for ip(x. Equation (13.d m +imo9 .e.3) as a partial derivative d/dk: d lKx dkeAkx{imQg)—4){k. (13. (13.t) h2k2 d ~J(k.
(13.268 CHAPTER 13. Thus we have first (the second term arising from the upper integration limit. (13. Solution: We apply the two operator expressions on the left of Eq.10) k by m0gt back : k —> k h The solution of Eq. Linear Potentials This is now an ordinary differential equation of the first order and can immediately be integrated to give an exponential. (13.gh [ d t ' j f c + ^ ( t ' . we have to reverse the substitutions and replace in (13.10) Note the amplitude function ipo(k) in front.12) . We define the wave number kt as kt.1: Verification of momentum representation solution Verify by direct differentiation that the solution (13. Thus addition of both indeed leaves the expression on the right hand side of Eq. We have thus obtained the solution of the transformed equation.6) to the solution (13.t In Example 13.11) solves the partial differential equation (13.t).t) = i>0[k 1— ) exp 2m0i J0 k+ ^(t't) (13.11) *[k?f.t ) = ipQ(k)exp Jo fdt' k+ m0gt' H (13.6).1 we verify t h a t this solution indeed solves Eq. (13.e. To obtain the solution of the original equation (13. On the other hand the second operator expression implies dtj>o imog^—ip(k.11).6). the following result h 2moi tp[k\ ^—.i) .6).11) in a neater form. the third term from the integrand) > / mog \ o exp dk h J h dt y ' 2mo k2i>(k.t) ok lm og^— exp dk •ghf dt'fc+^(t't)W(fc. Our next step is to reexpress the solution (13.t ) ' U ( f c .6). t ) . (13.6) is then $(k. E x a m p l e 13.= k rmt. We observe t h a t t h e solution is actually a function of kt — k — mogt/h. i.
4a).— [It(k) . we can write the integral / = l h l h (hr + 7 7 (h 3m0g 3 m0g \ 1 H 3m0g With the definition « * ) • ! ( * = * . we obtain rp(x.} fA ' _ _tkt 2 k+ t) 2 k . Inserting ip(k. _ k t  h2 3 Inserting here the explicit expression for fct.I0(k)].14) (13.19a) .11).18) h =k mogt p h S' h . (13. m0g (13. we have ip(x.t) into this integral and recalling the property (13. 27T ^ . . .+^f (13.17) = J_/ C X ) h2 [It(k) . m0g ~7T' k Z k _ r ^ h t Y + i n Smog (13.t) = 1 / dkipo ( k mogt 2 h ( 1m\ig \ With .= 1 / f°° dkelktx^{kut) dkipol k V —t 1 exp iktx K ) (13.2„2+2 mfaH 3^3+3 .13.t) = .16) Next we evaluate the Fourier transform of ip(k.o o 2 —  exp iktx h mZgH' . (13.11) can be written j'A ^.2 The Freely Falling Particle: Quantization 269 Then the integral in the exponential of the solution (13.t) defined by Eq./„(*)] 2m^ig Inserting the explicit expression for the argument of the exponential.o ut = 2mo —kf 2mo A..13) rn0_t+2 + mlg2t3 .15) we obtain / = .gfci + 2 g2m0 2/i 2 ' (13.
t) =y= dkip0(kt)) exp i< fctx V27T J . 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.t)\2 for t — oo (distant future or past).18) into a • different form in order to be able to perform the integration with respect to k.19a) exactly as for a falling particle. 13.270 and CHAPTER 13.o o Jo This result can be interpreted as follows.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 . Linear Potentials f dt'.21) In other words. 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.2 Probability distribution at large times Next we inquire about the behaviour of the probability density \ip(x. To this end we convert Eq. (13.20) (13. the relation (13. 7 u>t'dt' (13.t) as a decomposition into de Broglie waves. fm0gt\ \ 2h ) (13. The fcdependent terms in the argument of the exponential are T:=ikX^{^fk2+r^k 2rriQig [ n nr (13.The associated kinetic energy of the particle is mv+ = huJf.2. Such a decomposition is possible only for homogeneous force fields.19b) h3 3 we can rewrite ip(x. These wave vectors vary in accordance with Eq. (13.gk— + 2h 3 2?7lo 2 2 h mQgk ^ _ mjfo2 2 i + ft2 fcr 2mf)g 03mg £3 . (13.ut> = Jo t2 g2m0 i 3 k t .23) .v = kh/mo.t) as 1 I00 ip(x.20) expresses the wave function tp(x.
It follows that ip(x.25) The amplitude ipoi^t) then becomes for i — oo: > 4>o(h) \t\—»oo 4>o[ k \ .t) exp i\ mogxt mox2 — h 2h 2ht mo<?2£3\] r (vn§x 24ft mogt\ (13.28) (for a verification replace a by a + ifi. fi > 0. (13. (13.26) d£e ie £ kt^ m0' 0.e. ftf fti m0gt\ 2h J' " J' m0x ht mo5* 2ft + 2mo ~ftT f.2 The Freely Falling Particle: Quantization The integral (13.24) We now set fit / 2m 0 V" 0 V m0a. integrate and then take the limit fi —> 0).QX ht V ht 2ft For i — oo the definition of £ defines a distance xc\ given by > k i.26) with the formula 1/2 (13. m0xcl ht m0gt 2h gt O v: °° 2hti (13. TOO#A ft and. /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. being now independent of £.t) 1 exp r . 2h h 24ift ' mogt^ r m. Then ip(x.18) thus becomes ip(x.13. rnogt m0g2t3 tX^r~.29) .t) = 1 1 oo 271 27T /. We evaluate the Fresnel integral in Eq./m 0 a.QX 2 ^ 0 7 (m./oo dkip0(kt) exp l x ixI exp oo . (13. may be put in front of the integral.27) xd = vt Thus xc\ is the distance travelled by the particle as determined in classical mechanics.
dA 4>o(h (13. xc\).3) the timeindependent Schrodinger equation i^ + ^ [ E + mogx]<l>(x)=0. 1/2  ±^(2mlg/h^x^ / 2i (13. .1) has only a continuous spectrum.30) We see therefore: The large time asymptotic behaviour of the wave function is determined entirely by the classical values of x (i.t) = eiEtlhct>{x) (13. 13. V(x)=m 0 gx V=E Fig. Linear Potentials Replacing here x by the large time value x c i.33) oo for E <C \mogx\.e. These are incoming and outgoing waves.32) For x — oo the function cf){x) behaves like > 4>{x) ex exp exp dx { 2mo (E + mQgx) \ H2 .31) and obtain from Eq. For stationary states we set xl){x. this expression can be brought into the following form: if *\ 0 ip{x. kt and ut 13. For x exponentially increasing and decreasing solutions are possible.1 A turning point at x = —XQ. (13.272 CHAPTER 13. (13.3 Stationary States In this section we show that the Schrodinger equation with the potential (13.t) ~ W —exp i V irlt m ktxd J cot.
The point — XQ.t) = dt (13. which is not possible for real particles. (13. so that ih dtp{k. E)dk.13.t). exponentially decreasing. H2k3 4>{k) — cexp Ek (13. We determine the constant from the condition that the continuum solutions are to be orthonormalized to a delta function. its momentum p would be imaginary there. and hence is continuous and extends from —00 to +00.35) It follows that /W tm0g \ 2m0 i.e. t) = Eil)[x. probability. i.e.t). at which E — V = 0.e. 1 f01 dketlcx^{k. Fig. 13.4b). i. Akx (13.3 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. J \ „ (13.34b) It therefore follows from Eq.36) m0g \ 6m0 with c = const. The Fourier transform is given by the relations (13. but also d 00 dt Eijj(k.e. is reflected. Thus the spectrum does not contain states which are normalizable over the entire domain x e [—00. we . i. In view of the simple form of the potential in the present case the equation can be integrated in the momentum or k representation. However quantum mechanically this is possible with a small. and we can establish for the resulting continuum solutions the orthonormality and completeness relations.1. i.e.t) — (p(k) > E +^og~Mk) = 1 (h2k2 ^m. t). 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.e.00].4a) and (13. t) = — = / V27T J . cf.0 0 dkelkxE^(k. i.34a) Then d &k 1 f°° ih—il>(x.6) that also for 4>{k. at which the classical particle bounces back.
i. / dE(j)E(x)(t>E(x') = ^ s .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 . so t h a t oo / <fiE(x) J dke ikx s/2nmog exp m0g \ 6m0 Ai(s).36) into (13.37) (eikxdk. apart from a constant phase which we choose to be zero.L [ dkeikxMk). / <>~ikx VZ7T J (13.. (1340) i. the relations oo /*oo dE4>E(k)4>E(k') = 6(kk'). </>E(x) = .£') = <*(#£')• 1 27rm0g' = /o .43) y=2^x+2^E.(k) = / J ~oo dx<t>E{x)<\>EI(x) = S(E  E') Inserting (13.40)) into (13.. h? h2 . 13.36) (with c replaced by (13.e.274 have / Using d(x) = — one verifies t h a t with CHAPTER 13.M *K Joo Setting dt cos ( ts n Jo st). (13.41) J—oo Next we insert (13.39) we obtain dkcc*exp m0g It follows t h a t cc = &(# .e.38) we have for <j)E{k): OO /"OO / oo (13.39) dk4>*E{k)4>E.a : ' ) . In a similar way we can verify the validity of the completeness relation. V 2lT J Mk) = .38). (13. (13.L dxeikx(f>E(x). Linear Potentials OO dx<ffE(x)<i>E>(x) = oo 8(EEf).
(13.43) / • '**«(» + * ) * • < * + *) ds dtexp i{s(y —TIT (27 + x) s3 ^Y J ds_ f dt_ f J exp — i< t(x + z) r 2^rJ 2W dxe^Qe^^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.46) This follows since with Eq./ 2m 0 E (13.^ . In the next section we derive the important asymptotic expansions of Ai(s) for both positive and negative values of s. For instance dxAi(y / • + x)Ai*(x + z) = 5(y — z). and we shall see t h a t one branch has periodic behaviour.44) is then expressed in terms of the Airy function: . 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= ^ . whereas the other is of exponential type.13..45) One can now derive normalization and completeness integrals for the Airy functions.J * .^x + ^ E (13. . and in Chapter 15 in the computation of energy eigenvalues for the threedimensional linear potential. / s M . the function 4>E{X) i and setting s y = — = 1 . (13. For a further solution Bi of Airy functions we refer in Chapter 14 to Tables.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.
TT < 30 < 0. .2 The contour C for s < 0 with ends in the angular domains (a) and (b).43) contains the phase (with x — z): > *(*) := sz . i. We want the integral to exist and so desire that lim e**^ = 0.50) Im z domain (b) / ' ^ ^ ~C domain (a) 0 ^ ^" ~ ^ \ x Zs=i^S Fig.276 CHAPTER 13. Thus we must have sin30<O. Linear Potentials 13. (13.49) (13. (13.\zz.e.r 3 cos 30 J + i ( sr sin 9 .TT < 6 < 0 for r > oo. (13. i.r 3 sin 30 ). (13.47) Consider the following integral along some as yet unspecified contour C in the plane of complex z: / := i .4 The Saddle Point or Stationary Phase Method The integrand of Eq. o (13.51a) and since with sin 30 < 0 also sin(30 + 2ir) < 0: TT < 36 + 27r < 0.. or n <6 <TT o for r > oo.48) We have $(z) = sre1 1 A S„3i6 r e sr cos 6 . .e. 13..f dzJ*W 2TT JC with z = reie. (13.51b) .
(13.51b).S )V2 (13. (13.57) $"(z s ) = 2% s.e. at which the phase becomes stationary.13. oo] as indicated in Fig. i. Then Ze TeC : V^se±i7r/2 = ±i\fzrs. where the derivative of the phase function vanishes. Then the ends of the contour lie in the domains specified by Eqs. Such a piece can be found in the neighbourhood of the socalled saddle point zs.z3s and V^s + . and we set with a real: z := zs + a (S)V4' (13.56) 2n(s)^Jc But I daefr"?*"^.53) Since we shall have the contour C in the lower half plane. provided its ends at infinity satisfy these conditions. we obtain da 2KJC (*)V4 c 1 e **(Zs) exp i{$>{Zs)+l(zZsf<$>"(Zs) + (13. we choose zs = iy/^s. Inserting (13.54) i ¥~s)3/2' (13. (13.e. (13. 0 = &(z8) = s~zt s = za „2„2i6>. 13. so that exp(±2i9s) — 1 and — s — rl. and the main contribution to the integral comes from the a region around the saddle point. Hence _ 2 / s)3/2 /•oo 2TT( s)!/4 doe J—oo CT2 i*3 3(.55) s < 0.55) into (13.54) where a is the new variable with —oo < a < oo. (13. —i ${zs) = szs .45) exists for the contour C.58) . We now choose the path of integration C at fixed O z = —iy/^s with a e f [—oo.4 The Method of Stationary Phase 277 The integral (13.51a) and (13. Then 6S = ±7r/2. It is reasonable to choose the contour in such a way that only a short piece of it contributes substantially to the integral.2.52) = exp(±i7r) Let s be real and s < 0. i.48).( \ / ^ ) 3 2z.
32) with the solution (13.61) 2m0E = f 2 \1/3r 2 2 2 fi h \h mog2 Equation (13.45) 4>E(X) ~ Ai(z). Linear Potentials where the contribution oc er3 is obtained from &"'(zs) whose evaluation we do not reproduce here.42) into an equation with a more appealing form.491. where — with e = (2/h2mog2)1'3 — we have /2^m3g3y/3 x\ = e[E + mogx}. Ryzhik [122].60) The reason for describing the point zs as a saddle point is that the integral in Eq.278 CHAPTER 13. With the help of the substitutions (13.63) The Airy function can be reexpressed in terms of cylinder or Bessel functions Zv(z).44) we can convert Eq.2 we show that for s — oo: > Ai(s) 1 . . (13. 2 s 3/2 . Sec.64) tSee I.44) and (13. 8. (13. cos . (13. Gradshteyn and I. According to Eqs. V^s / 4 V3 (13. This is an important aspect in the WKB method to be considered in Chapter 14. For (—s) — oo the integral converges towards T/TC._ 1 . S. (13.s ) V 4 exp \(s) 3/2 (13. as one can see from the following equation^ ri' + Ffz •2^2. V 2h2 (13.56) decreases exponentially around a — 0 for real values of a. > so that Ai(s) 2 v ^ ( . but would increase exponentially for imaginary values of a — thus describing a surface around that point very analogous to that of a saddle.. (13.2/32u = 0 (13.32) therefore becomes dz2 + zd) = 0.62) This is the Airy differential equation with one solution 4>{z) = Ai(z). We observe that the Airy function has a periodic behaviour in one direction but an exponential behaviour in the opposite direction.59) In Example 13. M.
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. cos  .4 The Method of Stationary Phase with solution u = z1/2Zi/2/j(7A 279 (1365) With expansions (13. 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 . the conditions (13.. Hence the saddle points are given by 0 = * ' ( z s ) = s — z2.13.60) 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. Thus there are two symmetrically placed real saddle points on the real axis.51a) and (13. Then.^ e ^ / V2 4 [°° Jo dae'2^.43)) Ai(s) = — [ dzeiit(z\ *(z): 1 Ai(s).z ± ) 2 * " ( z ± ) . It follows that for s —> oo: Ai(s) ~ . Example 13. Eq. (13.59) and (13. we obtain f° Joo dp=—[a+ i\a\]da. kl dpe*' = ~e^'4 V2 f° Joo daeW2 and f°° dpe'^ JO = . * " ( z ± ) = =F2Vi. changing to the variable p. 27T Ic Jc 3 This time s is real and positive._ . . Choosing the contour C to lie along the real axis. P Introducing a new variable a by setting p = ai\a\. with *"(z) = 2z.s 3 / 2  .
Jo Jo \ . i.e. n!= / e~zzndz= / e~z+nlnzdz. one obtains n! ~ where we used f^° dxexp(—w2x2/2) = V2^nn+1/2en. Then.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). the method of steepest descent evaluates the integral J^dz by expanding f(z) about its extremum at ZQ with z — ZQ = ix. 13. (11. Solution: Briefly.*. ~ 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. z o = n .280 CHAPTER 13. 13.3. since f'(zo) = 0 and (z — ZQ)2 = —x2. and hence integrating as described above through the saddle point indicated in Fig. Linear Potentials E x a m p l e 13. Thus here / ( z ) = — z + n l n z .Be 2 Fig. / " ( z o ) = —n/z2 = —1/n. ^/2KJW1.111).3 The saddle point at z = n + ix.
for misunderstandings pp. A. of misunderstandings and other aspects may be found in the monograph of Dingle.* The method had.e.* Descriptions of the method can be found in most books on quantum mechanics and in some monographs. Kramers [153] and L. *R. Kramers and Brillouin. Froman [99].1 Introductory Remarks One of the most successful methods of solving the Schrodinger equation is the WKB method. i. where *G. Dingle [70]. B. Brillouin [37]. and the role played by h in our considerations here.317. A critical overview of the history of the method.Chapter 14 Classical Limit and W K B Method 14. Jeffreys [144]. ' T h e most prominent earlier expounder is H. Froman and P. ^ The central problem of the method is the topic of "connection formulas".g. The central issue of WKB solutions is their continuation in the sense of matched asymptotic expansions across a turning point (i. H.295. Wentzel [282]. § See e. The WKB result is therefore frequently described as the semiclassical approximation.e. will there be that of a coupling parameter. It goes without saying that such comparisons are very instructive. the linkage of solutions above a turning point to those below. For the history see pp.§ 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 . O. 316 . 292 . been developed previously by others in different contexts. named after its first promulgators in quantum mechanics: Wentzel. however. N. 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.
• {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 .1) (14.g. A(r) and S(r) being real functions of r. It is then possible to derive in a general form the relation known as BohrSommerfeldWilson quantization condition.2 Classical Limit and Hydrodynamics Analogy In this section we consider the limit h —• 0. for which the observation of position and momentum is independent of the time at which the observation takes place. For stationary states ^/. the derivation in Example 18. In this equation we set.2) dip(r = I VA{v) + ^ ( r ) V 5 J eiS^lh.* We illustrate the use of this condition by application to several examples.3) . V>(r) = A(r)eiSWn. 14.t) = eiEt/h7P(r) and \V(r. We shall see that in this limit the > Schrodinger equation describes a steady stream of noninteracting Newtonian particles. (14. We start from the Schrodinger equation ih <9* =HV at with H=£— p2 2mo + V(T).t)\2 = \iP(r)\2. the energy has a definite value E with V(r.V(r)]^(r) = 0. so that _ Aip AA+^V• h AA+CVAn (AVS) + IVS h VS + AAS) iS(r)/h (14. the function S being called phase function. Classical Limit and WKB Method E — V changes sign).7.282 CHAPTER 14.
4) (14. we obtain the following two equations: E .6) 2mo dt now with A also a function of t. (14. h2 ih— = H$ A + V(r) * .+ VAVS + AAS = 0. h h (14.13) . V(r.1) and separating real and imaginary parts. We set in this equation.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. These equations are equivalent to the Schrodinger equation.10) m 0 ^ .14. (14. (14.Q A 2i 8A 1 (14.2 Classical Limit and Analogy with Hydrodynamics 283 Inserting this into Eq. h h Taking real and imaginary parts of these equations we obtain the equations: + dt and 2rriQ 2m. We can identify Eq. (14.t)}2. (14. (14.t) = and obtain dA i_dS_A + dt h~dt 2 h r A(VS)2 AA h2 2mo ih A(r.11) We define as probability current density the vector quantity J:=K where V* = ijf* V* rnio i_ „ *.V(Y) + and 2VA • VS + AAS = 0.12) VA A (14.t)eiS{r't)/h.10) with a continuity equation by defining as probability density p:=*** = [A(r.* VAVS+AAS + V(r)A.
Classical Limit and WKB Method This means # * — V * = — AVA + — and so J = — A2VS.15) and (14. TTIQ (14.16) this becomes m 0 1 rr + m o V . we have mo—(A2) + V(A2)VS + A2AS = 0. %(1"1' I<2) and V J = — V • (A2VS) mo Prom Eq. + h2 AA V = — ^ . m0 But now (14. (14.9) as 9S 1 mo^ 2 + T. (14.20) With this we can rewrite Eq.18) ot ot This is the equation of continuity that we encountered already earlier. (14. Since we obtained this equation from Eq. mo <™> (14. (14. ^ + V • J = 0.J = 0.16) BA 1 moA—. ot 2 Multiplying this equation by 2 and recalling that the function A was introduced as being real.14) A2VS.9) for the phase function S.. the implications of this equation are also those of the equation of continuity.+ AVA • VS + A2AS = 0.21) . ie. . (14.19) m0 m0 it is suggestive to define the following vector quantities ± = at v := I p = JVS.17) With Eqs. Since J = — A2VS = PVS.10). (14.284 CHAPTER 14.10) we obtain = — [V(A2) • VS + A2AS}. (14. (14.n. Next we investigate the significance of Eq.01. (14.
which for very small values of h yield the dominant contributions.2 Classical Limit and Analogy with In the limit h2 — 0 this becomes > Hydrodynamics 285 ^ + W 2 + F = 0. that the quantum theory implies an essential singularity at H = 0 (i. 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. (14. or that of equilibrium in electrodynamics) and dS/dt = —E (from ^ oc exp(—iEt/K)). This expression with h in the denominator shows. and hence''' ^(m 0 v) + m 0 (v. In the first • place we require the ansatz (14. This is the analogy of the Schrodinger equation for h2 — 0 with hydrodynamics. d — = (14. v dt dt so that the equation can bedt written dr dt ~(m0v) + W = 0. they are asymptotic expansions.1/h. Such series are typical for expansions in the neighbourhood of an essential singularity. so that V ( v • v) = 2(v • V ) v + 2v x rotv.22) (14. the wave function \I/ describes effectively a fluid of particles of mass mo. V • J = 0. . with the help of Eq.20). > In the case of stationary states we have dp/dt = 0 (which is analogous to the condition of stationary currents. However. We deduced this equation solely from the phase S in the limit h — 0.25) dt We recognize this equation as the classical equation of motion of the particle of mass rriQ. 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.e. The equation of continuity then describes the stationary flow of a fluid. An expansion in ascending powers of h therefore starts with contributions of the order of 1/h2. (14.V ) v + V 7 = 0. > The motion of the particle or rather its probability is altogether given by the above equation of continuity. This result is very general. r o t v = ( l / m o ) r o t g r a d S = 0. This means. we obtain ^ V 5 +  v ( v v ) + V V = 0.24) d 1 dr d •— = d Lv _ . The particles have no interaction with each other.23) (14. dt 2 Constructing the gradient of this equation.14. a singularity different from that of a pole). where.e.2) for the probability amplitude. i. 'Recall the formula V ( u • v) = (v • V ) u + (u • V ) v + v X rotu + u X rotv.
27) (14. B. Dingle [70].286 CHAPTER 14.5)) and 2 ^ f dx ax + All=0. is an asymptotic expansion. i = eiW(x^h. Sec. As remarked at the beginning.3 14.3. i. 318.1 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. We consider the onedimensional timeindependent Schrodinger equation y" + ^. or h2/2mo(E — V).2. There are also other more specific descriptions.In A(x). Messiah [195]. the method is widely familar under the abbreviations of the names of Wentzel.[EV(x)}y = 0. (14. p.26) We observe already here that an expansion in rising powers of k. 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. (14. h2 l 2m0 {EVy corresponds to an expansion in decreasing powers of (E — V).§ 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. 6. The expansion in powers of h2. We set.e.4) and (14. similar to our earlier procedure with A and S real functions. Kramers and Brillouin.30) ax *For details see R. . Classical Limit and WKB Method 14. (14.28) The calculations we performed at the beginning imply the following equations (compare with Eqs. such as "LiouvilleGreen method' and "phase integral method'. §We follow here partially A.
Li This expression can be substituted into Eq.2mQ(E . (14. (14.34) Equating coefficients of the same power of h we obtain (S'0)2 = 2m0{E .33) 3 (S")2 .10)).2rn0(E V) = h2 lS'A 2S'0 (14.4 (S>)2 1 S'" 2 S> (14. Integrating this second equation we obtain ll[dX so that = \l^dX' A = c(S')" 1 / 2 .14. (14.29) becomes (note that this is still exact!) (S')2 . Since the expression (14.32) s" = sz + h2s'( + •••. r//\ 4V^ 2 Up to and including contributions of the order of h2 Eq. + (14.31) \a.28). (14.3 The WKB Method 287 Equation (14.V) = h2 In the WKB method one now puts S := S' = S0(x) + h2S1(x) + (h2)2S2(x) S'0 + h2S[ + ••• .V) and (14.30) corresponds to the equation of continuity (i. 3 / 5 .A = const. .35) can be integrated.35) .36) Equation (14. (14. Eq.32) is therefore (S'0)2 + 2H2S'QS[ .\ In S'.e. 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.
33) to make sense as an asymptotic series.19). y/2mo(V(x) .and So = ±h / — — A J k(x (14.40) rx dx v "7 W)\. (14. (14.41)? We note first that the expansion in powers of h2 does not converge. B. The solution is seen to be periodic provided k(x) is real. 1 S e e R.± ln(S'0 + h S[) + In c (s'0y/i and hence exp :5 0 + 0(h2).41) We now ask: What are the conditions of validity of Eqs. . it is reasonable to set (cf. Dingle [70]. Chapter XIII.E)' Then y(x) = V^(a 7 exp dx + 6 exp T(x) (14.288 CHAPTER 14.e.39) and (14.27) we obtain now with Eq. Classical Limit and WKB Method has the form of a momentum.39) where a and /3 are constants. For V(x) > E (this is the classically forbidden domain) we set l(x) = ^==^====. y = c'{k)1/2 exp or y(x) = a \ A ( z ) cos I / dx W) dx + 13).I In S' + In c h 2 2 2 exp ^{S0 + h Sx) . i.38) h y/2mo(EV)' (14. Eq.' For the expansion (14. (14. p = h/X) k{x) :Then S'n +. (14.37) For Eq. E > V(x). we expect that in any case So » ft2Si. (14. (1. and instead is asymptotic which we shall not establish in detail here.31): y = exp = ~ ~iW(x) h exp L nH™ exp ^S .
36).42) [(^).14. (14. Setting R := A' = \moKV'{x)\ \2m0{EV{x)\W .45a) .3/2 s&T + 3 {So Il« we have (cf. Eq. Since k(xy 1/52\S'0 (14. E > V(x).36)) 2s'0s[ = [(s&r 1/2 ra) 1/2 With S£ = ±ft/A. 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 (14.43) The condition So 3> ^2<Si therefore implies that /f>^i/^)k = H/y/2m0(E is sufficiently small.V{x)) ^ This condition is satisfied provided E or E — V{x) is sufficiently large or (14.3 The WKB Method We have 289 S0 = ±h and Si follows from (14.1 / 2 ]" [§0SS).
14.45b) so that the condition So » h2S\ is also satisfied by demanding that i ? C l . (14. We can derive the equation whose exact solutions are the WKB approximations of the Schrodinger equation. h = 1) . Roughly speaking one therefore requires E to be large in both cases. Classical Limit and WKB Method we can rewrite the inequality (14. In the neighbourhood of E = V(x).290 CHAPTER 14. How do the solutions of Eq.44) as Idxsj2mQ(EV{x))l 1 + ^R2j » ^Rh. In terms of X the original Schrodinger equation is y" + = y = 0.48) behave in the neighbourhood of such a point called "turning point'.e.V(x) = V(a) . at which E — V{x) changes sign? We note first that in this case near x = a: E . (14.2 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). V(x)) (14. and hence (with UIQ = 1/2.48) The (dominant) WKB approximations are exact solutions of this equation. In a similar way we have for E < V(x): l'(x) <C 1.3. (14.47) Differentiating this twice — we omit the details — we obtain the equation y" + l A2 (A 1 / 2 )" *i/2 y = o.46) The desired equation has the solutions y(x) = aXll2 exp ±i dx / (14. around the classical turning point. the WKB approximation is in general invalid (see below). *2 A(z) = V / 2m 0 (£ .V{x) ~ Or .a)V'{a). i.
a ) 9 / 4 ' 291 (v^r 1 (x — a>2 These relations imply that Eq. 14.48) possesses a singularity of the form of l/(x — a)2 at every turning point x — a. We define solutions ?/i and ?/2 with branches to the left and to the right as follows (for . We assume we have a turning point as indicated in Fig. 14. since this requires different approximations there. In the immediate neighbourhood of such a point the Schrodinger equation can therefore not be replaced by Eq. in the domain E < V the WKB solutions are exponentially increasing or decreasing. i.14. 14.48). The transition is provided by matching relations. in the domain E > V the WKB solutions are oscillatory. 14. 14.1. (14. As indicated in Fig. and on the other side. On one side. We shall not derive these conditions here in detail but rather make them plausible in Sec.3.3 after stating them first here.1.1 there are WKB solutions on either side of the turning point and some distance away from it.1 The regions around the turning point at x = a. It is therefore necessary to find other solutions in the neighbourhood of x = a and then to match these to the WKB solutions in adjoining domains as indicated in Fig. which connect one domain with the other.e. (14.3 The WKB Method with the derivative relations 1 1 a)5/4' 4(x 1 5 16 (x. V=E Fig.
H— J =^ v/exp + m (14. has the same asymptotic behaviour in the domain i « a a s Ay\. (14. in order to make the formulas independent of whether the potential is rising or falling. a interchanged. Note also the extra factor 2 in yi. as in Fig. one could simply replace everywhere /•a / pa • • • b y / Jx J x . 14. 14. Thus.1) is then: VI exp {~[T) Vxcos fx dx TV Ja ^ _ 4 (14. Classical Limit and WKB Method definiteness note the + signs): Vtexp yi ~ S + Jx I X for for i « a . (14.50) x S> a. and E ^ V for a ^ x the same formulas apply with the same direction of the arrow. (i x dx 7T ~~k~ 4 / One should note t h a t the solution ^4?/i + By2 for A ^= 0.292 CHAPTER 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. This means the asymptotic part suffices only in t h e exponentially decreasing case.49) x 3> a Vxsin and 2 exp V2 VACOS (L X dx A ra dx X for for i « a .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 / —. however with the limits of integration x. P u t differently: We could add a lot of contributions to Ay± without affecting the asymptotic form of Ayi.52) For a potential rising from left to right. The first W K B matching condition in the direction indicated by the arrow (and for the potential decreasing from left to right.
2 The overlap regions around the turning point at x = o.3 Linear approximation and matching How do we arrive at the matching relations given above? We observed above that Eq.3 The WKB Method 293 In that case Eq.14. (14. the equation V approximately linear around turning point V=E Airy solution WKB .e. whose exact solutions are the dominant WKB approximations. the equation becomes approximately y" = (xa)xiy. If higher order contributions are to be taken into account.Airy overlap* Fig. One should remember that the matching relations above are those for the dominant terms in the WKB expansion. i. as we saw. y" + Toy = o.54) .46). 2 1 X (14.3. (14. 1 ~ (x .48). the matching relations have to be altered accordingly. 14. Thus around that point we can not use the WKB solutions and hence have to find others of the original equation (14. is singular at a turning point x = a.53) Since at x ~ a.51) — always with the decreasing exponential on the left — is valid in both directions. 14.a). Therefore we consider now this original equation in the domain around x = a. (14.
J (X (i4 58)  Replacing in Eqs. Stegun [1]. formulas 10. — a ^> 0. i I 772 \3 + > for ia>0. A. > for a. In the following we require the asymptotic behaviour of the solutions Ai(—z). or to the literature:H 3/2 2^{zy/* Ai(z) ~ < ~~FL~T77 y/TTZ1/* 1 1 s e K*) n for ~.^ 4 I 7T x — a <C 0. . (14. In particular we encountered a solution written Ai. (13. p.1 / 4 by \^.291.a)X\/3. or M. p. Cf.4. With the substitution z = (x . (14.56) We have discussed some aspects of solutions of this equation in Chapter 13. B.57a) and (14.59). R. as > may be seen by referring back to Eqs. (14. Dingle [70]. (14. (14.57b) UV2 3 11 by J dx .57b) To the same degree of approximation as the equation y" = (x — a)xiy.Bi(—z) for \z\ — oo. we have r x * f x ' /2(x a)i/2d*=^(x . Abramowitz and I. (13. Classical Limit and WKB Method where xi is a proportionality factor which we choose to be constant.a)3/2 ^ /2 =^ 3/2 o Q. This is given by the following expressions.57a) COS 2^2/3 _ SZ 1 1 ?(z) /2 3 es^ Bi(z)~< 1 1 for x — a C 0.60).294 CHAPTER 14. 448.55) the equation becomes the Airy differential equation d2y dz2 zy. A second solution is written Bi.5960.
As an illustration of the use of the WKB solutions we apply these in Example 14. we enter the domain of validity of a WKB solution.3 The anharmonic potential well. At and around the turning point the solution of the Schrodinger equation is therefore given by Airy functions.1 to the quartic oscillator. 14.52). These solutions. Example 14.g.l. F T' dx (Z)1^ by ^ harmonic oscillator V=x 2 +x 4 V=E linear approximation Fig.59a) . and hence must be proportional (in view of the uniqueness of the solution there).(zf2 by J we obtain the matching relationsX (14. 14. by Ai(z).51) and (14. As we approach a limit of this interval in going away from the turning point. (14. In this sense we have now verified these relations.14. The latter domain is a small region where the solutions from either direction overlap.3 The WKB Method and 295 •. In the direction to the left as indicated in Fig.l v/2m0(EV(x))dx 7r = (2n + l). n = 0. In this way different asymptotic branches of one and the same solution are matched across a turning point.1 the Airy solution becomes proportional to the exponential WKB solution and to the right to the trigonometric WKB solution. e. We summarize what we have achieved.2.2.1: 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 neighbourhood of a turning point the Schrodinger equation becomes an Airy equation (in the leading approximation). 14. however.. are only valid in a small interval around the turning point at x = a as indicated in Fig.
.51). In the neighbourhood of the minimum at x = 0 we could approximate the potential by the harmonic oscillator. Hence we search for solutions which are exponentially decreasing in the far regions. 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 ( . the continuations of these functions into the central region 2 are: ya(x) = c v A c o s I / J.(x) = c ' v Acos I / — I for a < x < b. yz{x) = c'vlexp ( [x dx\ I — / — ) for x 3> b. (14. Sometimes this distinction becomes imprecise. for which the WKB approximation surprisingly yields already the exact energy eigenvalue as one can verify by evaluating in its case the BohrSommerfeldWilson rule. regions 1 and 3 in Fig. 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.3. (14. 1 . The condition for this is ya{x)=yb{x) for xe(a. (14.b). Let E = V(x) at x = a and x = b. 14. 14. . j/i.44) this implies large values of n. . n = 0 . According to Eq. The potential now has to be inserted into this condition and the discrete eigenvalues En of the problem are obtained.61) Evidently these functions have to continue themselves into each other (since the wave function has to be unique). 2 . b are the two turning points. extends only over a length of very few wavelengths and is therefore valid for n small.y o dx TT T provided 6 _ 4 /. .^ . (14. . . I " ' (K rx dx iz\ _ ( (b dx_ 7T _ fx dx _ • f / fb dx\ f* dx\ ( fb dx\ .59b) Solution: We consider an anharmonic potential well as depicted in Fig. We argued earlier that the WKB method is suitable in the case of large values of E.l. Then around these points the potential would be well approximated by a linear potential (as we saw above).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 .. whereas the approximation of the eigenvalues by comparison with the harmonic oscillator in the domain of the minimum. /n = s u ^ TJcos^ya TJ+cos^a T J . (14.296 CHAPTER 14. The corresponding eigenfunction would be approximations of the proper eigenfunctions around the origin. 2 n = 0. 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.e. In these domains the dominant WKB approximations are 1 /j/i(x) = cVlexp / fa dx\ I —/ — \ r for i « a.60) Using the matching relation (14. Here we are interested in the discrete states (the only ones here).2. i. 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. Classical Limit and WKB Method where x = a.3. ^ A = (2n + l ) .
equivalently.x number of turning points h. In this socalled "old quantum theortf (i. . Consider the Schrodinger equation in the abbreviated form d2iP(q) dq2 f(q)t/>{q) = 0. 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. Wilson by supplementing classical mechanics by Planck's discretization. A.1.4 Three cases with different pairs of zeros.14.63a) or. B.x number of turning points TTH. This relation is sometimes wrong. (14. Dingle. . (14. and knows of no published form or script.63b) *The author learned this in lectures (1956) of R. Sommerfeld and W.e. 2. 2 . (c) V Fig. . . Bohr. with n — 0.64) (14. pdq = n + 1 — . The corrected form* is.4 BohrSommerfeldWilson Condition 297 14.1. n = 0.4 Bohr—Sommerfeld—Wilson Quantization A quantization condition which is very useful in practice and in a wide spectrum of applications is — and remains in spite of its old fashioned reputation — the quantization condition established by N. 14. ri2 I Jqi :pdq n + 1 — . .
66) J<1\ Thus in this case the socalled old quantum theory is correct. this is an exponential zero.e the relation (with a shift of TT/2) 1 2/V4 exp Jqo f9^W)dq 1 1 1 4 (Z) / sm Jqo (14.67) Case (b): Next we consider.e. i. this is a trigonometrical zero.65) We use the knowledge of these solutions to find the eigenvalues of the equation. 14. together with q — — oo. (14. Since sine and cosine have zeros spaced at intervals of TT. / • n an integer > 0. (14. this means dqyf—f = (n + l)7r.e. i. where /(g) is positive. We have the case of trigonometric solutions. (14.51) the linkage between the trigonometrical and the exponential solutions across the position go in space at which /(g) = 0. Jqi ^ fpdq J =2 pdq = (n + l)h.4(b). —/ = K J h It follows that / pdq = (n + l)Trh= (n + l)h. 14. where /(g) is negative. In quantum mechanics we have dq2 + Ki/j(q) = 0. if /(g) is negative. 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. 0(g) ex . the wave function is to vanish at points q = q\ and qi with the function /(g) remaining negative in between as illustrated in Fig. There are three cases.e. In this > case the wave function for g > go is . as illustrated in Fig. Before we continue we recall from Eq. where n —. i.298 CHAPTER 14. i. the case of tp(q) — 0 at q = gi. Case (a): The case of two trigonometrical zeros.4(a).
e.77T (where we reversed the order of integration to obtain a positive integral and multiplied through by minus 1). 14. Since both cases refer to the same region qo < q < q'0. (14. the sines must be proportional.63b). Hence we have pdq= i n +  .14. Solution: As in the case of the quartic potential.69) pdq = (n + l ) " 4 h.63b) to obtain the eigenenergies of the quantized harmonic oscillator.70) We can therefore summarize the results in the form of Eq.67): • the wave function for q > qo must be proportional t o (Z)1/4 sin [ ^W)dq+>x Jqa . ri\ / J do y/f(q)dq (n + 1) 1" (14.e.2: The harmonic oscillator Use the relation (14. E x a m p l e 14.68) (14. 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 .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. ^~f(q)dq Jqo LJqo + 7T J'% On fq / i 1 V~f(<i)dqjK (n + l)7r. i.4(c).63a) or (14. (14. Hence we obtain the condition I'o Vf(l)dq 90 (n + 1) IT. Thus in this case the socalled old quantum theory is wrong. i. For the wave function to be exponentially decreasing to the left of qo and using Eq. (14.2 and 14. and • • for the wave function to be exponentially decreasing to the right of q'0. this is the case of two turning points.3. Case (c): Finally we consider the case of i/j(q) = 0 at q = ± o o exponentially as illustrated in Fig. As illustrations we consider Examples 14. Therefore their arguments differ only by (n + l)ir.
where g is the acceleration due to gravity. so that E = hu{n + 1/2).2 3 J x (n + 3 / 4 ) 2 / 3 . Calculate its quantized energy. :: V=m0gq Fig.300 With potential V(q) = 2ix2mov2q2. whereas a turning point does. Thus there is one turning point at E = mogq.28 x 1 0 . permit this although with rapidly diminishing probability.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.5.63 x 10~ 3 4 J s . 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. Solution: This is the case of one trigonometrical zero of the wave function ip at q = 0 and an exponential zero at q —> oo. one obtains E ~ 2. ^Thus a trigonometrical zero is the condition of absolutely no penetrability beyond it. in principle.V(q))dq The result is E/u. The WKB Method this requires evaluation of the integral with qi = ^' E/2K2mQu2 / 2 m 0 ( B . g = 9.807 m s . Example 14.5 A particle above a flat Earth. 14. 4 / x CHAPTER 14. and h = 6.2 . .71) For m 0 = l x 1 0 _ 3 k g . as illustrated in Fig.
5 Further Examples 301 14.74) 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 . ra 2/ = 0. 1 .2. d dE V)={n+)n.73) Jx! rp2a dx I^<E 2 v) = ^ n — = . E x a m p l e 14. (14.= — <*— = h(n+\.) M0 N = l + n+1 : + • 2/B Hence dE K 2 d dE[ M0 2VB.( 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 (14.72) \\2dxp=Qdx^{Eh Hence.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 ( — = .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 £ = .14. h M0 _h 1^3/a ~ 4 0 / 2 J V \ 3 _ 2N3 \M0) ~ ~Ml ' .f V ft h JXl h 2TT (14. (14. Solution: The proof is contained in Example 18. and u> is the oscillator frequency in either well.l. In the case of screened Coulomb potentials (cf.7. 2TT W 2K J p dE V where p = ^/2mo{E Solution: We have — V). 9 .AE(q0 2 = 2n + 1) = — exp 7r dzy/E zn + V(z)/h where q=zo are the left and right barrier turning points.75) which verifies the formula immediately. with p = modx/dt. Chapter 16) one obtains the quantization relation (for Mo = const.5 Further Examples E x a m p l e 14.
y2)(y . and (b) evaluate the WKB exponential exp[—2/b arr j er ] and the WKB prefactor 2/ w e u.133)).~ \ (14.6. Solution: The potential has a finite minimum at q = 0 and a finite maximum at q = 2a2/3 as indicated in Fig. It follows that the energy between the minimum and this maximum lies in the range 0 < E < £ m a x = ^aSb2.a . 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 .6: WKB method applied to the cubic potential Consider the cubic potential V(a)=1b\\a*~q). (14.6 The cubic potential. where .78) (a) We determine first the three turning points at qi = <?i. Fig.79) .y3) = 0.302 and T CHAPTER 14. Example 14. where 1 2 • 1 2 y = Q.e. Eq. q = y+ a . 14. The WKB Method ^ 2 JV 3 2TT ~ ~Mjf (14. (14. i.77) Determine (a) the turning points. (11. which determine the imaginary part of the energy. 14.<?2i<?3 for E > 0 given by E — V = 0.76) in agreement with the result obtained from the classical Kepler period for the Coulomb potential (cf. The q term in this equation can be removed by transforming the equation to a cubic in y.
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.a 2 [ l . (b) W e see t h a t gi a n d 93 m e r g e t o 0 w i t h e — 0. tan60°=%/3. 92 = a2[l .q)(q .\tan0^ v/3.c o s O + cos(6» . h = 1) ^barrier := / d( }J^2~(E~v) = ^~= dq^/(q .08. D .93 2a2 [ . a n d t h e a n g l e 0 = 0° t o E = £ m a x . H e r e cos(<9 ± 120°) = . 92.81) m a 91 . B y r d a n d M .T h o m s o n [196]. g 2 ~y{3}>0.2/3 i n t o q = y + a 2 / 3 .ui=K(k).1 / 2 .2 c o s 0 ] . p . 78: sin 120° = v ^ A c o s 120° = .93).9 2 [ . 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. M .85) T h e angle 9 = 60° c o r r e s p o n d s t o E = 0. ' P . 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 .cos 0 + cos(6> + 120°)] = a2 92 . 80 a n d 361. F .0^6O°.2 c o s ( 0 . 212. we set 0 = 60° — e. 93.g i ) ( g 2 . 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. 14. ui=sn 1 \l— — = l. (14.80) I n s e r t i n g t h e r o o t s yi. we h a v e qx = . i. (l_fc' 2 a + fe'4)l/4. q3 = a2[\ .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 . F r i e d m a n [40].g i < 1. § T o h e l p we cite H. (14.2 c o s ( 0 + 1 2 0 ° ) ] .c o s ( 0 + 12O°) + c o s ( 0 .= s i n t 3 v 3 (14. (14. gi ~ y {V3e} < 0. p .120°)] = a 20? 2a2 3 2a 2 • cos 0  91 . V3 + t a n 0 2 . p .04.' : ± 6 0 ° w h e n E = 0.gi = 1. V cose' + sin 8 (14.{ V 3 e } > 0. M i l n e .9 3 — = = 2 " 3 " [ . B . 91 < 93 < 9 < 92(14. 6 = 0 i m p l i e s E m a x of E q .6. p . 0<6O°.78)) c o s 30 •• 4P3 ( ' \ aeb2} .84) 92 .§ ( c o s 0 ± \ / 3 s i n 6 0 a n d § sin 6 sin# (14.k 2 2tan# \/3 + tan! . 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. 37.9 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 .9 l ) f f / du s n 2 u en2 d n 2 u . e > 0. 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 . q3 ~  . W e o b t a i n ^ fbarrier : = 6(92 ~ 9 3 ) 2 ( 9 2 .3/2. f o r m u l a s 236.5 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. (14. D w i g h t [81].< 1.120°)] = .14. .1 2 0 ° ) w i t h 0 < 6 < 60° for 0 < E < £ .79) a n d t h e n s e t t i n g m o = 1. _ \ / 3 — tan(9 2 V92 .e.
72. E(u = K(k)) = E(k).89) This expression will again be obtained later with the help of configurations called "bounces" (cf.= —^ = k a2[cos0+^] v ^ + tanfl V=". V?2 .91 . sinV = q3 qi ~ = 1.2)*w+2(fe4+k'2)E{k)] v ' G(k) For one complete round from one turning point back to it.91) a2[cos0^] y/3tanfl ? L .87) Inserting G(fc) together with the expressions for the prefactors into (14. A:') = # ( * ' ) • (14. Next we evaluate the required integral across the well from 91 to 53 at energy E.^ + ^ ) (l + fc'2)2 . as in (14. 2 2 . Thus Jwell : = [13 1 f13 rf 1 / dq J^^EV) 1 y/2 1 V 2 f[13 = — —. V 93 .) .2)K(k) + 2(fc4 + k'2)E(k)) k ^i — . Friedman [40]. G(fc) =86aS (ITS£ li[fe.S . E(u) the incomplete elliptic integral of the second kind. Thus Iwen=lgF(ip.^ V3 / V V3 y a / c o s e J.84) we obtain Carrier = 2a 2 sin6> 2 b( . E(u = 0) = 0. Whereas the integrand of the above barrier integral is effectively a momentum.sm t %/3 £\. 0 (14. the integrand of the integral across the well is effectively the inverse of this momentum.?)(?3 . Byrd and M.fc'2 + fe'4)"1/4.304 CHAPTER 14.90) hi JlnynEv) V(qqi)(q2 Vz b Jq q) I" bJqi where P fe = 9391 9291 = . D. 91 < q < q3 < 92 bJc.qz) (14. (14. we obtain in the limit of E = 0: j d ( ^ ( E „ y ) = 2/barrier = ^ ! . Eq.k). V3 cose + ^ = V3 2 1 + 72 v/l + 3 7 4 C (cose) 2 = 1 + 3 7 4 = 4 d .32)).~ y ^ ) a< [ COS 5 + .86) 15 tan. The WKB Method elliptic integral of the first kind. (14. formula 233. p. (14. 1+72 = and J o s 0 + ^ = (1 .92) ''P.85). ./ dq~ ^2 b JQl 1 \/(qqi){q2q)(q. We can evaluate this integral again with the use of Tables of Integrals.00. (24. E(k = 1) = 1) (•u\ = K{k) i Q G{k) = / Jo Next we set s dusn2ucn2dn2u = j[k'2(k2 15fc4 . .91 F(TT/2. with K(k = 1) = oo. F.2(fc2 . g= 2 .
** However.93) where w is the harmonic oscillator frequency in the well. in fact through logarithmic contributions contained in the argument of the exponential for k ^ 1.88) in rising powers of fc' .96) Expanding the coefficients of the elliptic integrals in Eq. we obtain 15 . (14. .113.97) 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. (14.4 2n + 1 fc' = 4 — ^ — .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.99) **For the potential approximated around the origin as V ~ a2b2q2/2 the eigenvalues are E = fat)(n + 1/2) with it) = ab. formulas 8.3. we cannot expect the ratio exp(—2Jt>arrier)/2J'well oc w exp(—27barrier) evaluated at k = 1. we obtain > .80). (14.2. pp.5 Further Examples Hence (using K(0) = rr/2) 305 ab ab (14.l. and hence at E = 0. 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.3 and 8.85) (14.94) C °S3 2 (lfc2 + fe4)3/2  l+ g f c ( l + fc) + Comparing this equation with Eq. . (14. 2a 5 b K(k) + 2 E(k). With q = 1/z the integral (14. Gradshteyn and I. to represent a physical decay rate.14. M. Thus one obtains from (14. 906. S.63b).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. Ryzhik [122].2 IE 4 fc' ln 8 (" «V iU^fc' 4 . 905. we obtain E a6b2k'4(l + k'2). 32 (14. To obtain the latter we have to use the quantum mechanical expression approximated by E = hw(n + 1/2). a°b n = 0. The same expression for E is obtained by applying the "method of poles at infinity" to the BohrSommerfeldWilson condition (14.! With these expansions one obtains ^barrier = 15 ahb 2 + 8A .114. (14..
It follows that (for one complete orbit back to the original turning point) . +3 the result becomes exP[2Jbarrier] = ± i _&g_ (25a5br+J e .( n + l ) e _ JL a * 6 and we obtain for the ground state (ro = 0) .e. (14.. .102) .7 ..j .e.178)) call the Furry factor set equal to one.e x p [ . oa ba Thus with the W K B method we obtain exp[2/barrier] — barnerj = 8 s.100) 'barrier = ~0°b In [ ^ — j .73).J + 0(VoJb).101) (14..Wigner formula (10.2 7 b a r r i e r ] Z Z7T where (cf. n ~ V27T .11)) dq 2TT y/2§?(Ev) It would be interesting to derive the same quantity with the perturbation method and to compare the results.306 CHAPTER 14. i.„.fc'2 + fc'4)1/4 * ± t ^ . V (14. _ T T • (14. also Eq. 7 ~ —.103) 27 w e l l 27rVn+y ' fn With Stirling's formula in the form of what we later (with Eq. this has not yet been done. (18. 2n + 1 — / 26a5b \ . _ exp[2/barrier] >. (0) i hw En = E\> . so that we obtain 4 ./^~^l^ab ^aub . (20.— (14. /25a5b\™+5 e exp[2/barrier] ^ ( r ) Correspondingly we obtain for the full period 2/ w e ii: 2/well = ? K ( f c ' ) ( l . 15 a . We observe that with \j2je as 1.> tf = = 2/wfiii = ±tv8a°b——P. i. The WKB Method For k' —• 0 the first and the third terms dominate. the result agrees with the ground state path integral result (24.96) for the imaginary part of the energy as in the Breit.. .ba / 25a5b\n+2 ±l_l eT5 a f >. 8 Bfc .105) As argued in Chapters 24 and 26 one expects this W K B result to agree with the one loop path integral result using bounces.
Rosner [269]. Quigg and J. that this is indeed the case. which permits only bound states. D. the Coulomb potential. Lane and T. Thacker. after the realization that quarks as their constituents might not exist as free particles and. C. B. The study of this potential. K.Chapter 15 Power Potentials 15. L. now known as quantum chromodynamics.e. at least indirect. T. This would mean that the force binding the quarks together would not decrease with increasing separation as in the case of e. There are numerous.g. that it may not even be possible to extract individual quarks experimentally with any finite amount of energy. [233] and H. indications from the nonabelian generalization of quantized Maxwell theory. Quigg and J. See particularly E. This potential became widely popular in the spectroscopy of elementary particles. Rosner [230]. Eichten. [231]. K.* 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. i. L. led to a spectrum which contains only scattering states. 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. C. 307 . In the present chapter we consider briefly the threedimensional potential V(r) oc r = r. [232].1 Introductory Remarks The particular linear potential we considered in Chapter 13. Yan [82]. the classification of the various states of nucleons and mesons and other particles.M. along with inclusion of angular momentum and spin effects and their interactions. Gottfried. describing free fall under gravity. in fact. Kinoshita.
</>)R(r).2) tt(r) = Ylm{9.2 The force The Power Potential F = . 15. Power Potentials general power potentials.e. We leave consideration of the logarithmic potential to some remarks and Example 15. In order to have a clear starting point. A > 0.308 CHAPTER 15. mi + m2 and r is the relative coordinate. v > 1. m\m<i M= : . R(r) = ^ . and u'(0) > [u(r)/r]0. These particles therefore cannot be separated by any finite amount of energy.5) u'(0) u[r) r=0 = R(0). we consider immediately the more general power potential V(r) = \r\ 'We have u(r) = rR(r). and so u'(r) .e.r acting between two particles is constant and directed towards the origin (of the relative coordinate). we return to the Schrodinger equation in three dimensions. h2 2n V 2 * ( r ) + [V(r) . and obtain 2/x 1(1 + l)h2 u"{r) + EV(r) u{r) = 0 2 2/ir (15. i.u ~ r i + 1 (/ + l)R(r).4>)ru{r) = Ylm(9. For a central potential V(r) we write (15.£ ] * ( r ) = 0. where \x is the reduced mass of the two particles of masses mi.6) = (I + l)u(r)/r = R ~ rl.1. f dr{u(r)}'2 1.1) i. (15.4) Instead of considering the linear potential.3) with the boundary conditions''' u(0) = 0. This force maintains this value irrespective of how far apart the particles are separated. (15. ~ (I + l)rl (15. (15.V 7 = const. and the normalization [ dr\y(r)\2 = 1.m2.
(15.1.8b) This condition implies that we have only one pair of turning points.. which has been reduced to an effective onedimensional problem in the polar coordinate r.9) into the threedimensional case in the following form: / Jo C dry/2n(E . x > 0. or f l H " ^ ) ! .63b) for the onedimensional Schrodinger equation. (15."MA.V(r)) = (2nl) +1 N.. (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. We also assume that V'(x) > 0.2. (14. Then k(x) = k(x). (15. (15. d5. We therefore proceed as follows. 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. is to be interpreted as the principal quantum number in three dimensions.. . This means that we transcribe Eq.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 (14.2 The Power Potential 309 In Chapter 14 we obtained the BohrSommerfeldWilson quantization condition (14. so that with Eq. . . (15. 3 .8a) ^ JO = (2Ar + l ) £ . odd integer ^=(nll)vh.4) in the threedimensional case.15.63a).10) Here rc is the turning point and n — 1.63b): 2 and V(0) = 0 (15. N = 0. In order to to obtain the .7) k(x) = h/y/2m0(EV(x)). and the turning points are at a — — b := — xc. We now want to find a quantization condition for the threedimensional problem.2.
12) and (15.6) we have to evaluate the > integral / : = [ ° dr^2n{E\r»)= Here the turning point rc is given by E .y) T(x)T(y) _ T(x + y) ("!)/" /•! / d"(1"")/. Power Potentials eigenvalues E —• En for the radial potential (15.\rvc = 0 . (15.14) Prom Eqs.17) The integral on the right is the integral representation of the beta function B(x.13) we obtain _ X_ „ _ so that A£_ (15. . dr = .310 CHAPTER 15. _ —vr" *dr. s. Now. (1516) (15. (15. 1W or Xv \E I It follows that (I) / ^""^(l*)1/2.12) i/z Jo We set dr 1 l A v ~ E \ (Et\l/v so that (15. E A fv.13) dt . c (nj\irh. (15.15) I ' = ^ 1 (( 1) B{x.11) rc = (15.y) defined by (xl)\(yl)\ y ^ (x + y .'(1*)1/2= J1F\lt)y1dt.
Weinstein [281]. T(z + 1) = z\. 15.1.15.* In this limit the eigenvalues become:^ E„ (nh^ni) Lr(f)r(i)^7^ rv"ir •n2 2/i 2^2 (15. (15.20) These are precisely the eigenvalues which one obtains from the vanishing of the (periodic) eigenfunction at the wall of the square well (of course n — 1/4 has t o b e replaced by n since the W K B approximation is only accidentally correct for small values of n ) . 15.1 Approach to square well with v —> oo.2 The Power Potential 311 so t h a t by comparison y = 3/2 and x = 1 + (1 — u)/i/. and n — — ]7r/i.1)!. ^ '"Compare with Eq. $T(z) = {z. (12.19) or En = _ 1)^(3 + 1)^/^2^+2) r()ir(i)^7^ • v 0 1 Fig. for instance simply graphically as indicated in Fig. One can convince oneself now.18) \v \E or 3 E2 r(§)r(i)V2M (n ' (15. The case v = 4 is that of the pure anharmonic oscillator also discussed by M. .55). t h a t in the limit v —• oo the potential approaches the shape of an infinitely high square well. i r ( i ) = r ( l + i ) = ( i ) ! .
61) h2 = I. the case of the linear potential. the harmonic oscillator. (13. we obtain En = (n . (6. since these are the ones we are interested in (cf.21) (4nl). x > 0. Physically it does not make sense for the probability amplitude to have a discontinuity there.mo = \.1 + \)hx> = {An • l)£fiw. (15. In the onedimensional consideration these are precisely the odd wave functions as illustrated by an example in Fig.2. = (n>r(§)A 2/3 3TT \h n 2/3 L r(§)v^7^ J (15. (6. so that the appropriate Schrodinger equation becomes d2(j) + (Edx2 x)<t>(x) = 0 .//L This result agrees with that for the onedimensional harmonic oscillator r2 in the form" cf>"(r) + ^(E\r2)cf>(r) =0 (15.25) "Comparison with Eq.> r ( 2 ) A V 2 in \)*\W \it^T& (15.22) 0 which selects from the usual and with the boundary condition 0(0) eigenfunctions the odd ones.oj2/2. we obtain E„. 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).g — —2. Power Potentials In the case of v = 2.24) is discontinuous at x = 0 (this means the derivative there jumps from positive to negative). discussion at the beginning of this section). 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. For the following we set in Eq.2) implies A s fj. > and with Eq.oj = ^/2X/JX. eigenvalues (JV + h)tkj — (2n . The potential V(x) = X\x\ (15. 15.23) The eigenvalues for the linear potential can also be obtained from the zeros of the Airy function.312 CHAPTER 15. In the case v = 1.41) for the .
E > V or s > 0.x)\x=0 = 0. 15. we must have as a result of the boundary condition <p(x) = 0: 4>(x)\x=0 oc Ai(E . s = E.2 The Power Potential 313 V(x) V(x)=x turning points exponential fallofl Fig.15. (14.x). W i t h z :— E — x this is d24>{z) + z<f>{z) = 0. and for s —> oo the trigonometric behaviour of Eq. i. i.2 Behaviour of an odd wave function at the origin.28) In particular at x — 0.60) applies. In the domain —a < x < + a . (15.56) and (14.e. (13. i.26) For z = s := E — x —> —oo. this wave function has the required exponentially decreasing behaviour at infinity. \(s?» (15. Eq. (15.e.59) applies. (13.57a) one solution of this equation is the Airy function <l>(z) on Ai(z) = Ai(E . Ai(s) 1 2v^(s)V4 exp x —> +oo.e. dz2 According to Eqs.29) .e.e.27) i. Ai(s) "v^ G ' cos s 3/2 _ t (15. i.
93528 12.03914 11. The expression (15.08181 5. Table 2.82878 1 2 3 4 5 6 7 8 9 10 2.e.94249 9. = im — (7r/4). p. *This Table is reprinted from C. (15.29) implies that cos (  ^ . L. copyright of NorthHolland Publ.00852 11.. of course. 201..82814 i.= ) = «.30) ! ^ / 2 .314 CHAPTER 15. Table 15.28) is really only valid for s large and s — oo. the eigenvalues En are determined by the zeros of the Airy function.00767 11. Rosner. i. with permission from Elsevier.33811 4.08795 5. n = l.e.94413 9. Quantum Mechanics with Applications to Quarkonium. (15.3.31) (an odd function has an odd number of zeros!).^ = !(2nl).02265 10.78671 7. so that n — 1/4 ~ n. Co. It is possible.1: SWave Energy Eigenvalues for V(r) — r (with h = 2/x = 1) from f 3 7 r ^n 2 ( i>j2/3 .52056 6. [231].51716 6. Quigg and J.1 demonstrates the quality of the WKB approximation by comparison with the exact values. .32025 4.En from Ai(£„) = 0 2.7 T Tl * 2 V 4 Actually this expression is only valid for E and hence n large.93602 12.2..e. but for large > values of E Eq. i. Power Potentials Table 15.78445 7. to obtain the zeros of the Airy function numerically. (15. (2/3)E3/2 or 3 / 1 2/3 E — En — . It is interesting to note that in the dominant approximation both methods agree.02137 10.04017 11..
In order to be comparable with experimental data these investigations. r <53) 1 3  dr cos 2 () (15. The resulting masses agree in most cases very well with those extracted from experimental observations.** r ( # ^ e e ) o c *(0) 2 .34a) **The most frequently quoted reference for this result is R. had to be supplemented by inclusion of relativistic corrections as well as other contributions arising for instance from spin and angular momentum interactions. We shall encounter WKB normalization constants at numerous points in later chapters. which are inversely proportional to the lifetimes. van Royen and V. i. F. D. ee. P.e. e.15. dr[u{r)f " f** «" (i £>i)4 Here we replace the oscillatory part cos (. See also J. In general these quarkantiquark pairs. of course.3 The ThreeDimensional Wave Function 315 15. can be shown to be proportional to the modulussquared of the particle wave function at the origin.* We can determine the constant N in the leading approximation by demanding that = 2 / Jo i. We saw previously (see for instance Eq.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. Weisskopf [239]. e. (14. have only a finite lifetime and decay into other particles such as electronpositron pairs. in Chapters 24 and 26. i. cos2()^— / 1 fr° 2 > w .e.e. *Note that this makes N a W K B normalization constant. The decay widths T.3 The ThreeDimensional Wave Function The linear potential has in particular been used in the investigation of the spectrum of heavy quarkantiquark pairs. Jackson [140]. . the charmanticharm meson ^ .• •) by a mean value. We indicate briefly here — without entering into further details — how these quantities can be calculated.g.g.
e. A = (15. (15.1 )n ..\ \ * n = l.2. (15. i. (15.36) \i dEn irh2 dn According to Eqs.. & (15.39) in / .2) and (15.4) we have for I = m = 0: ^(O) 2 = n'(O) 2 y O o(^0) 2 = ^ K ( O )  2 .316 In accordance with t h e relation CHAPTER 15.35) u T (•T=2TT/OJ dt cos 2 cot 2' (15.38) N : sin V ^ Jr dr' 7T (15.40) VWY .34b) VMEvy Variation or differentiation of Eq.37) (15. (15. dE a—— / dn J0 so t h a t with Eq.35) N2 . Power Potentials this averaging implies a numerical factor like 1/2.. (15.10) 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.Jo dr' n M?) ~ 4 in sin sin JTCldrW2v{EV)j l \ 7T N 7W)sin = ( . so t h a t approximately l N2 / Jo k(r)dr .rcu w X{r)dr = Kh .l N n . From Eq.32) we obtain to leading order u'{r) and therefore «'(0) = N N (1JU0) N (15.
(15.* Using the result (15. .43) or the results of corresponding calculations one finds the following dependence of (^(O)!2 on the quantum numbers n = 1. G. J. Itt^pocn2*"1^"^.15. and hence with Eq. The case of arbitrary positive power potentials as above has been considered by R.23) for the energy levels of the linear potential. the expression (15. § S. E.45) 'See the references of Quigg and Rosner cited at the beginning of this chapter. MiillerKirsten [28]. 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. W.g.1 ^ ( 2 ^ „ ) 1 / 4 . (IM2) l*(o)r = M dEn dn irh 2 ~^WEn ~*T (15 43) ' Inserting here. 3 .41) I*(0)2=_L^)^. for various potentials: (a) For a confinement potential of power v. 'For one such investigation which includes the Coulomb potential in addition to a confining potential see e. S.3 The ThreeDimensional Wave Function Since we are considering potentials with V(0) = 0. Hence Eq.38) implies 317 (15. Hite and H.2. J. (b) for the logarithmic potential^ l*(0) 2 ex ^ v (15. Bose and H. for instance. . we have ^'(0) = ( . 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). Bose. S.l ) n . W. K. (15. Kaushal and H. J.44) (15. K. . W.37) we obtain: 1 (2M^)1/2 47r ft. MiillerKirsten [29]. MiillerKirsten [147]. 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. .
rewrite the equation as ^ z + [ . S.1. Example 15. E' = E + gln(r0). 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 . q = 2n + l.c = —a//3.. {S = ^ > 0. and there Eq..46) and the classical Kepler period of Eq. MiillerKirsten [29].^ ( 3 ^ + l)55p^«(3^l) + .2.1: Regge trajectories of the logarithmic potential Consider the radial Schrodinger equation for the potential V(r) = g\n(r/ro).I l n f ^ p . n = 0. Solution: Details can be found in the literature. ip = exp[(z — c)/2]. J. i.L2 h4 = 4/3exp[(2a — /3)/f3]. . We observed there in the case of the Coulomb potential that the period is proportional to n 3 .. (15. (11. W.318 (c) and for the Coulomb potential^ CHAPTER 15. *=^f. 7 With the substitutions r = exp(z — c). = 1(1 + 1).j . Power Potentials *(0) 2 oc .. = (I + 1/2) 2 . . Eq = i q . Landau and E. in agreement with Eq. U(z) =  dz zf3e2a/?e2*.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. Lifshitz [157].g ^£+(<*Pl»riy = 0.^ +•••.133). (14.e.75).— oo < z < oo." 1 11 L . Bose and H. K.5.L 2 + U{z)]<f>. . D. (15. M.
Phenomenologically the screened attractive Coulomb potential may be written V(r) = g2pr/ro .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.Chapter 16 Screened Coulomb Potentials 16. that no matter how far the particle is scattered away from the source of the Coulomb potential. Historically this potential arose with the realization 319 pVr . r in which /i has the dimension of an inverse length or equivalently that of a mass in natural units. r where r is the distance from the source charge and the charges and other constants are collected in the coupling g2. In reality. 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. The result is a screening of the Coulomb potential which thus attributes it a finite range ro. a charge as source of a Coulomb force leads to an effective polarization of the vacuum like that of a dielectric. so that like charges are repelled and unlike charges are attracted by each other. in microscopic and hence quantum physics with the possibility of real and virtual creation of particles and hence a vacuum state. This implies. and in fact. it is. A very similar potential is the Yukawa potential V{r) = g2 — . It is clear that for ro — oo the > potential becomes of Coulomb type. it will always notice the latter's presence.
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. however. all coefficients Mj are real and independent of the energy E = k2.* 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.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. In the literature reference is sometimes made to the socalled static limit of a relativistic propagator. Heisenberg [133]. 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. Since. The realization that mesons and baryons are made up of quarks does not really change that picture at lower energies. (16.320 CHAPTER 16. i. and that. 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 .3) where for the real potentials we consider here. in fact the parameter [i represents the mass of such a spinless meson. from the relation Pi + P 2 + M2c2 = 0.2 ) or (p = hk) 1 /" . A Green's function is effectively the inverse of a quantity called propagator which is intimately related to the expression in Eq.1) We encountered Green's functions earlier. Yukawa potentials and superpositions of such potentials play an important role in nuclear physics.e. one frequently resorts in calculations to the expansion of the potential in rising powers of r.ii k . the mathematical expression with an exponential is not so easy to handle analytically.1) and is written in spacetimedimensional Minkowskian notation l/{jpvpu + /i 2 ). 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.
only for the cases of n = 0. Regge trajectories were realized to play an important role in the high energy behaviour of hadronic scattering amplitudes.r)=0. if>oo. Bethe and T.Q.1 in the above.e. Froissart [223]. for the Regge trajectories or Regge or /plane poles of the 5matrixll / = ln(K) 'Standard references are the monographs of R. . Thus in the present case it turns out that it is easiest to calculate first the expansion for the expression l + n+1.e. 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. (16.{l. .e. i. C. (16. Naturally it is easiest to familiarize oneself with these by studying solvable potential models. Masson [180].1 Introductory Remarks 321 in this context — are functions which interpolate integral (i.5) where K — ik and A n is an expansion in descending powers of K. (16. (16. Potentials expandable as in Eq. bound state eigenenergies and the 5matrix. . Cheng and T. l+n+l = ^P. Wu [47]. f H.e. $ In the following we consider screened Coulomb or Yukawa potentials of the type of Eq.k. S S. Kinoshita [21].r). ''The approximate behaviour of Regge trajectories for the Yukawa potential has also been calculated by H. Lovelace and D.§ 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. physical) values of angular momentum as functions of energy E. i. S. £+**PVV t/. See also the other references below. . "These were first considered by C. Squires [258]. to be of the order of 1/k. 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.16. Mandelstam [185].3) have been considered by various authors. J. Omnes and M.k. T.4) where E = k2 and h = c = 1 = 2m.mo being the reduced mass of the system. where n is an integer and we referred to these already in Chapter 11 in the simple case of the Coulomb potential.1. Frautschi [97] and E. A. 2 .^ Regge trajectories arise as poles of the Smatrix in the plane of complex angular momentum. i.3) in the radial Schrodinger equation for the partial wave ip(l. They are usually written I = an(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. however.
2 Regge Trajectories Since our treatment here aims at obtaining an S'matrix as a generalization of that of the Coulomb potential. Under these conditions the S'matrix is meromorphic (i. Longoni and T.t Considering such superpositions of Yukawa potentials with /•oo / a(/j. H. The conditions for this have been investigated in the literature* and may be summarized as follows: /•oo V(r) = / d/xa(//)e^7r.e. Squires [258].** The energy is later obtained by reversion of the resulting series. 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. M. Screened Coulomb Potentials as a function of the energy. (16. W. Bottino. Miiller and K.4) to z = —IKv and set TP(l. for all n. o rV(r) I dpp\V{peie)\ regular at r = 0. W. Section 11. *See in particular A. A. tf H . Regge [31] and E.6) 2K{M°~ An(K))x+ 2K^(^K) MiX ' (16J) **H.)/j. W. 16. as may be seen from Eqs.k. 'In the case of the Coulomb potential (cf.z) Then x is a solution of the equation VaX= (16. . starting with the power r _ 1 of the Coulomb potential.9) the cut starts at E = k2 = 0. J. J. J. < const.19). in the present case at fc2 = 0 or k2 — Mi = 0.k.322 CHAPTER 16. has only simple poles) in the plane of complex angular momentum and in the complex &plane (E = k2) cut along the imaginary axis. < oo for all 0 < vr/2. J. Proceeding now as in the case of the Coulomb potential we change the variable of Eq. Muller [204].z) = ez2/2zl+1X(l. one can assume an expansion of the potential V(r) in ascending powers of r.15) and (16. In the following we follow mainly the last two of these references.ndp. Schilcher [203]. Muller [201] and [202]. 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. (16.
a+ l)$(a + 1) + (a. J.10) zm$(a)= ^ j=m Sm(a. (16. (a.a + j)Smi(a.a + j)Smi(a.z) = Ha). (16. so that to leading order we have VaX{0) = 0. The function <&(a.z) is known to satisfy a recurrence relation which we write here for convenience in the form z$(a) = {a.a — l) — a — l. where $(o.9) we obtain m l)$(a .j) = (a + j l. (16.a)${a) + {a.9) (16. (a.awhere (a. all other So(a. Eq.11b) ^ The associated boundary conditions are: So(a.0) = 1.5)) a = + 323 . Sharma and H. j)Smi(a.11a) The coefficients Sm(a.5) is equivalent to a = —n.1).b.j) for \j\ > m are zero.j)$(a + j). {0) X = Ha. as in the case of the Coulomb problem.j) (16. K.L ^d {bz)a l+ l + ^ ^ l 2K =  n &ndb = 2l + 2 = 2n^4^ K (168) The right hand side of Eq. b\ z) is seen to be a confluent hypergeometric function which for reasons of convenience we abreviate in the following as 3>(a). W.j +(a + j + l. . and all Sm(a. MiillerKirsten [249].j) may be computed from a recurrence relation which follows from the coefficients (16. anharmonic and cosine potentials have been derived in L.j l) + (a + j.7) is seen to be of order 1/K.a + + l). a + 1) = a — b+ 1. (16.b. We also observe that the ansatz (16.10): Sm(a.16. By a repeated application of the recurrence relation (16. which means that the hypergeometric series breaks off after a finite number of terms. a) —b — 2a.2 Regge Trajectories where d2 Va = z^ and (cf.i 0) = 0.* Recurrence relations for coefficients of perturbation expansions for Yukawa.
aj 2 \ [a. + (16.e. (16. #(a + ra) fjL$(a + n) V.' — .n.7) and hence in Ra may be cancelled out by adding to x a contribution /x<I>(a + n)/n except. Screened Coulomb Potentials Substituting the first approximation x^0* = 3>(a) into the right hand side of Eq. (16. so that 1 1 2K— ' (2W[a'a]2+(2^F[a'a]3 1 r i .5) the Regge trajectories I = ln(K).aj 2 (2K)A _ [a. oo _ . This equation is seen to be 0 = 7n7L a ' a Jl + P a + n $ ( a + n) = 0. [a.An{K). Va+n = Va . 1 .j).12a) where [a. of course.a + j]i+l = MiSi(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). i. i R »] = 2K[a.324 CHAPTER 16. 0 < \j\ < i. We first cite the final result . when n = 0. [a. . (16.<&(a + n) on the right hand side of Eq.a]l^a) + ^2(2K)i+1 Y.7). for the coefficients M^> of the expansion Details can be found in the references cited above. [ a .a + l ] 2 [ . Any term /j.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.1] 2 .a\4.[a>a + J]i+i*(a + 3). Evaluating the first few terms of the expansion (16. This follows from the fact that D a $(a) = 0.l. a . (16. and hence Va$(a + n) = ra$(a + n). the latter can be written . (16.[ a + l.o]i = M 0 .13) one obtains the quantity An(K) and hence with Eq.13) + ••• • One can now construct coefficients with their recurrence relations for the individual terms of this expansion.
15) The same expansion may be derived by the WKB method from the BohrSommerfeldWilson integral for three space dimensions as explained in Example 16.l)n(n + l)(n + 2) + 2M 3 M 0 (3n z + 3n . P.1.1) +6M 2 Min(n + 1) + 2M2M02 + SM^Mo] + g^s [3M 4 M 0 (n 2 + n . it is clear that one expects the trajectories to be finitely closed curves with asymptotes given by those of the Coulomb potential. 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. i.1) +6M 2 Mm(n + 1) + 2M2M02 + SM^Mo] .§ This is an important observation which indicates the equivalence of the methods.2. The result is AnW = M0'[n(n + l)M2 + MlM0] <2» + ^ ^ 1 + — ^ [ 3 M 4 ( n . of course.( 2 1 ^ 6 1 ) [ 3 M 4 M 0 ( n 2 + n . (16. G.e.2 Regge Trajectories 325 and then demonstrate the calculation in Example 16. With numerical methods one can achieve more.1) + 3M3M02 + Mln{n + 1) +4M 2 MiM 0 ] + 0(K~8).^ 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. including an exploration of the domain of small energies E. Obviously the expansion — which.16. Thus the expected appearance is that shown in Fig. Ahmadzadeh.5). we obtain the expansion for the Regge poles. Analytical Observe there the quadratic form of the centrifugal potential! "See in particular A. However. Burke and C. is an asymptotic expansion for large values of the energy — is not very useful in the very interesting domain around energy zero. .14) +4M 2 MiM 0 ] + 0 ( K ^ 7 ) . Tate [6].1 by evaluating the first two terms. . Inserting this expansion into Eq.^ p [ 3 M 4 ( n . 16.l)n(n + l)(n + 2) + 2M 3 M 0 (3n 2 + 3n .1) + 3M3M02 + M22n(n + 1) (16.
so the imaginary part ai of I — an represents its width in angular momentum. a] 3 = M 2 S 2 (a." Re I 1 0 1 2 Fig. [a.0) = (a. 0).1 Typical Regge trajectory for a strongly attractive Yukawa potential. where the imaginary part of I is very small. satisfies the relation ajAO ~ h. p. For a resonance with a long lifetime. 0) and Prom Eq. However. Solution: We evaluate the first three terms of expansion (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.O) = Other terms vanish since So (a. through which the particle orbits during the course of the resonance. (a.12b) we obtain [a. For a bound state a j = 0 and the orbit becomes permanent. 16. ±1) = 0. Frautschi [97]. ai is small and A6 is large.1). Hence Si (o. From Eq.326 CHAPTER 16. just as a decay width T represents the width of the resonance in energy. C.—" + In = — . one can expect a resonance with the lifetime determined by this imaginary part.a) = b . and the angle A#.a)So(a.11b) we obtain Si(a. The conjugate variable to energy is time.2 a = In See S. 114.^ .13). Example 16. In fact. . (16. (16.0). 16. and the lifetime At of the resonance satisfies the relation TAt ~ h. Screened Coulomb Potentials expressions of the behaviour of a Regge trajectory in the immediate neighbourhood of E = 0 are practically unknown. a] 2 = M i Si (a. . this is a particularly interesting domain since at the position soon after this point at integral I (as at I — 2 in Fig. Similarly the conjugate variable to angular momentum is angle.
. .a + l)S0(a. 0) + (a + 1.2 Regge Trajectories 327 Analogously we evaluate S2(a.0)+ 0 + 0 = (a1).a)5o(a.13) we obtain 0 (Mo . 0) + a S i ( a . Thus (with terms which are 0): Si(a. 1).1 ) + (6 .1) + (a .n ( n + l Inserting these expressions into expansion (16.b)(a .6 + l)a + A.0) Si(a.11b). a)Si (a. Solution: Ignoring higher order terms in r.An) • [ M i M 0 + n(7i + l)M 2 ] 1 A n = M 0 .00 B + Cz 2ni VA + 2Bz + Cz2 B 2niI —= + vC with ambiguous signs of square roots I. We obtain the quantities S i ( a .1.0) + 0 + 0 = ( a . 1) Hence 52(a.4) with mo = 1/2 and around the classical orbit the integral in the following relation I h = <b ^J pdr: dr •K2 Mo {l + \)2h2 1/2 ^classical orbit The integral is most easily evaluated by the method of "poles at infinity".2 a ) .0) + 0 + 0 = ( 6 . 14.0) (a .1) Si(a. (n+l)U+^pJ . A" An An A" = = = (a.1 ) + (a. .al)S0(a.2 V~Az and z \ zz dr\ A 2B C dz V'A + 2Bz + Cz2 (6.6)Si(a. a)Si (a. One sets z = 1/r so that.(ra + 1) [n + (2K)3 A n+1+• n K K (M0 .2o)Si(a. using the Cauchy formula (6. i ) again from Eq. 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 . 1) (a . (16.^ 2 M " + ! ) M 2 + M i Mo] + • E x a m p l e 1 6 .6 + 1 ) . (a.65) 2m dz V'A + 2Bz + Cz2 • Z— 0. a)Si (a. we have to evaluate (cf. Sec.16. (a.A „ ) H Mi 2An 2K if n + (2i^): r M 2 2K Hence 1 .65) and observing that in z there is one pole at the origin and one in approaching infinity in view of the expansions VA + 2Bz + Cz2 _ VA y2 .0) = (a .
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.(iO mp~ ~ ' ' ' '+ l)M2 + M0Ml] (16.e. We know that corresponding to every Coulomb Regge trajectory we have a corresponding one in the Yukawa case.5) together with the expansion (16. This shows that 1(1 + 1) has to appear in the BohrSommerfeld—Wilson integral as (/ + 1/2) 2 . I. 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 . Example 16. by expanding the right hand side in rising powers of v and evaluating the individual integrals. (16. i..l. ±2HK in agreement with our expressions above.t First. In the second paper the second order WKB approximation is used and shown to yield complete agreement with the terms given in Eq.328 Thus here dr K' CHAPTER 16. Boukema [34] and [35]. The turning points in this case are at r = 0 and r = oo. Here we do not perform this procedure.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.1)1(1 + 1)(Z + 2)M 4 + 2M3M0(3Z2 + 3/ . n = 0. *J.16) = n n=o 1 2 With this inversion we obtain = M0^^[l(l + ^ [ 3 ( Z . we can use Eq.* 16. tSee Chapter 11 and V. (16. Screened Coulomb Potentials Mo . Solution: For details of the solution we refer to papers of Boukema.2. Thus we can write down the S'matrix for the present case by exploiting the limiting case of the Coulomb potential.1) +61(1 + l)M 2 Mi + 3M2M02 + 3Mx2Mo] + 0(K~6).14) for An(K) in order to reexpress the latter in terms of I so that i+i+ A. however. Singh [252]. . t?{l + \f M0 ±2K 1/2 2TT 2J 2V^2 (n + l + l)h = 2TT Mo h.15).
18) of the Smatrix now to explore further aspects. .4 The Energy Expansion We observe that Ai(K) has the property MK) = MK) 329 (16.16.4 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\. This has been done§ and one obtains . the behaviour of the scattering phase in the domain of high energies.2. Now reversing the expansion (16. Paralleling the case of the Coulomb potential in Chapter 11..l. One can use the explicit expression (16. i.e. such as.l (two additional terms are given in the literature). for instance.* 16. we can now write down the S'matrix in terms of the scattering phase 5i as the expression TV/ I 1 I ^l(J<l) S ( W = e r(/ + i^#>)e ' (16 18) ' We observe that the poles of the S'matrix are given by l + l + ^p = n. J.19) +2M 3 M 0 (3n 2 + 3 n . which are precisely the expressions yielding the Regge trajectories of above. the following *See e. 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 .19) we obtain the energy. n = 0. . W.l)n(n + l)(n + 2) + 2M2M02 . and could then have carried out the perturbation procedure not in inverse powers of K but in inverse powers of y/K2 + M\..17) and is therefore real for real values of the potential coefficients Mj.g.0[(KZ + Mx)'6} (16. We observe also that the S'matrix is unitary as a consequence of the result (16. Miiller [204]. J. W. H..17). § See H. Miiller [211].14). Expanding the square root VfC2 + Mi for \Mi/K2\ < 1 one regains ln(K) with the expansion (16.
^ 16.< 3 M 4 M 0 ( n .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 .l)n 2 (n + l) 2 (n + 2) 1 + • • • (16. J. Warburton [279]. Sommerfeld [256]. in the following expression with an "See in particular G.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. Iafrate and L. C.l)n(n + l)(n + 2) . A. 282. B. 107.1) + Af3Mff + 1) ^ lOM 6 M 0 2 (n .330 CHAPTER 16. . Mendelsohn [135] and E. k) as coefficient of the outgoing spherical wave in the asymptotic behaviour of the wave function tp(r).e. 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. i. 'According to S. Watson [280] in 1918 and later resurrected by A. the transformation was introduced in the form given below by G. 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) ^ — .2)(n . p. Zauderer [286]. MiiilerKirsten and N.11) + 2M22M03(9n2 + 9n .20) Extensive investigations of the energy eigenvalues for the Yukawa potential can be found in the literature.1) + 2M2M. W.10) + 12} + 4M3M05 +2M 4 M 0 4 (6n 2 + 6n . E. p.f 24(2n + l)(Z + n + l ) 5 M4M0(nz M06 M$n(n ^ M9 + n .3 M 2 V ( n + l) 2 M(o I +2M 3 M 0 2 (3n 2 + 3n . N. Large coupling expansions derived in H. J.* We recall from Chapter 11 the definition of the scattering amplitude F(9. Screened Coulomb Potentials expansion.l)n(n + l)(n + 2)(n + 3) +2M 5 M 0 3 {5n(n + l)(3n 2 + 3n . Frautschi [97].
k) 0ikr (16.25) Setting Note the minus sign in the argument of the Legendre function. Actually 5j lrr/2 as in Eqs. I = n + x. For this purpose we consider the following integral taken along the contour C shown in Fig. .187).k) — e 2i5[(k) ~" p—iirl (16.184) to (11.. Eq.''' The idea is now to consider each term of the expansion (16. n = 0.kmcose).2 in the complex Zplane: *<«•*> = s / c c K <u(21 + 1) vf{i.iJ 11 —[S(l.26. so that sin KI = sin7r(ra + x) ~ (—l)nKX = (—l)lir(l — n).fe)P . c = 1. dn Iml 1 I c ( \ x o x 1 A 2 A 3 x4 X 5 Rel Fig.187)) (16.k)~l] = ~ k" PiSiW sin 5i(k) and S(l. \x\ <C l .2 The integration contour C. 16.5 The SommerfeldWatson Transform 331 ingoing plane wave in the direction of z (here with h = 1. z=o in which (cf.24) Si(k) being the phase shift. The scattering amplitude possesses the partial wave expansion F(0. (cos 9). (11.k) = 2zfc L " v "'' v .k)\2.k) = ^T(2l + !)/(*.TOO= 1/2): ip(r] r ~*°° Jkz elkz + F{0.21) The scattering amplitude determines the experimentally measurable cross section o given by da (16.2.23) f(l... (11..1. ~:' sin nl (16. 16.22) = \F(9.23) as the contribution of the residue of a pole in the plane of complex I of some suitably constructed contour integral.
—.3. in which case c = a and d = b. For the momenta indicated we set (with metric +.k) Zi Jc TT{L . In order to see the relevance of Regge trajectories in a reaction of particles. 16.fc)P„(cos0). Screened Coulomb Potentials Then integrating with the help of Cauchy's residue theorem. 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'). Such a direct reaction and its "crossed channel reactions" are shown schematically in Fig.332 CHAPTER 16. as indicated in Fig.n) oo {l)1 Pt{cosO) v ' (16.28) . 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. u={pq')\ (16.—) s = (p + q)2.3 Reaction channels and their respective Mandelstam variables. n=0 which is the usual partial wave expansion. we obtain m® = ± I dl^±^f(l. a + d — • c + b. d(q') (16. for instance.—. a + c —> b + d.27) a(p) © b(q) Fig.26) P((+COS0) = 2 ( 2 n + l)/(n. t = (qq')2. be elastic scattering of a off b.3. 16. Thus the direct reaction may. 16. we have to digress a little and introduce a few simple ideas which played an important role in the development of particle physics.
F. except for poles and cuts which characterize the reactions in the three channels.16. A(s. etc.c o s 0 s ) .4.The variables s. (a) Mandelstam's hypothesis: All three processes (described by the s. the momentum k there would correspond to qs) s = ~4(ql + ml). the only kinematics we require in the following is given by the relations (e. B. u = . T See e. A(s. For a Lorentzinvariant normalization of the initial and final states. which is an analytic function of the variables s. u.t. Chew [48].g. L. In the Yukawa picture of a reaction the dynamics is described in terms of the exchange of mesons IT (called pions). One now makes two hypotheses.u are not independent.t) = y^Fs(s. the dynamics is described by the exchange of quantum numbers. Blankenbecler. Treiman [25]. one can write the amplitude as depending only on two.et) (16. N. N. s = 2qf(lcos0t)(16. Mandelstam [184] and R. Ft = £ ( 2 J + l)F.0s). Khuri and S.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). S.5 The SommerfeldWatson Transform 333 and for simplicity we assume here that all particles are spinless and have the same mass mo.. one then has A(s.t). Symbolically we then have the situation shown in Fig. if s describes the square of the total energy in the s channel.u channels) are determined by one and the same relativistically invariant scattering amplitude A(s.§ Since the three variables s.31) i l Taking all particles to have the same mass mo.32) t = 4(q? + ml).29) and with selfexplanatory meaning for the cross section of the reactions ~ = \Fs(s. M.t.u are known as Mandelstam variables. t = 2q£(i . In a Regge theory on the other hand. § See G.u) (if any of the external particles has nonzero spin.^ (b) Chew's hypothesis: All (composite) particles lie on Regge trajectories. there will be several such amplitudes which together describe all three processes). 16.t. e. .t. t. Thus.g. if we had been considering the s reaction originally. (16.g. Goldberger. the variables t and u would describe momentum transfers in the crossed channels.(<fe)^(cos 6t).9s)\2.t) = VtFt(t. (16. the quantities carrying these quantum numbers are the Regge trajectories.
It is known from the Mandelstam representation that the amplitude possesses a branch point at cos 9 = 1 + TOQ/2(^.5 The Lehmann ellipse.5 for Yukawa potentials V{r) u: lm cose dr a(/j.4 Yukawa versus Regge theory. So what can one do? 1 H. the Legendre polynomial Pj(cos 9) does not possess a cut. .334 Yukawa CHAPTER 16. For s — oo : cos 9t = 1 + s/2q^ — oo. Screened Coulomb Potentials Regge oc(t) Fig.r 1+m 2 /2q 2 o ^t "" Re cost: Fig.((fc)/Kcos0t) schannel ichannel schannel variables the partial converges only within the socalled Lehmann ellipse^ which is shown in Fig. Thus one is interested in establishing a representation of the amplitude in terms of ichannel Regge trajectories.e. 16. 16.)e~ •fj. 16. But > > wave expansion X)(2Z + l)F. 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. However. Lehmann [162]. But for integral values of I.
contour C / / D ^ ^ .t). the contour representation (16.33) . 16. N F(6.^ B 1 A Fig. „ i.6. and there is only a finite number N + 1 in the domain of 9W = . 16. sin nl [16. Again we have / Jc> • • • = 1ni y^ residues j3n „ For Yukawa tions A and f(l. Thus consider now the same integral but taken along the closed contour C shown in Fig.6 The integration contour C".5 The SommerfeldWatson Transform 335 T h e answer is. i.16.1 / 2 . Then / JDD' ••• + j JBB' • • • = 2TTI V ^ residues f5n. Moreover.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.26) but for a different choice of the contour.e. all poles of Ql > 0. 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 Iml 1 \ i All \ r\ 0 / closed .e. to use the SommerfeldWatson transform.
this behaviour has been confirmed in high energy hadronic reactions. It follows that under these conditions. i. of course. for high energies in • > the schannel). In particular. Apart from more refined details. if ao(t) is the Regge pole with largest real part. The slope of the Regge function an(k2) is an " Simple cases are the harmonic potential and the squarewell potential. Potentials with a shortrange repulsion more singular than 1/r 2 have been considered by N. i. Regge trajectories have. for plots and discussion see R.35) This result is valid for s — oo and t negative (cf.e. as we discussed briefly in the above. MiillerKirsten [29]. at high schannel energies. K. Limic [177] and E. (16. The > result demonstrates that the high energy behaviour of an schannel reaction is determined by the leading Regge trajectory in the crossed ^channel. . Omnes and M. Regge [229]. 42. (IM4) This is the asymptotic Regge expansion of the invariant amplitude for s —>• oo. 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. we have P\/2+IR{Z) ~ 0(l/y/z) — 0 for s — oo (i. Foissart [223]. the amplitude can be represented as a sum over ^channel Regge poles. Screened Coulomb Potentials Since for s — oo : cos Qt = 1 H n ~* °° > 2?* and (from Tables of Special Functions) farMoo: Pa(z)* ^+l]\(2zY. J.32)) and finite.t)&B(t)sa°V. The logarithmic potential has been investigated in a way similar to the Yukawa potential above by S.336 CHAPTER 16.e. we have (16. W. Predazzi and T. Eq.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. In subnuclear physics they played an important role before the advent of the quark idea and thus of quantum chromodynamics. 16. ^M)^E i 2 a f: a „t ( t ) ^ m <^'F(s. Bose and H.e. p.
J. . like a meson or a baryon. Related aspects are discussed in Sec. and also as socalled "fixed poles" . For an overview see e.** Whereas the former are related to absorptive properties of a reaction. W. Regge poles are not the only possible singularities of a scattering amplitude in the plane of complex angular momentum. Singularities appear also in the form of cuts in the plane of complex angular momentum. The above treatment of screened Coulomb potentials is incomplete. 20. The small imaginary parts of eigenvalues or lifetimes of resonances have not yet been calculated.g. The way to do this is similar to calculations in Chapters 18 and 20.1. MiillerKirsten [209]. H. the latter are related to the distinction between "' elementary and "composite particles". a composite particle being one with structure.16. and is composed of quarks. called "Regge cuts".6 Concluding Remarks 337 important parameter in string theory.
This page is intentionally left blank .
and are themselves periodic with period e.Chapter 17 Periodic Potentials 17.0 < k2 < 1.or 27rperiodic trigonometric functions on the one hand and the nonperiodic hyperbolic functions on the other. that the Mathieu equation lies outside the scope of equations which can be reduced to hyper geometric type. 2K or AK. The reason for these difficulties is.) which are functions that interpolate between the n. The Schrodinger equation with this potential. separation of Laplace's equation Ai/> = 0 in ellipsoidal coordinates leads to the Lame equation (no sn 4 z term). since many analogous relations hold. motivated mainly by the regularities of the crystal structure of matter. are not much harder to handle than trigonometric functions. example 14. 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. however.lP = 0 in ellipsoidal coordinates leads to the ellipsoidal wave equation with sn 2 z and sn 4 z terms. 19 and p. 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. Thus Jacobian elliptic functions also pro*To be precise: Separation of the wave equation Aip\x2.1 Introductory Remarks Prom the beginning of applications of quantum mechanics periodic potentials of various types were immediately considered.g. See F. called elliptic modulus. p. depending on the parameter k. In essence these periodic Jacobian elliptic functions. 339 .* These elliptic equations involve Jacobian elliptic functions like sn(a. M. 25. Apart from the discontinuous KronigPenney potential consisting of a periodic repetition of rectangular barriers. K being the complete elliptic integral of the first kind. the most immediate candidate is a trigonometric form like that of the the cosine function. Arscott [11]. Taking in the ellipsoidal wave equation the limit k —> 1 (fc elliptic modulus) and putting tanhz = sin 6.
. which one looks up in books like that of MilneThomson [196] or Tables when required. i. or 5wave Schrodinger equation with the cosine potential is conveniently written d2v —  + [A . Like the trigonometric functions sine. Muller [73].2h2 cos 2z]y = 0.1. Both of these domains are important for a host of other considerations. these periodic potentials may be approximated by series of degenerate harmonic oscillator potentials.7. 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. need not be afraid of them. 8. In fact. such as sn 2 z + cn 2 z = 1 and double and half argument formulas etc. W. For large coupling. We might mention already here the much less familiar but more general Lame equation.2) where K2 = n(n + l)k2 and n real and > —1/2 and 0 < z < 2K. 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. Periodic Potentials vide periodic potentials like trigonometric functions.1) where h2 is a parameter and — IT < z < IT.. The Mathieu equation. k2 — 0 which means that in > > this limit the Lame polynomials degenerate into the periodic Mathieu functions. (17. Dingle and H.. See also the discussion in Sec. W.340 CHAPTER 17. those of integral order. 17. Lame equation: H. The first of these conditions is already tMathieu equation: R. and to relate these to the weak coupling bands or regions of stability. the Lame equation possesses 2 n + l polynomial solutions. Here the Jacobian elliptic function sn z is a periodic function analogous to a sine function and has real period 2K} For n = 0. [205]. J. which does not possess any polynomial solutions. which is conveniently written ^ + [\2K2sn2z]y = 0. ''The reader who encounters these Jacobian elliptic functions here for the first time. (17. B. 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. the three Jacobian elliptic functions s n z . This is a significant difference compared with the Mathieu equation. c n z and d n z are handled with analogous formulas. each with its own characteristic eigenvalue. The most important formulas for our purposes here are collected in Appendix A. Miiller. cosine and tangent. 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. J.4.1.2 . or Sw&ve Schrodinger equation with elliptic potential.e.
. Although the mathematics literature on the subject uses the physical concepts of stability and instability. 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). Correspondingly quantum mechanics — which requires the electrons of an atom to move around the nucleus — explains the stability of atoms. y" + \y = 0(h2). i.e. Here an important aspect is the determination of the domains of stability of the solutions and the boundaries of these domains. since the solution there is of the form cos vAz. Thus a bounded trigonometric solution is indicative of stability and an unbounded exponential solution of instability. The orbits of planets are stable in the sense that deviations from the recurring elliptic orbits are small. a bounded function.e. the case of h2 small. these concepts are rarely explained as such there. Looking at the Mathieu equation for h? —> 0. The instability of an orbit would be evident either from an unbounded spiralling away to infinity or from a collapse into the centre. we see that for sufficiently large values of h2 the periodicity or boundedness of the solution is destroyed and hence becomes one of instability. 17.e. 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. In classical mechanics stability is treated for instance in connection with planetary motion. Adding the negative term — 2h2 cos 2z to A.1. the semiaxis A > 0. h2 = 0 belongs to the domain of stability. Salem and T.2 Cosine Potential: Weak Coupling Solutions 341 seen to rule out polynomials. i. we compare the For interesting related discussions see M.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.2. i. i.17. as in Fig.2 Cosine Potential: Weak Coupling Solutions We consider first the weak coupling case. Vachaspati [242]. and one can imagine that a nontrivial parity factor exp(in7r) then turns into a complicated phase factor exp(iz/7r). 17.e. we see that in a plot of h2 versus A.^ 17.
2h2 cos 2z]y(z) = 0. y~(z) = 6sin vXz.e. ° ^ [1 + 0(h2)}.342 CHAPTER 17. d2y(z + 7r + [A .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. d(z + TT) cos 2z We see that there are solutions which are proportional. Periodic Potentials equations ^ p + [A . i. Setting z = 0. y(2vr) 2 a = —7—^ = 1. (17.B and a.e.4) with (for convenience in connection with later equations) y'+{z) y'_(z) = avAsin\/Az = = bv\cos^f\z = —y(z). We set (here and in the following frequently apart from contributions of 0{h2)) with constants A. (17. 2/(0)' yfr) . 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). (with Tn as translation operator) Tlxy{z) = y(z + TT) = cry{z). A solution of the second order differential equation is determined completely only with specification of boundary conditions which determine the two integration constants. we obtain y(vr) y(2?r) a = — — a = —^^r. a = const. i. — • . b: y(z) = A cos vAz + B sin \f\z and y+(z) = a cos yXz. sin V Az or linear combinations. y+{z) . TT.2/i2 cos 2(2 + vr)]y(z + vr) = 0.
whereas the solutions y+ and y_ have here been chosen specifically as even and odd around z = 0 respectively. 2/'(7r) = a ^ W + ^ .y{ir)y'+{K) = abVX in agreement with its value at z = 0 above. i.4).W or fy+(n)aa J/_(TT) = W ^ ^ J a ^\ = n For linear independence of a and /3. 2/_(0) = 0.{ay'_{ir) + bV\y+(ir)}a Now from Eq. and by'+(?r) = —ay/Xy{n). This means. (17. \y+>y\ = y+{ir)y'{ir) . y'_(Q) = bV\ with Wronskian W[y+.y+(7r)j/_(7r)} = 0. 7 y\z) = ay'+(z) + (3y'_{z).3) we obtain y(z + ir) = a[ay+(z) + @y(z)].e. yV(0) = 0.y_] = y+(0)y'_(0) — y_(0)y+(0) = ab\f\.aa][y'_{n) . with constants a and /3. we obtain V (TT) = ay+ (TT) + (3y _ (TT) = < [aa].a2b2\ a± ~ 2a6\/A .17. ay'~(ft) — bVXy+(n) and the Wronskian w + {y+(7r)y/_(7r) . we choose these with the following set of boundary conditions: y+(0) = a. (17. This expression is equal to the coefficient of a2 in the quadratic equation for a.2 Cosine Potential: Weak Coupling Solutions 343 The solution y(z) is an arbitrary solution.oWX] . y'(z + n) = a[ay'+(z) + py'_(z)}. the determinant of the matrix must vanish.y+(7r)y_(7r) = 0 or abVXa2 . [y+(v) . y{z) = aty+(z) + 0y(z). Setting in these last equations z = 0 and using the previous pair of equations and the boundary conditions. Using Eq. The roots a± are therefore obtained as 2bV\y+(ir) ± J4b2\yl(Tr)4. We can write therefore.
1. this implies y+(Tr)/a = ± 1 .6) We can derive various terms of this expansion perturbatively with the method of Sec.5) a The parameter i/ is known as Floquet exponent.0 This series may be reversed t o yield the Floquet exponent. 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 .22)) the boundary conditions of periodic solutions.e.455A2 + 1291A .7. **This point is not immediately clear from mathematics literature. 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.5) is that for h2 = 0 we have v = A/A and hence more generally v2 = \ + 0{h2).344 Setting / = 2by/\y+(Tr). where the constants are usually taken as unity from the beginning. (17. This is the easiest application of t h a t method since only simple trigonometric expressions are involved. i. Setting cr+ = e . <T__ + <T_ = 2y + (vr) .1169)/i 12 64(Al) (A4) (A9) 5 2 + 0{hw) r 16 (17. <7__cr_ = 1.8) "Observe that for integral values of v (in lowest order of h2). Eq. cr+ = 1/<T_.. (5i/ 2 + 7)/i 8 vz + — . i. (17.+ 2 2(i/ l) 32(i/2l)3(v24) ( 9 ^ + 58. ** An important consequence of Eq. 2 _ hA 2(A1) (13A25)/i8 32(Al)3(A4) 12 ift (45A3 . This calculation is demonstrated in Example 17.= — . (17. (17.2 + 2 9 ) ^ 2 + 64(1/2 _ 1 ) 5 ( z / 2 _ 4 ) ( l / 2 _ 9) + u ^ ^ ^ . Periodic Potentials g= 2aby/\. we have <7± — CHAPTER 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. . _ / ± v 7 ^ V .e. Prom this condition we determine later (cf. 8. T h e result for A is the expansion A = h4 2 . .
(17. (17.35A + 15A2)/l8 /. which can then be rewritten as Dvy = 2fe 2 (A + cos 2z)y.e. . rr +h* 3 =7rsin7rvA 64(A1) (A4)AVA +Q(h12). With a perturbation theory ansatz for y+ around h2 = 0 as in Example 17.1: The eigenvalue A for nonintegral v and h small Use the perturbation method of Sec. Example 17.2 how the expansions for small values of h2 may be obtained perturbatively. n an integer.9b) TT2 Next we observe that if we demand that the solutions y+ (z) and y_ (z) satisfy the condition (17.„„ x n. 124. we must have a2 — 1 and hence e 2iv/A7r = 1} i.1) 15A2 .3).jl2. VA = ±n.7 to obtain the eigenvalue A as a perturbation expansion for nonintegral values of v and b? small.10) A = n2. To lowest order the solution y is j / ° ) = yv = cos vz n where d2 Dv:=—~+u2. W. (17. 8. — 2/i 2 A and insert this into the Mathieu equation y" + [A — 2h2 cos 2z]y = 0. Schiifke [193]. It follows that in these cases i? = n2[l + 0(h2)]. as will be done later.9a) 4(1A)>/A 64(A4)(A1)3A^A One observes immediately that these expansions cannot hold for integral values of v or A. (17. (17. These cases therefore have to be dealt with separately.35A + 8 . cos TT\/A 32A(A1) 2 . Meixner and F.1 below one obtains the following expansion for cos7rz/ which is fairly obvious from Eqs. (17.11) or smvz or e±ivz.12) S e e J.17. Solution: We write the eigenvalue equation A = v2.5) and (17.2 Cosine Potential: Weak Coupling Solutions and so 345 (8 .9a) and is therefore not derived here in detail^ ri/ = VA + rCOS7Tf = COS7TVA + /l 4 h4 7rsin7rVA = 4\/A(A . = J—^ + o(hr). p. For these integral cases of v we demonstrate in Example 17.
for instance. i. 17.14) = = = cos(y+ 2)z + cos{y— 2)z. as shown. since j / 0 ' = yv leaves unaccounted R\. (17. up to 0(h2) we have A = 0.16) *These smallh expansions are convergent. i.e. v . (17.2)j/„_ 2 + {v.2) _ __ _ (y. by J. and (v. ("• ^ + 2) (o) 2(2^ . v)yu + {v.h2).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.346 CHAPTER 17. We observe that in these cases 2cos2z cosvz 2cos2z smvz 2cos2ze±il/z i. (17. of course. h2). Meixner and F.e.11). We assume for the time being that a ^ 0 and 2v + a ^ 0 (the latter case requires separate consideration.13) The first approximation y(°) = yv leaves unaccounted on the right hand side of Eq.h2).14) therefore leads to the following nextto leading order contribution (v.2) — 2 "^ 2(2i/ + 2) " + 2 vanishes. (17.(2) + .16) directly into the original equation in the form of Eq. when a = 0 or 2v + a = 0. (17.„ + £ h2i £ P2i(2j)yv+j (17. se and me as* oo i y = „«>) + yW + i.2) which applies when v is nonintegral.!/) = 2A We now observe that DvVv = 0.2 — we can write down recurrence relations for the perturbation coefficients pa(2j) together with boundary conditions. or A = 0(/i 4 ). The right hand side of Eq.3. '. W. h2(u. siniy + 2)z + sin(i/ — 2)z.11) and so in R}?' may be cancelled out by adding to j / ' 0 ' the new contribution a(2v + a)' = 0.v±2) = l. in each case 2cos2z2/„ = 3/„ +2 + j / „ _ 2 .e.2. (17.. so similarly yW leaves uncompensated / # > = ft2 ("• v ~ 2 ) R (o) . e ± i ^ + 2 ' 2 + e±i("2)z. = . Periodic Potentials The complete solutions of these cases are written respectively in selfevident notation cev(z.v + 2)yu+2]. Next we treat terms yv+a in j / 1 ' in a similar way. see Example 17. where except. v) = 0. v . With our perturbation formalism — as demonstrated in cases considered in Sees.3 and 19. (i/. (17. . .e. so that also Dv+ayv+a and hence Dv+a = Dv + a(2v + a) and Dvyv+a = a(2u + a)yv+a(17.11) terms amounting to i?i 0 ) = = 2 h 2 ( A + cos2z)y„ = 2h2Ayu + h2(yl/+2 + y1y2) h2[(v. The recurrence relation can also be obtained by substituting the right hand side of Eq. i. Proceeding in this manner we obtain the solutions ce.15) Hence a term fMy„+a on the right hand side of Eq. Schafke [193]. . sev(z. me±v(z.
v)\ . * Note. . For v = 1 we then have yi = cosz = y~\.. .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.2—r——— (i/ + 2. the solutions are not yet normalized.u) + h4 (".1 with A = v2 — 2h2A but with v = 1. Following the first few arguments of Example 17.. Solution: We start as in Example 17. fo^ + 2) . (17. The higher order terms naturally require a little more algebra..18) Again applying Eq.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)+. + y~i) (17.e.. „ .2 Cosine Potential: Weak Coupling Solutions 347 together with the equation determining A. In the case of integral values of i>. Hence we obtain 0 = /i 2 (2A + l ) f h4 + ••• . Meixner and F.7).. These expansions are actually convergent with a definite radius of convergence as explained in the mathematical literature. Example 17. i.g. .R{ .2 ) v(y2.14).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]. J. See e."2). We consider the specific unperturbed solution yv = cosvz which is the dominant contribution j / ° ) of our solution y. p. W. (17. 121. v) ' ( 2 v + 2) (17. (17. . thus determining the quantity A. however. 0 = h2(u. one has to deal with each integral case separately as in the following example. 2 ( 2 i / . Schafke [193].17) Inserting 1 for the stepcoefficients of Eq. This completes the determination of the solutions for nonintegral values of v in the domain of small values of h2. we obtain immediately 0 = 2Ah2+ 2(^T)+' which verifies the term of order h4 in Eq. is set equal provided the coefficient of the sum of the contributions in 3/1 contained in R\ to zero.17.
and v is to be found in ascending powers of h2.e. m. *" Jir Thus we have and uses / dz cos mz cos nz = — <5 mn . (17. 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 . 120. In the above we considered the case of v an integer and calculated the eigenvalue A. the case of A an integer. 123) (there the solution is called cei(z. We introduce a normalization constant c and rename the normalized solution y then yc.19b) This expansion agrees with that given in by Meixner and Schafke [193] (p. Periodic Potentials h4 r ••• . 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]. Schafke [193]. h4 (17.* The associated solution is ym+ym. Inserting this into A = v2 — 2h2A 8 for v = 1 we obtain (cf.n integers.24c) below) . Meixner and F. Eq. see below!) one obtains the additional terms given by Meixner and Schafke [193]. h2). i. . p.19a) This expression agrees with the result given in the literature (there the eigenvalue is called ai). 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. We consider this case again in an example.348 From this we obtain 2h2A = h2 CHAPTER 17. The specific normalization here is taken as I /*7r rn 1 = — / dzce2(z. W. With normalization (not to 1. J. The reverse situation is later of importance. 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.
3 1 A + 8) ( A .2 Cosine Potential: Weak Coupling Solutions Example 17..l ) 2 (A . fe47T2(A4) 4%/A(A .Q . on the left hand side of Eq. We thus obtain the expansions cos"\/A7r sinvAvr = = cos27r + (A — 4 ) ( — s i n V / A 7 T ) A = 4 — = + . S.9b) v — 2 + <5. . (17. W.3 5 A + 8)TT2(A4) _ 3 / / 64(A . MiillerKirsten. J. 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 . so that 2 2 cos nu = cos 7r(2 + S) = cos 27r cos TC8 = 1 and comparing with the above. Manvelyan. we obtain^ S 3 V 2 and therefore i/ = 2  iy/E ( h 3 \2J 7i . H.4)Av A2v A ~ 32A(A . Hashimoto [124].9b) about 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. .3 1 x 4 + 8) 274233 (.4 ) + . ^S. o 1• • • . (17.('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 . J . (11A2 .4) 2 7T 2 /l47T2 + • 8 h87T2 8A(A . We set A — 4 ~ 4e and expand the cosine and sine expressions appearing in Eq.17. Gubser and A..1) .1) (A . (17. Liang and Yunbo Zhang [189]. Substituting these expressions into Eq.1)2/A TT2 +h8 (15A2 .9b) and considering the approach A — 4 gives > (A4)2 + .• 1 + 5n h 2 s 2932 Setting in cosm.= l 2vA sin27r + (A — 4)(cos vA7r)x = 4—= + ••• = j= h •• • .
= and s+w 2tW±I.2. bV\ y+(0) 2/40) 07. These are defined by specific boundary conditions.1 Boundaries of domains of stability.w^. Periodic Potentials Our next immediate aim is to specify the solutions for which Eq. for instance. solutions for integral values of v. 17. Fig. concentrating again on the leading term in the even and odd solutions y±(z) for ease of understanding. in F. 17.4) into account. . as also conditions (17.e.21b). i.6.350 CHAPTER 17. we have" y+{z ± 7r) = a cos vA(z ± TT) = y+(z) cos vXir = TV{Z) F sm v<\7r. (17.29. Arscott [11]. "These conditions.W4W=UtWSta±I(. We verify these for the large/i 2 solutions in Example 17. M. 28 .2*0 y~(z ± 7t) = 6 sin VA(z ± TT) = y_(z)cos VA7r ± 6 cos vXzsin v\ir. We now derive these boundary conditions.10) applies.. (17.21a) and (17. pp.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. Taking in particular Eqs. may also be found.
.2 Cosine Potential: Weak Coupling Solutions We select the equations with lower signs and put z — TT/2.17. f^\y+(j) /vr\y_(7r) and for the derivatives ^(i)^^'(D^ .4(f) (17.y V+ \ 2 7"^ V 2 fla^+U a cos V A  ] 6^A + a&( sin\/Aj ^A = abV\. ^\y±M /vr\yL(7r) . we obtain y+(7r) _ i  2y_(f). 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 _ . Then y+ y 351 U . 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+. /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. /7r\j/!_(7r) y L l 7TJ . which after some rearrangement can be written y+(vr) _ . .
We see that there are four possibilities and hence four different types of functions. (17.24a) ao = hA h6 b\ = 1 — h —— + — — • • • . . These are defined by the following boundary conditions.1. .21a) and (17. y > se2n+i=qo.5) that the left hand sides of Eqs. y > se 2n +2s 9o +i The functions so defined have respectively period 7r. (17.23) We can now see how the domains of stability and instability arise. (17.T + .3. of v is even or odd. A 2 y • ce2n=q0i.1). These may be subdivided into classes depending on whether the integral value n — 0..2): hA 7h8 29h12 + Y l28^04. (17. (17. y > ce2n+i=qo.' fh2 V2ceo = l . 2K.2. Finally we cite here some expansions of the eigenvalues along with those of the associated solutions (an explicit example. se2 = sin 2z 2 12 13824 ' h2 z —— sin 3z + • • • .. IT. 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.— — — • • • . (17. Periodic Potentials and hence** a afc\/A For integral values of v (in lowest order of h2) we know from Eq.+ . in this order..21b) must be ± 1 .352 CHAPTER 17. 17. where the notation for the corresponding eigenvalue is put alongside on the right: y+(f) = 0 with y+(f) = 0 y'{\) = 0 with y _ ( § ) = 0 with with A .. This is shown schematically in Fig.24d) v 12 ' **To avoid confusion we emphasize: In Sec...• a2n(h2). Mr7nA (17. A *• b2n+2(h2). 17. was treated in detail in Example 17.24c) 8 h2 sin 4z \ . The ordering of the eigenvalues which then results is a0 < bx < ai < b2 < • • • < bn < an < • • • for h2 > 0. ce± = cos 8 64 h4 5h8 b2 = 4 1 . the case of ai with solution cei (of period 27r). A > b2n+i(h2).2 we require this relation for nonintegral values of v. 19.24b) 8 h2 z —— cos 3z + • • • .22) * a2n+i(h ). 27r. and whether the function is even or odd. sei = sin 8 64 /i4 h6 a\ = 1 + h2 — .1 for the first few eigenvalues which are boundaries of domains of stability (as discussed in Sec. There the second term on the right is nonzero.
the walls are not infinitely high and hence tunneling occurs from one well into another. . Miiller [73].17„. 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. . We can see from the inverse of Fig.> (17. of course. J.3 17.2 suggests that for very large fluctuations the potential can be approximated by a number of independent and infinitely high degenerate harmonic oscillator potentials.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.hz — 12 .3. 17.17. B.2cos4. 353 . Consequently one expects the eigenvalues in the large h2 domain to be given approximately by those of the harmonic oscillator.2 or by calculation that the potential 2h2 cos 2z has (harmonic oscillatorlike) We follow here R.1 Cosine Potential: Strong Coupling Solutions Preliminary remarks In the case of large couplings h2 we observe from Fig. Dingle and H. The cosine potential cos 2z depicted in Fig. Thus in Fig. In the case of the cosine potential. . 17.1 that the boundaries of regions of stability merge to lines. 17.24e) 17. 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. 17.3 Cosine Potential: Strong Coupling Solutions „ 5/i4 a2 = 4 H 12 763/i8 .2 The cosine potential. W.z3 ce2 = cos 2z . 17. 13824 .
2/i cos z V = 0. Thus we rewrite the Mathieu equation as y" + A + 2hz cos 2 \z ± 7T y = 0 and take c o s 2 [ z ± . we therefore set 9 A A = 2h2 +2hq + —.2. Periodic Potentials minima at z — ±ir/2. 8.26) 17.. n 0. This equation has normalizable solutions for X + 2h2 2h q0 = 2n + 1. Sees..3.26) into the original Mathieu equation and obtain y" + 2h q + rrxr . where A / 8 is a remainder. (17. Thus in this case the quantity on the left is not an exact odd integer. as in the case of the cosine potential.3) and is.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 = 0.2 and 8. Taking the remaining terms of the cosine expansion into account. but only approximately so. (17.354 CHAPTER 17. and we set this equal to q. Ion (17. We therefore wish to expand it about these points. of course.27) .25) This is a Weber (or parabolic cylinder) equation which we encountered earlier (cf.. simply the onedimensional Schrodinger equation for the harmonic oscillator..1.2 The solutions We insert Eq. there is tunneling from one barrier to the next. In the case of finite heights of the potential barriers. 4 (17.
16/i/ (17.+2h': 2h level splitting oscillator 2nd excited state associated solutions A^9 h=4 oscillator 1 st excited sate ^1 oscillator ground state A^Oh 2 large h 2 =0 Fig.29) is known.30) D A): COS « = 4z + 1 iq sin 2.3 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 . z).29) the equation can be written in the form D^A where 1 2 6 /i (16A" + 2 A ^ ) . (17. (17. 2 (17.3 Schematic picture of Mathieu equation eigenvalues.27) we )A = 0. another solution.31) We make the important observation here t h a t if one solution of the type (17. (17. the linearly independent one.28) A" ±4hcos zA' + 2h[ q =f sin z \ A. 17. and substituting this into Eq. Choosing y(z) = A(z)e2hsinz.17.
35) But since and so a term iiAq+n on the right hand side of Eq.34) The leading approximation A^ = Aq therefore leaves uncompensated terms on the right hand side of Eq.9) (9. 2 A A = («. (17.35) can be cancelled out by adding to A^ the next order contribution A& = 1 (g. (17. of course. Periodic Potentials obtained by changing the sign of z or alternatively the signs of both q and h throughout.30) amounting to Ri ] q = . 2[(92 + l) + A]. (9.32) cos^'+i) [\K . D{q%Aq+4i = 0. Aq+4 2Jh .94) = (9 + l)(9 + 3). ? . (9l)(93).356 CHAPTER 17. 2 /i D^Ag = 0.4)A.9 + 4) = (9.94) \ — Aq4 (17. . ? . (17.4 ] .4 ) ^ . (17. ^ ) A + ( ? . (9. when i = 0 — the terms not involving Aq in (17.33) (17.Q)Ag + ( ? . 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.(°) + A^ is the approximation of A to that order provided the coefficient of the term in (17.9) = 0.\z To lowest order the solution A is A^> = Aq. 1 + 4 ) ^ + 4 + (^. With some algebra one finds that 16^ + where (9.e.30) can be cancelled out by adding to A^> a new contribution — fiAq+4i/2i — except.^ 6T [(«>« + 4 ) A ? + 4 + (Q.35) not yet taken care of is set equal to zero.36) Then ^4._ 4 .g + 4) . i.
Thus in particular A(z. q. q.37) and so on — left uncompensated — must vanish. h).q 4)Aq+4 q + 4 1 q + + + + + <^(.g4) 4. h) = A(z. In its turn the contribution A^1' leaves uncompensated terms on the right hand side of Eq. Then for A = A(0)+A(D+A(2) + . q (17. . +4. . q . ) ^ 4..g + . Thus 0 = (g. Q . (17. h) = A(z.4) 1 1 (q._ 8 (17.3 Cosine Potential: Strong Coupling Solutions 357 This equation determines the quantity A in the eigenvalue equation to the same order of approximation. except those involving Aq. q . We note here again that if one solution y = A(z)e2hsmz 2hsmz is known. q .17.q + 4)(q + 4. q. (17.39) which is the equation from which the quantity A and hence the eigenvalue A is determined..8)A.4 .g + 4)(g + 4.30) and hence y = Ae2hsmz to be a solution of the Mathieu equation.39) is obtained by changing the sign of z or alternatively the signs of both q and h throughout. the sum of the terms in Aq in Eqs. q. . (q. (17.35) and (17.<J+4 Aq+& + 1 2 1 . an associated solution y — A(z)e~ with the same eigenvalue and thus the same expansion (17. A(z..4.38) Aq8 + 1 2 Clearly we can continue in this way.4) (q .q 8)Aq+8 (o)' ^±V(q 1 4. (q.q + 4) (g.)+ & ! = « ( .g) 26/i L_ 2 13 /i 2 ^±A{q + ^q) + —1 ^A{q^q) '17. A.30) amounting to RW 9 = fag+ — 2?h 1 ^ 2™h? +< 4 ) R(o) \ .4.4) (q . (17. can be taken care of by adding to A^ + A^ the next order contribution A^2\ where A (2) 2h u 2 (q. h). to satisfy Eq.40) .37) Here all the terms.
We observe that for the large values of h we are considering here. Note different definitions in use. We return to Eqs. .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 . The differential equation there is Eq.45) {B) We recognize Dq as the differential operator of the Hermite equation.67).d2B w dw2 + w(l w dB_ dw d dw 1.44) where D(B) 9 = dw2 $_ !)• (17." 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. (17. 2 w2 + H* (17. and discussed there.46) 2s(«. it is found that D(B) z+ — ]B = 0. Before we study the eigenvalue equation in more detail.i ) We encountered Hermite polynomials earlier in Chapter 6.43) 9 2eh . a solution Bq(w) of the equation Dq Bq(w) — 0 is Bq(w) oc He^^^iw). this means the solutions are valid in particular around z — 0 since then l / \ / 2 3> l/Vh ~ 0. (17.) i ( Q . where Hen{w) is a Hermite polynomial when n is an integer. 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. 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.e.358 CHAPTER 17.42) (17. i.41) so that the early successive contributions decrease in magnitude. (6.
7 T H 1 2 Z 4 <1.34). ..47) where the coefficients are the same as before. 16hJ (17.51) . Comparison with the case of solution A now shows that the form of that solution can be taken over here except that everywhere Aq{z) has to be replaced by Bq(w).e. the expansion for the quantity A being identical with that of the previous case.d2Bq + w{l dw2 "» 2 >^(» 2 + ^ = {q. i.. (17.qi)Bq_4.3 ) \He* ! (9+1) H. i. A{wz (17. with the equation C" + 4h cos z C + 2h ( q . i. The solution with contributions B^°\B^>. 1 COS I . which is an amazing and unique property of the Mathieu solutions — cf.4 (using the known recurrence relations of Hem{w)) the perturbation remainder can be linearized.17. also the same coefficients in the case of functions C below.49) Then the dominant order solution is C^°\w) = Cq{w) with and — again choosing the multiplicative factor suitably — Cq(w) = 2i(" +1 ) ^ .3 Cosine Potential: Strong Coupling Solutions 359 because then as shown in Example 17. Finally we obtain a third pair of solutions.50) C = 0.q + 4 ) £ g + 4 + (q.e. forms a rapidly decreasing expansion provided that \w (z)\ « Vh.e.sin z + by changing the independent variable to w(—z) = 4v/icos I 7r z (17. (17. y — Cexp(2hsinz) and Cexp(—2/isinz).48) This shows that this solution is valid in particular around z = ir/2. q)Bq + (q. 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.e. i. (17. as in Eq..
. These domains of validity are indicated in Fig. ie 1 COS I .(w2 + A)Hen. (17. q)Cq + (q.h). q + 4)C 9 + 4 + (q.44).4: Hermite function linearization of perturbation Using the recurrence relations (6. Again the solutions have the same structure as in the previous two cases and the eigenvalue expansion remains unchanged.q. He'n(w) = wHen(w) Hen+i(w). each of them associated with one and the same expansion for the eigenvalue.54) where the proportionality factor is complex. 2 J where n=(ql).q.h) ocB{z. (17.52) and the coefficients are again the same as before.w2)He'n .59d) of Hermite functions Hen(w).q.(ql)2(q3)HeX (95) i(« 2 + i) + iA Hei (91) . 17.7 T 1 2 Z 4 <1. show that the perturbation remainder of Eq. Solution: Inserting the dominant Hermite function solution into the right hand sid of Eq. can be reexpressed as 72 = (n + 1 ~(q + l)Hen+2 l)He. (17.n2(n l)Hen2 .4)C g _ 4 . we have a remainder •R:=w2He„+w(l Using the recurrence relations wHen(w) = Hen+i(w) + nHeni{w). (9+3) (2n 2 + 2n + 1) + . (17. Periodic Potentials since then (using the known recurrence relations of He^) ~Mw2^ +w(l+w2)^ + (w2 + o dwz dw V 2 (q.4. A closer look at that expansion is our next objective.53) This solution is valid in particular around z = —7r/2. 2' the remainder 72. Example 17.360 CHAPTER 17. We have thus determined three pairs of solutions of the Mathieu equation. 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. (17. We note that C(z.44) can be linearized with the same coefficients as in the case of the trigonometric solutions. The solution y = C exp(2/i sin z) is a decreasing asymptotic expansion provided that w(z) ^ v^/i.A Hen .
r j=r. ) .j^0 e 2ft sin z A + ~ww 1 7 i + ^7ir{ p i(9' ^ ^ + 4 + Pi(q. q .1i .q4) (17.i ) ( w ) 2l(. l)Ag4 + P2(q. e. (q.4) = (q ..q + A){q + A. (q.55) M2 = M.l)(q .l)[l(q _!)]. .56) (94.. We can write the solutions obtained previously.3). in the following compact form with coefficients Pr(q. (q. with (. ~l)Aq4} 2h 1 { P 2 ( < Z 2 + + P2(<? lAq+i ) 8 A ) ' " ' +P2(q. 17.)Bqi(w). (?) 1 1 + (9.q + 4)Bq+4(w) + (q.' 361 in 2*f*1>[i(gl)]! = (q. (b) The allowed moves are +4.j): y = = = e e2hsinzA{z) oo 2ftsinz r=0 V . q) = 2[(q2 + 1) + A].17. 2 + ^ M 3 + (17.0 and —4.g) 1 1 and so on. 2)Aq_8} + • (17. no intermediate move to or from q being allowed.3 Cosine Potential: Strong Coupling Solutions It follows t h a t with the definition of Bq (w) as Bq(w) that He K . the solution y = exp(2/i sin 2)^(2).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). q + 4) = (g + 1)(? + 3). ^ 1 Hv" '^ ^' ^ ' .q 4.g4)(q4. .57) .q + 4) (q + 4.g. q)Bq(w) + (q. We see that (a) every coefficient Mr results from a sequence of r moves from q back to q. The coefficients have thus been constructed in a way which now allows the formulation of a recurrence relation. In addition (c) each move except the last has to be divided by the final displacement from q divided by 4.3.
j)] = 2j]jPr2(g. the factor j removing a duplication of factors in the denominator at the junction. i. . Thus we have. g + 4j)P r _!(g.g + 8)(g + 8. . j + 1) (17.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. W. 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) = (l)rPr(q.j)Pr+1(g.l ) .j). P2(q 2) Pl(g. 2) 1 2 ' ' ^ ' ^ 1 (g. x)_ (g. for instance. r ^JtPr^jOPr+lfejV^rfejO^r+l^j)] r 2^jPr(g.i (g.j).P 2 ( g ._1) = Mzil. p = (<?><? + 4 ) ( g + 4 >g + 8) .g + 4)(g + 4. l ) .g + 4) ^3(9. j) +(g + 4j + 4.60) M2r+l = = *R.g + 4 ) ( g + 4. (17. 0) = 1 and all other P0{q.g + 4)(g + 4.j).1) + (g + 4j.g + 12) 1 2 3 (g.g + 8)(g + 8.j)^2(9.j). g + 4 j ) P r .e.g + 4) 2 1 (g. Dingle and H. (17. as jP r 2 (g.4.g + 4) + 1 1 1 and so on.g + 4)(g + 4. Periodic Potentials PlM)=(M+i).j). M2 = P 2 ( g . g + 4j)P r _i (g.58) with P 0 (g. J. B. and the contribution from a sequence of r moves from q + 4j back to g. j . and in general* M2r = J]j[P2(g.34) we deduce that Pr(q. Miiller [73].j ^ 0) = 0.g + 4)(g + 4.3) ^3(9.362 Clearly CHAPTER 17. j) = (g + 4j .59) One can now write down formulas for M^r and M<ir+\ in terms of the coefficients Pr{q. From the coefficients (17.
as a result of tunneling between wells. This means that the in approaching such a domain the appropriate solutions must become proportional in view of their uniqueness. t Terms up to and including those of order l//i 5 had been obtained by others and by different methods. remains to be determined next. of course. Up to > > this point. h — —h. as pointed out earlier.3 Cosine Potential: Strong Coupling Solutions In this way one obtains the eigenvalue expansion A = 363 _2h2 + 2hq^(q2 1 . Its precise deviation from qo. These domains in the interval —7r/2 < z < 7r/2 are indicated in Fig.(5q 4 + l)^Jiq(q2 + 3) + 410g2 + 405) + 34q2 + 9) . *E.62a) f R . We observe that there are regions. a= (8/i)3(9x) [1(91)]! 1 . [116]. [137].4. It will be seen in Chapter 18 that in the case of anharmonic oscillators. q is only known to be approximately an odd integer qo = 2n + 1.^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. B.3. L.284g + 57) + (17. Dingle and H.92q3 + 70g2 .^ . 17. .5 the following proportionalities there: B[w(z)] = aA(z).* We observe that the expansion remains unchanged under the replacements q — —q.4 The level splitting Above we obtained three pairs of solutions along with their domains of validity as decreasing asymptotic expansions. S. 17.( 3 ^ . J. where solutions overlap. In fact we demonstrate in Example 17.2 * +3) + ^ 2 (9g4 . W. Ince [136]. Miiller [73]. Goldstein [115].61) +288161796g2 + 130610637) These are the terms given explicitly in the reference cited above.17. the solutions and eigenvalue expansions again have the same type of symmetries.
7T H Aq(z) = = C O S ^ . Periodic Potentials + _(3<?+2<? + 3) + 2 15 /i 2 (9g 4 + 92g3 + 70q2 + 284g + 57) + • • • (17. We write and expand this function as follows: . and determine the constants a and a.4 Domains of solutions and their domains of overlap.(17.62b) *z domains of overlap Fig. a .1 ) ( 7T+ z 1 — COS 2 1 \l^+1> 2 Z C O S S ^ .COS 7T 2!42 \4 2 COS /l 7T V4 Z H 1 2 Z H 1 2 .1 ) ( 7T + .32).364 and C[w(z)] =a~A(z)._ ^ l CHAPTER 17.Z (g+1) 1 H 1!4 4 2 1 (g + l)(g + 5) . 17. Solution: We begin with the solution containing the function A{z). The first term of the expansion of A(z) is the function Aq(z) given by Eq. Example 17.1 4 .5: Proportionality of solutions in domains of overlap Show that in their common domain of validity B = aA and C = aA.
!)(. Hence we have to construct even and odd solutions and impose the boundary conditions at z = 7r/2. _ 1 " 2 ( 7 r / 4 + z/2) in the result is seen to be a = 1 2 /i 7 fa. These are the conditions of Eqs.e.g. s .g + 4) (q + 4. In order to link our asymptotic solutions with the Mathieu functions of integral order defined earlier in the consideration of small values of h2. A*1) given by Eq. (17.92q 3 + 70q 2 .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.q) 2 7 /i 1 1 4(g + 3) 1 (g + l ) 2 ( 9 + 3) + 29ft ti(9l)]l i (8/1)4 ( 9 .3 Cosine Potential: Strong Coupling Solutions 365 Inserting this into A = A(°> + A' 1 ) + • • •.i ) 1 (g.36).63) In Example 17. For further details and explicit expressions of higher order terms we refer to Dingle and Miiller [73]. with e. 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 . with (8h)li''1) (3g2 . Since the solution with A is obtained from that with A by a change of sign of z. As we saw there. \ql)/2(w) 2i(9D[i(._i)]! w*te» 2l(91 (. (17. the coefficient of the dominant factor c o s ( .2 for the case of small/i2 solutions and thus insure that in both cases the same boundary conditions are obeyed.46). (17.6 at the end of this section we derive for these \axgeh2 solutions the conditions similar to those given in various equations in Sec. This is essentially the expansion of the parabolic cylinder function with the exponential factor removed — see Eq. these are conditions imposed on even and odd solutions at the point z = it/2. i. (6.i ) [(91)]! It follows therefore that over their common range of validity B = aA with a = b/a.284g + 57) + • The other relation is obtained similarly.61) for the connection and Eq.2.2g + 3) 1 [£(91)]! 2?h 215h2 (9g 4 .53) for the expansion.17.22). (8. we have as even and odd solutions y±ocA(z)eZflsmz±A(z)e 2/isin. we have to impose the appropriate boundary conditions.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. we obtain Bq[w(z)} H. 17.
a ± C[w{z)}c_2hsinz^ a (17. W. p. give the following expressions for polynomials with argument zero: ~19)"(^)!.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\ ! ' . He'v(0) = Hev+1{0).66) ^E.66) then follows from the recurrence relation (6. The functions defined as in Eq.64) These are now the solutions around z — ir/2 which permit us to apply the boundary conditions (17. He'u{w) = wHev{w)He„+1{w).22) and to correlate the solutions to those for small values of h2. one obtains the first of relations (17.59d).63) to the domain around z = ir/2 by using the proportionalities (17.e.62a) and (17. (17.e. Then y± oc ^ M e ^ i n . The second of relations (17. i. i. With the help of the duplication and inversion formulas of factorials.» a \dw _ C[w = 0}c_2h a w=0 = Q (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 = 0.366 CHAPTER 17. Periodic Potentials We can extend the solutions (17.62b). 80. He2n+1 (0) = 0.66). . Thus y+ = ce and y_ = se.65) For the evaluation of these conditions we require the following expressions involving Hermite functions which we obtain from Tables of Special Functions:^ ^ ( =0) = 2TT2I//2 Fi(^ijii' He2n(0) = ( Ur w=0 •W+m' (17. Magnus and F. and then replacing 2n by i>.g. Oberhettinger [181].
67c) 4 2 \dw)w=Q C[w = 0] [\(q .17.( g .67d) Considering now the second of conditions (17.» 2 6 /i the condition reduces to the equation cot{ .82g2 .Qo) + 0[(q .67b) <f ~ 82g^ .\4 q + m { 4« VJL ^[±(g3)]!sin{f(gl)} [J(ffl) "(^L^»{^': _ _q_ ¥h + g . B[w = 0] 2TT _ _g_ ¥h+ g 24/* _ JL g4 .67c).3 Cosine Potential: Strong Coupling Solutions 367 With these expressions we obtain by insertion into B and its derivative.67a) and (17. and similarly for C and its derivative. from which one obtains "1 1) 4<*3) 2(9!)/2 2 ^ \b. f^(16h)i/2e4h 2 [Uq1)]\ 3(g2 + l) + 2^h2 (9g4 .! ) ] ! [ . the solutions would be given by q = go = 3.106g .65) we have B[w = 0] _ a C[w = 0] ~ ~ ^ 4/i (17.87 4 2 ¥w (17.136g + 9) + • • • .40g3 + 18g2 .69) If the right hand side of this equation were exactly equal to zero.11.67a) g 106g 87 2 14 /i 2 (17. .1) 7T .39 (17.7.39 2~^h? (17.g0)3]. Hence expanding the cotangent about these points. and using the reflection formula (8.68) Inserting here expansions (17. (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 . we have cot j ^ ( g ~ 1)} = ~\<(l .
= _a a 6 _4h (17. i..71) = 2 13 /i 2 fc{l6h)i/2e4h 1 3(g2 + 1) 2 [4(51)]! L (9g4 . we use \(q)~\(q0) + (qq0)(^J .408g0 + 1089) 2 /i +• (17.40gg + 18gg .70) + ^ 2 (9?o .136q + 9) + .70) except that now qo = 1.9.72) This leads again to the result (17.136g0 + 9) 2 13 /i 2 1 4 2 19 3f(9gg .40g3 + 18q2 .120g£ + 467g .65) leads to the same result except for a change in sign.136g0 + 9) + • Clearly the last of conditions (17.528gg + 3307Q . Periodic Potentials ' " Vn [i(gol)]' 3(g02 + 1) 26/i .73) We now return to the eigenvalue X(q) and expand this around an odd integer qo.. Considering now the first of conditions (17. .65) yields the same result except for a change in sign. (17.(17.368 Hence we obtain qq0 a [2(16h)i°/2e*h CHAPTER 17.65) we have (dB/dw)w=0 (dC/dz)w=Q Proceeding as above one obtains tan{(ql)\ 7T . The remaining condition (17.40gg + 18gg .e. . Thus we have 2 (16/i) qo/2 T2 4.h qqo = * [£(*>!)]! 1 3(902 + 1) 26h +^^{9q^ ..5..
results from the boundary conditions. 9o = 2n + 1. (17. minus sign).17.74a) we obtain for the level splitting in dominant order* x r ^ \ / N 2(16/i)>+1 _4h A+(go) . Miiller [73].3 Cosine Potential: Strong Coupling Solutions Differentiating \(q) of Eq. p.2. Dingle and H.. solution of the Schrodinger equation for the basic and nontrivial periodic potential is important for a thorough understanding of its quantum mechanics.61). i. Goldstein [116]. (17. *In Sec.3 we call this difference 26E„n. + ^ 2 (9<?o + 89o " 78g02 . In the literature one finds frequently the statement that the splitting is a nonperturbative effect. . (17. W.87) = Azpteo).^ ( 3 g 0 2 + 8go + 3) A ( Q 0 ) T (87r)V2[i ( g o i)]l L2 w .e. plus sign) give the socalled level splitting AA(go) as a consequence of tunneling indicated schematically in Fig. Effectively the degenerate eigenvalue expansion results from the anharmonic terms contained in the power expansion of the cosine potential. example 3. B. M.. 120.3. J. Arscott [11]. and that for the odd Mathieu functions se qo+ i and se9o (lower. . quite apart from applications. (17. The complete. In particular we saw that the perturbation expansion yielded only the degenerate eigenvalue expansion. n = 0. The tunneling effect.1. 17. 20.88q0 .e. made evident by the splitting of the asymptotically degenerate (harmonic) oscillator levels.. though approximate. i. See also F.A_(g0) ^ . We see here that *R.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. and not the separation of these as a result of tunneling. In the literature that we follow here* in all cases several more higher order terms have been given.3. this becomes 369 = (16/i)^+1e4h l . just as any solvable problem is important for the insights it offers in a concrete and transparent form. The dominant contributions were also found by S. n ' r 1 / rrre 4/l ^odd(go) Seven (<Zo) Wfaqom (16/i)T n\V2n e~4h.74b) The above results for the cosine potential are in many respects important.t From Eq.
§ The boundary conditions we imposed here are those of a selfadjoint problem which has real eigenvalues. (17.2. results from the imposition of boundary conditions. y^{z) = Aq(z)e~2hsin *.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.6: Translation of Solutions For the largeh 2 even and odd solutions (17. evident through the exponential factor exp(—4/i) in Eq. As mentioned at the beginning. From Eq. T y'_(0) y'A±*) y{z). . and one can compare the methods.32) we obtain Aq(z ± TT) = (l)±i<9^'Aq(z). The splitting can also be obtained with some methods of large order of perturbation theory. y+(0) 2/_(0) Solution: The derivation requires a cumbersome tracking of minus signs. The explicit derivation of the level splitting (17. yA{z) = Aq(z)e2hsin '. 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.2: y+(±ir) J4(±TT) y+{z±n) y(z±7r) = = ± —y+(z)± 2/+(0) . 17. M±TT) . the exponential factor we encountered in the semiclassical method where it has the structure of the factor exp(classical action/ft). 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. _3/+(z) + ____2/_(z).370 CHAPTER 17. 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. . Thus we do not reproduce every step. We begin with the solution containing the function A(z). which one can then try to solve in order to obtain the behaviour of the coefficients of the late terms of the expansion. in effect. (17. The splitting can also be derived by the pathintegral method as will be shown in Chapter 26. 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. Then we obtain first y±(z) = yA(z)±y^(z). the Schrodinger equation with the cosine potential is a special case of several other more general cases. Example 17. Periodic Potentials this nonperturbative effect.74b) is also important for various other reasons. The exponential factor represents.74b). ^ ± n ) = _ ( _ 1 ) T i(<JTl) A 9 ( 2 ) (the extra minus sign in the second equation coming from Aq(z ± 27r) = — Aq(z)). Naturally one expects the solutions there to be related to those of the periodic case considered here.
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+.3 The Ellipsoidal Potential Replacing VA.yV+\z^ = [j/+(0)] 2 (<? .VA DV 371 1 V±.1 Elliptic and Ellipsoidal Potentials Introduction Our intention here is partly to deal with other periodic potentials.4ft) sin j ^n(q . Finally for the derivative of the odd solution we obtain y'_{ir) = K ( T T ) . A ?q( Z = ±TT) = .4ft) cos i TT(<J + 1 ) 1 . — j / + ( 2 ± 7 r ) = ± i s i n  ^ ( g = F l ) U + ( z ) .. Similarly we find y(z±ir) = = and by putting z = 0: V.4ft) ~ 8(q . (17. Aq(z) The derivatives can be handled with the help of Eq.) cos < 7r(q 1) [ = .g A ? ( z = ±TT). (l^ifr^i^O + Cl^ifo^j/aCO cos  . which is easiest by comparison with the foregoing treatment of the Mathieu equation.4 17.y]z=„ = [y+yL . where Aq(0) = Aq(0) = V2.z ) .T ( ? =F 1) p + (0).7 r ( q = F l ) } j / .31). A' (z = ±ir) = (. etc.( ± 7 r ) = c ° s  .( z ) From this we deduce for z = 0. Replacing sines and cosines by the functional expressions. y+(0)~2V2.1 )  . and y+(z = ±TT) = V2(q .q).4ft. = ±ir). we obtain — with Aq{z + 7r) = (1)^ ^ 2 A .[A'qM + 2hAq(n)] = iV2(q . y'_ (0) = \/2(4h .%/2(g .• Then y'+ (0) = 0.2hAq(n)] .c o s  ~(<1 T 1) f y . 17.4. ( .( 2 ) . we obtain the conditions stated at the beginning.17. The Lame equation has not been a widely known equation of mathematical physics so .7 r ( g ^ l ) U + ( z ) q : i s i n { . the operator having solution Thus ^ ( 0 ) = (q/2)Aq(0) = A~'q(0) and A'q(z = ±TT) = ~qAq(z qAq(z = ±7r). but also to enable a brief familiarization with the Lame equation. since A q (0) = \/2 = Aq(0): y+(±n) = ±isin^(9=Fl)J3/+(0).Ah) ~ 32ft.
* As Arscott [11] (p. and was continued by Erdelyi. J. M.5 < z < 5. W.Q. *A.75) ' T h i s means double well. Periodic Potentials far. 17.1.0. Liang. Tchrakian [165]). however. 194) remarks to parallels with the Mathieu equation:". . inverted double well and cosine potentials (J. (17. This line of investigation. Erdelyi [85]. p.0. 194) remarks. but also cubic potentials. one arrives at three equations of which one is the ellipsoidal wave equation pt + [A .0 and . the nonoccurrence of the Floquet exponent..K2sn2u . Arscott [11].. § F .. If one separates the wave equation § V 2 * + J2* = 0 in ellipsoidal coordinates (which we do not need to consider here).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.372 CHAPTER 17.fi2A." Fig. as alluded to at the beginning. so that Floquet's theorem cannot be immediately applied . H. k) for k = 0. f E . 19...98. for general n the solution of Lame's equation is not singlevalued owing to the singularities in the finite part of the uplane.5.75. L.. Arscott [11] (p. 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 . MullerKirsten and D. But more recently it has been observed to arise in various contexts.0. the new stage of the development of the investigation of the Lame equation was really initiated by Ince^ around 1940.5 The potential sn2(z. Ince [138]..* and the work of both prepared the ground for presentday investigations. The most conspicuous difference compared with the Mathieu equation is. H.
1.3 The Ellipsoidal Potential 373 Here K2 and A are separation constants and U2 = l2(a2 — c2). The second step is to insert (17. . Although the results have been calculated for the ellipsoidal wave equation.2 = n(n + l)k2. (17.17. The function sn2u is plotted in Fig.4. . N = 0. The range of the independent variable u is 0 < u < 2K.78) We follow in this brief recapitulation the description in H. Jianzu Zhang and Yunbo Zhang [206]. MiillerKirsten. cmx and dnu. For barriers of finite height the parameter q is only approximately an odd integer qo in view of tunneling effects. i. in front of the second derivative has to be kept in mind (mo being the mass).\k\ < 1. In order to distinguish the above equation from that with the Lame potential. K being the complete elliptic integral of the first kind.K) = qK+^±.e. (17. i. W. J.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. in the case n2 — oo. 17.K2sx?u}y = 0. In comparisons with the Schrodinger equation. and of the derivation of the level splitting'. 17. where n is real and > —1/2 (and n is an integer in the case of solutions called Lame polynomials) where k. (17. .75) reduces to Lame's equation and one writes K.e. the equation ^  + [A . . Eq. 2 . If we put fl = 0.a > b > c being related to the lengths of the three axes of the ellipsoid in a Cartesian coordinate system.77) where q —> go = 2iV + 1.2 Solutions and eigenvalues In the following we sketch the main points of the method of deriving asymptotic expansions of the eigenvalues and eigenfunctions. The first step is to write the eigenvalue A as A(q. for very > high barriers (harmonic oscillator approximation around a minimum of the potential). we consider mainly the Lame equation.2k.77) into Eq. (17. is the elliptic modulus of the Jacobian elliptic functions snu. the usual factor —h2/2mo. we refer to the potential consisting of the two terms with sn 2 n and sn 4 n as the ellipsoidal potential.5 for several values of fc. (17.76) and to write the solution y = A(«)exp  f Ksnudul = A(u)[f{u)]K.
i.q.80) The domain of validity of these solutions is that away from an extremum of the potential. ll^ ^A(u)[f(u)]^k±A(u)[f(u)}^k.374 where CHAPTER 17.K). Thus in the third step two more pairs of solutions B. _ Solving the resulting equations iteratively as before. A. /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.77). dnu± cnu K Thus one can construct solutions Ec(u). k' = Vl~ k\ \dmi±cnu/ 17. one pair in terms of Hermite functions of a real variable.. more precisly for dnu = cnu F 1 » . one obtains again the same expansion for A. A into equations in terms of the variables z(u) = ^—{k' . i.q. .e. which are respectively even in u (or snu) or odd. (17. dnu =F cnu . (17. \/8K (dmz = cn« \ ' F . C replacing yl.K)=A(u. .K — > A(u)=A(u + 2K). determine their proportionality factors) in domains of overlap (their extreme regions of validity). Periodic Potentials . A(u. / z J . but solutions B. ^ (17.81) Since these expansions are not valid at the extrema of the potential (where the boundary conditions are to be imposed). one has to derive new sets of solutions there and match these to the former (i.82 . A are derived.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.e. A second solution is written y = A(u) exp < / nsnudu >. C which are valid for dnu±cnu <1. B and C.Es(ii). the other in terms of those of an imaginary variable. (17. by transforming the equations for A.e.
C = a~A.75). 17. Erdelyi.2)4(33g4 + 410g2 + 405) 24/c 2 (l + k2)2(7q4 + 90q2 + 95) + 16fc4(994 + 130g2 + 173) +512f22fc4(l + k2)(q2 + 11)} . p. F.6(5g4 . *A. a of _ _ B = aA. who obtained this expansion for the eigenvalues of Lame's equation (i. Miiller [205]. W. as well as \duJ2K 11 (17. 64.4. .85) The first three terms of this expansion were first given by Ince [138].^ ^ { ( 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£.3 The Ellipsoidal Potential 375 In their regions of overlap one can determine the proportionality factors a. Es(2f0 = Es(0) = 0.e.38g2 . (17. W.384n2k4(q2 + 1)} q r {(l + 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) . J. Tricomi [87]. This expansion is found to be in the more general case of Eq.e.63)} + ••• . the ellipsoidal equation.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. one sets at it = 0 and u = IK altogether** Ec(2A") = Ec(0) = 0. (17.17.86) \duj0 o h r 2K > \duJ =[sr 0 =0\duj (1787) H. G.3 T h e level splitting In the fourth and final step one applies the appropriate boundary conditions on these solutions. i. £2 = 0). Magnus. (17. Oberhettinger and F.
Ec£ 0_1 and Es2° of periods 4K.:J {3(1 + k2)2 (9q* . (17.i Here the upper sign refers to Ec*> l or Ec£°.g 0 ) ( ^ . 2K and 4K respectively. 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).j 26K2 (1 85) = A(g0) + (q .l n 2^+ i J + 0(«i) (17. W. 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. Finally expanding A(g) ~ A(g0) + ( g .136g0 + 9) n 3. from which the difference q .1 / 2 (2TT) 1 /22« l + (l_fe)^«( i . Muller [205]. Periodic Potentials These conditions define respectively functions EcX°.90) + 3. + fc )(3g + 8g0 + 3) k2)2(9q% + 8g^ .go)« qo(l + k2 22K {3(1 + k2)2(q2 + 1) . J. go being an odd integer.78g^ .88g0 . 2K.Es^ 0+1 .89) one obtains the eigenvalues from which the level splitting can be deduced.87) (17. and the lower to Es*>+1 or Es*5.2 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)*. 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 .91) tt H.2 /c +256fc2g0(g2 + 5)} (17. .4k2(q2 + 2g0 + 5)} + • • • ].376 CHAPTER 17.g0 is obtained by expansion around zeros.40g03 + 18g02 .
2h2 = A. .4 Reduction to Mathieu functions Under certain limiting conditions the ellipsoidal wave equation reduces to the Mathieu equation.2h2 cos 2x}y = 0.Ah2 sin2 u}y = 0.17. in which they dubbed the classical configurations Lame instantons.* M G .e. (17. Rao [79].86). Dunne and K. Ince [138].f ln[(l + fc)/(lfc)] _ e 4h) ( 8" \ ^^ __ (16h)«. J. h . A = 0. that calculations with the Schrodinger equation are not only easier.93) (17. K = 2/i. i. this equation becomes y" + {A . A . One can see. L. then snit — sin u and Lame's equation (f2 = 0) becomes > y" + {A . W. however. (17./2i and hence A^ A/ A AU I2 .J1 O. (17. Thus if k — 0 and n —> oo in such a way that > K2 = n(n + l)k2 ~ finite. (17. *H.* Apart from the choice of notation. V. it = x ± ^ . Replacing u by x ± IT/2. Hence the conditions fi = 0. Thus in the case of the dominant contribution of Eq.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. 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. + E.92) reduce the periodic ellipsoidal wave functions and their eigenvalues to corresponding Mathieu functions and their eigenvalues.4 h ( i e M 9 0 7 2 \. .90) one has in this limit: /l±±\ ~K/U = e . K2 = h2 (say).74a).4. 17.3 The Ellipsoidal Potential 377 This result agrees with a result of Dunne and Rao** who calculated this expression using instanton methods. but also more easily generalizable. Miiller [206]. A(g)>A(g0)T4/i\/e ^ [Hqo ~ !)]•' in agreement with Eq.87) agree with those of Ince.2h2 . the conditions (17.
e. H. Thus.** The associated Mathieu equation appears also in string theory in connection with fluctuations about a L>3brane (see Chapter 19). 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. i. W. Ouvry [276] and Jianzu Zhang. Cho.5 Concluding Remarks The potentials we considered above will reappear in later chapters. .g. J. Rana [287]. Quian [123]. W. Gu and S. D. N. H. Sukhatme [148]. S. See Chapter 25. de Veigy and S. § . for instance. Liang. MiillerKirsten and D. MiillerKirsten and J.§ This is therefore a very important equation which in the limit of infinite period becomes a PoschlTeller equation. A generalized associated Lame potential has been considered by A.378 CHAPTER 17.W. M. A host of related elliptic equations has recently been discovered and studied.^ The equations considered above also appear in diverse new problems of physics. thus revealing an unexpected significance of this not so wellknown equation of mathematical physics. Tchrakian [165]. J. Periodic Potentials 17. See e. Kan and K. 1 1 See Z. This is an interesting problem. Ganguly [103]. Shiraishi [49]. all basic potentials. H. In particular we shall encounter the Lame equation as the equation of small fluctuations around classical configurations associated with cosine. **Y. Khare and U. 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.Q .Y. J . A. double well. inverted double well and cubic potentials.
Wiedemann [4] and a revised version of parts of this reference by J. Lee [98] referred to it as a "long standing difficult problem of a quartic potential with symmetric minimd'1. Wu. Bender and 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. J. The recent work of R. In the cases treated most frequently in the literature the anharmonic *We follow here largely P. 379 . Liang and H. There is no end to this: An entirely new approach to anharmonic oscillators was recently developed by M. both of which make the calculation more difficult. Bender 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.". W. This seemingly simple problem revealed extremely rich internal structure . [19]. f C . Priedberg and T.. T. Achuthan. D. ' T h u s A. W. T. In particular the investigations of Bender and Wu. MiillerKirsten and A. H. M.1 Introductory Remarks The anharmonic (quartic) oscillator* has repeatedly been the subject of detailed investigations related to perturbation theory. MiillerKirsten [163]. demonstrates that derivations of such a quantity are much less familiar than calculations of discrete bound state eigenenergies in quantum mechanical problems. Weinstein [281].t which related analyticity considerations to perturbation theory and hence to the large order behaviour of the eigenvalue expansion attracted widespread interest.. Wu [18]. J. Turbiner [273] remarks: "It can not be an exaggeration to say that after the seminal papers by C. in nearly a thousand of physics articles the problem of the anharmonic oscillator was touched in one way or another.Chapter 18 Anharmonic Oscillator Potentials 18.Q.
V(z) = \\h*\z2 I 2i 4 \cr\z . which are nonetheless linked as a consequence of their common origin which is for all one and the same basic differential equation. Anharmonic Oscillator Potentials oscillator potential is defined by the sum of an harmonic oscillator potential and a quartic contribution. and thus lead to very different physical situations. Case (1) is obviously the simplest with the anharmonic term implying simply a shift of the discrete harmonic oscillator eigenvalues with similarly . 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.1 The three different types of anharmonic potentials. *~z (1) (2) (3) Fig. (3) Complex eigenvalues with tunneling: In this case. V(z) = \\h*\z* + \\<?\z*.1. 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. 18. with the potential described as an inverted double well potential. These contributions may be given different signs. described as the case of the double well potential.380 CHAPTER 18. 18.
It is this expansion which led to a large number of investigations culminating (so to speak) in the work of Bender and Wu who established the asymptotic nature of the expansion.18. The imaginary part will be rederived from path integrals in Chapter 26. since the potential decreases without limit on either side of the centre. 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. We therefore consider in this chapter and in Chapter 20 in detail some prominent examples and in such a way. The eigenvalues of this case are given by Eq. In this case our aim is to obtain the aforementioned complex eigenvalue. This type of potential allows tunneling through the barriers and hence a passage out to infinity so that a current can be defined. Thus we are mainly concerned with the double well potential and its inverted form.86) below. The level splitting will be rederived from path integrals in Chapter 26. Case (3) is seen to be very different from the first two cases. Calculations of complex eigenvalues (imaginary parts of eigenenergies) are rare in texts on quantum mechanics. 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. Case (2) is also seen to allow only discrete eigenvalues (the potential rising to infinity on either side). as is sometimes hoped. (18.175) 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. We begin with the latter. it will not be possible to obtain the exponentially small contributions with convergent expansions. The shift of the eigenvalues is best calculated with straightforward perturbation theory. and this behaviour is that of asymptotic expansions our treatment largely terminates this muchdiscussed topic. except for a change of sign of c 2 . that the general applicability of the method becomes evident. 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. (18. We do not dwell on Case (1) since this is effectively included in the first part of Case (3). The boundary conditions are nonselfadjoint and hence the eigenvalues are complex.e. i.1 Introductory Remarks 381 normalisable wave functions. The eigenvalues of this case are given by Eq. If the barriers are sufficiently high we expect the states in the trough to approximate those of an harmonic oscillator. however with decay as a consequence of tunneling. 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 result is an expansion in descending powers of h?.
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 .1 and more specifically in Fig. We take here h = 1 and the mass rriQ of the particle = 1/2.2 The inverted double well potential with (hatched) oscillator potential. 18. (18. we can pass from one case to the other by making the replacements: 4 E1/2 = 2Ei. Anharmonic Oscillator Potentials 18. If suffixes 1/2.2.2 18. 18.382 CHAPTER 18. h 4 2 2h1> ~1/2 2c?. a point which has to be kept in mind in comparisons. 18. The potential in this case is given by V{z) = v(z). and the Schrodinger equation to be considered is C L^ + [E + v(z)]y = 0. v(z) = ^h4z2 + ^c2z\ (18.2. (18.1) for h4 and c 2 real and positive.2) We adopt the following conventions which it is essential to state in order to assist comparison with other literature.1 refer to the two cases.z Fig. 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 .
and a variable w defined by setting E= b* 1 + A A and w = hz. The result will be that derived originally by Bender and Wu.2) as Vq(u >)y(w) = with T)Jin\ 1 (A + c2w )y(w 0 (18. _i\(w) and q = qo = 2n + l.. 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.. 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 = 0.7a) The problem then reduces to that of the pure harmonic oscillator with y(w) a parabolic cylinder function Dn(w).18.2. (18.2. the harmonic part of the > potential dominates over the quartic contribution and Eq. (18.e. i. (18. y{w) oc Di. The problem here is to obtain the solutions in various domains of the variable..2 The Inverted Double Well Potential 383 Introducing a parameter q and a quantity A = A(qr. (18. n = 0. /i).1. 18.d2 .5) becomes . then to specify the necessary boundary conditions and finally to exploit the latter for the derivation of the complex eigenvalue.4) we can rewrite Eq. and h large the eigenvalues are essentially perturbatively shifted eigenvalues of the harmonic oscillator as is evident from Fig.5) (18 6) "^'"dw* ' * 2' 2 In the domain of w finite. to match these in domains of overlap.1/7  Vg(w)y(w) = o(J^j. although our method of matched asymptotic expansions here (which parallels that used in the case of the cosine potential) is different..8) h8 v"{z±) = h and V(z±) = 5 2 2 c ' 2 2 Thus for c > 0 and relatively small. \h?\ — oo and c finite.
these solutions are not valid around z = 0. as encountered and explained earlier.9) z =0 these Thus these c2z4 1/2 + • 2 (18. •A~2 hrz" 2^\ c^z V2 —— + (18.12) Then A(z) is found to satisfy the following equation r /iV c V V/2 A"(z)±2<^. This observation allows us to define the pair of solutions yA(z) = .13) — that one equation (of the two alternatives) follows from the other by changing the sign of z throughout.9) we obtain 1 h4z2 c2z4 A L2 (18.+ ^ . as we shall see. Inserting (18.4(2.2 We are concerned with the equation d2y(z dz2 where ^ 1 .2 A (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 ) .14b) . Before we return to solutions we derive a new pair which is valid in the adjoining domains. be and A = A(q.14a) 1/2 VA( ) = A(z)exp (18.) exp Z ifdz[ifdz[ h4z2 4 h"z2 + + cV cV 1/2' (18. We observe — before touching the square roots in Eq. 1 d r A!{z)±iA{z)±{ 0. (18.10) into (18.2.11) y(z) = A(z)exp ± z dz h4z2 4 + E hAz2 + cV y(z) = 0 (18.13) + 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. Anharmonic Oscillator Potentials T h r e e pairs of solutions 18. (18. y 4 2/i Here again q is a parameter still to 2 determined from boundary conditions. This construction is simplified by the consideration of symmetry properties of our solutions which arise at this point. In order to arrive at solutions we set in Eq.384 CHAPTER 18. h) is obtained from the perturbation expansion of the eigenvalue.10) E= qh2 ^ j.
20a) These expansions are valid as decreasing asymptotic expansions in the domain \z\ >0 h.4 .13) TzA'(z) T ^A(z) + qA{z) = O We define Aq{z) as the solution of the equation zA'(z)(ql)Aq(z)=0.16) For large h2 we can write the Eqs. we do not pursue their calculation. (18.Aq(z) approximate the solutions of Eqs. (18.34) below) — to be derived in detail in Example 18. Thus we now have the pair of solutions yA(z) = exp yA(z) = exp dz I^4 + l„2„4 1/2. We take the square root by setting z2h^1/2 +.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.. 385 (18.19) We see that one solution follows from the other by replacing z by — z.18.20b) .13) and A(z) that of the lower of these equations. (18. Clearly Aq(z). 1 2 4 2 Aq(z) + 0 (18.zh2 =()*2"(18. (18. 1 (18. (18. Since these higher order contributions are of little interest for our present considerations.1 and as a verification again in connection with the solution y# — as the other solutions.15) where A{z) is the solution of the upper of Eqs.13) and we can develop a perturbation theory along the lines of our method as employed in the case of periodic potentials. /H 4 Z h +~C Z 21. / c z Aq{z)+0 1/2 . One finds that these solutions are associated with the same asymptotic expansion for A and hence E (given by Eq.e. i.2 The Inverted Double Well Potential with A(z) = A{z) and yA(z) = yA(z).
„ .20a). away from the central minimum. W i t h proper care in selecting signs of square roots we can use the solutions (18. . A = .l)A(z) 1 = tfA"(z) A .[yA(q. 03 = . (18.. Vq+ii = Vq + 2i.2 . A A _ 3 .21) Example 18.. X » / A q + 4 i = Mg+4i .e. „f 1 \ S o l u t i o n : We rewrite the upper of Eqs.17) and (18..:=*£ +i(5l). z)). .20b) to construct solutions y±{z) which are respectively even and odd under the parity transformation z —> — z (or equivalently q —• —q.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). h2 —> —h2)..1: Calculation of eigenvalues along with solutions of type A Use the solutions of type A.13) in the following form with power expansion of the square root quantities and division by h2: 1 + (q . i.• Using Eqs. (18.e. Anharmonic Oscillator Potentials i.^A(z) °° /1r2z2\i + J2 { fir ) \**A'(*) 1 + jA^Wl. to obtain in leading order the eigenvalues E. h2. (18.2 ( * ) 2 and from this or separately A"(Z) : 1 (q .20b).386 CHAPTER 18.20a) and (18. h2. (18. 1 . 04 = 5 .f .e. zA'(z) where the expansion coefficients are given by ao = l. i.! ) ^ . one observes that with £>. z) ± yA(q. * = — . (18. 2 . a2 = 2 1 .18) we obtain A » = H^rL^z) 2 z .v ' = o (« . c*i = 1 1 . ft = . we write > V±(z) = . 128 8 16 A = i. As always in the In this way Rq is expressed as a linear combination of functions Aq+4i(z). procedure.
^f( g + i)(±H2^ + 1 > \ — 1. those in Aq(z). (18. Hence in the present case In its turn A^ leaves uncompensated terms amounting to A „ /2c 2 \ 2 9 „ M 2ft4 2 { 4/i2^ .(q + 3)(q + l)Aqvri(q '^ ' . [1(9 + 1)]! . w = hz.23) . We return to Eq. r ( CJw) = ^ . i. +1J±iw) (observe that Eq.22) are parabolic cylinder functions D i . (18. (18. _1J±w) and D_\.18... and the coefficient of terms with i = 0.'2H9i) ^faDfr™) and . give an equation from which A is determined.2 The Inverted Double Well Potential Thus a term /j. Hence to the order we 0: {^)^) [{ 1){ 2 ^ ^ ^3)^ 0 +1 ^ + 2^ + 3)]+0 {h) 2ft6' A = 24 1 2 ^ + 1>2^+°G9A= . 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.5).22) is invariant under the combined substitutions q — —q.e. of course. The solutions yq(z) of the equation Vq(w)yq(w) = 0. (18.. when i = 0. c V + It follows that l) + 0 (i The same result is obtained below in connection with the solution of type B.—77 [i(9l)]. Rq '..Aq+4i in Rq 387 can be taken care of by adding to A(°> the contribution — _ ^ 4 ' except.w — > > ±iw) or functions R B w q( ) . 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). = •>— r. In this way we obtain the next order contribution A^1' to A^°>.
0) = = = ±(q±3)(q±7). See Example 18. l(q2 + l).388 CHAPTER 18.26) The first approximation y(w) = y(°>(w) = yq(w) = Bq(w) therefore leaves uncompensated terms amounting to 1 4°) = where [q. .q + 4j}yq+4j.123. since Vqyq = 0.23) have been inserted to make this recurrence relation assume this particularly symmetric and appealing form. (18.0).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 + . 115 . Vq+4j = Vq + 4j. Actually these factors can also be extracted from W K B solutions for large values of q.28) (A + c w )yq(w) E 2 4 1  2 ^ [q. (18. (18.24) The extra factors in Eq. (q±2)(q±S). Tricomi [86].q}=A + c2S4(q.^ For higher even powers of w we write i w2l V<i = Y. G. As an alternative to Bq(w) in Eq.27) Now. pp. 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 . F.±4) S4(9. Oberhettinger and F.l)2/?4.29) ' A .q + 4j] = c2SA{q. and so Vqyq+Aj(w) = Ajyq+Aj(w). Comparison with our notation is easier if this reference is used. S2i(q.Aj)(18.±8) 5 4 (g.25) where in the case i = 2: SA{q. (18.( < ? . and for j ± 0 : [q. we also have Vq+^yq+tj = 0. W.3)y g _ 4 .3.4j)yq+Aj. Magnus. (18. (18. j=i (18. Erdelyi.
(!) («.e.• q. (18.) ty^^w) to be a solution to that order we must also have to that order [q.) + j/ 2 ) (W) + • • • with the corresponding equation from which A can be obtained.[rp ) Y.—v*+*j ( 18 .1. A = ^(q2 + l)c 2 + o(J^.18.e.^ g ( 4 g + 29) + O(^ 2 1 2 Qr2 2 4 2 / 1 \ J. Thus the next order contribution to yq is y H = ie 2^ —z.2 The Inverted Double Well Potential 389 Hence a term (iyq+4j on the right hand side of Eq. (18. 2 . . i. In Case (3) the parameter q is only approximately an odd integer in view of tunneling.10) we obtain E(q.35 ) . . n — 0. This type of invariance is a property of a very large class of eigenvalue problems. (18.31) Proceeding in this way we obtain the solution y = yW („. (18. so that the entire expansion is invariant under the interchanges q . and c 2  replaced by — c 2 .34) is the expansion of the eigenenergies E of Case (1) with q — qo = In + 1. i. p i(9»9 + 4 J'WK7».) + j. .34) We observe that odd powers of q arise in combination with odd powers of 1/h2. ( 18 .3 °) For the sum y{w) = y(°)(«.27) can be removed by adding to T/°) the contribution (—^/4j)y<j+4j. h2 —> h2.q]=0.—A(q + 1) .32) Evaluating this expansion and inserting the result for A into Eq. Equation (18. We can now write the solution y(w) in the form y{w)=yq{w) + Y. (18. and even powers of q in combination with even powers of 1/h2. h ) = qh .
i.—. > "p. and for j ^ 0 all other P0(q.e. We observed earlier that these are obtained by making the replacements q —> —q. 5 + 4j) = 0 for \j\ > 2% or \j\ > 2i + 1. We thus have the following pair of solutions VB(Z) VB(Z) = B w . o.argz=0 (18. ? T 4 ] [ g T 4 .37) For further details concerning these coefficients. Anharmonic Oscillator Potentials ±4 ±4 ±8 ±4 P2{q. we may infer that given one solution y(z).38) [yB(z)]argz=7T = [?/s(2. w — ±iw.11) is invariant under a change of sign of z. q + tt)= J2 i=2 p il(?> Q + ±3 + 4*)[q + 4j + 4t. (18. Again we can write down a recurrence relation for the coefficients Pi{Q. H. q + 4j) = 0.i 1 + 4j) in complete analogy to other applications of the method. J.g±4] ±4 ±4 [g. 2 4tPi{q. g ± 4 ] T4 ±4 and so on. Our third pair of solutions is obtained from the parabolic cylinder functions of complex argument.390 where for instance v J CHAPTER 18. W. q) = = 1.)]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. MiillerKirsten and A. Wiedemann [4] . Achuthan.q±A) = MM!i±MM+[M±8][g±8. ^VO(<L<?) ^i(?." Since our starting equation (18. there is another solution y(—z). their recurrence relations and the solutions of the latter we refer to the literature. q + At] (18. / l \ 1i i( ) + J2 ( ^e ) J2 P ^q + 4j)Bq+lj{w) iu=/iz.36) with the boundary conditions p o(q.
Such contributions arise.2 _4 A } 2cVl1/2 h4 8c 2 1 in the solutions of type A which are not valid around z = 0.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 . 4 .5) are known in some mathematical literature as "stretching variables" and are there discussed in connection with matching principles.39) are with the same coefficients Pi(q.q + 4j) as in ys. Mauss [192]. one has to stretch each by appropriate expansion to the limit of its domain of validity. . we obtain: exp 8c2 3 < 2cVl3/2 /i4 h2z2 „(zA exp 12c2 (18.g.3 M a t c h i n g of solutions We saw that the solutions of types B and C are valid around the central minimum at \z\ = 0. the solutions of type A being valid away from the minimum.2 The Inverted Double Well Potential These solutions are therefore defined by the following substitutions: VC(Z) = [yB(z)]q^q.40) "Variables like those we use here for expansion about the minimum of a potential (e.g. so that the eigenvalue expansion remains unaffected by the interchanges q —> —q.z2 . J.h^ih. (18.18. Thus in the transition region some become proportional. h2) in which odd powers of q are associated with odd powers of h?.h^ih 391 (18. 18.4)). h2 — — h2 as long as corrections » resulting from boundary conditions are ignored. in fact. (18. see e. with normalization since the normalization constants are also asymptotic expansions.2. In this bordering domain the adjoining branches of the overall solution then differ by a proportionality constant. In order to be able to extract the proportionality factor between two solutions. We add parenthetically that all our solutions here are unnormalized as is clear from the fact that the function in the dominant term (e. like w of Eqs.g. Vc(Z) = [V!B(Z)}q^~q. Integrating and expanding as follows since h? is assumed to be large.The solutions yc. Aq(z)) does not appear in any of the higher order terms.
45) .44) we see that in their common domain of validity yA(z) = ~yB{z) a (18.2)i (27T) 1 /2 e f(9l) 1 eb Uq + m (18.41) with the solution ys(z) of Eq.(«. 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. not around that point) VA(Z) VA{Z) = ehVl2c 2 e\z*h* z^) + o(± zfr+i)+o(±\\. .44) In the solution y~g(z). ^Jw) has a similarly complicated expansion for 1 5 — — > argw > .ir_I + M + 1)]! (2w*y U i\[~W • i 5 with 7r > argw.. we would have to substitute correspondingly the expression (18. 2 ) i ' ir! i iHim 1 3 j argw < 7T.0x ! [ £ ( g .43) W + 1)]! w^+V A .l ) ] ! ( . > n. 4/T 4 Prom (18.42) we obtain for the solution ys(z). (18. (18. We do not require this at present. Thus from the literature [86] we obtain Di_{q_1){w) = w*{ql)e\w* but D Wi) ^=. The function Di. y^(z) we see that in the direction of z = 0 (of course.2 « . (18.42) H W2 (ql)e$W I „ . Comparing the solution VA{Z) of Eq.392 CHAPTER 18. eh* /12c* e\z*h? (18.) ( ^2 „22)\ z £(</!) e 4re z 4 4 = hz: [i( 9 l)]!24(«D l+O h? (18. 2 ^i![i(g4il)]!(2«.4 i . with z in ys(z) replaced by — z.w yB(z) ~ B.7 r .z).43) (since z — — 2 implies > argz = ±7r). Anharmonic Oscillator Potentials Considering the pair of solutions ju(.
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 h2.5. i.41).2.y'_(0) ^ 0.47) yA(z) Comparing this behaviour of the solution yc(z) with that of solution of Eq.18. we see that in their common domain of validity yA(z) where a = =Vc(z).49) d yc{z) Again there is no such simple relation between yA(z) 18.50) and y+(0) ^ 0.9.e. — 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. (18. the ratio of yA(z)iVB(z) i s n ° t a constant. Looking at the potential we are considering here — as depicted in Fig.50) will be seen to imply qo = 2n + 1 = 1. hence we have to .. The first of the conditions (18. (18.46) However.21) as even and odd about z = 0. (18.yc{z)pansion (18. ^^+i)Ah^ze* [_i( g + l)]!24(9+D Inserting the ex l +O h2 (18.1 ) e 12c 393 Kgl)]^"1) l + O h? (18. At z = 0 the solutions of type A are invalid. large probability for the particle to be found thereabouts.2 The Inverted Double Well Potential with a 2 (K ^ ( 9 .48) (h2y 4(«+i) [i((? + l)]!24(^i) l +O /i2 an (18. 18. Recalling the solutions y±{z) which we defined with Eq. We proceed similarly with the solutions yc(z).7.42) into y~c(z) w e obtain Vc(z) = (h*. and the second qo = 3.11.. we see that at the origin we have to demand the conditions yV(0) = 0 and y_(0) = 0 (18.4 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..
= = V^ >—= = — i i ^ i [1(93)]! ^_ sin{_(q + i)} (18. with the reflection formula (—z)\{z — 1)! = IT/ simrz.l ) ] ! [ .5 ( 9 + !!)(*"')• i ( 1) .i)H = e."*(«!) ^ ' ~ [ .. Expressions for C 9 (0). Stegun [1]. ( 9 + 3)]! s i n { . e.( g + 3)}.55) Solution: Prom the literature.. IT 4 dio In fact.(0) + ^ ( 0 ) l z—>0 2.^{(9+1)}. 4W >S [ I ( ( ? _ 3 ) ] ! s i n { f ( g + 1)} and hence =^= 1.(0) (18.) [i(«l)]![i(?+l)]! 2TT V5F[i(gl)J! V (18.2: Evaluation of yB(0).SxW) = \ i»fl(0) .(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)}.( g + 3)}.i ( g + 3)]l (18. (18.2. + 1)]!' D ' j .! ( .oi(. y'B(0).F ? .g.(0) = . [C.l)(°) Thus with the help of the reflection formula cited above: D B.58) diu [  ( g .56) D 4(. 1 [7(93)]! n «(°) = .52) # 1 = ^ y' c (0) a and 2/B(0) = a (18.u=o follow with the help of the "circuit relation" of parabolic cylinder functions given in the literature as (q r — ) Di(. Anharmonic Oscillator Potentials use the proportionalities just derived in order to match these to the solutions valid around the origin. C. Z (18. .. one finds that [ 1 IT (q + 1)1! = .54) dio q(w) .51) and 0 = V(0) = Jim i»xM .i ) .' '. M. we obtain i V7T24( g . Abramowitz and I.53) yc(0) a' Clearly we now have to evaluate the solutions involved and their derivatives at the origin./o^Tpif (q+i) .. Then imposing the above boundary conditions we obtain 0 = y'40) = limJ[y'A(z)+y'A(z)] = i»i.s i n { .394 CHAPTER 18.(0) = and htoi)(°) ^—'.1)('») + ^ r + l)]! ^.fc(0) Thus we obtain the equations (18. A.(«. [ .57) [i( 9 _l)] ! 2 i(9i) B. — C.(iu)]. y c ( 0 ) . Example 18. We leave the detailed calculations to Example 18.
53) in dominant order and insert the appropriate expressions for a and a from Eqs.60) (1861) [C>)]o = iV2i [1(9 + l)]![i(9 " 3)]! (18.sin{^(g+ 1)} = J . 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) _^_ . Starting with the derivative expression we obtain (apart from contributions of order 1/h2) lsin{f(g + 3)} i Sin{(g .6 ) } ^ — i t a n < — (q + 3) >.49). (18. 4 r T (18. (18. + i)] V* H(?+l)]'[£(9l)]l Similarly we obtain [°'i(.18.3)}.3)} (h?)faV[\(q [Uq U + l)]\2*(q+V _£ e 6c^. (18.+1iH]«=o = .46).56). ~ ^ 1 5(9+1) 2i(''. isin{f(g + 3 .+i)(°) [ _ I ( ._l) From this we derive C.1/2)!.2 The Inverted Double Well Potential From this relation we obtain 395 ^^^i^w^vy ^(9i)(°) (18 59)  Inserting from (18. we obtain 2 i(»+i)[_i(.(0) = *>i(„+i)(0> 2 .i .62) . + i)]!2i(«+ ) 1 [(«3)]! V5F[i(9l)]! sin{(q + l)}.sin{^( 9 . using the above reflection formula and the duplication formula ^/TT(2Z)\ = 22zz\(z .i ( « + i ) [ _ i ( . W . _ l)]!2l(91)(/J2)(9+D We rewrite the left hand side as 1 sin{f(g + 3)} r„. 7T 4 V 7T 4 (B) Evaluation of the boundary conditions We now evaluate Eqs.)[i(g3)]! and ^i(.
we again obtain (18. For c 2 < 0 and the solution y(z) square integrable in —oo < ?Rz < oo.9.53).2) with potential (18. the energy E is real.11.63) vanishes for q = go = 1. .( g 0 + 3)  + • • • ~ (g .64) the right hand side is an exponentially small quantity. + 3)} = .. ^ . Anharmonic Oscillator Potentials Then the derivative relation of Eqs.. Thus we have to determine these conditions first. .. (gg 0 )^±^ y e5?.1).e. (18. In fact the left hand side of (18.. and the left hand side of (18. Recall > the original Schrodinger equation (18.. i... (18.1. . 18.5.63) and (18. With a Taylor expansion about go the left hand side of (18.64) for q = g0 = 3. 18. 7 .^ >^_>m e J*.L ^ i .64) In each of Eqs. z2 > z2. 7. The analytic continuation of one case to the other is accomplished by replacing ±c 2 by =Fc2 or.65) Expanding similarly the left hand side of Eq. (18.53) becomes 8 m(I(. this is the case of the purely discrete spectrum (the differential operator being selfadjoint for the appropriate boundary conditions.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.. This is Case (1) of Fig. .396 CHAPTER 18.. (18. by the rotations E > einE = E. z^ ein/2z. we obtain cos((g + 3)) = ._ e =*.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. 1 1 . the vanishing of the wave functions at infinity).9.^ .9o) ^ cos  .2.63) becomes (? . .go) ^ (1) It follows that we obtain for the even function with q — go = 1.. equivalently. 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.64) about go — 3. (18.i . (18..5.63) Proceeding similarly with the second of relations (18.
Thus exp exp (<£ff} [+<£ )%} sin * 0 for $lz —s. the case of complex E). or — < 6 < . we see that the solution with the exponential factor is exponentially decreasing for \z\ — 00 provided that cos 39 > 0. i.e.e. + i s i n 3 6 0 i y ) ~3~^ c2\l/2\z\3 oc exp <M — J — . we therefore demand that for $lz — +00 and • —7r/3 < arg z < 0 the wave functions have decreasing phase. i. i. Then c 2 y/vi ^p^My) yf = exp _ r HvcV2!*!3 cos36. Now. y{z)~exv\i[) 7T dV'V.66) z3 = \z\3em. (18. for fiz * ±00 we have y(z) ~ exp < ± i / dz In order to decide which solution or combination of solutions is compatible with the square integrability in the rotated (c2 reversed) case. 2 c2>0. 6 6 In the case of the inverted double well potential under consideration here (i. Thus in our case here the behaviour of the solutions at infinity has to be chosen such that this condition is satisfied. 1/2.e.+00 in arg z € (!•" * 0 for 5ftz — —ex) in arg z € » N> (18.67) Rotating z by vr/2. replacing sin 30 by " 0 + ^) 1 IT = .cos 36. (18.68) .2 The Inverted Double Well Potential 397 One can therefore retain c2 as it is and perform these rotations. It is then necessary to insure that when one rotates to the case of the purely discrete spectrum without tunneling. the resulting wave functions vanish at infinity and thus are square integrable. in the domain — TT/2 < 36 < > 7r/2. —}.18. i. we set z w r = \z\eie. if TT/3 < 6 < 0.e. r / c 2x 1/2^3 = exp{ ± i — —\.sin 3(9 This expression vanishes for \z\ —• +00 if the angle 0 lies in the range — ir < > 36 < 0.e.
68) for c 2 > 0. Anharmonic Oscillator Potentials V(z) Fig. We do this extension with the help of WKB solutions. i.Z\. Equating to zero the coefficient of the term with sign opposite to that in the exponential of Eq. This is not the asymptotic behaviour of a wave function of the simple harmonic oscillator. and there impose the boundary condition (18. This is the boundary condition also used by Bender and Wu [18]. (18.3 The inverted double well potential with turning points ZQ. (B) Evaluation of the boundary conditions The following considerations (usually for real z) require some algebraic steps which could obscure the basic procedure. (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. Our procedure now is to continue the even and odd solutions (18.3 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. we match the WKB solutions .e. We have to remember that we have various branches of the solutions y(z) in different domains of z. We therefore explain our procedure first.68) on y±(z) (by demanding that the coefficient of the solutions with other behaviour be zero). for z > Too. Our even and odd solutions y±(z) (cf. 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. For c 2 < 0 we have correspondingly y(z) ~ exp I ± ( — 1 — y c2 < 0.21) to + infinity and to demand that they satisfy the condition (18. Looking at Fig.398 CHAPTER 18. 18. Eq. 18.Z\.21)) were defined in terms of solutions of type A which have a wide domain of validity.
1 2i4 1 2 4 / ~ 2 4 ~ 2 C * (18. (18. . ^ 2 4 •K + —(r. ^2? 1 2qc2 /2? . A J24 1 ? 2 • "^ •z h \—c z ~ — q h . The distant turning point at z\ as indicated in Fig. equations (21). Messiah.71) We now come to the algebra of evaluating the integrals in the above solutions. 291.=c z 4 dz 1 L2 9 + 2 4 1/2 x exp .. Dingle [70].2.z\) ( \ y^NKQyz) qh2 z 4 dz h H—c z 2 x sin ^ .B. Dingle [70].4) for E and ignoring nondominant terms) t o .^ T rz\ 1 + yh* . p. p. z2h4/A > c2z4/2. B.e.zi)/ \ 1/4 ywKB\ ) z — g/i* .18. *R. 4 2 T ' i. 6.V 2 iL44 + <?z / „2 4 X COS < f / il/2 G?Z g/l 2 1 2L4 z 4 M 1 2i.3 is given by (using Eq. 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\. (18.4.70). 291. h2 (18.2 The Inverted Double Well Potential 399 to the left of z\ to the solutions of type A. (22) or A. t To the right of the turning point at z\ these solutions match on to (r.e. Sec.70) where z < z\.69) The W K B solutions have been discussed in Chapter 14. f R . 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. In using these expressions it has to be remembered t h a t the moduli of the integrals have to be taken.4 h + c z 2 1/4 < . z\ ~ . i.*l) Z 1 2L4 1 qh? + zzh* 1/4 _2i4 VWKB\ ) . Vol.^ 4 1/2 + icV 7T Jz + (18. We begin with the exponential factors occurring in Eqs. I [195]. 18.
Eq.2.4 In 2c 2 J + 1/2 ^1 2c2 1/2 21 and hence (with use of Eq.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.2^3/2 2_(h^\ll2 2czz 4 N 1/2 8?3\ 2 ~hf ' =F9/2 ^ 2 ^ .^ i ^ ^ c 2 . (18. . We obtain this factor by expanding the expression in powers of zjz\ (since in the integral z < z\).^ I 2 2 1/2 zV = F 8c2 Jz [i . Thus the above factor yields '2c2\l/2 h4 so that (cf.72) Here the first part is the exponential factor contained in 2/A(Z)) J / A W respectively (cf.40)) E± = exp _/^_2 f ± In A 2c2 .. Eq. Anharmonic Oscillator Potentials E± = exp = exp .400 i. (18. In the remaining factor we have (looking up Tables of Integrals) 21 2 dz h 4 2 2c.\qh2 + \z2h± .73) .e. CHAPTER 18. (18.211/2 1 \ /j.i c V 8c T 2 1/2* ti>_(2 ~ exp 8c \3 2 r< 2cV 2c2 z2 i24£Y'2u h* 3/2 >.2.±9/2 = ^ V M z^l2 21 2c2 exp ± i dz 1 4 1 1 V2 2 (18. (18.40)).2 \ 1/2 .^ ] V 2 and so /^2f 8c2 3 1 2cVl3/2 h* f q fZ\ ck h^y/ E± = exp T 2 jJ Z ^ [ g . zi 2q 1 c 2 [1 .
we see t h a t in their common domain of validity VA (z) = /3ywKB 0)> where .76a) P P r 2/1 2 " 9 ( .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) .*l) (3y^(z) (18.78) Inserting this into the solutions (18. > » Returning to the even and odd solutions defined by Eqs.\z2h4 + J Z\ nl/2 \M fZ .Zl) ( \ yWKBV^J \z h* 2 1/4 and ywKB(^) \z h* 2 1/4" (18.*l) (18. h2 — — h2 is maintained.2Z 4 C S ' + ( ± ) e x p I i\ cz / cz3 V3\/2 U^2 h6 12c 2 12c2 J (18. CZ2 JZ1 cz 3^2 V2 h6 12c 2 ' Ac2z2 cz" 3y/2_ (18.79) +S_(±)exp . In these expressions we have chosen the signs of square roots of h4 so that the conversion symmetry under replacements q — —q.(J*i) VA 0 ) = PVwKB (Z C.20b).21) we now have V±(z) 1 2 \yA{z) ± yA(z)} = \W^(z) WI ± _(J.1 ) 9 / 2 _(2c 2 ) 1 /2_ 1 (18.77) Now in the domain 2 — oo we have > \qh2 .76b) apart from factors [ l + 0 ( l / / i 2 ) ] .71) and these into (18.20a) and (18.74) Comparing these solutions now with the solutions (18.18.2 The Inverted Double Well Potential 401 Thus at the left end of the domain of validity of the W K B solutions we have (l.21 1/2 .75) 9/2 1/2 0= or '(2c )V2 2 and (3 = — 2 2 (fr2) (2c 2 ) 1 /2_ 9/2 (18. (18.
(18. i / 3 ± .^(q2 + 1) .82b)) /j6 \ 9o/2 («*) = W? mam ''*• with qo — 1. (18.^ j .3. (18...6 The complex eigenvalues (1883) We now return to the expansion of the eigenvalues. we see that we have to demand that 5 + ( ± ) = 0.81) Inserting expressions (18.h2) + (qqQ)h + (18. (18.84) Expanding about q = qo we obtain £(g..76a) for (3 and /3.402 CHAPTER 18. Anharmonic Oscillator Potentials where 5+(±) 5_(±) = = Q / ? ± ^ e x p ( ^ ^ T ^ ) e x p ( . Inserting this into the latter equation we obtain (the factor "i" arising from the minus sign on the left of Eq.h2) + {qqo)(^pj = E(q0. this equation can be rewritten as or as the replacement + / u6 \ 90/2 (/i 2 )W 2 =^(_) 2®"1 f — J .85) + . i. h2) = \qh2 . i. .68).2.^ § ( 4 g 2 + 29) + O ( ^ ) .65)... (18. E(q.e.5. (18.82b) This is our second condition along with Eq.80) Imposing the boundary condition that the even and odd solutions have the asymptotic behaviour given by Eq./ ? = 0. 18.34). Eq./i2) = E{qQ.e. (18. (18.
!*WKB(Z) ~ 2 . Solution: The turning point at ZQ close to the local minimum is given by qh2 2* 4 z2hA c z ~ 0.3. [19]) for ft = 1 and in their notation 90 = 2 ^ + 1. In the above we did not require the matching of WKB solutions at the lower turning point ZQ to the solutions yA(Z). zo} means "to the right of the turning point ZQ" . above) (1.2 .ZQ) .l ) ] ! > + = ^ 2 . ^ 2 4 y qh 2H ""{f' z h H—c z 4 2 1/2 1 ^ 4 + IC2Z4 dz qh2 4 2 .h2) The final result is therefore E () i ( 2 7 T ) V 2 [ l ( g o .4 1 2 4 1/2 ft + i c V 2 1/4 + f}' z.86) will be dealt with in Chapter 20. h2) of the relations KWKB (Z) =PVA(Z).36) of their Ref. in comparison with the work of Bender and Wu [19]) we deal with this in Example 18.e. S W K B ( 2 ) = PVA(z). ft6 ^ = e.~p{q.3: Matching of WKB solutions to others at ZQ Determine the proportionality constants p(q. h2). 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. Example 18.2 The Inverted Double Well Potential 403 Clearly the expression for (q — go) has to be inserted here giving in the dominant approximation / h6 \ <?o/2 . Since this is not without interest (e.yA{z). i. where the superscript {r.g ( .g.h6/6c2 E = E(q0. The large order behaviour of the eigenvalue expansion (of Case (1)) which Bender and Wu derived from their result (18. o 2 + l)^(4.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 (3.18. * _ qh2 1/4 *• 2 j _ 4 i 1 2 4 z h H—c z dz qh2 2 1 2L4 •{/.
. and we have 1/2 dz h?_ 2 J /i2 qh2 + iz2h4 _ ^ 4 J dz zn qh2 + z2^ 1/2 . 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 . N./ ^ 2 X 1 / 2 exp(ifi.zo)/ \ qh2 2 ZO + z2hA 4 2 ." / 4 hj (ft2 2 2)( 9 +l)/4 J'W'KB (Z) 2\1/2(fe222)^1)/4 h) exp(±h2z2) ^2ey/4 * Solutions of this type (which are asymptotic forms for q — 00) have been investigated in t h e > literature.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. Schwind [247]. cf. 2 z 2 ) / 2 e \ ..] { hz combined with the particular constants contained in y ^ ' ^ g .g In other words.O. 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. the (approximate) solutions* 1/2 exp • ± i ( .zo) % ' K B (Z> 1 / 4 2 l / 2 h z V Q / 2 7i/2 4 2 e x p ( ^ e .404 CHAPTER 18. Anharmonic Oscillator Potentials where zo > z. J ^ K B match on to the W K B solutions with exponential behaviour to the right of 20.ic224 2 4 qh2 + i^fe 4 . 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. 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.
in Eq. as in the case of the Mathieu equation.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.89) . (18. (18. ~p requested at the beginning are given by /1y/2[j(gi)]i2^c^ h 6 /12e3 /ly^ih+iwi"" w 18. „ / M 1 / 2 i .3." / 4 ( ? « / 4 . 4" ) (h?z2)^q+1) for q j£ odd integers.3 The Double Well Potential 405 It may be observed that 3/Y/KB ( z ) corresponds to solution (3.1)! = ( 2 7 r ) 1 / 2 e .3 18. In this form the WKB solutions reveal their similarity with the solutions VB i VB > Vc i Vc a n c ' demonstrate that the factors which we inserted (e. J/J Hence ^ and h so that \]q] \2"'2 ~ ( 2 7 r ) 1 / 2 e . (18.g.z ± ) = .5 "l _ o f .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.23)) appear quite naturally.^ 1 . But there are significant differences.kyil(. i.J ' i J . (j . Then .DfJ»v we see Comparing now the expressions (18.13) of Bender and Wu [19]._ . we shall employ basically the same procedure as above or. We can reexpress these relations with the help of Stirling's formula.87) The minima of V(z) on either side of the central maximum at z — 0 are located at h2 h8 z± = ± with y(. h4 > 0.. We consider the following equation ^f& with double well potential V(z) = v{z) = jz2h4 + ^c2z4 for c2 > 0.(r.e. in fact.zo).41) for yA(Z).yA{z) with expressions 27WKB i J'WKB ' that the factors p.88) + [EV(z))y{z)=0 (18.
the previous equation (18.93) w± = h±(z .z±). Anharmonic Oscillator Potentials and V{2\z±) V(4)(z±) = = h\ 12c2. 18.91) (18. (18.i>5.94) (18.90) V(i)(^)=0.91) becomes T>q±(w±)y(w±) = o(^)y.95a) . (18. .z±) y = 0.V(z±) l{z z±fhA + 0[(z .4 The double well potential.92) (thus we sometimes use h4 and sometimes Ii±) and EV{z±)=lq±hl With the further substitution + ^. 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 .406 CHAPTER 18. vW(z±) = ±Qch2.z Fig. (18. (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. 18.87). is again seen to be invariant under a change of sign of z.e. i. we conclude that in the dominant approximation q± is an odd integer.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. Thus again.2.95b) with the equation of parabolic cylinder functions u(z) = D„(z). d2u{z dz2 + v+ 2~r u(z) = 0. (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. given one solution. (18..i 4 f W » = o. we obtain d*y (18...3 The Double Well Potential where ^ 407 d2 •« 2 2  d8.. we obtain another by replacing z by —z.100) . (18. (18.96).95a) and (18.3. (18.97b) Our basic equation. (18.93) for E into Eq. where U{z) = and near a minimum at z± 4_ 4 jp[V(z)V(z±)].96) dz2 + 5 « 4 +  . Eq. Inserting the expression (18.95b) By comparison of Eqs.97a) U(z) = (zz±)2 + 0[(zz±)3}.1. n = 0.
101b).105) "8 + 4 ^ . (18. z) may be obtained from the solution yA{q. h2 h4 2c' with g+ = q. z) = yA{~q. yA(q.z) by either changing the sign of z throughout or — alternatively — the signs of both q and h2 (and/or c).1 6 ) . Aq(z) = Aq(—z) = Aq(z). The dominant approximation to A is then the function Aq given by the solution of the first order differential equation VqAq(z)=0.h2) .101a). ~h2. (18.h2.104) Both solutions yA{z). i. n± = V2h2. i. z).103) Looking at Eqs.e.e. h2. we observe that the solution y^iq. these equations can be rewritten as + (qz+z)A{z) = . Integration of Eq. (18. (18.c W + 1) .^ A"{z) + 2c V2 *A(z) (18. h2.102) yields the following expression M*) I z2 _ z2 11/2 *+ z + z+ q/2 z+ 1(91) \z + z+ 1(9+1)' (18. Anharmonic Oscillator Potentials and selecting z+ (z2+z2)A'(z) h z+ = —.^q(l7q2 6 4/i + 19) + (18. The result is given by A and E(q. h2. We leave the calculation to Example 18.408 Since CHAPTER 18.c W + I ) 2C yfi 2h 5 2 A V2c* g(17g2 + 19) + 0 ( / t . 8hw (18.4.101b) To a first approximation for large h2 we can neglect the right hand sides. (18. z) = yA{q.102) We observe that a change of sign of z in this equation is equivalent to a change of sign of q.101a) {z2+z2)A\z){qz+ + z)A{z) = ^ A"{z) + £A{z) (18. but the solution is a different one.106) .yA(z) are associated with one and the same expansion for A and hence E. Vq = {z2+z2)— + (qz+z).
(18.103) for Aq{z) is very similar to that of the corresponding solution in considerations of other potentials. <4q+2 — z — z+ Aq+2iAq+2j (18. however. . (18.101a) with A replaced by A . We know the first derivative of Aq from Eq. z • . = Aq+2i+2jAq.108) We wish to rewrite this expression as a sum y coefficientiv4.113) 2 . ± 1 .18.102). First. ± 2 .(l + + 2  z y* .Aq 1 + 2^± 2 Aq+2 (^M^f (18.106). Solution: The structure of the solution (18. (18. Similarly we obtain _z+_ Z _ Aq + 2 — Aq Aq+2 + Aq 12 ^g+2 A ± ± _2A 2 ± 6 + (18.3 The Double Well Potential 409 Example 18. J = 0. .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.. A Differentiation yields » = ^^w (18.z2 ) N t.qz+)2 .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 . . The first such step would be to reexpress the right hand side of Eq.110) Aq+2+Aq2 9+2 A. as a linear combination of terms Aq+2i.z\) + 2z(z . . (18. Thus it is natural to explore analogous steps.111) {Aq+2  Aq2)2 (4zz+)2 {z2  2 .107) KV) ~ Aq(z) {z*z\y Aq{z) (z 2 z\f (z .{z2 .qz+) [2z2 .109) We observe some properties of the function Aq{z) given by Eq. we reexpress A'q in terms of functions of z multiplied by Aq.103): .*l) (* ~ "+> ) A (z)exp q /2 \ 3 4c J (18.112) From these we obtain. 92 ZZJf (18.2 z\) • Aq$4 £Aq + Aq—4. + Aq+2 + Aq Aq+2 . (18. i.114a) . (18. . for instance by componendo z z et dividendo.4zz+q+ z\{q2 + 1)].e.+ 2 ^ ± l + 2^±i + . . such as periodic potentials.105) and (18.
34).4)A„_ 4 + (q. (17.116) into Eq. (18. _ 2 + 6 ( 9 2 + l)A ( ] + (q + l)(q + 3)Aq+4.115) and (18.(0) V2 ' 2c K + h4+Ao . q) = 0.aAg+S 12+ 16 (18.115) (z*z%) 2 i2 (z 2 .q + 6)A<.+6 + • • • ] . (q. (18. QA?+4 _ 10 A + 6 .l ) ( g .z%) 1 n z+z 2 \4z+J [Aq+4 [Aq+4 .116) [Ag4 .2\2 (z* . •••]. In particular the dominant approximation of A is obtained by setting (q. <?) = 6(<j2 + 1) + z: 2 2A +72"' (18.4 ( g .112). Vq+2iAq+2i=0 and Vq+2i = Vq + 2iz+. i.2A g + A g _ 4 ] .118) 2s/2c ~^j[(?.e. (18. Since VqAq = 0.q + 2) = 4{q + l ) 2 .119) It is now clear how the calculation of higher order contributions proceeds in our standard way.9=F4) = (g=Fl)(?=F3). Anharmonic Oscillator Potentials A 1 . (18.4 4+* + . 1_4^±^+8^±±12^±^ +. (q. A(°~) = Aq.114b) . cf.l ) 2 A . (18. Eqs.4]j.q2)Aq2 + (q. .101a) the contribution R .2 ^9+2 „^9+4 + \4z \4z+J z+ Finally we have also [A q _4 — 2Aq2 + 2Aq+2 — Aq+4 (18. .33) and (17.q + 2)Aq+2 (18.4A ? _2 + 6Aq .z\Y Z2 + {~)\Aq+42Aq 1 224 Aq.3 ) A . q + 4)Aq+4 + (q.112) CHAPTER 18.410 and from this = Hence with Eq.4Aq+2 + Aq+A Inserting (18.2Aq + Aq4] 1. leaves uncompensated on the right hand side of Eq.117) 4(q + l)2Aq+2 Here the first approximation of A.108). ' T h e reader may observe the similarity with the corresponding coefficients in the simpler case of the cosine potential.<? +{q.q)Aq + (q. we obtain 5 K = ( ^ f ) [ ( 9 . (18. where the lowest coefficients have been determined above as (9. _ 4 .
The first approximation A(°> = Aq leaves uncompensated on the right hand side of Eq. h2. thus yielding the next approximation of A as given in Eq. Terms jj.94) 8z+ (9.e. *)] (18.z)±yA(q. 9 + 4) ^+48z+ (18.28.g) Az.Aqjr2i in this may therefore be eliminated by adding to A^0' the contribution A^1' given by (9.g) 8z+ (g.g) 4z+ {q q)+ 2r (9. which reduces to 2 \/2c 2 0 = 2(3g 2 + i) + — A + ^ g . (18. y'_(q. z)\ = 2^A^ h2 '>z)± VA(Q.123) ^[yA(q. 9 + 2) Aq+2Az+ (9. i.0) = ^ . g .e.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.92) ig4 ' H ~ 23/X4 iz+ An2 H + (9.105).101a) the contribution Rq .124) .121) This coefficient of Aq set equal to zero yields to that order the following equation 0 £) ' {^) (g.(0). (i) _ / V2c\ [ ( g .0)=0.z) Considering only the leading approximations considered explicitly above we have (since Aq{0) = l/z+ = A.h2.h2.yA(z) derived above are valid around z — 0. where Rq is the sum of terms left uncompensated by AW.122) The solutions yA(z). in the domains away from the minima.g2)(g2.4 ) ( 0 ) ( g . (18. A' (0) = q/z\) y+(q. g) 8z 4 .0)=0. \ZZ±\ > 0 ( y We can define solutions which are even or odd about z = 0 as y±(z) = = £ fe^te' h2> z) ± y A. g .h2. of course. y'+(q.94)(g4. i. h2 2y^ Age? h4 (18.h2.h2.3 The Double Well Potential we have T>o 411 2iz+ J Aq+2i except.0) = y^(q.g ( 1 7 g 2 + 19). for i = 0. (g. g + 4)(g + 4.{^ h2. (18. h2.g + 2)(g + 2.2 ) (0) (18.
125b) It is clear that correspondingly we have solutions yc(z).^ % [i(?l)]!24(«. Moreover.91).99) for z — z±. in view of the factor "i" in the argument of Cq the solutions VAJVC have the same exponential behaviour near a minimum. Bq[w±{ ^ .95b) — 1 T>q(w±)y(w±) = j£ Thus we write the first solution yB(z) = Bq[w±(z)] + 0(h±2).23) with appropriate change of parameters to those of the present case. the equation is — with differential operator T>q as defined by Eq. (18. (18. and another VB(Z) ± 25^ch3wl + c2wi . This means.125a) = VB{~Z) = Bq[w±(z)] + 0(h±2) = Bq[w±(z)] + 0(h±2) (18.2A y(w±). Anharmonic Oscillator Potentials Our second pair of solutions. (18. We see already from Eqs. but w±(z±) = 0. we have w±(—z±) = —2h±z±.Evaluating the exponential factor contained in VA{Z) of Eq.412 CHAPTER 18. in this case we use the Schrodinger equation with the potential V(z) expanded about z± as in Eq. 18. 2(9+) .yc(z) with complex variables and Cq(w) given by Eq. + 0(hl2).126a) hjW + l)]! VM = Vc(z) = Cq[w±(z)] (18.yB(z) is obtained around a minimum of the potential. (18.1 ) D (18.93) and setting w± = h±(z — z±).95a) and (18.3 M a t c h i n g of solutions Next we consider the proportionality of solutions yA{z) and VB(Z). (18.126b) These are solutions again around a minimum and with yB(z). we > .yc(z) providing a pair of decreasing asymptotic solutions there (or correspondingly y~B(z)iyc(z)).We draw attention to two additional points. Since w±(—z) = —h±(z + z±).95b) that the solution there is of parabolic cylinder type. yB(z). Thus yc(z) Cq[w±(z)} = = Cq[w±(z)]+0(h±2). Inserting (18. (18.3.
125a) we see that (considering only dominant contributions) in their common domain of validity e ^+z+ l + O 1 yA{z) = yB(z). (18. we have \z ~ Z+\^Q ^ l f e 2 2 2 _kh2( ~ i . a= ^ . 24(9D[i(gi)]! hj (18..18.ZI and find that these are given by z+ + 0(l/h) for finite q (cf.z±)dz . 2 An+z+e 4w+ (18.97b) ~ exp .172)) we add here comments on this approximation.• • " . (18. . We observe that for z — z± the approximation yields exp[±/i? t z 2 t /4].• )2 (^)itoi) ( 2 z + ) 5 ( 9 + i )e ' yA(z) z+> lft2 2 _ I . the dominant term in the power expansion of the parabolic cylinder function is given by Dv(w±) ~ wv±ew±>4. .128) Similarly we obtain in approaching z+: VA(Z)  ( 2 ^) (9 " 1} e. (18.147)). (18.129) .frW. (18. and comparing with Eq. .^/w + ± 1 .3 The Double Well Potential (cf.164)) only in a nonleading contribution and hence implies the equivalence of the exponentials in the relation (18.146). ein+z+e in+(z 22+(<?+i) .« + ) 2 (9+1) e (18. Thus for h very large these turning points are very close to the minimum at z+.^(±>(^*4 ^\h\{zz±f exp 7>2 r 2 ^n±z± For later reference (after Eq. Later we calculate the coordinates of turning points ZQ.97b)) exp 413 '\hlJQ'uV\z)dz = exp = exp (18.i*l(«^ 4'"+^+ 4"'+v ( « .172) obtained later. . 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.127) Recalling that around arg w± ~ 0. Here z± = ±h?/2c are the positions of the minima of the potential. (18. Eqs. Allowing z to approach z+ in the solution IJA(Z). .
even above the turning points. basically. Since it is more probable to find a particle in the region of a minimum than elsewhere. K. S. and this means at both minima. J. g . H. but it is clear that none of the above solutions can be used at the top of the central barrier. S. they assume large quantum numbers. Furry [102] . 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. de Deus [66] and W. K. P.(18. Various investigations^ therefore struggle to overcome this to a good approximation. 18.130) Therefore in their common domain of validity a a = dte+vM' {2z+)^. Thus 1 E . We have to impose boundary conditions at the minima and at the origin. The involvement of these WKB solutions leads to problems.131) We have thus found three pairs of solutions: The two solutions of type A are valid in regions away from the minima. 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.z+)]fr+V ' (18. The next aspect to be considered is that of boundary conditions. Bhattacharya and A. Anharmonic Oscillator Potentials D_h{q+l)(iw+(z))2i(o+V Vc( ) z 2i(g+1)ei^(z_.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.7r and the solutions of type C around argz = ±7r/2. The pair of solutions of type B is defined around argz = 0.3. Thus it is unavoidable to appeal to other methods such as the WKB method to apply the necessary boundary conditions at that point. Rau [22].\h\)^+l\\{q l + l)]\ 1+ 0 J_ . The solutions in terms of parabolic cylinder functions have a wide range of validity. Bhattacharya [23].414 and CHAPTER 18. since.+)2 [1(1 + I)]' [i(q+ l)}\[ih+(z . D. R. we naturally expect the wave function there to be similar to that of the harmonic oscillator.
3 The Double Well Potential 415 the most basic solution would be even with maxima at z±.Zii yB{z±) a ± a yciz±) (18.132).135) Hence the conditions (18.132) Fig. However. as indicated there.5. At dtz — ±oo > we require the wave functions to vanish so that they are square integrable. We have therefore the following two sets of boundary conditions at the local minima of the doublewell potential: y'+(z±) = 0. and y+(z±) = 0. as indicated in Fig. and y'±(z±) V'B(Z±) ± =y'c(z±) (18. y(z±)=0. (18. (18.133) y+(z±)^0.134) y'+{z±) ± 0.136) . 18. as indicated in Fig. The odd wave function then exhibits a correspondingly opposite behaviour. We have y±(z±) = ^VA{z)±yA{z)]z^. 18. ^(*±)#0 (18. y'_(z±) = 0.18.5. an even wave function can also pass through zero at these points.133) imply yB(z±) yc(z±) « a' and y'B(z±) y'c(z±) a a (18.5 Behaviour of fundamental wave functions. y{z±) £ 0. 18.
.416 CHAPTER 18.9.134) 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 . Eqs. 1 1 . 7 . Thus sin {^ +1) } = ^^>^30fe^i.138) Using formulae derived in Example 18. .5. .m Using Eq.2 we can rewrite the second of (18. Anharmonic Oscillator Potentials (B) Evaluation of the boundary conditions Inserting into the first of Eqs. _ 8)} s . . (18. (18. (18. also (18.136) the dominant approximations we obtain (cf.«*{=(. (1. +1)} = 4 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..136) as  M .„*{!<. (18..142) (<Z9o)J(l)^°+1)(l)1/2 .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..
18. V(z) — • z E=V Fig. y+(0)^0. Hence we have to impose at z — 0 the conditions j/_(0) = 0.1 " <813 1 4)  In Example 18.5 B o u n d a r y c o n d i t i o n s at t h e origin (A) Formulation of the boundary conditions Since our even and odd solutions are defined to be even or odd with respect to the origin.3. y'+(0) = 0. we matched them to the solutions of types B and C which are valid there and hence permit the imposition of boundary conditions at the minima. We do this with the help of WKB solutions.6 Turning points z$. we must also demand this behaviour here along with a nonvanishing Wronskian. 18.5 (after determination of the turning points) we rederive this relation using the WKB solutions from above the turning points matched (linearly) to their counterparts below the turning points and then evaluated at the minimum. We deduce . y'_(0)^0.144) Thus we require the extension of our solutions to the region around the local maximum at z = 0. Since the type A solutions are not valid at the minima.18. (18. We emphasize: We needed the solutions of type A for the definition of even and odd solutions.3 The Double Well Potential Thus altogether we obtain 417 ^ • .z\.» ^ B ^ ' .
We also note that the height of the potential at the turning points is h8 qh V(z) 20.24 C Z O 1/2 ZQ = Ac2 + + °(r A_ 2c 1 2 2q\ 1/2 2 / cq 5 + h± + 0(h~ ) h%) (18. h8 25c2 (18.96) that the two turning points at ZQ and z\ to the right of the origin are given by \qh\ i. one finds that .148) ("f * "The superscript (I.\qh\ + \h%U{z) (18. (18. the dominant terms of the WKB solutions are* 1 yWKB\ ) Z 1 1 ~l/A 1/2 ~ qh\ x exp + h%U{z) . (B) Evaluation of the boundary conditions We now proceed to evaluate the boundary conditions (18. + £\h%U(z)=Q. In the domain 0 < z < ZQ.418 CHAPTER 18.146) and Iqh* 1/2 z\ Ac" + +o ^ (2g2)V4 2c + /i ' (18.144).e.e. 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. ZQ) means "to the left of 20" . As a consequence. Anharmonic Oscillator Potentials from Eq.^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. to the left of ZQ where V > E. i.145) \qh\ + \zW Using z+ = h2 /2c. 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.
(18. we have to match yA(z).i (2q ln (18. (18.^ j O e .150) Evaluating this we have ±h ~ h2 GMln 1 In 2g q In +4 1 (z .18.(29/^)1/2 1 ^(zz+){(zz+) 1/2 2 2q_\ 1/2 hJ 2 l l q+ q  ^ ( ^ . Thus we consider in the domain \z — z+\ > {2q/h2+)l/2\ h 1.3 The Double Well Potential and 1/4 419 ywKB\ ) z — \qh\ x exp I / + \h\U{z) 1/2 I. )+^*+)2lr} rnD h\z'2)}z Q )\ = .148) and (18. 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) .y^(z) to 2/VVKB(Z) anc ^ ^ W K B ^ ) .^ + ) 2 + igln2(zz+).^ e therefore have to consider the exponential factors occurring in (18. 20 dz \qh\ + \h%U{z) .151) In identifying the WKB exponentials we recall that y^KB is exponentially increasing and ywKB is exponentially decreasing.* 4+ . .149) and consider both types of solutions in a domain approaching but not reaching the minimum of the potential at z+.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 >'^.149) In order to be able to extend the even and odd solutions to z = 0.
We see therefore. This is the aspect investigated by Furry [102]. we can write the exponential as exp (27T) ("f 1/2 I dz Z+ 1 qh\ + q/A 1.125a) and (18. 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.153) This expression is valid to the left of the turning point at ZQ above the minimum at z+. However. In a corresponding manner — i. since there is no way to obtain an exact leading order approximation with Stirling's formula for small values of q. l/2 ~h\U{z \(zz+?h\ (18. The only way to relate these solutions is with the help of the Stirling formula which converts t h e product or ratio of such factors into factorials such as those inserted from the beginning into the unperturbed solutions (18. Thus using the Stirling formula z\ ~ e~(z+1\z + l)z+2y/2~rr. the results necessarily require adjustment or normalization there in the q—dependence. of course.420 CHAPTER 18. Since correspondingly 1/4 2 l/2 \*% we obtain + 4/1W*) (*^)i/'(M)i/*' 27T 1 / 2 {h\)^{\{q1)]\\2"+ *+ 1(91) ~hi l2 q/A e \{zz+)*hl for \z — z+\ > m 2q \ (1 ' (18.125a).ZQ) .126a).154) . t h a t the Stirling approximation is amazingly good even for small values of the argument).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. \ yWKB\ ) z 1 fl(g3)]! V*(h%)V*\zz+\h(*+V\2 q/A h + e* \(zz+)W+t (18. 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. using Stirling's formula (and not the inversion relation) — we have —(1.e.
104).156a) and 7 _ i\(Qm •K •/2(M )l/4 ^ 2 n« q/4 (2*+) •(9D e ^^4.158) Applying the boundary conditions (18.e.154) with the typeA solutions (18. multiplied by (1 + 0(\/h\))) the proportionality constants 7 .K?+1).e. this result is somewhat imprecise.143) which was obtained with our perturbation solutions from the boundary conditions at the minimum.3 The Double Well Potential 421 where [\q}\ was written as [\(q — 3)]! for q large. Comparing for z —• z+ the W K B solutions (18.157) is used in Example 18. (18. it is our philosophy here that the factorials with factors occurring in the perturbation solutions are the more natural and hence correct expressions.e. However.5 for the calculation of the tunneling deviation q — go by using the usual (i. and is shown to reproduce correctly the result (18. Returning to the even and odd solutions (18.159) 7 i(«i) r(9^3) ^2~ j(9l) . (18. we obtain in leading order (i.103) and (18. linearly matched) W K B solutions. 7 of t h e matching relations y^NKB\z) i. The relation (18.144) we obtain fi(0) Tin the present case as 1 (91) 1 — 5 (18.18. . (18.n('>2o) 3/WKB(Z) =1VA{Z)I (18. \hl. (18. 27T1/2 9/4 1VA{Z).123) we have V±(z) 1 [yA(z)±yA(z)} = ^y^l(z) ± ^j#&(*) (18.153). [\{q — 3)]! are really correct replacements of [\q}\ only for q large.155) (2z. .157) Since the factorials [^(q — 1)]!.99).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 (18.143) T^^*)) 27T (18. as the results also seem to support.
(18. p.163b) The integral hi?) .. n + uY 2c „ ^6 8V2c2 (18.161) we can rewrite the integral as r I2{z) = = dz^{a2 .. Friedman [40].19. 60 and 361.3 evaluated at z = 0 is then given by = y/T+u[E(k) uK{k)} (18.05.148). 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. .z2)(b2  z2).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.iV2 ! i .149) near z = 0. G2 K [i 4.164) / 2 (0) z ca [(1 + k2)E(k) . K(k)} 3 ^ 2 y/T+u[E{k) uK(k)\. Byrd and M. D. 3G2 12V2c ^See P. .422 CHAPTER 18. (18.162) T h e integral appearing here is an elliptic integral which can be looked up in Tables. 213.163a) (18.160) Setting 6 =4?^T"z°' rb Ac2 + VQ—> (18.. F. p.k'/2. formulae 220. Anharmonic Oscillator Potentials Thus we have to consider the behaviour of the integrals occurring in the solutions (18.
1 " e " O T (18 169)  Y^V / 4 tfo. (G\ . Here we are interested in the behaviour of the integral in the domain of large /i 6 /c 2 .3 The Double Well Potential 423 where K(k) and E{k) are the complete elliptic integrals of the first and second kinds respectively. The nontrivial expansions are derived in Example 5.1. 1. n = 0.2 A( L ) 4 s / and z dz yWKB\ J and d^WKBW 2=0 (*?) a ( . .18.. ± .r1)/10//2 ( f g 1^ yS» /L ( ^ ) l^4 .1)]' Gfo. /2n + l 2 ± — . ^ ^ . 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 .165) in agreement with Ref.e T ^ 5 U W ( 2 „ ) i / 2 ( ^ i W4C • (18 170) (18 1?0)  ^The expansion of I2 used in Ref..§ Since the integral is to be positive (as required in the WKB solutions) we have to take the upper signs.± ' 3G 2 ' T ( ^ (18. . n. which implies small G2. [167].2. [4] misses an n—dependent power of 2 in the result.( ft 7..) = 2 \. .3 )l ! .( 2V + l)ln V 4 / 4( 2 n + l ) l nV 2 n . ..166 ) ^^iffil/^(2*)1/affiii£ Correspondingly we find ^ (18168) S W I 2 = o . where the result is shown to be (with q ~ q0 = 2 n + l .
157).143) with qo — 2n + l. . (i*™) We obtain the energy E(q.3 concerning the exponentials. (18.3.n = 0. Expanding this around an odd integer qo we have ( 8E\ h? —J Inserting here the result (18. with the help of Eq.h2) + (qq0)^=.3.173). (18.. . h2) the split expressions 29o+1/i2(—W2 h6 (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.1. Anharmonic Oscillator Potentials With these expressions we obtain now from Eqs.106). .»W.172) A corrresponding relation holds for h+ and z+ replaced by /i_ and z.6 Eigenvalues and level splitting We now insert the replacement (18. (18.424 CHAPTER 18. (18. 2 . 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.h2) and hence the splitting of asymptotically degenerate energy levels. Inserting the expressions for h+ and z+ in terms of h. we can therefore impose the boundary conditions at z = 0 by making in Eq. . 18..i ) ] ! e ~ s A ' . 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).143) the replacement: {hlY'2{2z+fe^zl + 2 9 ( ^ y Z e"A.. and obtain ( * . (18. 18. (18.175) ~E0(q0.174) .^ W ^ / 4 t o . (18.172) into Eq.
1 7 425 2 .^ e 21/^6c2 fe6 ( 1 mass mn = (1 8 . which evidently supplies some correction terms. .7 =( e N n+1 . The result (18. (18.157).176) is described by Bhattacharya [23] as a "modified well and harried result AjfWB.1 7 6 ) (2n+9)/4^/ft V7T«! V C / \ eiI72^.143) with (18.178) "In S. c 2 /2 <> A — given as E0 = 4\ h(2fc) 1 / 2 o y J H q2 4/c > ^— + q ^ 2s c2 V2 .. the following expression for large q ~ 2n + 1. in the WKB restricted sense) defined by ^( n +2^KB(0) 2 Here the derivative dE/dn corresponds to the usual oscillator frequency.fr2) is given by Eq. 1 u2 c 2 (3 9 2 + l) ^ « b . i.h2) / h 6 \ —K />6\n+l/2 q o / 2 2«>+2h2 — —^= m 1. T \ *n ! 1 ' (18. Bhattacharya [23] the "usual WKB approximation" of this expression is — with replacements h4/4 <» k.q2 7. the level splitting. i. i n U n / ! Thus AE(q0..h2) = E_(q0.106).n = 0.18.. K. 2h4' i.^ 2^ h* .V2^\ \n+ i .174) and (18.e. the difference between the eigenenergies of even and odd states with (here) qo = 2n+l. 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 / » : = \ . for finite h2.e.3 The Double Well Potential where £0(<7o.e.h2)E+(qQ. (18. Combining Eqs.e..1.e.2 = 2 . can be given by In the work of Bhattacharya [23] the WKB level splitting is effectively (i.2. the pure WKB result of Banerjee and Bhatnagar [14] (i.
h2)\mo=i = . (18. we identity the parameter k there with ft. The Furry factor represents effectively a correction factor to the normalization constants of WKB wave functions (which are normally for mo = 1/2. 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.181) "W. however. H." Of course. We see therefore. (24) there is k 2 (i0n+i3)/4 //c3/2\n+1/2 V2fc3/2 6XP n\ ft \/nn\ in agreement with AE(q0. Bhattacharya [23]. and A there with c 2 / 2 here. A>0. This.h2 <> — h2. Anharmonic Oscillator Potentials which is unity for n — oo with Stirling's formula.h )/2 2 2 ( — ' 3 6\/2c A h6 n+l/2 h6\ 2{2n+5)/if \ exp 2 \ c / of Eq.178).6 \9o/2 e~h&l^\ with (m 0 = l) (18. (66). the Furry factor corrected WKB result follows automatically. that if this symmetry is taken into account from the very beginning. . Eq.176) together with Eq. these deriva> tions do not exploit the symmetry of the original equation under the interchanges q <> —q.^ 2 + ^h2 If in addition the potential is written in the form A / . h = 1.3)). K.** Had we taken the mass mo of the particle in the symmetric double well potential equal to 1 (instead of 1/2).. Then E^h2) and AE becomes 2^+U 2 (^)W 2 h6 = Eo{q0. h4 and c2 replaced by 2E.180) h8 E0(qo. (18. Furry [102].2X2 V(z) = \z2tL) .h2) * ^ 2 q o / ^ q l _ m ^  (18179) l~o o». (18. 2/z4 and 2c2 respectively (see Eq. ** Comparing the present work with that of S.6) to yield an improvement of WKB results for small values of n. 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. Then the splitting A ^ W B of Eq. as we do here. (18.4/4 here. we would have obtained the result with E. The factor set equal to 1 represents a somewhat "mutilated" Stirling approximation.426 CHAPTER 18. as is also explained by Bhattacharya [23].2 / 7. supplies the bridge to the perturbation theory results and removes the unnatural appearance of factors e and nn.
h>) * 2V2^^^y/2e^3/3A.* A similar correspondence is also found in the case of the cosine potential. 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.15). Miiller—Kirsten [167] the potential is written as V2 VW = \ for mo = 1. . MiillerKirsten [186].§ 18.180) agrees similarly also with the result for arbitrary levels obtained with the use of periodic instantons* and with the results of multiinstanton methods.88) h4 = 2/x2. The comparison with Eq. D. Lee [98].182) implies AlE(l. Bhatnagar [14]. P. 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. however.c2 = A/2. carried out along the lines of the corresponding calculations for the cosine potential and thus of the wellestablished Mathieu equation. Mansour and H.W. Patrasciociu [110]. We considered above only the symmetric twominima potential. §J. (18. J.—Q. M. m 2 <• p 2 / 2 . W. ^To help the comparison note that in J. See R. D. ZinnJustin and U. (2.34) and (E.11).18. J. Eqs. J. Jentschura [293].3 The Double Well Potential 427 a form frequently used in field theoretic applications. D.^ *See E. and g2 is given as g2 = r]2/m3. 'This will become evident when the perturbation theory result of Chapter 17 is compared with the path integral result in Chapter 26. (18. Friedberg and T. first paper. Banerjee and S.^ Thus for the case of the ground state Eq. K. H.7 General Remarks In the above we have attempted a fairly complete treatment of the largeh2 case of the quartic anharmonic oscillator. formula (4. these are presumably not of much interest in physics (for reasons explained in Chapter 20). In principle one could also consider the case of small values of h2 and obtain convergent instead of asymptotic expansions. Gildener and A. Jentschura [293].3. M. Equation (18. (18. ZinnJustin and U.181) therefore implies the correspondence rj1 <> A/2. Liang and H. so that by comparison with Eq. provided their result is multiplied by a factor of 2 resulting from a corresponding inclusion of antipseudoparticle (antiinstanton) contributions.
we cite papers of the Uppsala group of Froman and Froman [100]. though not exclusively numerical. Perturbation theoretical aspects are "See e. in both symmetric and asymmetric form. In fact. though incomplete list of references in this direction has been given by Garg [105] beginning with the wellknown though nonexplicit (and hence not really useful) ground state formula in the book of Landau and Lifshitz [156]. and numerical studies are presented to support this claim J It will become clear in Chapter 20 — as is also demonstrated by the work of Bender and Wu [18]. Very illuminating discussions of double wells and periodic potentials. 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. There is no reason to view the anharmonic oscillator differently. The wide publicity given to the work of Bender and Wu made pure mathematicians aware of the subject. There is no way to obtain the exponentially small (real or imaginary) nonperturbative contributions derived above with some convergent expansion. in his recent paper on simple uniform approximation of the logarithmic derivative of the ground state wave function of anharmonic oscillator potentials. Martin. has also been the subject of numerous numerical studies. 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. The double well potential. Pradhan [182]. [19] — that the expansions considered above are asymptotic. mostly in connection with instantons. can be found in an article by Coleman [54]. Santi and N.428 CHAPTER 18. Sophisticated mathematics — like that of the extensive investigations of Turbiner [273] over several decades — seems to approach the problem from a different angle. for instance. N. As references in this direction. like those for anharmonic oscillators. as he discusses. as some relevant references with their view we cite papers of Harrell and Simon [129]. .g. The ground state splitting of the symmetric double well potential has been considered in a countless number of investigations. Anharmonic Oscillator Potentials Every now and then literature appears which purports to overcome the allegedly illnatured "divergent perturbation series" of the anharmonic oscillator problem. B. P. Simon and Wightman [178]. [132] and Loeffel. Ashbaugh and Harrell [12] and Harrell [130]. Tables of properties of Special Functions are filled with such expansions derived from differential equations for all the wellknown and less wellknown Special Functions. Mahapatra. 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.[131]. A reasonable. B.
f + • • •) 2u\ . Saxena. which is assumed to be small. Thus k' and hence k' = v ^ ( l .^ + • (18. fc ~ 1 . Biswas. Cooper and Strottman [221]. + (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.\G\ + • and 1k2 =2Gv/2q4. Gutschick.G2q + . We have 4 ~k' 2u(l . X +2 + ' ^ (18. Example 18.184) Hence we obtain for G close to zero: 1Gy^g l + Gy/2q~ and 1 .18.186) (18.183) S o l u t i o n : We have k2=1^./4 E{k) = 1 + • l n In (18.190) Hence ln]=ln : l n ' 7 2 ^ + 2 (18. [122] as 13 1. T h e r e f o r e it is convenient t o deal w i t h this here.191) Our next objective is the evaluation of the elliptic integrals E(k) and K(k) by expanding these in ascending powers of k' .192) 'F 12 16 . 2 a. We obtain the expansions from Ref.189) =2u 2u2 (18.2G^2q + 4G2q .3 The Double Well Potential 429 employed in papers of Banerjee and Bhatnagar [14]. 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.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 . Srivastava and Varma [24] and Bhatacharya and Rau [23]. (18.u)1'2 = V2u{ 1 .. k! (18. 1+ u : Gy/2q. q ~ 2n + 1.187) We now reexpress various quantities in terms of u.5: Evaluation of WKB exponential with elliptic integrals Show that (l + u)^2[E(k)uK(k)}'. (18. Datta.185) (1 + u)^2 = (1 + GVW/2 = 1 + GJ±. Bender.
3 u .u)} + — u ( u .u){24 . 8 .1)(1 .13)} + 12u 3 (u .u) 2 {24 .u){24 .4. ) J 4 (18.u)(u .{ M 2 ( 1 .u ) (^){« + 5^ 1 )} + 5.u 3 ( u .« ) ^ { 2 4 .6 in the denominator and in the last step pick out the lowest order terms in u (i.193) 2u(l ..3) = = ~ .13)} + ±u3(u .194) and (18.u){24 .u) + 2u} + i u 2 [ 2 + (1 .{l u ( l u) / W 2 + 2} .430 and CHAPTER 18.197) .196) Now consider the last line here without the factor 2. (18.3 « .13)}.e.3 .195) we obtain E(k) 1 + ln uK(k) .1)(1 . (18.3) .u) + 2} + i u ( u .1)(1 .u) + uf\ 2 + ju2(l 16 + (u W 2 12 4u2(lu)2 +  l ) u ( l .u)} .3u(6« .2{«(1 .3 u ( l .3u(6u .18u 2 + 39M} + 12u 3 (u .« 2 + 3u] In [ .3u(6u .6w2 + 13u} + 12u 3 (« .^ L ) H[(i _ U ){4 + 3u(l . 8 + 6 u 2 .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}] 1 H 1 o u(u .195) With (18. Anharmonic Oscillator Potentials *« = M £ H Consider first E(k): E{k) = 1 + .3 y . 4 \ u In —= I + V W 2 1 + ln x(l u) 31/ +.3u(6u .u) 2 {8 . 1 3 « 2 = .u){4 + 3u(l . (18.1)(1 .13)} + ~u3(u .2)1 4 uA"(fc) In ( ^ = ) ^{«(1 .u ( l .3) .3).u ( l .u)} u{u(l .3) 1 + In ^ . .«) + 2} + V [ .— ) . 8 + 9u 2 .194) ln '^J + 2 In i/2u )+H l + 2u(l .a = (l«)(jl 2u/ 2 ) . up to and including u2): u(u .
l2 = ^ ( l + uf/2[E(k)uK(k)}. where w±(z) — h±(z — z±). [(1 .4u . V^LJ 4 2 16 ^ . ' 8G 2 q = 4 2 q ln f 26/2 \ o .3 The Double Well Potential Now consider the bracket [.] in (18..it)} . / to + 3GH~ \ 8 . Inserting here from the above expansions the contributions up to and including those of order u2. .J? dz^\E V(z)\ \ 87T2 / lEVtz)]1/4 Solution: From Eqs.zo) HWKB normalized _ ( h* y^eM. VB(z) = y\hTB(z) with VB(Z) ~ Bq[w±(z?  ^ [±(q  J l)}\2^V ..e. From (18.200) Remembering that u = Gy/2q.4u .6: Normalization of WKB solutions Show that the normalized form of the WKB solution (18. 2 .18. i.e.199) We now return t o the expression on the left of Eq. (18.2{w(l u) + 2}] = u(3u2 .195) with (18.198) 9 «. (18.155)) <*VA{Z) = VB{Z) a n d VWKB(Z) = 1VA{Z).197) we now obtain 4 1 • E(fc) .u H I .196).156a) we obtain a and 7 and hence the ratio (always in the dominant order) T2~ ) w i t h h += ^ h • Since (cf. this becomes _2 2 u 2 / j M 2 / 3^2s 2 ~ 3G 2 2G2 H \ / 2 H / .2 l_ln(1 JL\ V {2.3) .2 (18. i..196) and (18.128) and (18. (18 201) = Thus 3ffl« (2i/aGi/a2i/V/V"4 3ffl2 (Gii7»J4  '• . (18. we obtain: h ^ —^ 1 + 3G 2 V 2 8 2 /„ u 1 . we have.«){4 + 3u(l .uK(fe) ~ 1 + In I — = Mu2(3u.128) and (18. (18.148) (for particle mass 1/2 and in dominant order) is given by H. + iu^Zu2)^~u+ — +0(u3) l .2 /o 2 431 (18.3). Eqs.173) t o be evaluated.^HfHGH Example 18.
*o)/ ^WWKBOO (18. ° \Edzy/\EV(z) V(z)V4 VWKB. i. 82. W. Anharmonic Oscillator Potentials With the standard normalization of parabolic cylinder functions. Eq.hA/2 = h^/4 this implies . normalized = We see that the normalization constant is independent of a quantum number.204) . i. and obtained as Eq.3)]!23<«D [5(91)]' >A. 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. 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. The constant obtained here will arise in Chapter 26 in the evaluation of the normalization constant of the WKB wave function (cf. (18./ .158).g. (26.zo). by using the periodic WKB solutions below the turning points. Magnus and F. Oberhettinger [181].147).l)]![i(0 . We start from Eq. Solution: The turning points at zo and z\ on either side of the minimum at 24. oo .146) and (18.e.f hi\ VWKB. 1 7(l. pp. (18.** oo / dwD\.7: Recalculation of tunneling deviation using W K B solutions Determine the tunneling deviation q — go of q from an odd integer go.432 CHAPTER 18. Example 18. (18. y±(z) = IVA(Z) ± yA(z)\ = — i M O O ± 1 Ji.Aw) = s/2i 2("!) V (41) and [(9 . Then ((r. 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 [ .35)).e.143).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+. are given by Eqs. 93.
18. (18.e.208) and cot L />G + 1/2 ^(ig4^+^))"" + J 1/2 2qh%h%U{z)\ 27 +  :± (18. of course.e.e.Jz0 sin ^o J LJzo 4 Jzo 2 y •••4y ••• = .J^Wj +(18.6..i ) N c o s / •••+! provided the BohrSommerfeldWilson fZ1 dz^/E J zn quantization 4 condition holds. Sec. 7T\ + ^ 4 7r\ dz .. from ZQ. i.V(z) = r J zo I ZQ dz[\qh\ \ Z h%U(z) (27V + 1 ) .212) tt See e.r + ^=co S [r.3. JV = 1.iK h\U{z) + / 4 (18.209) "27 In the present considerations we approach the minimum of the potential at z+ by coming from the left.2..^r + + .205) We now apply the boundary conditions (18. z{) we expect tt (18. approach the minimum also from the right. \h\U{z) 1/2 .210) Choosing the point z to be z+.( . We could..132) and (18. Messiah [195].133) at the minimum z+ and obtain the conditions: 0 = — sin 7 and 0 = — cos 7 Hence £ + 1/2 dz(±qh\\h%U{z) 4_ ± — cos 7 />G^ 2 2 +i4c/(z)) 4 (18. i.g. ^(7(z) r t J z0 •I Zi I V2 dz \qh\ 1. 6.207) (18.g..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. A. from z\.211) 7 z l . 1/2 .. i. this implies sin cos sin cos Thus e.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. Then at any point z € (zo.
W "Cf.209) and (18. where u> is the oscillator frequency of the wells. i.143). One may note that the exponential factor here is not squared as in decay probabilities (squares of transmission coefficients).Zi / r. D.exp ( \/2h2T V2 7r 7 dz l / qh\ + \h\U{z) (18. + l ) j } ^ 1_ 27' (18. Bhattacharya [23].90 — =F— for q0 = 1. (18.215) which results from the factor of 2 in front of the sine in the WKB formula (18.149). Anharmonic Oscillator Potentials Similarly under the same condition (l)Jvsin J Zn where it is understood that qh2/2 ~ Ey(z±). . K. The latter give 7 7 1/2 exp dz dz \qh\ \qh\ + + \h%U{z) 1/2 exp I — / \h%U(z) (18. We note here incidentally that this agreement demonstrates the significance of the factor of 2 in (18.434 CHAPTER 18. The prefactor 2\/2h2/n is. Bhatnagar [14]. 7r"y in agreement with Eq.215) we can insert for 7 / 7 the expression given by Eqs. the level splitting is given by AE(q0. 5. (18. .148) and (18. Lifshitz [156]. S. they become c o t { ( .215). and cot< (q + 1) '!}» for q = 50 = 1.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. (18._ £/>(£<•>•) tan{(g + l Since tan ) ^ ^ . 2IO/K. Landau and E. ' /^i .205).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. 2Sm (2 g /ft2)i/2 in agreement with the right hand side (the last step following from Eq. 1 1 . We now see that Eqs. .214) {«+"i} 0 for q = go = 3. P. in fact (with our use of qo). 9.. K. (18.213) The integral on the left can be approximated by 1. (18.3.216) Inserting this into Eq. (18. M. In Eq.146)). Banerjee and S. .210) assume a form as in our perturbation theory.215) we can expand the left hand sides about these points and thus obtain . (18..h2) = = h2 h2 1 y 1/2 2{EEo)=2^{q~q0)=2^±l V2 V2h2 . and L. 7 .27 9 .5.e.2 . The formula thus agrees with that in the literature.
MiillerKirsten [9]. M. since no probabilistic interpretation is available for the field theory matrix elements in virtue of creation and annihilation processes during the interaction. 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. Morse and H. 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. in an early investigation Case^ showed that potentials of the form r~~n. Feshbach [198]. f K . W.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. H.Chapter 19 Singular Potentials 19. Giittinger and H. 'For a review from this perspective with numerous references see H. are not as troublesome as one might expect. M. In particular Case pointed out that for a repulsive singular potential the study of scattering is mathematically welldefined and useful. 435 .* However. n > 2. Some decades ago — before the discovery of W and Z mesons which mediate weak interactions — and before the advent of quantum chromodynamics. Aly. J. However. The centrifugal potential oc r~2 is generally considered as exceptional and is treated in detail in wellknown texts on quantum mechanics.* The physical analogy between singular field theoretic interactions and singular potential scattering of course breaks down at short distances. 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. W. Case [45]. also in P.
Gubser and A. Callan and J. W. D referring to Dirichlet boundary conditions) in 10dimensional string theory. 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. that with the hyperbolic cosine replacing the trigonometric cosine).e. In fact. Singular Potentials theories in mind. Hashimoto [124]. R.2) § C . K.2 19. S. but also that the Smatrix can be calculated explicitly in both the weak and strong coupling domains.2. In view of the unavoidable use of Mathieu functions expanded in series of Special Functions. H. N. some also permitting further research. Manvelyan. 19. S. This property singles this case out from many others and attributes it the role of one of very few explicitly solvable cases. D. Tamaryan. roughly speaking a brane is the higher dimensional equivalent of a membrane visualized in two dimensions from which the DS brane derives its name. .436 CHAPTER 19. MiillerKirsten and Jianzu Zhang [226]. It is therefore natural that we study this case here in detail. J. In the concluding section we refer to additional related literature and applications. this particular singular potential plays an exceptional role in view of its relation to the Mathieu equation. We shall see that not only can the radial Schrodinger equation be related to the modified or associated Mathieu equation (i. Maldacena [41]. (19. G. MiillerKirsten.Q. J. S. H. the attractive singular potential 1/r4 was found to arise in the study of fluctuations about a "brane" (the D3 brane. Singular potentials arise in various other contexts. § 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). W. Liang and Yunbo Zhang [189]. Park. Recently.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. This chapter therefore gives the first complete solution of a Schrodinger equation with a highly singular potential. our presentation below is deliberately made elementary and detailed so that the reader does not shun away from it. J. for instance.
4) dz2 + In the literature this equation is known as the modified Mathieu equation or associated Mathieu equation.28) below). MiillerKirsten.3b) 7 = % j.g. The following substitutions are advantageous: y = r1/2cf>. and so =4 5 and n ez = r = ~ = e^4fir. \[\ in the correspondence is nonintegral. (19. . This has advantages compared with higher dimensional cases. (19.2 The Potential 1/r4 — Case of Small h2 437 where E = k2. This is rarely possible and therefore this case deserves particular attention. W. h2 = ikg. [l+2) ~ * A> ~ the equation converts into the periodic Mathieu equation of Chapter 17. J. using the method of Dingle and Miiller of Sec. The radial Schrodinger equation then assumes the form 2h2 cosh 2z . We begin with the derivation of various types of weakcoupling or smallh2 solutions which we construct again perturbatively. h = ei^y/k^. Manvelyan. ig h \j g (19.3a) In the case of the attractive potential (as in the string theory context referred to above. 19. In the following we study the solutions of Eq. We observe that with the replacements p2 z^iz. and as we have in mind in Sec.mo = 1/2 and h = c = 1. Eq. H.19. It may be noticed that since I is an integer.4) first for small values of h2 and then for large values of h2. It is clear that we draw analogies to our earlier considerations of the periodic potential. R. 8. Our ultimate aim is the derivation of the /Smatrix for scattering off the singular 1/r4 potential.( I + 0.7 that we employed also in previous chapters. Liang and Yunbo Zhang [189].^ where \f\ can be an integer and thus leads to singularities in the expansion of v (see e. Subsequently we derive the same 5matrix from the consideration of largeh expansions and calculate the absorptivity in the case of the attractive potential. 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.Q.3 below) g2 is negative and hence h? is real for E > 0. (19. (19. r = 7 e z . J.
.2 A / i 2 . —Zu — (Z„_i + Zu+i). Singular Potentials Small h2 solutions in t e r m s of B e s s e l functions We now develop a perturbative procedure" for solving the associated Mathieu equation in the domain around h2 = 0. (19.I . Aly.6) terms amounting to 1 „ 2 d2Zu A V (19. i.7) We shall see that this parameter is given by the same expansion as in the case of the periodic equation. and H„ (w).Hv (w).2. (19.g) This equation is wellknown as Bessel's equation or as the equation of cylindrical functions. where these are the Bessel function of the first kind.Nv(w). Proceeding along the lines of the perturbation method of Sec.11) We follow here largely H.4) becomes dw2 w2 w2 \ dw2 w dw \ j Next we define a parameter v by the relation V2 v2 = (/ + .^% dw1 1 1 —— = dw = ][Zv22Zv 4 w Zv + Zv\ = (Zv\ 2 + Zv+2].i l . MullerKirsten and N. . (19. w 2 .438 19. H. H. J.e. 8. — Zv+i). Eq.2 CHAPTER 19.1.5) so that Eq.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. VahediFaridi [10]. (19.7. (19. W.6) becomes to zeroth order </>(°) = <j>v of <f> in h2 Dvcf)v = 0. First we make the additional substitution w = 2/icosh z. where £>„ : = —_ + i _ + J l .9) leaves uncompensated on the right hand side of Eq. Neumann or Bessel function of the second kind and Hankel functions of the first and second kinds respectively. The zerothorder approximation = Zv[w) (19. dwA w dw { wA ) (lg. The solutions Zv(w) are written Jv(w). (19.
2 ) * ( z / .u + 4fZu+A}.i/4)*Z 1 / _ 4 +(u. where We observe that DUZV = Q. v2)*{y2. wz The expression (19.u (19.18) Now </>(°) = Zv left uncompensated Rv . v .20) 2)*ZV_2 + {y.10) is now particularly simple: i?i°) = h2[(v.10) in terms of functions G„ defined by Gv+a = —nZ v+a . v)Gv + (y. when a or 2v + a = 0. (19.13) (19.2 ) * i ^ 2 + (!/. The next contribution to (^°) + t^1) therefore becomes 0(2) = /i4[(z._2 + {v.i/) = 2A. Du+aZu+a but Dv+a — Dv so that DuZv+a = a(2l/ (19. v + 2)*Z„ +2 (19. (19.6).12) (19.13). z . therefore 4>^ leaves uncompensated i#> = / i 2 [ ( i / ^ . of course.14) (19. (i/. .19..21) +(i/. where the starred coefficients are defined by (19. I / ± 2 ) = 1. The terms (19.v + 2)*{u + 2.u + 2)*Z„ +2 ]. it is more convenient to use these relations in order to rewrite Eq. . v .2)*Zi.2. can be cancelled out by adding to </>(°) the new contribution liZu+aja{2v + a) except. v + 2)*(v + 2. We assume in the following that 2u + a ^ 0 (the case of 0 has to be treated separately). (19. However.13) therefore lead to the first order contribution jfU = h2[{v. ~a(2v + a)~ w2 * a> Zv+a = a(2u + a)Gv+a.2 The Potential 1/r4 — Case of Small h? 439 we can rewrite the expression (19.10) as a linear combination of various Zv.15) (19.17) Thus a term fiGu+a on the right hand side of Eq.16) ( I / . v + 2)Gv+2].!/+ 2)*4°J 2 ].2)G„_ 2 + (y. and this means in (19. = §.
.23) These coefficients may also b e obtained from t h e recurrence relation P2i(2j) = p2i2(2j . etc. + 2(y2\) — + 32(i/ 2 l) 3 (^ 2 4) —^ (9. v ( 4 5 a . ^ 64(a_1)5(a_4)4(a_9) + 0 ^ )• (1928) 1 6 . P2*o(0) = 0.I/±2)*(I/±2. . v — 2)*(v — 2. j=i.v2)*{v2. Evaluating t h e first few terms.j^0 (19. Zv + Y^h2i i=l J2 p2i{2j)Zv+2j. . . we obtain 9 h4 = a 2(al) 3 2 (13a .v)] (19. v) +(i/.25)/i 8 32(al)3(a4) .. we obtain t h e expansion . p 0 (0) = 1.v + 2)*(i/ + 2. (19.1169)/i 12 .22) where p2(±2) p4(±4) = = (i/. p4(±2) = (i/. in Eq.17)).I/±4)*. i/)] + • • • . i/ ± 2)*.l)(z/ 2 . Adding these terms and setting the coefficient of Gv equal t o zero.v) + {v. i/ ± 2)* (i/ ± 2. Reversing the expansion and setting a = (I + 1/2) 2 (not to be confused with a e. subject to the boundary conditions P2i(2j) = 0 for  j  > i. (I/. (19.26) iy — 2.440 CHAPTER 19.v) + hA[{v. i/ + 2j) + p 2 i _ 2 ( 2 j ) ( ^ + 2j.i/±2)*. .4)(»/2 . we obtain 0 = h2{v. 1 1 \ 2 + 2) ~ ) s X 2 A = v 2 hA (5v2 + 7)h8 v . and p 0 ( 2 j ^ 0) = 0. v + 2)*{v + 2.25) Finally we have to consider the terms in Gv which were left unaccounted for in Rv . v + 2j) (19. . Ri.455a + 1291a .v + 2y{v + 2.2 + 2 9 ) ^ 64(^ 2 . i/ + 2j).2)(v + 2j . . (19.9) 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.24) P2i2(2j + 2){v + 2j + 2.4 + 58.2. Singular Potentials Proceeding in this manner we obtain the solution ^ = 0(0) + 0 (1) + ^(2) + .g.
33) = = cosh(^ + 2)z + cosh(^ — 2)z.v + 2)<t>v+2].i .7) into Eq. — We return to this question later but mention here that this problem has already been dealt with in Chapter 17. (19.h2[(j)u+2 + 4>v_2] h2 [{u. Thus v is real for small values of  h21. apart from a normalization factor which we have chosen (so far) such that the coefficient of Zu in <f) is 1. The solutions 4> of the modified Mathieu equation are now completely determined. .2. we may say that the first approximation <^°) leaves uncompensated terms amounting to R^ = 2ft2 (A . (19.34) .31) It follows that Dv+2n4>v+2n = 0.cosh 2z)4>.3 Small h2 solutions in t e r m s of hyperbolic functions For later purposes we require yet another type of solutions.30) £>„&. v)(f>v + {v. 19. we can rewrite the latter as j ~ .19. v .cosh 2z)4>v = = 2h2A(/>„ . We are still left with the question of what will happen if 2v + a = 0 or v — ± 1 .^ + e±(v~2)z.2)<t>„2 + (v. Substituting (19. ±{ v+ z e = sinh(v + 2)z + sinh(z/ — 2)z. £>„ = — .An(u + n). so that J2 (19. 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 .v2(j> = 2/i 2 (A . (19.4).2 The Potential 1/r4 — Case of Small h? 441 We note that this is an expansion in ascending powers of h4. ±2.32) (19. (19.29) sinh vz or e±uz. = 0. Thus to O(0) in h we have (jr ' = 4>v — cosh vz. 2 . (19.
38) (f> with (\>v = exp(vz) — Meu(z. 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. v ± 2) etc.37) where p M (±2) = (I/.4 Notation and properties of solutions We now introduce standard notation as in established literature.h). Singular Potentials (I/.h). Denning (19 36) (">" + <*) =o^7Ta)> we now have the solution  p(z. Arscott [11].2. (19.7 — we refer to Meixner and Schafke [193].I/) = 2 A . M.* The solutions of the modified Mathieu equation which we are considering here are written with a first capital letter.h) = <(>„ +Y.442 where (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. Meixner and F.h2i *=1 J2 j=i. .h). W. (19. The use of the symbols (v. > These solutions correspond in the periodic case respectively to the solutions and me^. In fact we could have obtained the same Ru by starting with the modified Mathieu equation for h2 replaced by — h2. 19.J/±2) = . etc. Schafke [193]. in the present context should not be confused with the same symbols having a different meaning in the case of solutions in terms of cylindrical functions since it is generally clear which type of solutions and hence coefficients is being considered. > (f> with </v = sinh vz — Seu(z. The solutions in terms of cylindrical functions are written *J.35) The form of i ? ^ is seen to be almost identical with that of the corresponding expression for solutions in terms of cylindrical functions. 8.I/±2)*. In order to avoid confusion arising from the use of different equations. F. • (19. we prefer to adhere to one equation with different solutions.J7t0 p2i(2j)<f>u+2j.1 .
h) = av{h)M^ (z. p. (19. Magnus and F. For the relation of Hankel functions used below. (19. Also from the expansion of Jv{z) in rising powers of z. h). 4>v = Hi. W. Schafke [193]. h). introduces the dominant order function into higher order contributions''' — oo i 2i Meu(z. h). (19. Mi. 16.h) and oo i M^\z. h). 'See e.39) Writing two solutions out explicitly we have for example — apart from an overall normalization constant. i. Oberhettinger [181].44) implies.43) and hence similarly M^\z + inir. see this reference p. i. — oo cv2r{h2)e^+2r>. (19.h) = Jv{2hcosh z) + ^h 2i J^ j=iJjLQ p2i(2j)Ju+2j(2hcosh z).h).19. Meixner and F. (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) .h = £ p2i(2j)(l)j exp[(^ + 2j)z] (19.44) (19.h2) oo = Y. which. h) are therefore proportional to each other as a comparison of Eqs. (19.45) T Below we frequently write the normalized solution as in J. h) = exp{vniri)Me„(z.40) Cev{z.e.42) The solutions Meu(z. h) = exp(inwn)MP(z. . we obtain$ Ju(2hcosh(z + inir)) — Jl/(2hcosh z exp(m7r)) — exp(inuTr)Jv. 17.MP(z.h).h). whereas the — as yet — unnormalized solution (19. this normalized solution possesses contributions exp(i/z) in higher order terms.e. As emphasized in Chapter 8.40) does not. as we remarked earlier. Me„(z. 443 (19. ' (z.h)±Sey{z. W.42).g.(2hcosh z). as Mev{z.41) i=l We see immediately that Meu(z + niri. <f)v = N„(2hcosh z) MJ?\z. 4>v = Hv (2h cosh 2) * M^\z.h) = exp(vz) + Y. {2h cosh z) • MJ?\z.h).
As mentioned earlier. H™(z) = e^HfXz). The functions Me±v{z. oiv{h) = (19. h) .50) E.e. . we have the following for a change from v to — v.4) we obtain oo £ r=—oo [(i/ + 2r) 2 4.e. MJ>>(0.h) .43). Mi 3 . the explicit evaluation of the S'matrix element) we elaborate here a little on the computation of coefficients of normalized Mathieu functions. r=—oo (19.\c\r + h2(c^2 + 4 r + 2 ) ] e ^ + 2 ^ = 0.8. ^ . 4) = exp(±^7r)i\4 3 ' 4 \ M^3'4) = M™ ± iMJ?\ (19. i.46) H^z) = e^H^(z). Equating the coefficients to zero we obtain [{y + 2r)2 + \]cv2r = h2(cv2r+2 + c£ r _ 2 ).e. 178) to be convergent for  cosh z\ > 1 but uniformly convergent only when  cosh z\ > 1 for otherwise complex values of z. h) = M™(z. i.exp(TM"r)MW(. 130) to converge uniformly for all finite complex values of z. we use the notation of Meixner and Schafke [193] and so write the expansion of the function Mev(z. the first relation is obtained as follows: 2M<£1 = M^l+M^l. For later essential requirements (i. (19.48) The series expansions of the associated Mathieu functions M„ (z. Singular Potentials where clearly MeJQ.47) With this equation we obtain for nonintegral values of i/:§ ±isin WK MJ)3A\z.h2) = ] T cv2r(h2)e{y+2r)z.444 CHAPTER 19. p. Meixner and Schafke [193].49) Inserting this expansion into the modified Mathieu equation (19.g.h) can be shown (cf. Meixner and Schafke [193].h) Using further properties like (19. s (19. h).h) on the other hand can be shown (cf. These points are important in our derivation of the S'matrix below.h2): oo Mev(z. p. 2M™ = eiuvMi3) + e~ivlrM^\ .
2 The Potential 1/r4 — Case of Small h? which we can rewrite as fc2^±2 c 445 = [A _ („ + 2r) 2 ] .19. W. Alternatively taking the inverse of Eq.^ c 2r 2r 2T~2 or 2r ^2 /l2[A_(I/ + 2r)2]£2l± 2 . 117. . Thus now c 2r . (39). + 2 r ) 2 ] . \ .T22r[\ A .53) /i2[A(^ + 2 r ) ] .54) 2 C2r 4 ^ h. h~2[\ .(/^.^ C 2r+2 (19. p. 2r (19>51) 2r For ease of reading we give the steps in rewriting this.* For 'Actually. Schafke [193].^ p 2 2 ^ or c 2r+2 2r2 1 (19. 122. see J.[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. eg = c~v = 1.51) we have 2r (19. (19. p. + o2„r 2 l2 ]  C 2r 1 2r c 2r2 or 1 c J =^[A(./ i 2 C ^ .(v + 2r)2} / z .52) in agreement with Meixner and Schafke [193]. . \) /. Meixner and F.^ + 2r + 2 ) 2 ] .2 [ A . Eq.
W. .52) and obtain c% 1 eg fc"[Afr + 2)»)h_a[A_(. i/2 + 41/ + 7 7i6 + 0 ( / i 1 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.( " + 2)2] ~ h2 h. Then di 2 1 t" + 2(^1) + • • • . t For this purpose *We do this in the manner of the original derivation of R.446 CHAPTER 19.12) ^ 384(^ + l)(z/ + 2)(z/ + 3)  " ^rW^hy._{u+m h2 M ( " + D + 55^iy] + i ^ 5 j /i 2 4(.+0^"> < 19 . Singular Potentials nonintegral values of v examples obtained in this way are (see Example 19. H. . Here we insert for A the expansion (19. M. J. Aly. J .^ ^ + 2)(i/ + o> l)(i/ 3) 0(^10). + 0(/.C + l) 3 (i/ . Solution: We put r = 1 in Eq.+4)!ll_. Liang and Yunbo Zhang [189]. H. + i). Spector [257].5 D e r i v a t i o n of t h e S .. MiillerKirsten. 8 4 ( i / + 1) 128(i/ + l)2(v + 2)(vl) + 5 10 32(i/ + ^ l)(i/ + 2) + 768(^ „„ . and also R. but follow our earlier reference H. W.2..1: Evaluation of a coefficient Evaluate explicitly the first few terms of the coefficient Cj/cfJ given in Eq. J.[v^.27) and truncate the continued fraction after the second step. (19. H. (19.Q .55).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.l ) 19. Manvelyan.1 for the evaluation of a typical case): hb + 0(hw). MiillerKirsten and N. VahediFaridi [10] (this paper contains several misprints which we correct here). 128(i/+ l) 2 (i/ + 2 ) ( ! / .55 ' Example 19.
57) ^ 1 . we have W r = ip + ut — 7r/4' . 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.3a) and (19. and considering the propagation of this wavefront. p. 2 ) 2 1/2 ex Z/7T IT H=i —J TTW P l+O 10 (19.56) where by Eqs. In the case of the repulsive singular potential here. . We obtain the regular solution by choosing Zv(w) = H„ (w) for 5ft(z) < 0. Oberhettinger [181].19. Magnus and F. 22.58) The behaviour of y reg near r sa 0 is therefore given by 2 1/2 2/reg "\ 5 exp 1\TT r\ —r exp i . (19. \w\ S> 1 and — IT < arg w < IT is known to be given by* (19.. W . I / + 2 2 l\TT +E ^ i\ E j=i. (19. Then y reg = = r rV2M®(z.59) which tends to zero with r.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. and then the continuation of this to infinity. this wave function near the origin is the wave transmitted into this region (as distinct from the reflected and ingoing waves).5) w — 2h cosh z = [ kr \ r Thus r —> 0 implies \w\ —> oo. The asymptotic behaviour of the Hankel functions H„ ' (w) for \w\ ^> \u\.h) 1/2 w) (19.j7L0 P2i(2j)exp i[v + 2j + 2 2 .
17.62) Also with Eq. 4%r g >0 4r^r ' rg See W. we can derive y_ from y+ since one can show from the circuit relations of Hankel functions^ that M^\z + m. h) through the entire range of !&{z).61) In fact. .j^Q p2.A1"—2 z '+"2 A~~ n or rA .A14A. Now from Eqs. Meixner and Schafke [193]. '(z.#. The series M^ (z. We require therefore > > the continuation of Mi. Oberhettinger [181].60) Using the above asymptotic expressions for the Hankel functions.57) the condition  cosh z\ > 1 implies I cosh z\ kr + ig/r >1. but uniformly convergent only when  cosh z\ > 1 for otherwise complex values of z.j¥=o Ki(2j)exp(Tmj) (19. 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. (19. p. '2"4 . ^ 1 ) (e i 7 r z) = Hi2!>(2) e^H^iz). In a similar manner we can define solutions y± by setting for 9fJ(z) > 0: y± = rl'2Ml?>*>(z. Singular Potentials so that when t — oo : r — 0.448 CHAPTER 19.h) can be shown (cf. Magnus and F.h) = exp(Mr^)il4 4 ) (z.3a) and (19. (19.h) oo j 2 — rV flp^ + ^fc* £ i=l j=i. This means that the origin of coordinates acts • > as a sink.(2j)<gH • (19. p. AV .3b) ln • 7T Q"*4' r fg ° Vk (19. We now require the continuation of the regular solution y reg to solutions behaving like y± at r — oo.V + § 9 2 > 4hh* . 178) to be convergent for  cosh z\ > 1. Prom the relation r = (ig/h)ez we see that r — 0 > corresponds to $l(z) — —oo and r — oo to $l(z) —> oo. 2h and the square of this expression implies 2 2 ~ ' A./i)./l ~ 2 ~ 2z + " 4 > 0.
449 This condition.e. which has to be bridged by using another set of solutions. the domain r_ < r < r+ — as illustrated in Fig.63) Thus there is a gap between the two regions of validity — i.3b) and (19. A suitable set is the pair of fundamental solutions Me±v defined by (19.rl)(r2 . is satisfied only for r < r_ < r__ or r_ < r + < r. and in order to maintain this symmetry. r 7r for 0 < r < ro. Since the real part of z changes sign at r = ro.3a).1. 19. .38) or (19. (19. 130). Originally we chose the sign of m / 4 as in Eqs. we have to assign different signs to the imaginary part of z in the two distant regions. i.ijt/4 Rez . (r — r+)(r + r+)(r — r_)(r + r_) > 0. This interchange in the > solution My interchanges y reg with y+.62). 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. (19.r2_) > 0 with r£ = rg(2 ± >/3). i Im z z = In r/r0+i7t/4 . 19.1 The domains of solutions.e. (19. Thus we choose . p.2 The Potential 1/r4 — Case of Small h2 Hence (r2 .49).19.ft/4 _Z z = In r/rm/4 0 r 0 =(9/k) 1/2 Fig. Meixner and Schafke [193]. These solutions converge uniformly for all finite complex values of z (cf.
z=—in/4. and we determine these coefficients from this equality and the additional equality obtained from the derivatives. (3 = a . At r = ro the real part of z vanishes and then switches from negative to positive. a4[r1/2Mev(z)] dr a'^lr^MeAz)} It follows that (19. Singular Potentials Then starting from the region r ~ 0. i. to r^2[AM^ + BM^\ A.65) in/A a>d[rl/2Mei/{z)]+f3>d[rl/2Me_Az)] Since Mev(z.)](1964) The right hand side of the first equation now represents the regular solution in the domain r_ < r < r+. h) (as one can check or look up in Meixner and Schafke [193]. Thus. Hence we match the right hand sides of Eqs.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_. we express in the domain r_ < r < r+ the regular solution as a linear combination of solutions Me±v with coefficients a and /3.450 CHAPTER 19.l/2Me_i/(T.v(z)\ —m/4) + ^[r^Me. .(2)]z=w/4 + 0Me. p. + ^[r'^e^z)] dr z=—iw/4 + (31 ^[rll2Me.B^0.e. (19.[ r l/2 M.67) Next we have to continue the solution beyond the point r+ to a linear combination of solutions y+ and y_. h) = Me^u(—z. i. 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.^)] dr z=in/4 dr (19. as just explained.66) a = (3 . (19. at this junction we require also the imaginary part of z to change sign. 131).v{z)\ . Thus we set rl/2M^\r) = = rll2[aMev(r) + (5Mev{r)}. (3) (r)] aA[rl/2Mei/(r)]+/3i:[r.
68) \j3Me„(z) + aMev(z)] P±[Me„(z)}+afz[Me„(z)] = = [AM^\z) + BM^(z)}.) d_ _ 1 d dr r dz Hence for any of the functions Mv:  [ . multiplying the first of Eqs. (19.70) = a±[Meu(z)}+P^{Me_„(z)} A^[M^{z)] +B^[M^(z)}. (19.64) the relations M&Hz) jz[M^\z)] and from Eqs. (19. (19. (19.65) and (19.68) From Eqs. (19. (19. ^ ] = ^ + ^ ^ .3a) and (19. (19. (19.70) by Me'_u (the prime meaning derivative). (19.70) by MeU and subtracting .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.3b) we infer that (z = l n r + const. the nonderivative parts (from the first term on the right of Eq. for instance.g]:=f(z)*&g(z)WZ) dz dz Thus. and the second of Eqs. For the Wronskian W of two independent solutions f(z) and g(z) which is constant and can be evaluated at any point. we use the notation W\f.68) the derivatives with respect to r by this relation.19. A and B. dr BM^(r)]. (19. = aMev(z) + 0Mev(z).2Mei/{r)]+d[rl/2M] dr dr = = rxl2[AM^\r) + Ad[ri/2M(z)(r)] dr +B^[rl'2M^{r)]. with the replacements of Eq. Then. (3. (19.69)) cancel out in view of the nonderivative relations.65).71) We now determine the coefficients a. and we are left with relations expressible only in terms of z. Thus we obtain from Eqs.69) Thus if we replace in the derivative relations of Eqs. (19.67): rl/2[(5Meu{r) + aMe_u(r)} pd[ri.
.58) and considering large values of \z\.u.M^] W[Mev. i. we obtain the first of the following two equations. Singular Potentials the second resulting equation from the first resulting equation.(0.M^]= IT e~iv*av.Mev]' P l ' From Eqs.4] = . B = J 2 L _ e .e.Mev] = = = av. W[MJ?\ M^] W[Mev.57) and (19. 170/171).g. av{h) = Me. W[M^\Mlf)]W[Meu.71) we obtain in a similar way A B = = W\MP\ Me.M^} W[Me„.* « * r r z H . h)/M^\o.MiA\z)} = 2(i)(2/Trw)(dw/dz) = (4i/Tr)(2hsinhz/2hcoshz) ~ 4i/>.48) (or cf. 7T [1.73) We now use Eq. h).74) Moreover. With these expressions the quantities A and B are found to be A= f au ^ 2isin VK \ a_u 1 a. the leading terms of the respective cylindrical functions. or obtainable by substituting e. (19. (19.76) .4] = . (19. (19.j] given in Meixner and Schafke [193] (p.u}W[Me.^ ^ a „ Q _ .46). M^4)] + W[MJ>3).^ 47 9? MJ>4)] MeV] [3. (19.e.A] = W[Mi3\z). we obtain immediately W[Z.75) With these relations and Eq.47) we obtain (av defined by Eq.] W[Mev. and Wronskians W[M^\MIJ)] = [i. using Eqs. we use the following circuit relation which can be derived like Eq. (19. .Me^} (19. 7T W[Mev.Mevy W[M?\Mev] W[Mev.. i. Meixner and Schafke [193].77) av I 2ism VK \ ctv av ^Thus. 7T (19.452 CHAPTER 19.46)) W[Mev. and the second is obtained similarly: W[M?\Mev\ W[Mev. p. (19. = IT <*„. .3] = [1. Me. TV W[Mev..\ \ ( av 2iwKav ^ v . (19.M^] J™aV. 169) M0) = e«^Ma) _ isin U1X M(A) _ (19.
/) = (l)1/2R2^2i^)ei^)U^. in Chapter 16). B=tfD'"". (19.82) The S'matrix is defined in the following way in the partial wave expansion of the scattering amplitude /(#).83) The Smatrix element is therefore given by ifc*"* .80) 2^fc (1981) where \7r/ i t sin VK /(*.h) + BM^\z.78) we obtain Ay reg « _ .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. 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 . (19. as we recall from earlier considerations (e._ [ / ( f c . (19.(ite™)" 1 { 8 J ' f(k.46) together with the relation Mev(z. (19. Then f' . (19.h) = Me_u(z./ ( . 2zitsm z/7T zzitsin I/TT Inserting these expressions into Eq.(l)le^}Pi(cos 9). f c r .19. Z ) e . (19.85a) .h)] ( m / 9 \ 1/2 {L\ {Aeikrei(V+l/2)*/2+Beikrei{v+l/2)*/2^ g) We set fl = ^ = MM).79) where the second expression follows from Eq. eii/7r/2 (19.l) ' We note already here that with the substitution R = ei7r7.f c .h).g. Z ) e ^ ] .
9?. we can write this (multiplied > by 2ismvTT. and with the left hand side following from Eqs.58)) in the limit r — 0. (19. 19.2.86) . (19.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. the reflected wave and the transmitted wave the quantities: A Ar At = = Re1' R 2zsin 7r(7 + u). we can define respectively as amplitudes of the incident wave. R 2i sin TXV.81) in another form from which we can deduce the amplitudes of reflected and transmitted waves. R — — = 2i sin ivy.2 The repulsive potential and the various waves.56).58)) 1/2 2r e~ 2hir cosh z. V(r)=g2/r Fig. cf. We can rewrite Eq. Fig. (19. (19.454 we can rewrite the . h) (see Eqs. and recalling that yreg is proportional to a function Mc> {z. Thus in terms of the (3) variable z.Smatrix in the form s i n 7T7 Si = CHAPTER 19.56) to (19. Singular Potentials r(i+k) sin 7r(7 + u) (19.85b) We shall obtain the 5matrix in this form in the case of large values of h2 later. 19. (19.
in which v is complex.28)) and R = ey real.e. This implies basically the evaluation of the expression R of Eq. (19. (105) to (108). (^)l4lu{h2)Jir(hez)J±u+i+r(hez). J. i. Gubser and A.h) is (cf.84) of the ^matrix. (19." \R~Ji\2 1  tO _  2 i s i n VK\2 [  . (p. Eq. 180).m a t r i x Our next task is to evaluate the expression (19. unity minus reflection probability = transmission probability. Liang and Yunbo Zhang [189]. y We observe that this relation remains valid if the real quantity R = e and the pure phase factor eW7r exchange their roles. h) serves the use of the same coefficients as in the other expansions. i.Q.2COS2VTT = (R.e™*/R\2 ' ' .79). i.2 The Potential 1/r4 — Case of Small h2 455 One can verify that for v real (which it is here in both cases of attractive and repulsive potentials in view of Eq.6 E v a l u a t i o n of t h e S . (19.39)) the associated Mathieu function expanded in terms of Bessel functions of the first kind. H. Hashimoto [124] and R.e..h) The function Mo (0. / i ) = Y. (19. unitarity is preserved. M^l)(0. l=—oo 1 1 lite*"* . this expansion is given as 1 oo MJ> Hz.19. Manvelyan. The expansion is: 4 r ( / i 2 ) M S ( z . Jv+2r. S. The latter is what happens in the S'wave case of the attractive potential. and then exploit it for the evaluation of the quantity R. 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. i. (19.„„„X (1 Q Q'T'l l \Reiun e~iuw/R\2 ~ \Reiv* .88) where the factor Me„(0. MullerKirsten. " S e e S. (19.88) the power expansion of the Bessel function one realizes soon that the expansion is inconvenient in view of its slow convergence. however not here. therefore we cite it from Meixner and Schafke [193].In Meixner and Schafke [193] (p.h) = 1 Me V\ 5 / r J2 __ 0 O c»2r(h2)J„+2r(2hcoShz). Eqs. if R becomes a pure phase factor and eil/7r a real exponential.±)2 + (2sinj/7r) 2 .2. but in the 10dimensional string theory context.( i J e ^ ) " 1 ! 2 = R2 + ^ . 178).e.89) .e. By inserting in Eq.** 19. A derivation is beyond the scope of our objectives here. W. J.
94) . (19. (19.. i. /=—oo (19. r = 0 and 2.90).90) Here the coefficients c2"{h2) are the same as those we introduced earlier for the normalized modified Mathieu functions in Eq. h) in many different ways. one obtains *.6 2 ) m 6 (^_i)3(z. (19.g.49). (19.h) = Jo(h)J±Ah)^^Ji(h)J±„+i(h) c0 {n ) ^ ' C^r^J2{h)J±l/2{h) cr(h2) + •••.h) v 2 = ^2 (l)lc%ih2)Ji~r(h)JM+r{h). Singular Potentials 4r (h )M$(0./>) = Co> 2 ) £ 7 " = — OO j ± > .2_4)(l/2_9)iv2y) + <AA. we obtain M&(0.91) Inserting here the coefficients given in (19.90) is in some sense amazing: It permits the evaluation of one and the same quantity M±J (0.456 so that in particular for z = 0 oo CHAPTER 19. As a matter of introduction we recall the definition of the coefficients with Eq. + 2 u l\2j 2 fh\2 "2+2 2 MX4 {v l)(v 4)\2 2 2(^ + 4 ^ . Eq.3 9 ^ . Next we evaluate M^1}(0. u»^. The formula (19.g. setting r = 0 in Eq.90) and choose r = 0 (one can choose e. 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. by allocating different values to r. e./i) with the help of Eq. ^ (19. r = 2 as a check). /n ^ . h2 is complex but v2 is real (cf.e. (19. (19.49) from which we obtain for z = 0: Me„(0. Here in the case of the repulsive potential.28)).55).
!' or with the help of the inversion formula of factorials R = v\{y — 1)! smKv f h 7T 2v (^ . 4(z/ T H ^ 3 .95) These expansions imply for the quantity R: I/! 2v ri1 i .IV 2 2 2(v =F 3i/ . one obtains M£!(O. ./I) h u .111) / / i 5 4 3 1+ 4i/ 2 + +' (z.32i/ .96) we extract for later reference the relation sin7rz/ R IT h h 2\ 2 Av [v\(viy.12i/2 + 64u ....19. 2 (h\2 ^ l/2_1l2^ 2(i/ 2 +3f~7) /feu .55) and collecting terms in ascending powers of h? (here for the case of nonintegral values of v). + # "+" ( I / .4)(i/ . In this procedure our attention is focussed particularly on the quantity called absorptivity in the case of the attractive potential. _2_(h\2 .2^ 2 ± 59^ .7) F " (i/±l)2(^^l)(i/24)V2.1 ) 2 ( I / + 1 ) ( V 2 .1) 2u(4u + 15i/ .2 The Potential 1/r4 — Case of Small h2 457 Inserting the power expansion (19. From expansion (19.2 _ ^ 4 ( ^ 2 _ 4)2 (19.93) and the coefficient expansions (19.9) V2 1 fh ±v 1+ 2 + • (19. We observe that this approximation involves v and h4 and hence is real.97) + This expansion will be used below in the low order approximation of the absorptivity of partial waves.4 ) V 2 > 1 (I/)!V2/ [1+ 2_Mx2 2(i/23«/7) (h)4 (iy+l)2(i/l)(^24)' .23) (hxQ 2 3 2 2 V ± l) (i/ T 1) (^ .] 2 \v i + {v2 .96) With this explicit expansion we can evaluate the S'matrix.l) 5 (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. This is the case of real .9 / .84): Si ST 1 (11/R2)(11/R*2) .R*2 ^ * #2 \ 11 tf2 R* (19.458 19.98) (cf.100) Here we set 2^/3 = 1 + j / ? e 2™ = l + gt (19. _ i?* 2 2+ g # 2 i?* 2 2 3/(i + s)(4 + i?*2).7 CHAPTER 19.101) where / is complex and g is real.102) (19. (19. Eq.^ i 4 2+^ # 1 / •. 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.28) that v is real. (19. Then cos 2TT/? = 1 .2.99) 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/? }]"' (19.103) Then SiSt (1+5) i .101) is > zero. (19. In this case R — R* and so 1/i?2 ~ Ofji4). Thus we set v = n + i(ot + i(3) = (n — 0) + ia.28)).eiun/R2)(eiu*n ~ (19. (19. (19.104) We now consider the case of a — 0 implying that g of Eq. and sin 2TT/3 = ft/. + ( i + s w (E 4 . 9 / « i(2yr/3)2 ss i(sin 2vr/3)2 ~ 2 sin2 TT/3. (e iun e~iv*^/R*2)' (19.
In all comparisons with these papers — which consider predominantly the 10dimensional string theory context — one has to keep in mind that many formulas there do not apply in the threedimensional case we are considering here. f W . and t h e values of ip(\/2) and ip{2>/2) as given in Tables^ with Euler's constant C ~ 0.l ) 2 2 \) +0{h«) (19. (19. H. ( I l l ) ) in agreement with a result in S.61n/i} + ~9~ 0(h8) (19.107b) = Ah2[l + 0(h4)}. MullerKirsten. Here in t h e case of the attractive potential (as remarked after Eq. Hashimoto [124]. S. Manvelyan. With t h e help of Eq. Hence with Eq. (19. .* )  14(^V = 1 T h e absorptivity A is therefore given by A = 1 .^ ) ~ 2 2 459 1+ 9 / • & ( ! . 1 29/ ^ (1 . p.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).e.3b)) h2 is real for E = k2 > 0.105) sin7r/3 (19. J.107c) Here h2 is actually Vh4.6C .97) this can be written A 4TT2 [v\{v . = 1 . We can also evaluate now the amplitudes (19.1)\} 2 h \2 iv 1 + Au {y . Liang and Yunbo Zhang [189] (there Eq. (19.3.28). J.102) we obtain Sis. Gubser and A.107a) In this result* we now have to insert the expansion of the parameter v given by Eq. At = 2i[l + 0(h2)}.577.61n2 .106) and evidently violates the unitarity of S. 2 /l4 '•i Al=0 2[(l + 1 / 2 ) 2 .2 The Potential 1/r4 — Case of Small h2 parameters v = n — (3.Q.86) and obtain in leading order for I — 0: Ai = *[l + 0(h% Ar = ±[l + 0(h2)]. one can calculate the next term of the expansion so t h a t At=Q = 4 ^ 4h4 {7 . "This is the result effectively contained in R. i. (19.1 ] Oih* For S waves this implies an absorptivity given by ((—1/2)! = 1/7F) (19. Magnus and F. Oberhettinger [181].19. W.
Wannier [278]. H. (19.460 CHAPTER 19. Park. Miiller and N. H.h) (19.3. we again make use of the replacement (17. completely different solutions. N. This is therefore a highly interesting case which found its complete solution only recently.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. 2 2 a = [l + 2 (19.1. K. Aly.3 The function q(l.108) For the large values of h2 that we wish to consider here. We are again concerned with the associated Mathieu equation given by Eq. i. W. h) for large h and upwards I = 0.e.4 Fig. where the part of interest here is based on H. S. VahediFaridi [8] and G.4) which — for some distinction from the smallh2 treatment — we rewrite here as 2 + [2h cosh 2z . MiillerKirsten and Jianzu Zhang [226].26) and hence set a1 = 2hl + 2hq + A{q. 19. Singular Potentials 19. 2. J.109) ^We follow here mainly D.$ q(l. W.h) 10. . Tamaryan.3 19.a }tp = 0.H. J. H.3. 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.
to which we have to return in this subsection. i.y^] = ab\f\ is the Wronskian (in leading order) of the solutions which are even and odd about z = 0 respectively. with an essential mathematical step.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 17.113) 'We are considering nonintegral values of v.yJ\ — 1 and cos TTV .21b). (17. we defined the Floquet exponent v and observed with Eq.109) in this way enables us to obtain asymptotic expansions of solutions very analogous to those of the periodic case. (17.112) In Chapter 17 the unnormalized even and odd large/i2 solutions of the Mathieu equation were given by Eq. We begin. (19. 19.63).3.e. (19.h*y_(l. since it is clear that large h2 considerations require a knowledge of the large/i2 behaviour of the Floquet exponent v. J4(0)=0. y'_(0) = 1.h*y (19.§ a W[y+.5) that this is given by the relation cos iris = ^ M .e. We also found the boundary condition (17. y± oc A{z)e2h sm * ± A(z)e2n sm z . (19.21b). (17.3 The Potential 1/r4 — Case of Large h2 461 where q is a parameter to be determined as the solution of this equation. y(0) = 0.110) a where y+(7r) is the solution evaluated at z = TV.111) we have W[y+.l + 2y+(l. and a is its normalization constant. which is even about z = 0. The behaviour of q as a function of h for some values of / is shown in Fig. Therefore the second term on the right is nonzero. i.y} where VF[y+. and A / 8 is the remainder of the laigeh2 expansion as determined perturbatively.3. The replacement (19.19. With the boundary conditions y+(0) = l. 19. which therefore cancels out or can be taken to be 1. however. See also Eq. .
K [ (z)}2hsinz (19. which is \/2 at z = 0. (19. B.111) we have^ iVo = 2 3 /2 1 + °{k and K = ¥frh l +O (19. (17. W.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 . Miiller [73]. Kin B y± oc y+(z) = N0 A(z)e2h A(z)e2h sin z + A{z)e~2h . and hence with Eqs. Singular Potentials and their extension to around z — TT/2 by the relations C[w{z)\ ^_2h sir Mz)}„2h. (19.114) a a Our first step is therefore the determination of the normalization constants No. Dingle and H. B.462 CHAPTER 19. This paper contains the normalization constants of large/i 2 periodic Mathieu functions.39 2 14 /i 2 + (19. J. Hence we require (obtained from Eq.32). Miiller [73].112) from evaluation of Eqs.115) Setting z = 0 we obtain in leading order y+(0) = 2iV0A(0) and y'_(Q) = 4hN0A{0).e. (17. (19.A(z)e2h sin 2 sin z sin z (19. i.118a) "We emphasize again: For integral values of v the normalization constants are different.116) (in the first expression we have A(z) ~ Aq(z) with Aq(z) given by Eq. J. We now obtain the expressions needed in Eq. See R. in the second expression 2h is obtained in addition from differentiation of the exponential factor).43) we see that z = 7r/2 implies W(TT/2) = 0.117) a a Referring back to Eq. (17. Dingle and H. . 1 1 Or see R. 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 . (19. W.114) at z = 7r/2.
(19. (17.8 g + 3) +0 ^2 (8h)^[~l(q + l)]\[l(q + 3)}\ 2eh ' ~ (19. (19.117) are known from the matching of solutions in Eqs. .! ) ] ! [ .119a) This formula — derived and rediscovered in Park. p. (19. Thus in particular (8h) i(?i) a (3q22q + S) 27h + l\(qW " ' Inserting these various expressions now into Eq.62b) (see also the explicit calculation in Example 17.118b) The factors a.2g + 3) 27/i ( 3 g 2 .87 2 14 /i 2 [ i ( g . Tamaryan. (19.119b) COS TTV + l ire This result gives the leading contribution of the lavgeh2 expansion determining the Floquet exponent v.h .3 The Potential 1/r4 — Case of Large h2 and 463 as dz )z=n/2 dw (dw\ o\dzJz=n/2 1C%2 .112) we obtain COS TTU + 1 7T '2Wa[\(ql)]l[\(q+l)]\ 02h 1 ¥h + 2TT ' 2^ha [\(q .[_ l ( g 2h and hence 5 + 3)]! 'All Ah 1+2 (3g2 . a in Eq. With the help of the duplication and inversion formulas of factorials the result can be reexpressed as cos( g 7r/2)[£(gl)]! .! ) ] ! [ .\ { q + 3)]! 7re 4/i 24/i + (2/i 1 /2) ( 8 / l ) 9 / 2 [ _ l ( g + !)]. 210.1Q11QM cos irv + 1 ~ •==— e .i ( g +3)]! 1 24/i {2hl>2).62a) and (17. Presumably they extracted it from "between the lines" of a sophisticated paper of Langer's [159] which they cite together with others.3). MiillerKirsten and Zhang [226] — is cited without proof by Meixner and Schafke [193].19.
z). z) = ip(q.109) remaining unchanged. h. i = 1.z) or as tp(q.3 Following the procedure of Sec.120) The resulting equation for the function A(q. of course.464 CHAPTER 19. for h — oo). We define these solutions in terms of the function Ke(q. the expression (19.e.? s m h z) Vcosh z (19. h. (19. h. —h. z) are TP(q.h. it is still a solution but not with the symmetry property ip(q. h. h.h. In the following we require solutions H e ^ ( z ) .3.122) that given one solution ip(q. .4.123) We again make the important observation by looking at Eqs. Singular Potentials C o n s t r u c t i o n of largefa 2 solutions 19. we can obtain the linearly independent one either as ip(—q.z) = Aq(z)exp[±2hiswh z] ^~°° exp ±z/te ( ')eT^/4> V cosh z e x = Aq(z) exp[±2hi sinh z] ** ~~°° ^ h e ^ V cosh z e ^q/\ (19. h.3.3.z) i/>(q. (19.z) := = exp[i7rg/4] j===^Aq{z) exp[2/wsinh z i(77r/4 V2ih k(q. dA 1. With the solutions as they stand. z). .120) to (19. z) = ip(—q. h(q.h.109) into the equation (19.h.h.124) Since this function differs from a solution ip by a factor k(q.121) We let Aq(z) be the solution of this equation when the right hand side is replaced by zero (i.122) Correspondingly the various solutions ip(q. h. z) can then be written as . Vcosh z \ 1 . —z). 1 / l + isinhz\T9/45R^oo e^ / Ag(z) = ===[r^— ~ . V— 2in (19.s i n n z ± iq)A = ± K qj dz 2 Aih d2A dz2 (19.108) and set ip{q.h.2.z).2 we insert the expression (19. each with a specific asymptotic behaviour. Then straightforward integration yields > inq 4 „. 17.h)iP(q.z) = A(q.h) =?==. —z). . „ 1 A „ —A cosh z——H . —h.. tp(q.z)exp[±2hismh z\.h).
3 The Potential 1/r4 — Case of Large h2 465 Instead.h2).19. 2j<. 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. we have to match a solution valid at Kz = — oo to a combination of solutions valid at $tz = oo.<5±1) (19 125)  (the expression on the far right in leading order for h2 large). Ju(z) TTZ VK 7T cos z T~ 4 1 + 0 ^V. Me„(z. (19.h2) = OvMPfrh2).e. * ** = > rojKe( "./i 2 ) M (19. 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 . >• * F $ 0 * e".e. . We observed that these satisfy the relation (19. (44) of W a n n i e r . [ \ .e. av{h2) = ^(0J ^ Mi i . or — equivalently — by the appropriate expansion of the Bessel functions as given in Tables of Functions.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]. : 7r/2.fc 2lqk A + 1/2 + 4fc2 c o s h 2 77 A / 8 P . (0. 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].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 . in performing this cycle of replacements the function picks up a factor. This is. (19./h).^ s i n h 7?)]g. we h a v e ( u s i n g cosh 2rj = 2 c o s h 2 77 — 1) ry Jo ' = dr. the Bessel functions contained in the expansion of the associated Mathieu function M{. 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.fi2) can be reexpressed in terms of Hankel functions. as we saw earlier. i. / 2i<jfc + A / 8 \ ~ 2fc / drj cosh 77 1 + * _' Jo V 8fc2 c o s h 2 77 7 2fcsinh y + [ t a n . 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 .126) For large values of the argument 2/icosh 2.127) sin z T /'1 M TT\ — <>{ 4 . we h a v e 77 = 0—> 6 = 0. achieved with the help of the Floquet solutions Me±v(z. (z. . i.45). One can show that the quantity < o of Wannier* (see above) is related to q by $o = & iqir/2 + OO. i.
In this way we obtain in the domain .h) = e(z) exp ihez .q. 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.128) We now define the following set of solutions of the associated Mathieu equation in terms of the function Ke(q.q.h) = e(z)exp exp ihe^ 4 + i\ 12 i4 3^cO. 3te»0.130a). 3te»0.130b) r_0 rll2exp\g/r (ig) ' 19. (2) (19. HeW(z. exp[—ikr — in/4\ kr Re{2)(z.h).q.q.4 T h e c o n n e c t i o n formulas We now return to Eq.h.g./i).h) (3) = = = = Ke(q.h) He (z. He (z.h) = e{z) exp ihez + iexp[ikr + i7r/4] kr 7T .eW(z.466 CHAPTER 19.q.h )~ \ ot±v cos(2/icosh z = vir/2 — 7r/4) F V2/tcosh z (19.129) The solutions so defined have the following asymptotic behaviour (where e(z) = (2/icosh z)~ll2 and h2 = ikg): H. h. (19. z): Re(2\z.3.130a) Re^(z.130b). (19.h). KeW(z.g. (19./i) Be^(z.z).128) and consider the cosine there as composed of two exponentials whose asymptotic behaviour we identify with that of solutions of Eqs. (19.h) Be^(z.q.q.q. (19.vr .
* ^ / 2 H e ( 3 ) ( Z ) g? Mev(z. Eqs.130b)). e W 2 H e ( 3 ) ( ^ ? j ^ _ e M/*/2jj e (4) ( ^ q j ^ e .130a). q.135) sin irv 'This means.« * r / 2 H e ( 2 ) ( ^ g> ^ ft) _ e W 2 H e ( 2 ) ^ ^ ^ = a_ 2TT e . g .(z. and He(*> exponentially decreasing. q.i ^ / 2 H e ( l ) (z> ?) (19. h.(z.h2) a. Changing the sign of z.q.q. / i ) + sin7r(7^)He(4)(2.134) Considering now the first of relations (19.h2) = 2vr i a_ 2TT />) _ e ^ / 2 H e ( 4 ) ( Z j ^ Q (19.3 The Potential 1/r 4 — Case of Large h2 fc>0 the relations: = 2TT 467 Msv{z.g.h) is asymptotically small (cf./i) — sin 7T7 Re^(z.19.li 2 ) Oil. — sin 777 H e ( 3 ) ( 2 . h) — sin 7T7 He^(z. we obtain in the domain tfcz <C 0 the relations: Me. (19. (19. h) = sin 7r(7 + v) He (1) (z./i). ./i).129)) He (2) (z. z) and He (3) (z. and thus can be matched in this region. z) are proportional there. these are exponentially increasing there.g. q.131) where the second relation was obtained by changing the sign of v in the first. h) = Ke(q.h). h. (19. we see that the solutions (compare with Eqs. q.134) in the region where the function He^(z. (19.133) In a similar way one obtains the connection formulas — sin nv HeW(z. h) = Ke(q.h) sin irv = sin7r(7 + ^)He (3) (2./i 2 ) = Me_j. t With Eq.q. (19.131) and eliminating He") and setting again exp[m7] := we obtain the first connection formula ozwr — sin irv He (4) (z. (19. e x ^ r / 2 H e ( l ) ( 2 > g> />) _ e .132) These relations are now valid over the entire range of z.132) into Eqs.g.h2) Me„(z. (19.125) the proportionality factor can be seen to imply the relation exp •ffa + 1) sin7r(7 + "} (19. (19. Substituting the Eqs.
Squaring Eq.119a) and (19.e'i7ri 1. sin TXJ.133)) — sin TXV He^4' (z. Wannier did not have the relation (19.l ) (l„—ikr e From this we can deduce that Si (19.m a t r i x From Eq.138) TXVICOS TXV We obtain the behaviour of the Floquet exponent for large values of h2 from Eqs.( . H.1 . which in our case is the solution He^4^.3. h) ~ — sin TXV (1) sin 7r(7 + v)e kr ikr in 4 rl/2eg/r+m/A / (ig)1/2 ( . (60) in the work of G. (19.468 CHAPTER 19.e. (19. The quantity 7 is now to be determined from Eq. and therefore remarks after his Eq. Singular Potentials The factor on the left.l ) ' s i n 7 T 7 ^VK /2 ikr sin 7r(7 + v) _l)leikr i5i„—ilir/2 SxelKr . exp 2^(9 + 1) and solving for sin TXJ one obtains the expression sm 7T7 sm TXV = ~iei7rq/2sin le iqK • COS 7T7. Thus here the S'matrix is defined by (using Eq.133) we can now deduce the 5matrix Si — e2t51.137) We see that this expression agrees with that of Eq.(19.119b) as „4/i 1 + COS TXV ~ rrl ( g . (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". of course.136) ^ jrfl+l/2) (19:I35) sin 7r(7 + v) sm WKljrtla/2) S m TXV (19. .119b). (19.119a) or (19. is only the dominant term for large h2} 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 .l ) ] ! c o s ( W 2 ) 2 2 (8h)i/ 2TT29/ 2 +0 (19. cos TXV + sin TXV. Wannier [278]. q. (19. (19.135).135).135) is the equation corresponding to Eq.Sin TXV. (19.5 D e r i v a t i o n of t h e ^ . i.85b) obtained earlier.139) ''Equation (19. where 8i is the phase shift.
i.141) COS TTU irvj — i s i n WR s i n h TTUJ.109). as a comparison with Eq. 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 (19.139) could actually be neglected) implying  cos m>\ = cosh nvi. (19.19.140) From this we obtain the absorptivity A(l. This is different from the behaviour for small values of \h2\.138) we obtain. . the real part of v must be an integer.4 The absorptivity of the S wave (attractive case).28) shows. (19. (19. irrespective of what the value of I is) by cos™ + l = c o s ( T j^J y/e1. h) of the Zth partial wave.e. (19. we can approximate the equation for q ~ h (i. 19. T h i s is r e a l for UR a n integer. 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.e. for the attractive potential. A{l.§ 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. since cos TVU is large.e.8h COS From Eq.137) together with (19. 9/2 (91) with the approximation q ~ h obtainable from Eq. imaginary part uj (so that 1 on the left hand side of Eq. i.3 The Potential 1/r 4 — Case of Large h2 469 Since the right hand side is real for real h.
S . 19. Partial or other aspects of the weak coupling (i. and. one may expect corresponding results for other singular power potentials. C. **A. Limic [177]. N. Apart from the papers already referred to. Rosendorf [225].^ A highly mathematical study of the potential ~ r~2 as an emitting or absorbing centre has been given recently. Jones [78]. Tiktopoulos and S.107b) and (19. W Some of these sources can be helpful in further investigations. B. Singular Potentials with near asymptotic behaviour (for h2 — oo) > Afi . "G.470 CHAPTER 19. Pope and T.4 (there are tiny fluctuations in the rapid approach to 1). 11 J.e.139) 2Tr(16h)q /32\ f e Thus with Eqs. Eden [46]. Lii and J. Treiman [271]. **R.** 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. VazquezPoritz [61].k) in the approach to unity are too small to become evident here. Bertocchi. Apparently this is one of the very rare cases which permits such complete treatment. YuanBen [285]. N. such as phase shifts. The diminishing fluctuations of A(l.4 Concluding Remarks In the above we have considered the solution of the Schrodinger equation for the potential r . Cvetic. Fubini and G. (19. Shabad [248]. small h2) case have been considered by some other authors. 19." as well as other aspects. F. J. I r a n [61] and M. 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. In essence. Masson [191]. Lew [128]. L. D. S. Esposito [88]. Handelsman. Challifour and R. Dombey and R. . Pao and J.142) we can see the behaviour of the S'wave absorptivity as a function of h. This is sketched schematically in Fig. (19. Furlan [20]. A. in fact. Cvetic. E. 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. H. N. Paliov and S. see also M. D.4 .P. and G.' Rudimentary aspects of a singular potential with a general power n have also been considered previously. A. H. Lii. in further contexts beyond those already mentioned. Y. H. A.
Chapter 20 Large Order Behaviour of Perturbation Expansions 20. B.* In fact. e. B. Muller [74].g.* However. W.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. Wu [18]. This is indeed a remarkable connection. We concentrate in this chapter on the large order behaviour of asymptotic *C. M. Dingle and H. [19]. 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. T. Bender and T. how the large order behaviour of the expansion of the eigenvalue is related to the discontinuity across the latter's cut. J. f R . the cosine potential is more suitable for such studies than the anharmonic oscillator since its case is simpler. *R. the asymptotic solutions in different domains all have the same coefficients. 471 . 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. also convergent expansions are known and a lot of literature exists on the equation. Dingle [71]. 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. We shall also see explicitly.
2 and below. as evident from Example 17. 8.2 The function ur = r\/pr vs. It is a feature of a physicist's approach to a problem that he employs a method of approximation which is such that the first term of the corresponding expansion yields a rough value of the desired answer. Large Order Behaviour of Perturbation Expansions expansions using in particular the recurrence relation of the perturbation coefficients obtainable with the perturbation method of Sec.7.6. Presumably this has never been done so far.1 and cases in Sec. This approach leads . 20. 8. 20. Fig. Since the method is also applicable to convergent expansions.8. r for p = 4.2.6.472 CHAPTER 20. one could naturally also explore the large order behaviour of these in a similar way and expect the behaviour discussed in Sec. 5. r for p = 4. 10 12 14 16 18 20 Fig.1 The function ur = pr jr\ vs. 19.2.
in the power expansion of the exponential function). as we discussed earlier in Chapter 8. If the function E(g2) can be written as the Laplace transform E{g2)= dze'z/g F{z)=g2 dte^FigH).2) or. The behaviour of such terms is illustrated in Figs. In some cases the answer is affirmative.20.1 and 20. i. equivalently for g = l/h.e. It can be seen from these. 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 formal perturbation expansion of a quantity E(g ) .1 Introductory Remarks 473 automatically to asymptotic series.3) n=0 Thus for any given value of g . The traditional method of using an asymptotic expansion is to truncate the series at the least term.2. .J2 2n h E (20. for instance.4) Jo Jo its Laplace transform F(z) is called the Borel transform of E. and the integral E(g2) is called the Borel sum. 20. i. the function E(g2) is given only approximately by Yln=o En92n. (20.1) is an asymptotic series if for any N lim 92^o 1 g 2N N E(g )^2En92n 71=0 2 = 0 (20. even if each En is known. The question arises: What is the information about F(z) that can be obtained from the asymptotic expansion of E. For an authoritative discussion of these aspects we refer to Dingle [70]. this term represents the size of the error. and the expansion is said to be Borel summable. T V lim h2N 2 n EQ?) . E(g2) = f2 Eng2n = Eo + J2 n=0 n=0 E 2(ra+l) ^9 (20. 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. whereas in the case of the asymptotic expansion it is the opposite: a factorial in r divided by a power. 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.e.
In any case. . (and E0 = 0). n=0 (20. We observe that a function E(h) with asymptotic series ^ r=\ A has an essential singularity at h — 0. . many asymptotic series do not even alternate in sign (as in the case of the series in Eq. Q. . (20.1)? Asymptotic expansions originate from integrating a series (i. In fact.474 CHAPTER 20. J we obtain Now if />oo dteHz\ °° oo ' 0 dteH*\ TEn+1g^V . In general this is not so easy. (20. that of F(z)) over a domain that is larger than its region of convergence (see Chapter 8 and Dingle [70]). (20. n=0 (20.4).n E(9 )~ 2 n=0 ^ E0 +1) = Ea^2( °° oo /»oo = J2ang2^ „n 'O n=Q dte~H\ a_i = E0.n _ En+\ n! En+1 ^ (!)"«!. 1 we obtain oo F(z)£(*)" = — . Large Order Behaviour of Perturbation Expansions Eq. . In this ideal case the integral of Eq.5) then with the integral representation of the gamma function F(z) = (z — 1)!. El .6)). 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. Thus if CO F(z) = aiS(z) + J2 anzn. (20. We can therefore write down a dispersion relation representation choosing the cut from 0 to +oo: . .6) The power series of F(z) converges inside the unit circle about the origin at z = 0. 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. .e.4) can be evaluated and E(g2) thus recovered from the asymptotic expansion by Borel summation.
3 and g < 0. (20.. This is the case if the real potential has a hump of some kind.. and E represents an eigenenergy.. (20. .11) V.20. § where 7(ff) = To exp  . e.9) The question is therefore: How can we determine 9 E(h). (20. we have a quantum mechanical problem where probability can leak away.2../ \p\dr\.N = l.. Popov and V.. M. $s E can be calculated by the WKB procedure.10) r as shown in Fig. if the potential or a boundary condition is complex such that the problem is nonselfadjoint. if the potential is V(r) = +grN. N = l.1 Introductory Remarks 475 except for possible subtractions..2. The potential (20.g. Weinberg [228]...10) is considered in detail in this work.3 The potential V(r) gr». 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. 20. i. and with g = 1/h and it is found that§ En = 4 n ) ~ \lis). S.e. 20. ? When ?s E ^ 0. or can we determine Ar independently? Fig. for < ^ 0.
9) for Ar. as we expect on the basis of Eq.^ The logarithmic factor in Er is a novel feature of this case.11) is. An interesting application is provided by Example 20.2 Cosine Potential: Large Order Behaviour It is instructive to investigate various methods for obtaining the large order behaviour of an asymptotic expansion. since this is effectively the Mathieu equation for which — for comparison purposes — extensive literat u r e was already available. Equation (20.76) (see also the comment on the comparison with the Kepler problem before Eq. of course. as was (and still is) not the case for t h e equation 1 V . (20.1. Thus in the following we apply these methods to our typical examples. L. V. the BreitWigner formula (10.11). S. However. the cosine potential being t h e most completely investigated will also be seen in the following. the eigenvalue expansions of the cosine potential and of anharmonic oscillators.73) in the W K B or semiclassical approximation. Eletsky. Popov and Weinberg. the perturbation method of Sec.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. Example 20. 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 . M. Weinberg [83].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. (14. Evaluating the integral show that 00 r=0 v ' where n is the principal quantum number of the Coulomb problem as in Eq. Popov and V.476 CHAPTER 20. Here n is the principal quantum number of the Coulomb problem (see Eq. . (11.134)). 8. (11. (20.113)). Solution: The solution can be found in a paper of Eletsky. 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. Straightforward RayleighSchrodinger perturbation theory usually does not enable this in view of the unwieldy form of coefficients after very few iterations. 20. In the first such investigation it was sensible to consider the Schrodinger equation with the cosine potential.
h) = 2h2 + 2hq+j.14) Thus these coefficients can be obtained once the coefficients Pi(q. 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. This has been done" and it was found that (we cite the results here since we rederive them below in detail by other methods. obey the recurrence relation jPi(q.J) = (g + 4 j .2 Cosine Potential: Large Order Behaviour 477 with an anharmonic oscillator potential. The recurrence relation (20. Eqs.Bq. (17.61)) that the coefficients M2i of the expansion of the eigenvalue A = E(q. Thus in the case of the Schrodinger equation with periodic cosine potential we saw (cf.l(q. M2i = 2j2 J[Pi(q. 3)? . j .l ) / 2 ] ! i![(gl)/2]! • We have seen earlier (cf. Eqs. (20.13) represents effectively a partial difference equation. are given by i .l ) ( g + 4 j .13) In the special case j = i the boundary conditions stated after Eq.j + 1).12) (20.j) a n d M2i can be deduced.Cq.58) and (17. (17.34)) that the coefficients Pi(q. (cf. Dingle and H. J.2 A = Mx + ~M2 + 7^WM3 + •• • . % = Aq. W.l ) +2[(q + Aj)2 + l + A]Pl. Miiller [74]. (20. (17. With some simplifications for large values of i this equation can be solved approximately and the large order behaviour of the coefficients Pi(q. (17.j) are known.3 ) P i _ 1 ( g . i. B.j) +(q + 4j + l)(q + 4j + 3)i^_ x (g.M2i+1 for large i.58) eliminate the last two contributions and one obtains the exact expression ^ ' Z j _ 4<[2» + ( g . (17.j) of functions ipq+4j. arising in the ith order of the perturbation expansion of the wave function oo 1 i ip — const.e.26) and (17.60). or equivalently of the quantity A.20.55)) X = E(q. . Eqs. h).
19) ^nOW =2(9"1) We observe that successive terms do not alternate in sign.18. Hence exact Borel summation is not possible.•91 {[l(? + i)]![(? + 3)]!}2 (20. see also below) .17) We see that assuming i is very large compared with m is here equivalent to assuming m is approximately zero.th term iW~^q 1 {[i(gl)]!} 2 2 7 r(8/ l ) i . (20. Large Order Behaviour of Perturbation Expansions where the bracketed expression at the end is the binomial coefficient. In the first method we do this by first obtaining the imaginary part of the eigenvalue with the help of nonselfadjoint boundary conditions.l (20.15b) 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.l ( 9 + l)]![(Q + 3)]!}2 i!^" 1 1 or (with the help of the duplication formula.*.1 22 the result becomes i t h term (i + 2n)!2 2 " 1 where n 1 (20. In the following we concentrate on the cosine potential and rederive the large order behaviour obtained above by other methods.478 CHAPTER 20. .91 l)]![i(g3)]!} 2 { [ . also written lCj.16a) (20. 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±^.16b) With the help of Stirling's formula one can derive the following relation: i large 1 (i + m)\ \o (20.
The potential we consider here then has the shape shown in Fig. A means to achieve this was found by Stone and Reeve.2 solutions obtained earlier. 20. f H . . \ 1 \ \ \ \ \ 0.3 Cosine Potential: Complex Eigenvalues 479 20. J. t To enable comparison with the work of Stone and Reeve we shift the cosine potential considered previously by 7r/2. we follow here a formulation in the context of our earlier considerations of the Mathieu equation. Stone and J. .5 A i 1 1 r f \ \ M\ \ \ / ! \ \ 1 \ \ I 3 \ \\ \ /41 i \ \ 0. MiillerKirsten [200]. What changes as a result of different boundary conditions is the deviation of q from the odd integer qo. i. 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. W.4 The cosine potential shifted by TT/2. Reeve [262]. From our previous considerations it is clear that the formal perturbation expansion of the eigenvalue X(q) in terms of the parameter g. 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/?. 2 . the difference q — q$. but with different boundary conditions.e. this difference was found to be real.4. . At the risk of repeating what we explained earlier. Previously. 20. . l .* However.3. in the calculation of the level splitting as a result of tunneling.20. but now change the boundary condition at the neighbouring minimum in such a way that the difference q — qo becomes imaginary. remains unchanged. We retain the previous boundary condition at one minimum. This is a very instructive exercise which shows how the eigenvalues change with a change in boundary conditions. . n = 0 .3 20.5 \ \ / \A Fig. we emphasize a few points *M.
21) <u> + ( i « . a boundary condition which at the face looks somewhat abstract. but achieves exactly what we want. which means replacing z by z±ir/2.T K ^ a D_i_{q+1){±iw).480 CHAPTER 20. Now we change this situation at the minimum on the right. We can distort the potential there or alternatively impose a different behaviour there on the wave function.e. Effectively we consider the tunneling from the left well in Fig.4. Around z ~ 0 w e can expand and obtain i)"{z) + [A + 2h2 . The physical wave functions there are therefore the solutions which are real and exponentially decreasing. and thus obtain ip"{z) + [A + 2h2 cos 1z\i>{z) = 0 (20. To be more precise we recall the Mathieu equation we had earlier: ip"{z) + [A . Setting w = 2h1/2z. we shift the argument by n/2. but t h a t with complex argument there. Large Order Behaviour of Perturbation Expansions for reasons of clarity (since some considerations here may otherwise not be sufficiently clear). 20. Here we choose t h e second way.Ah2z2 • • • }^{z) = 0.4 (around z — 0) to the one on the right (around z ~ ir). Thus one chooses not the solution with real argument at the minimum.20a) As stated. (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. . 20. X + 2h2 = 2hq+—. 8 the equation can be approximated by d2ip(w) (20. and allows its imaginary part to tend to infinity. (20. i.20b) with potential V(z) = — 2h2 cos 2z as shown in Fig. the degeneracy of eigenvalues of neighbouring oscillators being lifted by the finite height of the barrier separating them. 20. as we found these for this potential.4. Thus around z = 0 the potential behaves like t h a t of an oscillator well with minimum there. and correspondingly we assume alternately even and odd oscillator eigenfunctions there.2h2 cos 2z]ip{z) = 0. Previously this situation was the same at the neighbouring minimum.
4. (20. (20. . 17. we naturally constructed the even and odd solutions with these (cf. that at z = 7r. 20. Thus we write the boundary condition at Kz = ir: tp(±iw) — 0 for > w — 2h}/2z K z— > i. by a different method. Thus.3 Cosine Potential: Complex Eigenvalues 481 As explained above. (where for the ground state q = go — 1) D_i{q+1)(±i2h1/2z) ' ± zoo (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. . considering around z = 0 the ground state of the harmonic oscillator for which go = 15 and replacing go by g. .> 0 for z > ±ioo. whereas here we construct these from oscillatortype solutions.e. 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 = 1.22). we see that ips ~ cos \j2qhz for z ~ 0. This somewhat abstract looking boundary condition is effectively simply one imposed on a complex solution of the equation giving the oscillator eigenfunctions. A (which we can loosely dub solutions of WKBtype). 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.63). i.e.23) . whereas here we start off with a solution even or odd about a minimum. . 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. we now choose an harmonicoscillator type of real solution around the left minimum of Fig.20. 3 . 2. however.24) Considering t m O i n Eq. the even or symmetric solution about z = 0 can be taken as *• = liDfrriVhWz) +Dh{q_qo)(2h^z)}. A derivation is given subsequently. Eq. such that this deviation becomes imaginary. We consider here explicitly only the case of the ground state around z = 0. so that the deviation q — qo results from the finite height of the barrier taken into account with boundary conditions at the next minimum. . Since the large/i2 solutions valid away from the minima were the solutions of types A.
30) 'See e. 2.•w) + 2ir T{u) <"+!)% Dvi(iw). p.we obtain the following circuit formula of parabolic cylinder functions Dv(w): Du{w) T DV{. tp(v). . p. Large Order Behaviour of Perturbation Expansions Although the two parabolic cylinder functions in Eq. Magnus and F. The differentiated function Y{y) is in Tables of Functions expressed in terms of the socalled psi function.25) qo) E2 0: (20. Oberhettinger [181]. 92. (20.w) = e T Dv{w) 2TT n») AvViDv^iw). We see this as follows.482 CHAPTER 20. (20.28) Extracting Du{—w) we have A.27) The differentiation of the parabolic cylinder functions with respect to their indices is nontrivial. As before we set A = 2hz + 2hqA where in the present case Ei = 2hz + 2hq0 + — and A & = Ex + E2 (20. 17.27) requires the differentiation of this expression with respect to the index v.26) Thus here we can approximate ips by expanding it about E2 i)s~DQ{2hll2z) + E2 ~{DE2/4h(2h ^ z) 1 2 + DE2/. 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.24) are linearly independent around z = 0 for q ^ qo. From Tables of Special Functions'. and then evaluation at u = 0. Oberhettinger [181]. (20. we know from the solutions of types B and C in Sec.h(2h1/2 J £2=0 (20. Thus the only nonvanishing contribution comes from differentiation of Y[—v). W.g.3. W.29) Equation (20.(. §See e. or logarithmic derivative of T(^) for which:§ ^(1/) r» (20. (20.g. this can be handled in a few lines. The arguments of the functions appearing in Eq. Magnus and F. However. Setting v = 0 in T(—v) = (—v— 1)! gives infinity.2 that one parabolic cylinder function is exponentially increasing in h and the other exponentially decreasing.27) differ by an angle IT.
.27) exponentially decreasing contributions. .+ v ^ v k^x k ^ + ^ where C = Euler's constant = 0 .3 Cosine Potential: Complex Eigenvalues From formulas given in the literature we deduce the limiting behaviour^ 1.e. . pp. 3 .20.//n and ij>{y) = C . (20. (20. i. which are valid around z ~ 7r/2. Magnus and F.. Reeve [262]. 5 7 7 . For u — 0 one obtains the result (20. . Oberhettinger [181].e.+w2/i IW Inserting this result into ips and retaining only the dominant contributions (i.)e" w /4 2 l ^Dx(iw) (20.4. Thus calculations for qo ^ 1 are not given by M. with the following expression for the largez behaviour'! (here v = E2/AK) Dv{w) ~ wue~w2/4 = evlriWe~w / 4 for  arg w\ < ^TT. Differentiation of this equation and subsequently setting v — 0 yields. 92. including the derivative of DE2ufl(2h1/2z)). Magnus and F.29) v=0 (in + In w)D0(w) + v2ire (m + lnu. (20.34) This is the behaviour of the symmetric ground state solution away from the origin towards the limit of its domain of validity. We now go to the WKBtype solutions. Stone and J.31). and "Prom W. those of type A.31) dT(u) r*(v) dv 1 i/>(v) ^ o l. This result is » not easily generalized to <jo > 1. we obtain the formulas (X) / \ 1 OO I » J ] (l+)e.32) which we can use here since all other problems have been resolved: d D {w) dv v (20.29). i/^0 483 T(v) It follows that d dv [T(u)\ (20. "See e. Oberhettinger [181]. W.g. r(i/) We now return to Eq. dropping from the derivative part of Eq. p. we obtain 4>s D0(2h^z) D0(2h1/2. 1E2 + ± 0(e~hz2) 2 Ah Ihir ehz* E2V'< 16/i2 V2^ehz2~ 2hM2z (20.33) = IV2~TT.
with expansion of cos z in the exponential) D0{2hll2z) = e~hz ~ e~2h / „2/i cos z \ — ). 2 c2e2hehz UJ2 . To this end we set z' = z — ix. (20. We observe that in the matching domain (i.29) and (17. We now return to the WKBlike solution (20.40) From Magnus and Oberhettinger [181] (p.e. (17. Recalling that we have shifted the potential by 7r/2.. 92) we take the formula: Dv{w) = T { y ^ \eiv^l2D^v^{iw) + e'^D^iw)] . that in approaching z = ir.40). cos z/2 ~ (20.cos z — 2/icos z ipA^ Tcos  and i n A —.38) in its limiting domain towards 2 = 0. near z = n the boundary condition (20.34). y COS c\ J (20.23) demands (with constant A) ip = AD_1(±i2h1/2z') > 0 for z' + ±ioo. (20.35) We construct with constants c\. (20. (20.sm I We now want to match this solution to that of Eq.41) 2TT . c2 the WKBlike linear combination e2hcos ^ W K B = ClipA + C2lpA ~ Ci z _—2/icos z — + C2 rT (20. we obtain the required solutions from the earlier ones. 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 (20.36) and consider this at the other end of its domain of validity. Large Order Behaviour of Perturbation Expansions extend these to their limit of validity to the left.e. by there replacing z by z + ir/2. i.36) cos  .484 CHAPTER 20.34) and (20. (20. where we can then match these to the solution ips. Thus in the present case and with q ~ qo = 1 we obtain „2ft.36) we can identify the dominant large/i2 behaviour of the constant C\ as c\ = e~2h.— 721 . Then V'WKB becomes (with cos z ~ — 1 + z' /2.36) V'WKB cie~2hehz' 9h  . i. x .e. Eqs.37) Comparing Eqs.39) Now. smf (20.
38) can be written TAWKB (20.complex conjugate] (2Q42) (2032) y /  Z e _ ^ + 2(. ^'mliw <2a47) For qo = 1 we recover the result (20. (20.34) we obtain E2 = i%\ = ±ih22Qe~8h.43) we obtain. (20. J Hence Eq.45) Comparing this result with tps of Eq.40) can be written ^ ~ AJ^e~hz'2 + A~(i2hl^2zTlehz'2.e. (20. i. (20. (20.e. with c\ = exp(—2h) from above. 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 .44) = DQ{2h / z) ± (2h7r) / e. .1/V)_le^_ Z With this result Eq. for both even and odd harmonic oscillator wave functions around z — 0 with qo > 1) is nontrivial.39) and (20.^. We derive the general result in the next subsection using our own method. A = ±i4/l1/2e4h ^ t c2 = T i2(2/ivr) 1 /2 e 6/ l .46) Repeating these calculations for the general case (i.46).3 Cosine Potential: Complex Eigenvalues Thus for v = 0 we have DQ(w) = —j=[D\(iw) V 27T 485 + Di(iw)] = real. 1 2 4 1 2 8h ^^2 2 (20.20. The general result stated by Stone and Reeve [262] is therefore presumably guessed. (20.43) Comparing now Eqs.2/.
e. we impose the usual type of boundary conditions as for the minimum of a simple harmonic oscillator there. Considering the original equation. (20.2 CHAPTER 20. the result is formula (334) there. V(z) . 20. •_(:)*<>. (20.z. 20. MiillerKirsten and A. Achuthan. 20.5 The cosine potential from — TT/2 to n/2. J. this equation may be approximated there as d2ip ~dz^ + E + 2h2 . H. On the other side of the barrier at z = —TT/2 we consider different boundary conditions. i. Eq.) + 7T V> = o. harmonic osc 71/2 E>V / E<V z+ O \ E>V TC/2 \ \ .E curre it \ j. Wiedemann [4].Ah2 ( z + . .*. (20. Thus in the immediate neighbourhood of z behaves as 2 1 2 ip ~ e ±i(E+2h ) / z ~ e 7r/2 the wave function i^{z) ±i(2hq)1/2z **We follow P. W. i. i.46) we consider a half period of the cosine potential as illustrated in Fig.48) These conditions provide the required quantum number qo = 2n + 1.e.486 20. Fig.5.20a).3. at z = TT/2. We have to impose two sets of boundary conditions.()#0. and expanding cos 2z around the point z = — TT/2.e. Large Order Behaviour of Perturbation Decaying excited states Expansions For our derivation of the generalization** of Eq.5. At the minimum to the right of the barrier shown in Fig. •+(=)* V(f) 0.
(20. i. B and C. three pairs of solutions of types named A. we have IPA(Z) „2hsinz Aq{z) + O il>A(q.51a) where H Bq[w(z)} w(z) = h(«v (w) Bq[w(z)} = Bq[w(z)}. For the continuation we use again the WKB formalism. (20. Recapitulating these from Chapter 17.50a) 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.e.h. (A) Boundary conditions at the right minimum We begin with the evaluation of the boundary conditions at z = 7r/2.Hz^+Ol^ Bq[w(z)}+0 h1'2 (20. we demand that ^(*)L<.20. (20. 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.21) into the Mathieu equation we obtain.' 4 / 1 1 / 2 COS ( Z + 7T \2 4 .3 Cosine Potential: Complex Eigenvalues 487 (using Eq.z).50b) which are valid as asymptotically decreasing expansions in the domain away from a minimum.51b) 2i(«1)[i(9l)]'. (20.49) 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.r/2 ~ e i(2hq) V 2 .h. Proceeding as in our previous cases and substituting the expression (20.z)=tpA(q.21)). as shown earlier. for COS I Z ± 7T 1 1 2 4 >o^\ The second pair of solutions is ipB(z) J2hsmz B.
(20. We saw in Chapter 17 that some of these solutions can be matched in regions of common validity.e.48) and proceeding as in Chapter 17.54) The even and odd solutions are defined as iP±(z) = tyA(z)±iPA(z)}.57) where go = 1) 3.52a) = Cq[w(z)}.qo .53) where a (8M («!)]« 1 + 0 { \ a = 1 (8/i) K9+ ) l + O (20. the relation q . 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. i.55) Extending these solutions to the domain around the minimum at z = 7r/2.56) Applying the boundary conditions (20.488 CHAPTER 20. Thus we have IPB(Z) = aipA(z) and VcO*) = O ^ A O 2 ) . These solutions are valid asymptotically with the first in the domain around z = ir/2 and the second in the domain around z = —ir/2. we obtain our first transcendental equation. \vw). Large Order Behaviour of Perturbation Expansions the function H!/{w) being a Hermite function.52b) The first solution is valid asymptotically around z = —vr/2. Here the upper sign refers to the even states and the lower to the odd states. We return to this condition after derivation of the .=F2 2\1/2{28h2y/4e~Ah l + O (20. and the second around z = 7r/2. we obtain 1>±W = \ a a (20. ^ W H . 5.^ i . (20. (20.
163 — but here we are interested in the WKB solution in a domain where it is proportional to a solution of type A. we see that the turning points at z± are given by 2hq . as we shall see with the result of Eq. formula 280.01.Ah2 cos2 z ip = 0. (20. C O S 1/2 ^" (l) ± ' 2<2+<2 (2 °. Jz+ ' T In principle the integral could be evaluated exactly — see P.20.Ah2 cos2 z ~ 0.2hq] / dz) We know that the solutions I/JA{Z). I^A(Z) overlap parts of the domains of validity of these solutions.61) [Ah2 cos2 z2hq]1/2dz.58) Ignoring the correction A / 8 . 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. Some nontrivial manipulations are again required to uncover the generation of important factorials. Inserting the approximate form of the eigenvalue into the Schrodinger equation we obtain dz2 + 2hq+ — .3 Cosine Potential: Complex Eigenvalues 489 second transcendental equation from the boundary condition at the other minimum. Thereafter we extend the latter across the turning point down to periodic WKB solutions.[ [Ah cos z . The solutions of type A are valid to the right as well as to the left of the maximum of the barrier. Our procedure now is to match these to exponential WKB solutions. i.49) and we can apply our boundary condition. and we obtain the proportionality by appropriate expansion. p. F. . Byrd and M.5 9 ) The WKB solutions to the right of z+ (where V > $IE to the left of the maximum of the barrier) are given by ttw . D. (B) Boundary condition at the left minimum The solutions of type A have been matched to the oscillatortype solutions at the right minimum.67). e x p ( . r ywKBlZJ ~ ~ — [4/i2cos2 2/i ]i/4 2 9 z 2 2 e x p ( £ [Ah2 cos2 z  2hq]ll2dz) ' 1 2 . (20. Friedman [40]. These then provide complex exponentials like (20.e.
Jz++7r/2 \ 1/2 2 V Integrating this becomes z+(ir/2) 2/i h 2 Z+ V IT 2h 4/i In ^ + 1/2 zA 2h 2 2h I ln z+ i) ( I 1/2 2 + 2+ 2 1/2 2/i J (20. we write cos2 z = l . 2/i In the same spirit we can set h[z + l 2h 1 — cosi I z + — 7r 2/i[l + sin z\.cos2 ( z+^) = (z+ .sin2 z = l . using the relation 1 + sin z = 2sin2(2i/2 + 7r/4).» _ i y v . (20.62) where we set — in accordance with our expansion *+ + 2 IT 1/2 <? + (20.490 CHAPTER 20.64) .62): L: = In ~ and hence L = ln In * f 2(z + f) {q/2hyn + + (* I In + 1/2 1/2 2hj 2 In In 2h 4(1 + sin z) (g/2/i)V2 cos z' 2(z + f) (qphy/^z+Z) !sin2(f + f) ( 9 /2fc) /22sin(§ + f)cos(§ + ^2 ' 1 4 = ln 4tan(f + f (9/2MV2 (20. (20. we obtain f ~ I = = 2 1/2 Ah [z + 1 2hq + o\h \z+ 2 dz J Z+ 2h I /MH)' dz 1/2 1/2 2h 2/i /•«"(.59) 2h COS Z+  ~ 0.62) as follows.) +0 z+ IT Inserting this into the integral I. We have for the following expression contained in (20.63) We handle the logarithmic terms in Eq. Large Order Behaviour of Perturbation Expansions Since cos(z + n/2) = — sin z.
we obtain I ~ 2h + 2hsm z Hence exp 491 V^m 2h\~'I z TT 4 — tan . (20.20.^ Proceeding similarly with the other WKB solution we obtain rWKB (z) ~ e x p ( . <$&?(*) 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 . We ignore this here. but mention that for q an odd integer: j(9D l + O and 1 r(93) 4(9+1) ! = ±7I .64) into Eq.66) The WKB solution therefore becomes (approximating the denominator of (20. 2*4(4tf)V« W ^ e[l( V D]! 1 /2y^ (20.50b).63) and (20.67) (27r) /22«+5(4/i2)V4V^y The last expression is again obtained with the help of Stirling's approximation. (20.60) by (4h2 cos2 z)1!4).62)./ * [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\"  .3 Cosine Potential: Complex Eigenvalues Inserting the relations (20. . . on using in the second step Eqs. . 1/2 (20.65) [Z[4h 2cos2z2hq]1/2d< e2he2hsin ^eg/4(g/2/t)g/4 2<?[tan(§ + f )}i/2 (20.68) ' T h e use of Stirling's formula here and below — essential to obtain factorials — assumes q to be large so that corresponding corrections would have to be calculated. (4^2)i/4 [ i( g _3)] ! V/ l (20.+ qj V2 4 A.50a) and (20.
(14. i 27r(8^/ 4 2/t /4 ' r(±) .KW4^8 4 . S1H (2/ KZ ) / (z f 2 (2^).1 ) ] ! 1 (8/Qg/ e4 2fe l[(g3)]!e^ .i ( 2 / i g ) 1 / 2 ( z + .+) + B^) HAh^/\ir. where _ 1 + [±) *) + £ (20. we obtain l(8/i)g/ 4 e..—^ (2^)^(16^2)1/4^ ^ _ _ [T+(±)exp{i(2/ig)1/2(z+_z)} 2(2/ig)V4 + r _ ( ± ) e x p { .50)).U ^K^. (20.™) .z)}]. We now return to the even and odd wave functions defined by Eq. Thus with IPA(Z) and TPA(Z) taken from Eqs. Large Order Behaviour of Perturbation Expansions again using the Stirling formula.l^l !T ^^* T2 7 r ( ^^ /.?/ 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.492 CHAPTER 20. io(8/*)g/V^i6ft2)V4[±(g3)]! .2 [ ? ) Imposing the boundary condition (20.69) l(8M9/4e2^ \2 [ i ( g .2 f e (27r) 1 / 2 (16/t 2 ) 1 / 4 2 (2/wz)V4[i(9_i)]! cos (2M1/2(^^)+4 7T .55) and match these to the oscillatory W K B solutions V'WKE:( Z ) > V'WKB (Z) t o the left of the turning point at z+. we obtain T _ ( ± ) = 0. 1 2 1 ±2 . Eqs. (20.z+)/ VWKBW 2\2j (2^72 l(8^/4e2fe(27r)i/22V2(4fr2)V4 2 < nz+) ^WKBW [\{ql)]\ 1 (8/i). i. (14.68) 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.e 2[J(«l)]l[j(?3)]! i l[(g3)]!e (20.T 7 77 r )V2 /4(2 .49).49).67) and (20.
3 Cosine Potential: Complex Eigenvalues Using the duplication formula \{1l) this becomes 7rV 493 \ ^ .74) This result agrees with that of Stone and Reeve [262].57) by the left hand side of Eq. who derived it for qo = 1 and guessed the form for general values of qo. the result (20. Ex = 2h2 + 2hq0 + A 1 (<? .l)j /2 e4fe(162/t2)^4_±l or ±i 7r\ 1/2 (16 2 /i 2 )9/ 4 e.h) + (qq0)( BE — 90 E{qQ. we obtain E(q.h) = = E(q0. Eqs. h) of Eq. for q an integer both sides are imaginary.3 ) I = }(»« !7rl/22I(. Combining these equations by replacing (28/i2)<?/4 in Eq.26).72) is our second transcendental equation.47). (20. (20. 20.25)) E = A = Ei + E2.20. In both cases we wrote (cf.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.72).21) expanded about q = qo. in fact. (20.4h (16 2 /» 2 )9/ 4 (91)]! = (16 2 /i 2 )«/ 4 = (16/i) 9 / 2 . (20.e.72) We observe that with the square root as on the right hand side both sides are complex. We now have the two equations which have to be satisfied. (20.h) + i l + O '4{[^ol)]!}2 E(q0.qo) El Ah' (20. (20.h)+i%Ego. (17. Equation (20.75) .h)+2h(qq0) (16h)i°+le'Sh E(qQ.3.57) and (20.72) we obtain qqo q0fiSh 2(16/i)*>e {[!(<?<)I)]!} 2 1+ 0 (20.73) Inserting this expression into A = E(q. Eqs. (20. i.
2 below. A = .9) the coefficients Ai of its perturbation expansion. we can obtain the imaginary part calculated above from the formula 4 (H) A ^ o = 27ri(<5£90)2. we recall that we obtained there for the difference 25Eqo between levels with the same oscillator quantum number qo. Bogomol'nyi and V. the discontinuity across its cut from zero to infinity.3. We had A= or _2ft2 + 2ft.77) One can say that this simple relation summarizes the intricate connections between the discontinuity of the eigenenergy across its cut. For definiteness we recall the appropriate expansions together with the definition of coefficients Aj. we can recalculate with the help of Eq. H.* The existence of a formula of this type was conjectured without an explicit derivation like that given here and only for the ground state. B. The same formula can be derived in the case of the double well potential.I g  ± l ( 2 „. .) {I6h)™/2e4h 1 + . 7 8 ) i=l V . Large Order Behaviour of Perturbation Expansions Referring back to the calculation of the level splitting with selfadjoint boundary conditions. f E . MiillerKirsten and A. cf. (17. Setting ng AEqo:=2iSfEqo.W [£(901)]! < (20. A.76) he Thus SEqo is the deviation of one of these levels from the harmonic oscillator of level.4 R e c a l c u l a t i o n of large order behaviour Now that we have obtained QE. (20. 25Eqo=8h[ 2\1/2 . Achuthan. J. Eq. * It would be interesting to see a derivation of the formula — reminiscent of an optical theorem — from first principles.+  .494 CHAPTER 20. its tunneling properties and the large order behaviour. by others. (20. W. i=l We can calculate Ai for large i from the relation Ai = * Jo 1 P°° ti^QE^dti (20. Fateyev [26].74b).e. Wiedemann [4] and Example 20.80) *P. i. 20.
(20. The integral is recognized to be of the type of the integral representation of the gamma function.1). We sketch this method here slightly modified and extend it to the final formula for comparison with our previous results. J). and the method can be applied in many other cases. W. i ^ = ^ + E?27MT E ^ifoj^W (2082) The coefficients Pi(q.20. (20. Dingle [74]. (20. j . so that an exact Borel summation is not possible.j) satisfy the recurrence relation (20.3 ) P i _ 1 ( c ? . (20. We also observe that these terms do not alternate in sign. 20. (20. there is nothing special about this case.74). B.l ) ( < ? + 4 j . f H .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.t We consider again the periodic cosine potential with (unnormalized) eigenfunction expansion oo . a = J + \<1> (2084) .l ) +2[(q + 4j)2 + l + A}Pi_1(q.j) + {q + 4j + l)(q + 4j + 3)Pi1(q.j + 1). the integration is seen to give (20. For convenience we write Eq. J. *R. Using again n = (go .14).13) out again: jPiiQj) = (g + 4 j . But as Dingle remarks.19) for the large values of i under consideration here.83) One now sets ftH = ( 2 % P * ( 9 .13) and the coefficients M2i or M2i+i of the eigenvalue expansion can be expressed in terms of these as in Eq..4 Cosine Potential: Another Method of Calculation 495 with $$E(h) given by Eq.81) We observe that this result agrees with the result of Eq. MullerKirsten [200].
h).31.86) one can convince oneself that Eq.e.(a) oc .h) +p(a + l. nx pi_i(a + l). Taking dominant terms. 0 4j(4j + 2g) pi_i(a .e.1) + 2 P i _i(a) + Pi_i(a + 1).85).e. Now.87) Taking terms of 0(h~l+1) of this equation. z2 (20. (20. Large Order Behaviour of Perturbation Expansions and considers large values of j for large values of i. (with a = j + g/2 in the coefficients) 8/? —Pi(a) = P i _x(a . 1  1 / 2 (4 (20.l. Whittaker functions WKjfl.1) + 2 (a — 1) a . 4j(4j + 2g + 4) + / .83) can be written j . Abramowitz and I. . (20.^(Z).h) = ^2pi(a) with p.ll{z)= L . which are solutions of the differential equation known as Whittaker's equation.91) + Wa+lil/2(z) = (1 .85) implies p(a . formula 13.l/2l Wa. (q + 4j) 2 ~ 4j(4j + 2g). we regain Eq. i. e. (20. (20.z)WaA/2(z) Setting V.496 CHAPTER 20.4.88) satisfy among other relations the following recurrence relation which is of significance in the present contextt {2Kz)WK>li{z) + WK+1.4 ) ^ . g < 4 j . i.27h pi(a) a = 4j(4j + 2 g . (20.85) Considering now the generating function of the perturbation coefficients. d2WKh dz2 + 1 4 K z zM2n WKtll = 0. a.g. A. p(a. (20. the recurrence relation (20. Stegun [1]. we obtain (2a .a R .h) 2+ 8h a p(a.1/2(z) (a1)! (20.89) Replacing here K by a and \x by 1/2.K + ^ ] L + K^)WK.90) See M. Pi(a) i.
h —> —h. Magnus and F . g(h)(20. (20.(« . v . > > ^See W. p.20.„ +1)2] 7 n\z \n which with K = a and /z = 1/2 is: WAz)~e^z«Y. We can infer the approximate behaviour of the Whittaker function for large values of z from the differential equation (20.h) ~/(fe) ' (a1)! .86) have to be chosen such that p(a. i. .87).88) which reminds us immediately of a radial Schrodinger equation with Coulomb potential.(« 1) 2 ]. 89. such that p(a.e.i/2(z) + Va+lA/2(z) = 497 2+ Va. ^Observe that if one solution of Whittaker's equation (20. Thus the approximate largez asymptotic behaviour is WKtp(z) ~ exp •FH V dz z e ^' 2 exp(/clnz). The more precise form obtained from Tables of Special Functions' is w*A* z 2 e / z" n=l [f . (20.j — —j..(« . h) is expressed in descending powers of h. > (20.88) is known.92) Comparing Eq. h) is expressed in descending powers of z = 8/i. Oberhettinger [181].Tl ss\{a)\{al)\ (z) s=0 s — a)\(s — a — 1)! for z — oo. Dingle discovered the solutions^ p(a.93) (a1)! ' where f(h).h) = \~\Pi(&) with pi oc h l.\)V . we have to remove a factor (recalling a = j + (7/2 of Eq.94) Hence to insure that p(a. (20. (20.i/2(z).84)) {8h)q/2( Ah for j < 0. a (20.g(h) are functions of h which in view of Eq.4 Cosine Potential: Another Method of Calculation we obtain Vai. another solution is obtained from this by changing the signs of K and z since this leaves the equation unchanged. In the present case this means the replacements q — —q.92) with Eq.
8 / t ) for j > 0. B.. Dingle and H. W. Comparing > > > this result now with Eq.95b) Then for j > 0 we obtain from Eq..l ) ] ! 11 R. J. this becomes / 1 \ (1)5'" ^ U + iq)<(s + iq1)\ 1 Picking out the term in hrl.. n = 0.94) here we have a series in descending powers of h.96) and a similar expression for j < 0 with /i — —/i. x 4i[2» + i(« — 1)1! il! [ ^ ( 9 . we obtain the result of Dingle: (20. (20. This expression can be obtained from the recurrence relation (20.94) with s — s — j : > x = J + y u ( 1)J ~ f (*+&!(*+*?pi i_ Using the reflection formula (for q = 2n + 1. 2. ^ .83) and is given by 1 1 .498 CHAPTER 20.1. g — —q. Correspondingly we set for j > 0 / 1 A pla = j + q.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. (20. .j _ g / 2  1 / 2 ( . i). Muller [74]. h\ oc (8h)q/2e~4h ( _ .1/2(8^) ( . Large Order Behaviour of Perturbation Expansions and write / i \ pl<x = j+ 2<7>h) « (8h)~q/2e4h v ^+9/2.). (20. j — —j.+ i f or 3 < 0 (20. _ i s.95a) Inserting (20.
Then [2i + ±(ql)}\2* Pi{qJ) [i + \q]\ (20. p.97c) We assume that in a leading approximation we can cancel the factors (i + \q).e. formula (30).20. B.4 Cosine Potential: Another Method of Calculation 499 We observe that Po(q. Ryzhik [122]. l{Q.157(4).e.J)} 2 i=i 2 [2i+l(ql)]!22il2 i![i(gl)]! i (20.{j + \<l). Dingle and H.97d) Then the coefficient M<n of the eigenvalue expansion.l) (i+y) 2*%(ql)}\ [2i + §(gl)]l i\[i+\ql]\ ' Inserting this expression into Eq. of Eq. Thus ° 9 (i + \q) 2«(i+Igl)! p .l ] ! (20.98) the duplication formula 22zl / y P*l>'7J<*l)'(»j»and obtain Mi% + \{q^)W + \{q. (20.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]\ (20.14).100) **R. J. note the misprint there: (r — s) ought to be (r — s)\.3)]!2 2 '+^. M. (20.1 )/ 2 2 2i ' ^i\{\{ql)]\ [2i + g .j) (i + lg)[2i + i ( g .0) = 1 as desired (although not really enforcible for large values of i). Gradshteyn and I.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. (20. Miiller [74]. i. A special case of the formula can be found in I. W. . formula 0. S.99) In the next step we use for both factorials [2i + ^(q — 1)]! in Eq.+i?_l]!}2 (20. 5. E* J'=I i+M 2 [2i + 9 .
Eq.' 1 i • (20.17).3 ) ] ! _ 1 Hence M or. 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.86) and set (cf. We refer back to Eq.e.l l ! with the large i behaviour in agreement with the result (20.500 CHAPTER 20. the result of Bender and Wu [18].]\.5 20. i.e. [19]. 20.81) for q — qo = 2 n + l ..101) This formula allows us to cancel all factorials [i. i. . i\ M i large 1 . since \{q — l) + \{q — 3) = \q..l. [^ + i ( g .. Eq...102) Thus the result is M m 1 2«Fi + g . i. i. Large Order Behaviour of Perturbation Expansions In the following step we use the approximate formula (20. n = 0. (18. replacing 2i by i: 2 ^2 8 ^ % + *"*}' (20.86).4)) £ = ^ = ^+ w and y= ^ > ( 20 .1 Anharmonic Oscillators The inverted double well In Chapter 18 we obtained the imaginary part of the eigenenergy E of the Schrodinger equation for the inverted double well potential.105 ) the latter because we see from the eigenvalue expansion (18. (18.Am i .e.o i • .34) that £ is an expansion in ascending powers of h 6 /c 2 .i ) ] ! [ ^ +  ( g .1.e. The imaginary part we obtained is interpreted as effectively the discontinuity of the eigenvalue across its cut from h2 = 0 to infinity.5.. (18.2.e. i.
and hence inserting here the imaginary part from Eq. i.20.109) or.e. i. The result establishes the series unequivocally as an asymptotic series.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.110) This is the result in agreement with the large order formula (4.5. (with E = h2£/2) my') = 290+1 (y')9o/2 e 2/'/3 (27r)V2[i(go_i)]!' and integrating with the help of the definition of the factorial or gamma function.(20. . one obtains the result Mr 2io+hr+1+f (l)r[r + f ] ! 7T3/2[1((?01)]! (20.86).4) of Bender and Wu [19]. h2 _ h222K+llHr+K+ll2{l)r+1 r+ K .e. the discontinuity across its cut. 20.l ) r . (18.77) can be verified. and then the relation (20.106) With £= E r=0 Mr (20. 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. (20. in terms of K where go = 2ftT + 1. We leave this to the following Example. in fact as one with Borel summability in view of the factor ( .108) Proceeding in this way.107) we obtain (i)r r 71" JO yirZ£{y')dy'. This can then be compared with the level splitting we obtained for this potential.5 Anharmonic Oscillator The Borel sum can now be formally written 501 (20.
J. J.77) was conjectured without explicit derivation and only for the ground state by Bogomol'nyi and Fateyev [26]. M. Liang and H.502 CHAPTER 20.Q . H. Achuthan. Cizek. it can be used to obtain the other. of International Workshop on Perturbation Theory at Large Order (Florida. J. . E. Harrell. Paldus and H. M. R. J.77) and its possible generality has been referred to by Brezin and ZinnJustin [36]. Large Order Behaviour of Perturbation Expansions Example 20.^ 20. (357) to (371). G. K. [19] can be traced back from articles in the Proc. Graffi. J. S. MiillerKirsten and A. V. Harrell.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. Gram. 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. A relation similar t o (20. E. 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). MiillerKirsten [164]. The relation (20. J. S. J. J. W. H. Cizek. An approach to arrive at the relation via instanton considerations has been examined by ZinnJustin [292]. M. MiillerKirsten and A. in t h a t when the quantity on one side is known. t p . *Some of the literature in this field following the work of C. " P . Wu [18]. Nakai. R. K. J. . W. + 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. Silverstone [62]. J. Grecchi. S. J.6 General Remarks The study of the large order behaviour of perturbation theory has become an individual direction of research. Damburg. J. W. Grecchi. Its usefulness has also been demonstrated there. Propin and H. as in later chapters. 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. J . Achuthan. R. *R. Wiedemann [4]. Bender and T. V. Damburg.* A more recent status evaluation is provided by the book of LeGuillou and ZinnJustin [161].77) has been obtained in the Schrodinger theory of the molecular hydrogen ion H^ . Paldus. 1982) [139]. Eqs. Propin. t The relation (20. Wiedemann [4]. Harris. Silverstone [52]. T. Solution: For the solution we refer to the literature.
our main interest in path integrals and their uses will focus thereafter on their evaluation about solutions of classical equations. 503 . Chapters 2 and 5. P. A brief introduction is also contained in R.Chapter 21 The P a t h Integral Formalism 21. We illustrate the method here by rederiving the Rutherford scattering formula with this method. Felsager [91]. Hibbs [95]. P. particularly useful in cases where the behaviour of a system is to be investigated close to its classical path. ^ Our interest here focusses on quantum mechanics. in general. can be found in the book of B. R. The path integral is. however.e. Feynman and A. Probably the most readable introduction. This formalism is particularly useful for evaluating scattering or transition amplitudes. For the application of path integrals to the hydrogen atom (i. A further wellknown reference is the book of L. However. P. 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. Thus we are in particular interested in rederiving the level splitting formulas obtained in earlier chapters with this method. S. 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. A standard reference is the book of R. which explains also difficulties. unknown). Feynman [93]. Feynman [94]. Coulomb problem) we refer to existing literature. 'See in particular H. *R. since in such cases the probability of this system choosing a path far away may be considered to be small. Kleinert [151].1 Introductory Remarks In this chapter we introduce the path integral formalism developed by Feynman. Schulman [245]. 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.
(21. ti).1) Let the wave function i/> at time t = ti. The Feynman Path Integral Formalism 21. =J dx(xf.tf\x. */ = U + (n + l)e.tf) = / dxK(xf.5) Sn = X / fc=l mo (zfe .x fc _i) 2 2e2 V{xk) Thus: Given xk(tk).t). (21. be ip{x. (21.2 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.e.3) We divide the time interval tf — ti into n + 1 equal infinitesimal elements e such that £() ti = = hi ti + e. we wish to know* ip(xf. i.x. We are interested to know the wave function tp at a final time tf > ti and position coordinate x = Xf. U.ti)ip(x. we can compute the action S1. ti) = 5(xf . Obviously K(xf.ti\tp).x). x.ti){x.ti).tf\ip) . 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). *In Dirac's notation we write this equation later (xf. 2mo dx2 (21.504 CHAPTER 21. (21.e—»0 Sn. the initial time.4) so that the timesliced action S becomes rtf «+i S = / 'ti n+1 dtL = d£L = n—»oo.e—»C lim V e m0xk fc=l  V(xk) = lim n—>oo.tf.2) where K is a Green's function.
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. The repetition of the quadratic factors in the argument of the exponential with indices differing by 1 suggests the construction of a recurrence relation.tf).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.ti) to its final position at (xf. 21. 21.21.2 The Path Integral and Green's Functions 505 Above we considered time intervals of length e. of which one is illustrated in Fig.t). Thus there is a countless number of possible paths that the system may choose in propagating from its initial position at (xi. as will be shown to be possible below. the Fresnel integral (13. The usefulness of the method therefore depends on the fact that some paths are more probable than others. Here and in the following we need one or the other of the following integrals (cf. We now consider intervals in position coordinates and in particular the following integral over intermediate position coordinates Xk which corresponds to the sum of exp(iS/h) over the uncountable number of possible paths from Xi — XQ to Xf = xn+i.28)): f dr\ exp ia a) ' (21.1 A path in (x.6a) .
mQf Xk L ~2e^ ^ .t ~Xk+l> + xk+i 2 ) +{xk .6c) In the evaluation of these integrals we replace a by a + ifi. (21.z f c _i) 2 } } . (21.0 ^7 (x f c + i .m0 .xk+x oo —oo oo .ti) '•= f g m ) (N(€\]n+i dxL.(x fc+ i .xk_i) dx exp i — { 2 a .xk+i) exp .. ie f me\ 2a\a~) 1/2 (21.s f c _i) + (xk+i . One therefore removes the troublesome factor by defining (Sn being the sum in Eq.Zfci) 2 } I +{xk+i x*. dxneiSn/n. 2 + 2a. We then integrate and finally let /x go to zero.8) Note t h a t in spite of the n integrations..Xi..2 + (x + x f c + 1 . dx exp z—{a. the power n + 1 in the denominator has t o be seen in conjunction with the n + 1 terms in the sum of Sn in "Thus. one has j ^ dr1eifa+i^ /£ = .5)): G(xf. for example.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 dx2.37)) dxk exp d(xk . The Feynman ia Path Integral Formalism f J—c J dr) r\ exp v = 0. J—oo dxneiS/h — • 0. with f™ dze"™ 2 * 2 / 2 = s/2ir/w2.ir 1/2 £dxeXp[i^{2(3 dxexp\i—<2lx+ 72 iV(e) exp Z7TT.6b) —c ia o drjr] e x p—V € (21. tf > U.* W i t h the first of these integrals we obtain (compare the result with the Green's function (7.506 CHAPTER 21.tf.x f c _ i ) 5 \m0 J (21. n > 0.
p. These integrations are indicated by the horizontal arrows in Fig.U) = / JXi=x(ti) = K(xf. See the square root in front of the exponential in Eq. Different constructions of these families of paths usually lead to different measures. we arrive at a path integral without the measure factor N(e) above. 183.e.e. 183. A mathematically rigorous definition of the path integral is difficult and beyond the scope of our present aims.1. 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. (21.Xi. Felsager [91]. p. (21. The additional factor is needed later to combine with the number n contained in the extra factor y/n + 1 (in Eq.tf.2). (21. in fact.5).§ 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. the Green's function K of Eq.g. G(xf. however prefixed with a new factor. § See e. (21. B. In the expression rx" / ?){x}eiI{^ Jx the quantity x" is only a symbol to indicate the endpoint of the path.5). Then one integrates over X2 from —oo to oo and obtains an exponential involving (x^ — XQ)2. B. and so on. Thus one first integrates over x\ from — oo to oo and obtains an exponential involving (x2 — xo)2. .10) can be looked at as the product of a succession of free particle Green's functions (or propagators). Felsager [91].tf. it may be noted that with a different philosophy concerning the summation over paths.tf. (21.26)) that — ignoring the potential — the multiple integral (21. It is neither a variable of integration in T){x} —> Ylndxn.Xi.2 The Path Integral and Green's Functions 507 Eq.7) for n — 1) to give a time interval. i.Xi.t In the following we demonstrate that the function constructed in this way is. and will therefore not be attempted here. which all vary from —oo to oo. as in our discretization in Eq. (21.24).9) T>{x{t)}eiS/h = ^ww^IdxlIdXnelSn/h(2L10) This is the formula originally given by Feynman.ti) for tf > U.' These factors N(e) were introduced to cancel corresponding factors arising from the Fresnel (or Gaussian) integration. 21. summation over all possible paths.ti) For tf > ti we now write* G(xf. (21. 11 For discussions see e.21.g. However. We shall see later (with Eq. i.
xi. however. 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.3 in the evaluation of I a specific path integral (see Eqs.Xi.ti).ti. 187 .tf + e. and one is calculating the Feynman amplitude or kernel relative to that of the free theory.U) imo 2 G(xn+i W)Idr. one with the full action S. the discussion in B. (23. I We employ this method later in Example 23. One way to avoid this problem with the integration measure is to consider the ratio of two path integrals.tf. 21.exp 2e n ieV(xn+2) G(xn+2 + V. we obtain the same as before except that there is no factor N(e). Consider a time t = tf + e for position Xf = xn+2. We have G(xn+2. pp. Then we have on the right the factor G{x+rj).1 and allowing n to go to infinity. Expanding this in a Taylor series around x. (21. Then the measure drops out.ti) = 1 j^p: f°° cte n +iexp } • fm0. .11) For reasons of transparency in the next few steps we replace xn+2 by x and suppress temporarily tf. so that the piecewise linear paths fill the entire space of paths.188.tf + e. Each of these linear pieces is completely characterized by its endpoints. the other with that of the action of the free particle SQ.xi. Performing the sum now in the sense of summing and hence integrating over the vertex coordinates in Fig.508 CHAPTER 21. Felsager [91]. we have G(xn+2. The Feynman Path Integral Formalism The factors N(e) can be avoided by instead summing over piecewise linear paths. It is for these reasons that — unless required — the measure is frequently ignored. We next determine the equation satisfied by G in order to verify that this is indeed the Green's function for tf > t{.72) to (23.Then we can construct a recurrence relation as follows. we have See.Xi.74)). 16 x2 Sn+l) \ 7T2~(Xn+2 V(xn+2) Setting now xn+\ — xn+2 = n.
tf. d and this means d i—G(x.Xi.xi.15) We see that G(x.Xi.ti)\ .xi.(21. tf.2 The Path Integral and Green's Functions so that G(x.xi.xi.tf.ieV(x) + ri< —G(x.tf)= dxG(xf. It follows that ip(xf.xi.tf.U + 0(e ).6c). Xi.Xi.Xi.ti) +\if(^G{x.ti)ip(x.ie 1 d2 2nV0dx^G{X.tf + e.xi. 2mo dx2 (21. U) + .ti) = 1 d2 + V(x) G(x. tf + e.6a).tf''Xi.ti) .— p G ( j .U ieV(x) 2i7re\ 1 / 2 m0 ie d2 G(x.6b) and (21.13) We integrate with the help of the integrals (21.t Thus + 0(e2 (21.Xi. tf. tf.tf.tf.14) = G(x.G(x.tf.ti) for tf > ti (21.tf + e.ti) 2 —l€ — ^2 + V(x)G(x.ti) = j^rr dnexp I ^rj1 .tf. ti) satisfies the timedependent Schrodinger equation.Xi. tf.ti) = G(x.U) +••• \ 509 G(x.16) .Xi.21. U) + Q(e2 = (lieV(x) + 0(e2)) G{x.tf + e.tf.U).x.ti) N(e) or G(x. Retaining only terms up to those linear in e.ti) +V(x)G(x.xi. .Xi. we obtain: G(x.tf.Xi. U) .tf.xi.tf. (21. U) = e^G + 0(e 2 ) dtf 1 d2G(x.
ti) = 5(xj — x).510 CHAPTER 21. as this is also called.15) is a wave function (cf. It follows therefore that {{W ~V+2m~0~^) K{Xh t/.U).x. We have therefore established that the Feynman integral (21. and the Green's function KQ\ S0= Here (cf.3.tf.Xi.U) = / T){x} eiS\ [fdt \mx2. X U) = i6{Xf ' ~ X)6{tf " U)' (2L19) Thus K is the Green's function for the problem of the Schrodinger equation.1)). is the Fourier transform of the nonrelativistic propagator.10) is indeed to be identified with the timedependent Green's function.tf. (21.U)6(tf . K0 = 9(tf .x. In a problem where V is small it is convenient to calculate K with a perturbation method.U. The result. as may be verified by inserting this into Eq.U)G0. (21.tf.8)) Go(xf. Eq. (21. 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. The function G which satisfies the timedependent Schrodinger equation (21.1 Configuration space representation In the case of V = 0 we have from the previous section for the action.x. The Feynman Path Integral Formalism also satisfies the timedependent Schrodinger equation. or K(xf.x.*<)' ( 2 L 1 8 ) 0(tf ~ li)QfG + 5(Xf ~ X)5^f since G(xf. Hence our original function K is the same as G. denoted by KQ. 21.17) Differentiating K with respect to tf we obtain —K(xf.20) . Eq.3 The Green's Function for Potential V=0 We now calculate the Green's function K for potential V = 0. 21.ti) = = 9{tfti)—G + G5(tfti) .16). now called So.U) = G(xf.tf. (21. (21.
z / c . TJ+1 dxn exp[i\{(xi 2 — XQ) + • • ^n) }] (21.x 0 )" n + 1 (in agreement with Eq.tf.i ) 2 .t) is the configuration space representation of the free particle (V — 0) nonrelativistic Green's function.8).tf. (7.tf.3 The Green's Function for Potential and hence Go{xf.22) • \ n/2 .25) T h e quantity Ko(x.tf.t) K0(x.tf.Xi. (21.21) fe=l and t = (n + l)e = tf — U and h = 1.Xi. f c . Then Go(xf.oo /*oo / oo cfcci / J—oo cfcc2 • • • / J—oo + (x. But t = tf — ti = (n + l)e.ti)= iX (o. 2t Go(xf.23) We now insert our expression for In into Eq.Z i ) \2KieJ \m0 J y/n + 1 1/2 m0 «mo (21.23) for Go and obtain Go(xf.Xi. . n + i ..Xi.ti) = = G0(x.Xi.tf.ti) = V = 0 511 m0 lim n—>oo \ 2irie oo (n+l)/2 /oo oo /*o / dx2. (21../ . We obtained the same expression previously — see Eq. exp 2 (21.21. (21.Xi) exp \2Kie{n + 1) 2e(n + 1) Note the square root factor in front which originates as discussed after Eq.ti)9(t).37) — by explicit solution of the differential equation which the Green's function satisfies.ti) Ko(xf. oo dxi 71+1 / J—c /_ dxn exp oo irriQ oo ~27 ^](a.24) ( s / .7) for here n = 1 (n+l)/2 lim ( —^~ K2meJ m /n A = 0 V 2e (21. exp where Xj = xo and Xf = xn+\ and there k = 1). (21. In Example 21.t) = ( m0 \2mt 1/2 irriQ and writing x = Xf — Xi. we have Go(xf.1 we show t h a t oo /.ti) lim n^oo lim n>oo m0\(?1+1)/V27rie\"/2 1 z —— exp 2e(n + l )(a.Xi.
x „ + i ) 2 = ( x n + 2 Xi + Xi .1: Evaluation of a path integral Verify Eq. If the potential V(x) of Eq.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. Solution: The result can be obtained by cumbersome integration.10) in a product form.)!} J — OO . (21.^(Xn+1~X0^2+iX(Xn+2~Xn+1^'i.tn+l•xn.x „ + i ) 2 = (x„ + 2 .xi)2 = 21 xi ^2o(„. In our proof of induction we assume the result is correct for n. i.ti. (21.e. For n = 1 (see below) h This result can be verified like Eq.25) we can rewrite the Feynman path integral or kernel (21.512 CHAPTER 21. Then oo 2 / (21. (21.26) with n integrations and n + l factors KQ.. ••• 2 / \ 1/2 \2iriheJ dxldx2•••dxnK0(xn+l.xo)2 + (x2 .)+(1"+2. the latter being a Green's function with interaction like that for the harmonic oscillator..22) oo Indxn+leiX<x"+*x»+^ 1/2 f°° dxn+ie^:. 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 (21.5) is carried along. n+l = lim / n ^{a^jexp fc=i ^°° Jxi = lim / • • • / dx\dxi • • • dxn I —1 nKx J J \2mneJ / \ 1/2 .x0.2y(xn+2 . as / JXi f r D{x{t)}eiS°lh .f_^y e. or else by induction. Eq.22). Example 21. we have instead of 5*0 and KQ expressions S and K._*R±HY+l_{xo_X2f J —! r dXieiM(xlxo)2 + (x2~xi)2 ei^(x2x0) and using Eq.tn)••• K0(xi. (21. (n + l)A"_ We set y := x n +i — Xi = x n +i — XQ and (x n + 2 .54). (7.7) by direct integration with ^ . ^ (xi .t0) (21.6a).x ^ 2 + y2 . e 2he ^{Xn+lXn)2 = \2iriheJ lim ••• n^ooJ J x(^*) e^() .
2 y ( a . A(n + 2) 21. rc+2 rl 1 / s2 513 so that n dz = dy = d i n + i . ) H — (xn+2Xi) n+2 n+l n+2 y2 .29) Hence with KQ from Eq.25): Ko{xJ) = = (S) e{t)(^X'2 y J . Using Eq. (21.2.a .i) .x2rriQ 1/2 toij_. i. We pass over to momentum space with the help of the following two integrals.3 The Green's Function for Potential V = 0 Next we set z : re+1 (x )2 — Xj). it .i) H —3/ .Xi) + ( x n + 2 Xi)2.2 M o m e n t u m space represenation Our next task is to derive the momentum space representation of the free particle Green's function KQ. and + 2 2 l n + 2 ( x n + 2 .3.max exp i 2t 6(t) 1/2 \2mtJ f°° 2KimQ J dp exp J—c ipx P t 2mo (21.30a) —6(t) / dp exp ipx — to Joo 1 i—t 2m.6a).28) exp (— V 2irimo J —( dp exp ipx — i p* 2mo (21. s 2 n + 2 2 „ / ^ n + 1 / \2 (x„+2a. This quantity is also described as the nonrelativistic free particle propagator. (21.. The first. e > 0 .e. (21. The second integral is an integral representation of the step function. n + 2 .21.27) '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.27) we have (with a = t/2m0) . iir(n + 1) 1/2 jn+1 n+l L(n + 2)A'n + l I 1/2 4 (n + l ) A . 0(t) Ti: i r Art alr This integral can be verified by applying Cauchy's residue theorem in the plane of complex r and using one or the other of the contours shown in Fig. n+l T h u s as r e q u i r e d 1/2 Jn+l ^ e J — oo 1/2 .Q .2y(xn+2 . a dp exp [ipx + iap2] = exp — i(21. (21.n+2a.i j ) H n +— z 1 . 21. (n+l)A n n j(a.
an expansion in rising powers of the coupling constant (contained in the potential). The generalization to three space dimensions (required below in Eq.31) is the nonrelativistic propagator which describes the propagation of the free particle (V — 0) wave function.4 Including V in First Order Perturbation We now proceed to calculate the first order correction in a perturbation expansion. 21.514 CHAPTER 21. Thus in first . The Feynman Path Integral Formalism Im t \t>0 Rex Fig. (21.Q (21.28) this becomes K0(x. Replacing 9(t) by its integral representation (21. This is the type of expansion most people describe as perturbation theory. 21.t) i(2 h?L L E iv 00 dr exnlipx T '2mo t + itr] — te ~U2^LdPJ ^E+£. The first order correction contains one factor of the potential V. ( 3 b) ° The expression E+ V — it 2m. the second order two factors of V and so on.51)) is fairly selfevident.*' where 2mo — r.e.2 Contours of integration. i.
^){iV(a.t) the first order correction is (G\ ~ G — Go) Gi(xf.33) xG0(xi.(xfc .. ^ = k ) = lim (S^ A H*\ nKX) *—J \2irieJ r "+1 />oo X / dsi / dx2.U) = — lim fv{x(t)}el —— js (n+l)/2 I dxi dx2.21) this can b e written as Coo oo /•oo />oo / oo dt / J —oo ^Go(cc/.t)] n+l . .ti.xi.32) Since — we recall this for convenience G(xf.t)\.t/.4 Including V to First Order: Perturbation Expansion h=l..i V e / D { a .20)) and with elS = exp i S i o / hi dtV(x.+i 6XP L — {xk XkiY {iV(xhti)} 1/2 X Em fc=l — 0 (xkXki) W i t h Eq. li f f 515 order (cf. (21._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 .xfc_ OO J— OO J—OO L J.Xi. x (n+l)/2 \ lim .^ J_00 V(xk. (21. } e °° f l f •/ oo /"oo F ( a ..ti)= n+l ri_> f V{x(t)}eiSo „ iSo f ' dt[iV(x.Xi.)} (21.t) JSo dtV(x. . dxr. (21.tf. .U). \2meJ n+l J.tk) J_( x exp fc=l = ie(—2(xkxk 1>{x(t)}exp iSoi dtV(x.i.tf.^. dxn exp i ^ —..21. dxi ( m0 V V27rie/ m0 \^He \ exp fe=.i.5) a n d (21. Eqs. .
. The factors of "V" originate from the series expansion of the exponential exp ftf . (21.x2. in writing down the expansion exp i dtV(x.t) =li dtV(x. Jti Consider the term of second order.x. / 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'.t2)V(x2.ti) t2)K0(x2.34) We observe that we can extend the ^integration to — oo since KQ contains the step function 9{t).t) ! Jti r*f dt'V(x.xi.t)K0(x.t2) xKo(x2. t) Ju Thus the term containing n times V contains a factor 1/n!. t2.e. The first two terms lead to identical contributions and therefore cancel the factor 1/2! in the expansion. OO /"OO / oo dt / J—oo dxKo(xf.37) Jti The last two contributions do not describe the time evolution from ti to tf.36) {if •j. This is cancelled by n! different time orderings.t2.tf.i / dtV(x.t) (21.tf. i. (21. i.516 CHAPTER 21. The Feynman Path Integral Formalism and hence correspondingly for the first order contribution Ky. It can now be surmized that the next order contribution to KQ + K\ is K2 = (~i)2 / dti / dt2 I dx\ I ti. = i dt2 / dx2Ki(xf.x2.U) xV(xi.I dtV(x. Xi.t)V(x.xi.ti)K0(x. (21.xi.ti).t2)V(x2.tf.35) We also observe here that the problem of timeordering arises only in perturbation theory. They must therefore be deleted.t.ti). t') + • • • . dx2Ko(xf.e.
t)= f dpdEeipxiEtV(p. Here.t2 K o (x 2'l2 . (21. x"^ X x"2.tf. 21. U) (21. For various reasons — such as illustration.tf. (21.tf. we are concerned with quantum mechanics. transparency or brevity — it is useful to introduce a diagramatic representation of individual perturbation terms. The diagrams represent mathematical quantities and hence are designed on the basis of rules. or in momentum space as mostly in field theory.40) The rules for the present case are given in Fig.ti) (21.t)V(x.t) ) K K K0 'V X K o Fig.t)Ko(x.ti) i dt = K0(xf. Such diagrams in the context of field theory are called Feynman diagrams.E). The Fourier transform of V(x.Xi.t) is V(p. x r l i 'V(x. 21.3 along with the diagramatic representation of Eq.3 Feynman rules and representation of the Green's function. of course.i ] P I dt J dxKn(xf.t)K0(x.t. and hence the diagrams representing perturbation terms here are Feynman diagrams corresponding to those in field theory.ti) = K0(xf.39). called Feynman rules.tf.x.tf.4 Including V to First Order: Perturbation Expansion 517 The full perturbation expansion of the Green's function K is now seen to be given by K(xf. tf. These may be formulated in configuration space as below.39) dxK(xf.ti).xi. This can also be written in the form of an integral equation which when iterated yields this expansion: K(xj.Xi.21.ti).t.Xi.38) . E) which is given by (in onetime plus onespace dimensions) V(x.x.t)V(x. .Xi.Xi.
dn = (L/27r) 3 dk.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. 2 The expression A 2 /T can be related to firstorder timedependent perturbation theory in quantum mechanics. and dn = dnxdnydnz. (21. in onespace plus onetime dimensions by A= dxf dxiil)*f{xf.41) We have to pass over to three space dimensions. The Feynman Path Integral Formalism 21.tj)K{xf.t) = . * k 1 • .e. The plane wave function normalized to one over a box of volume V = L3 is. The transition amplitude A describing this process is defined by the Green's function sandwiched between the inand outstate wave functions. for the allowed values of k: kx = 2imx/L. i. with p = Kk. .7 = e i k ' x _ i : l i with f i/}*i/jdx.e.= l. ^(x. (3) The transition probability per second is = \A\2 . mo V since ip is normalized to one particle in volume V.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. (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).ipf have to be properly normalized. here with E — k 2 /2mo.tf. to one particle in all of space. We recall for convenience from **Note that from the wave function we obtain. i.ti). T = duration of time for which potential is switched on.ti)il)i(xi.518 CHAPTER 21.k = 2wn/L.Xi. There the wave functions ipi. (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.
.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.e. Proceeding in this manner one obtains Fermi's "golden rule". Hence (h = 1) atot = f J j2^~k dp V Vm0\A(p. P(*) = j r (2145) (cf.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. and with these the following perturbation ansatz: ih^* with an(t) = HV.44) = u(Ek. V(x. The transition probability per unit time is . (21.21.47) .44) and f_oo can be replaced by fQ. 47T 1 3 7 (21. u = ^\H'kfp(k). (21. # = ^an(i)VnM) n = a^(t) + \a^(t) + . Eq. 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.— . in which T is a large but finite time.117) and Example 10. Then for stationary states M^t) and itm^ = f HUt'V^t'dt'.T)\2 f • (2L46) We take the following expression for the timeregularized Coulomb potential. H = H0 + \H'.k. (10. Then H'ki can be taken out of the integrand in Eq.4)..t) = e~^lT\ r where r = Ixl.
tfx. The Feynman Path Integral Formalism In calculating the Rutherford formula with the help of our previous formulas.39) OO /"00 / dt / OO J —OO dxK0(xf.t) (21. t. (21. t)K0(x.tf.tf). The lowest order approximation of A. Ai = dxf / dx. (21. (x*.tf)ip*(xf.ti).f. Thus with the ansatz Ko(x. ti) = co5(t — U)5(x — Xj) etc. This now has the form of the amplitude (21. £. (21. we have to replace the onedimensional quantities dxf etc. t. (21. We can relate this amplitude to the amplitude ark of Eq.Xi.50) Here x — xj — Xi and t = tf — ti.ti). (21.51) Inserting this into Eq. T a large but finite time. In our previous onedimensional considerations we had with Eq.49) together with the following wave functions (of asymptotically free particles) 1>i(*i.520 CHAPTER 21.U) = ~ e ^ l ^ \ ^(*/>*/) = _ L e * P .30a) K0(xf.44) by contracting the propagators to instantaneous point interactions.xuti) = ^ J°° d p e i p x _ i ^ * .t)V(x.49) i. The threedimensional generalization is (taking tf » T and ti <C —T.f. by threedimensional dxf and so on.i x (i)V(x. we obtain A1 = clS(ti tf) / dxf(i)V(xf.). (21.52) .tf)K0(xf. (21. so that 0(t) can be dropped) K0(xf.t. is now in three space dimensions A = Ai= / dxf / d^ty*f{'x.tf. dt dxip*f(xf.x.tf.44) (here in the scattering problem with initial and final energies equal).yLi.tf)ipi(xf.tf)Ki(x.e.* / .t)K0(x.ti)'ipi(xi.tf.. i..t) = ^ J*ie*<*'*><<*'*>. x. Xj.48) where from Eq.^ / .>/>.x. (21. also called Born approximation.
i ( k 2 .59) .k. Then we require the following integral.T) = JdtJdx(^\V(x.55) We evaluate this integral with the help of the following integral in which we complete the square in the argument of the exponential and set r = t .k.t)exp A(p.57) we have A(p. Then integrating over q and q' we obtain .p ) .r) = = y .k.21. £). At2 dteiat~^ a2T2 f°° =e " J dre^lT2 p2k2 2mo = (~)l'2ea2T'' / 1 6 . k2 ipx + i—t + ikx — i 1 2mo 2mo (21.2: piqx / Y iukawa •" dx e~Mr 47T q + M2' 2 (21.p 2. (21.53) 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.iaT2/8: L With 0 0 • .k.fdxe^rt* dxe< p) x fdte1 ' (4) 2m0 T2 e "^^J2" 16 .).f f dxi f dt f dx f dq / d q ' p V 2 (2TT)6 2 +iq' • (x — Xj) — i 2mo (t — U) + ik • x 2mo . which is evaluated in Example 21. this becomes A(p.5 Rederivation of the Rutherford Formula we have (for ingoing momentum k and outgoing momentum p A(p. (21.T) = ^JdXJdt%*t2/T2exp i ( k . P2 .56) (21. (21.54) Now inserting the expression for the potential V(x.T) = x exp L z/ q/2 k2 521 f dx. 2 „ x v~ m o (21.58) In doing the integration over x we introduce a temporary cutoff r = 1/M by inserting the factor e~Mr in the integrand.~ . .
63) we have: Probability per unit time = 77777.T2f 8 V/ 2 .522 CHAPTER 21.a ( — ) ^ . The Feynman Path Integral Formalism This means: The result of Fourier transforming the Yukawa potential ~ exp(—Mr)/r is the propagator (21. { I i _ F T l } .61) We can now Hence in our case \A\2/T has to be replaced by \A\2/(nT2/8)1/2. In our case this has not been done.62) We want to consider the limit T — oo.k.„ 2. integrating over the square of the ^dependent factor in V we have /"°° dt(e4*/"1*}2 = (^p) ' 2 instead of T. For this we require the following • representation of the delta function: —k2T2 rp —k2T2/8 S(k) = lim T . otot contains the factor \A(p. Using Eq.P ) 2 + M2]2 4 VvrT2. dE = PjV = ^ E ± . mo m (21 . 2 with 2mo J 2ir a2 Ct 16TT2 LU7I F [(k .^ .e. i.\ 2?7lo / 1 (21.< ^ £ \M VTTT2. 66) o . we obtain . e (21. k .T)2/T. y 2 [ ( k .„.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 (21.65) But E=*. Using the result (21. ( 21 .64) Inserting the result now into our expression for the total cross section. (21. <rtot. / / P 2 _ .59). k2 ." + M ] p) ' "•"• ^2 .51 — 2 /w2 y ~ —. T ) ~ . (21. This expression assumes that the integration over time t is normalized to 1.59) we can reexpress A as I /7rT2\1/2 e (P 2 k 2 )T 2 f Ajr ~) p * P . . e 8{k) = lim 7=*—^ (21.60) Now.3. write down the probability per unit time which is ( 8 \1'2 _a2 16^ .63) We establish this result for T — 1/e in Example 21. where M has the meaning of mass (in the case of the Yukawa potential the mass of the exchanged meson).
67) and (21.^ . 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 . This result may also be obtained from lyukawa of Eq.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 — .59) by replacing there q 2 by ( k — p ) 2 . i. (21. p 2 = k 2 . with k = y 2moE / we obtain the differential o?m\ 4fc sin 4 (0/2)' 4 cross section (21..67) T h e delta function here implies energy conservation.4. 2mo / dft. and inserting the expression thus obtained into the final result of Example 10. giving the potential a coupling constant. esfn (21.2: Fourier transform of the Yukawa potential Show that [ j / j eiqx dx—e .e.p ) 2 = 2 k 2 ( l .e .j .68) into Eq. —Mr 4w = .j .M £ M * = / m0kdn = m0^2m0E (k .^ Jo 47T q 1 M — iq 47T i q2 + M 2 Jo G— 9 Example 21. (21.e. 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.5 Rederivation so t h a t of the Rutherford Formula 523 M^f) . /Yukawa = ) Solution: We choose the 2axis along the vector q.= e " f c a ^ o 2v^fa /4a = lim £^o —. so t h a t (21.cos 9) = 4k 2 sin 2  .70) . setting M2 = 0.68) Inserting (21.^ . Then (x = r) foo „2jj. Example 21.69) This is the wellknown Rutherford formula.21. (21.* 9 ' d r e .65) with the cutoff M > 0.
71) with respect to k yields the same as °° 2 .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 .75) .71) Solution: In general one uses for the evaluation of such an integral the method of contour integration. ' i a = lim e —0 pk 2 /e2 e^TT 21. The Feynman Path Integral Formalism o° 2 / oo dxe~ax eikx. Since dxe" 1 1 1 = « / . Differentiation of (21. '<*> = . We have x> = I J ^ l p X p l x ) . 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 ) .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. We consider again nonrelativistic quantum mechanics. We write the state vector of a particle of threemomentum p in the momentum space representation: p). In the present case. a simpler method suffices.q).^0 a^C 2^/TTO. (21.x) (21. (yx) = 6(y . a. (21..* .72) co From (21.oo J / )eikx. Va we obtain g(k) = Je~k V a = 0. (21. 7 r°° d 9 xdxe~ax elkx = — dx — (e~ax oo 2a Joo dx Partial integration of the right hand side implies oo /. In configuration space the state of the particle is written x). however. 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).70) and (21. Direct integration /4a .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 .
Thus (qPp) = p(qp) = p(27r)3<5(q .82): The general Schrodinger equation has the state vector to the right. The time development is given by the Hamilton operator H.82) (21. (with ft = 1) d/dt\ip) = —iH\tp). (21. The relation between Heisenberg picture states and Schrodinger picture states is \ipt)s = e~lHt\ip)H.t)} constitute a "moving reference frame" in the sense that ip{x..81) Comments on signs in Eq. H.t). See also L. as had to be shown.80) (21.e.e. . i.83) (21. so thattt d — x.If '/'(x.£) = ( x V ^ ' .t).t'. i.e. (21.i'x. t\ipo)s. at initial time ti to its final position xy at time tf. (px) = e~ipx.e.x. so that \if>t=o)s = H>)H.6 Path Integrals in Dirac's Notation therefore imply 525 (xp) = e ipoc . Then the states {x. (21.We now introduce the Dirac bra and ket formalism..t).t\ = iH{x. (21. i.79) Feynman's principle is expressed in terms of the propagator or transition function (x'. t' — t. M^ip(x. Then ( X  ^ . which describes the propagation of the particle from its initial position x. d/dt\x.p). x' — x infinitesimal.i) = e iH4 x.t).0) = e""x). We consider first an infinitesimal transition. i.t) = —iHip(x.t) = +iH\x.78) Pp) = pp) is to be interpreted to say: The quantity p is the eigenvalue of operator P with eigenvector p).The time development in the Schrodinger picture is given by \tpt)s = e~lHt\4>o)s. ) H = (x</>o)s.t\.i) = K(x'.76) These relations imply the completeness relations 1 = J dxx><x.^ l x ) . Hence for the infinitesimal transition (x'. pp.e. The equation 1 = J ^p)(p. (21.21. t) = (x.t> =iH\x. Similarly we have a position operator X with Xx) = xx). 4) is to satisfy the Schrodinger equation.160. x. 159 .77) (21.t'x. i. Ryder [241].We define x. we must have d/dt{x.t) = e* H t x).
P V<P e l )  P > e P X <2184 = /(w/(^ ' ' *" '"'' *=£Uv«. 2mo Since (21.x). i. Pq) = J 7^3Pp)(pq) = J dppp>*(p .x') + V{x)5(x' .* ..85) the amplitude or expectation value or position space representation (x'#x) = .89) Using this and previous relations we obtain for the Hamiltonian (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).526 CHAPTER 21. (21.p'). (21.y<&(p'ix')(x'ivix)<Xp) = f dxe. 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.90) (P'IVIP) = y<&.V 2 < 5 ( x .i p '.e.x y(x)e i p .x = V(p .91) .88) Thus P has the operator representation P= /dxx)V(x.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). "  » We suppose that for a single particle of mass mo moving in a potential V(x) the Hamiltonian is (21. (2187) (21.
Thus with •"•/ — ^ n > *•/ — ^ni ^ 0 — Xj.t) (cf. EQ — Hi . tf subdivided into infinitesimal transitions (xj+i.j. (x'. We now consider the transition of the particle from Xj. t).82)).95) when applied to the bra vector (x. (21.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 '.tj+i\x.t') \^(27rf5(p 2mo +V(pp' where we used (21.e.p'). We have <x'. i.i'x.p') + V(p . for the amplitude of an infinitesimal transition.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 .92) We now return to the expression (21.6 Path Integrals in Dirac's Notation we have in momentum space the representation (p'ff p> = ^(27r) 3 <5(p . the signs are reversed when these operators are applied to the ket vector \x. i. But e'ipx + 0[{t . 527 (21.93) + ipx P r ip(x'x) iH(t't) (2TT) 3 (21.p) + i{t .21. ti to x j .fx.p') • • • j W < 5 ( p ' .84).tj). (21.92).94) In the above we have chosen the signs such that p = —iV and H = i— ot (21.
ti..i/xj. d x n .xk (21.tf\Xi.94) and thus obtain a phase space expression. We now insert for each of the infinitesimal intermediate transitions the amplitude (21.tixo.. x f c ) = ^ .X n _i)] h) (21.528 CHAPTER 21.98) tk — *fcl # ( p f c .. . d x „ _ i .a — —e/2mo and divide b o t h > sides by 2TT. .tj) = / ••• / d x „ _ i d x n _ 2 .tfei) fe=i .x 0 ) + p 2 • (x 2 . t n  x n _ i . x i .+ V(x f c ).x2)(t2 We can rearrange the factors here in the following form c„=x/ n—1 (Xf. . x n which the particle passes at times tQ.x i ) H h pn • ( x n .tn respectively.i /Xj=x(*i) f J [ ^ 1 dp2 3 3 J (2TT)3 (2TT) . i.xic) by fc — Xfej H(pk. i. The resulting relation is m0 2irieJ 1/2 m0x exp ze ) " / 2TT exp Pk 2 mo 1 (21. .e.xn)(tn in_i)].dpnl dpn (2TT) 3 (2TT) 3 exp [i{pi • (xi .. x i ) ( t i .t0>(2196) We define a p a t h in configuration space by a succession of n + 1 points x o . (21.ti) rx/=x(i/) d x i dx2 .ti) = x 0=Xi n^n fc=1 ? <^Pi (2TT)3_ exp / ^fa . x exp [ . d x i . dx. zmo T h e n t h e integration over all p ^ can be performed by using Eq. T h e integration over the n—1 intermediate positions.e.27) in which we make the replacements x —> eifc.2. .99) . . .tf\Xi....i { f l " ( p i . then corresponds to the sum over all paths.t 0 ) + H(p2. . t n _ i ) t„2><xi. d x i ( x n .. (Xf. = 1 X X<jPfc We now replace H(pfc.97) H hH(pn. The Feynman Path Integral Formalism we obtain (it will be seen below why we do not introduce normalization factors here) (x/.
tf\xi. i.100) Hence with all tk — ifci = e we obtain /n dpi i _(2vr)3 exp •.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. .p)]\.e. (21. to / m0 \ 3 / 2 las. > Thus the phase space representation of the pathintegral* (21.103) is another form of the pathintegral for which the integration with the help of Eq.] ex H ^I (.101) Then Eq.e.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.. Truman [254].21. m& ze ) J (2vr)3 exp it p fc • xfe IP* 2 mo (21. / {xf. (21.98). (21. G.tfci) fe=i £(**'**){ p***^} —tki m0±2k (21.fc=i 3n/2 r n m0 exp 2_^i{tk 2m(tk . A. G. i.98) can be rewritten as 3n/2 2^") x / n dxfe exp * ^o** ~tfcV rnl <i 2 m ° x fc ~ ^ (21.102) 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. Smolyanov.H(x. Tokarev and A. (21.ti) = J D{x(t)}D{p(t)}expi [' dt\pi.8) as N(e) so t h a t the integral does not vanish for e — 0.
pfc) = Pfc • xfc .tf\xi. (21. nl (xf. . pk) = L(xk. x. I hi dtL(x. Garrod [106]. We have now obtained the path integral representation of the matrix element ( X / . 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'.8).8).x) . £ /  X J . Clearly for a better understanding one also wants to explore the case of sandwiched operators in the path integral representation. t to x. (21.tf\xi.x) (21. We know from classical mechanics that this relation is a Legendre transform which transforms from variables x. t. Thus finally we have Feynman's principle: (Xf.H(xfe. operator (21. Note that we have n factors [iV(e)]3 but only n — 1 integrations dxk.V(xk) = m0±l .t'}.ti) where nl D{x(£)}exp '. £ J ) with no operator in between.iJ(x fc .ti) = ^lim^ N(e)3n / * JJcbc fc exp i / fc=i * dtL(x.e. p .104) where for the particularly simple Lagrangian L under consideration here m0±l . xfc).106) In order to be able to deal with this expression we first require the momentum operator representation of the operator x(t). i.530 CHAPTER 21. (21. We check below that this is given by x = / ( ^  p > i  ( p   <2L107) See also C.105) D{x} n ^ o o [7V(e)]3n Udx kk=\ In this notation the measure factor is contained in ©{x}. The Feynman Path Integral Formalism yields automatically the correct normalization factor or metric in the configuration space path integral^ in agreement with Eq. this agrees with the corresponding discrepancy in the onedimensional case of Eq.
With contractions and integrations over delta functions the expression for the matrix element then becomes <xVxx'. x fdxG  ^ P <p"x3)<x3e^V) x (21. ••• y (2^ x= f dp p) ^(p d We now consider an expectation value in which we replace the operator x (which is sandwiched in between exponentials containing Hamilton operators) by the expression (21.109) . so that we obtain a cnumber position coordinate which we can shifted across other cnumbers.108) (x 4 p)^^(px 5 )(x 5 e^ t 'x 6 )(x 6 p / )e.0 = J'^ j'^e^''(p"\e^''xe^'\p')e^'.6 Path Integrals in Dirac's Notation For the verification we use Eq. .76) we can rewrite the factor (x4p)^(px5) as (x 4 p)x 5 (px 5 ).*') = / I F / (2?FeiP"'x" / ^ 5x5 (p // K^> 5 )(. (21.t'}. With the wave functions (21.t"\x\x'.5 l^ t 'lp / > e ./ ^ I < « . (px).107). (21.* * . Thus <xV'xx'.t'} = = fdx5x5(x"\eim"\x5)(x5\eim'\x'} f dx5x5{x". . .77) and thus have 531 *!*') = *M = * 7 ^ I P > « .* * f dp = j (2^ p) ^ d .v ' x ' and thus {x".t"\x5){x5\x'.21.i p '• x .
we obtain with (21. t0).x ' ) (2^ i^+vM + D «p(x"x') (21.t'){x"\H\x') = X + •••} (21. We now consider the transition of the expectation value of operator x from Xj.• dx1 (x n .tf subdivided into n transitions (xj+i.85) (xVxx'.tj+i\x\xj. = x"6{x" . ] .90) we can rewrite this expression as (x".x') .£ n xx n _i.t n _i) x ( x n _ i .74) of the delta function.110) Recalling Eq. .x') .t"\x\x'.f ) i f e i p .94) for the case with no operator in between. (21. tixx 0 . . Inserting here the integral representation (21.i{t" .ti = to as before.i ( t " .i(t" ..112) where x" is a cnumber. (21.i t " ( x "  #  x 5 ) ( x V ) +.113) Inserting for each of the infinitesimal intermediate transitions the amplitude .£/xxj.t') dp .iit!' 1') 2mo + vtf) + <5(x"x'). £ n _ixx n _ 2 . x / / e . This is the expression corresponding to that of Eq.tf — tn and x$ = XQ.£'> = x"[«5(x" .it / / x / (x"ffx / ) + it'x"{x"\H\x>) + • • • .t') (xV'xx'. Thus with x^ = xn.111) 1 .0 dp x„ / (27T): 1 . The Feynman Path Integral Formalism In the second last expression here we expand the exponentials and obtain (xV'xK.*/<x"x5)(x5tfx'> + .tj) at equal time intervals e. (21. (21. we obtain (here and in the following the limit n —> oo at the end being understood) (x/.532 CHAPTER 21.0 = and hence /dx 5 x 5 [(x"x 5 )(x 5 x') .£j) = / ••• / dx n _idx n _2 . tn2) • • • (xi. ( x " .£j to Xf. We now proceed from there along parallel lines.
..21..x2)(t2 ~ h) + • • • + H(p ••• + p „ .t"\x'.t"\x'.t"\x'.t') sider the variation S(x". For simplicity we consider the case of one spatial dimension.{^( P i. Xn _i X„ xi dx.x 0 ) + p 2 • (x 2 .x 1 )(t 1 Z & ) +H(p2. We demonstrate first that the operator representation of P given by Eq.\Xi. tj \ xt.iyxxj.t') (21.7 Canonical Quantization from Path Integrals (21.t"\x'.t') .y (27F (27F •" (2^(2^)3 exp[.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..2 .ti) 533 rxf=x(tf) = / • • • / dXidx. (21...{x".t"\x'.2 . tj) = xf (x/.tf\x.105). . U). the string of factors xi X2 .10) or (21. .112).x n _ i ) ] .89) is consistent with the conjugate expression dL (2L116 * . dyi l xi x 2 .t') = and con 5x"Vx„{x". n—>oo *•»• 21.( x „ . tj  xj1 x». dxnn _i Xi X2 .114) x exp [i{pi • (xi . x n _ i x n i JXi=x(ti) /Xj=x(tj) 7 .115) where (there is no integration over x n ) nl 0){x(t)}= l i m [ i V ( e ) ] . (21.t') = (x" + 5x". (21. Before we can consider canonical quantization in the context of path integrals we have to investigate in more detail the role played by momenta.117) "•"One may observe here that if x = x^ = x„ in the integrand of Eq. we obtain* (xf...114). x n _ i x n reduces to the single factor x^.xi) + f f dpi dp2 dp n _i dp n The momentum integrations can be carried out as before and we obtain finally (x/. x (21.and hence may be put in front of the integrals to yield (xf. .x .« • > We reconsider therefore the transition amplitude (x".tj) = / JXi 2){x(i)}x(i) exp i I I Jti dtL{.3 " n d x f c .
118) JI(x)/h . (21. Hence w « = (t).t"\x'.120) However: In our case here 8x ^ 0 for t — t" (although 5x — 0 at t = t'). and we demand the validity of the equation of motion.(x". For the purpose of enabling some formulations below.i/(x)/fi ) j ^ ) dtL(x. .118) — = (21 5x"(x".tt). dL d5x / oft—5x + / dt dx dt Jt' ox Jt t' and hence <9L —oxdt + ox ox nt t. (21. " M{x)/h &(t ) = fa (L2) 2 11 and — using Eq. <9L' dt —dx + — 5 x ox ox f . t". dL r /•*" .l&cdi.x).122a) = 17) 5x"Vx/.t"\x'. with N = N(t" . The Feynman Path Integral Formalism with 8x" = 5x(t") ^ 0 and 5x' = 8x(t') = 0. Thus in this case of classical mechanics dL_ d fdV. Then.t').t') =  6— / 1 V°{x}t . jt/ 37 ( £T. dx dt \ dx _ (21. " (t).t'): 1 fx" 6(x". we separate the metric factors from T){x} in Eq. (21.10) and set D{x} = D°{x}/JV(i/ . V°{x}[i6I{x)/h]e* Jx' The following steps are familiar from classical mechanics: yV 51{x) S /1 dtL(x.119) The classical equation of motion follows from Hamilton's principle for which 5x = 0 at t — t'.534 CHAPTER 21.t i_(dL_ x't' h\dx J t„ (21. x) Jt Jt dL. dt\Ox w(x) = r dt d^_dL(dL dx dt \ dx °L 8x + ^ T!5 X 1 ox (21.
125c) Corresponding to our relation (21.tf\xi. we now have in one spatial dimension {x^tflxlxuU) = J^ff'DQ{x}x{t)eiIWh.t')(x'.1 (A)iV. . 535 (21.125b) and the measure property (3) N~\A + A') = A^.t"\p(t")\x'.89) we see that Having explored the appearance of canonical momentum in path integrals we can proceed to extract the canonical quantization.1 (A / ). / Jxi V°{x} / Jx' V°{x}.124) where with A = tf — ti. 6 (21. § The evolution of the system is expressed by the Feynman functional integral (xf.21.t'\ = l. The following properties are assumed to hold: (1) / Jti V°{x} = ^ x.122b) By comparison with Eq. S.t'). x' (21.7 Canonical Quantization from Path Integrals or JL(x»y\x. We consider this here in quantum mechanics in one spatial dimension but parallel to field theory.125a) (2) the completeness relation holds. (21.x). F. (21.U) = T777TT / &{x}eaW\ IXi=x(ti) ) iV(A) JXi=x(U (21. I(x) = / JA dtL(x. ^2\x'.115) in three space dimensions.1>) = ^{x".Xf = x(tf). (21. Here f*f D°{x} indicates that the integral is to be taken over all x with boundary conditions Xi = x(ti). Wightman [264].126) R. Streater and A..
t').t"\x'. .ti) = —!— p V0{x}xx • • • xneu^h. (21.t')..t') \ i ox" i ox"J or in commutator form [P.t>). But for t = ti we have (and equivalently for t = tf) as we can see from Eq.t') = ih(x.t')=x"(x". The Feynman Path Integral Formalism The time t is in general a time between ti and tf.129) = ^ y With Eq.t"\x'.. (21.ti) = x' 'Y^{xf. We can rewrite this expression as the commutator relation (ITU*" x"^](x". (21. (21. (21. (21.127)) (x".tf\xi\x'.133) = ih{x". Thus (xf.tf\T(xix2)\xi. Then the relation follows with the help of the properties (1)..132) The generalization to higher spatial dimensions proceeds along parallel lines.t"\x'. We consider the case n = 2.ti) = Xi(xf.127) We can now write down an expression for a timeordered product of operators. tn lie correspondingly in consecutive order between U and tf. We have (cf.t'){x\t'\x2\xi.536 CHAPTER 21.130) ~i5x7jX"{x".ti) f ' V0{x}xlX2eu^lh.tf\xi. (21.ti).128) where tf is later than ti and t\. we have the relation (xf.t"\x"\x'.t\x'.t"\X'. Eq.t').t\x'.t"\x'.122b) we can deduce the nontrivial canonical commutation relation arising here.t') (21. Thus if T denotes such ordering.114) (xf.tf\x(ti)\xi.t') + x"^J^{x".tf\T(Xl • • • xn)\xi. Taking the functional derivative 5/iSx" of this equation we obtain 'ilx7'^' '^"^x"\x ''*') = (21.x](x. (21.t"\x'..131) = ih(x". (2) and (3) above.
the modulus of the ratio of the (r + l)th term to the rth term is larger than one.1 Introductory Remarks In field theory a perturbation expansion is generally understood to be an expansion in ascending powers of a coupling constant. Whittaker and G. In field theory it has been known for a long time that expansions in ascending powers of a large coupling constant are fairly meaningless.g. nonzero radius of convergence in the domain of the coupling parameter. although the first few terms indicate a convergent behaviour.e. the series converges.g.Chapter 22 Classical Field Configurations 22. E.e. it has some finite. z = a + i/3 with a — oo or f3 —> oo or both. If the expansion is mathematically well defined. integrated over) beyond its radius of convergence. N.e. The vast majority of perturbation expansions discussed e. It was therefore natural to search for alternative methods of expansion. if a function is expanded in the neighbourhood of an essential singularity* or if an expansion of the integrand of an integral representation is used (i. i.e. i. In physical applications such cases are rare. but asymptotic. in the sense of rigorous convergence tests. T. for sufficiently large r. Expansions of this type arise. i.g. in the context of quantum mechanics are not convergent. see e. 102. as we have seen in Chapter 8. Such expansions are strictly speaking. W = e x p ( l / z ) . the exponentially small level splitting which occurs in "The expansion involves both positive and negative powers of the deviation. divergent. such as the fine structure constant of quantum electrodynamics. > 537 . Such an expansion parameter is usually known to be or assumed to be small in some sense. p. Watson [283].g. Besides. it was known from quantum mechanics that RayleighSchrodinger perturbation theory by itself does not yield e. An essential singularity is really an ambiguity rather than an infinity. Consider e. so that successive contributions to the first approximation do not invalidate this approximation.
g. cnumber first approximations. There.. the procedure discussed thus far) to systems with constraints was developed by Dirac [76] and is introduced in Chapter 27. In searching for other means of expansion.e. 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. 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. 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). higher order corrections are of orders ±. such as soliton theory. In the following we consider various typical examples. although this terminology is not precise.. However. These methods are often described as methods of collective coordinates. This type of expansion is decsribed as "semiclassicar. 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. A method which achieves this is in particular Feynman's path integral procedure.. and the corresponding expansion is called the "loop expansion". i. In particular it enabled nonlinear problems to be studied and led to a consideration of topological properties.h°.h.h2. cnumber. and has led to insights which previously seemed unimaginable. A further challenging task was the development of methods of quantization of theories which incorporate classical. such > a series begins with a contribution proportional to (e. i. 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. One therefore faces the problem of quantizing a system which is subject to constraints.. The classical.538 CHAPTER 22. Classical Field Configurations the case of the symmetric double well potential or in the case of a periodic potential.e. the dominant contribution is singular (has an essential singularity) for vanishing semiclassical expansion parameter (h — 0). The fundamental extension of the method of canonical quantization (i.) l/h2.e. and in general this change is accompanied by certain constraints. 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.h3. Many of these aspects are interesting by themselves . On the contrary. the latter's topological properties if any. nature of the dominant approximation does not imply that this describes classical motion.
may like to consult the article of D. It is evident that symmetries and their violation by the classical approximation play a significant role in the entire consideration. V{</>) = m§0V + A0(<^)2.22. Thus references to field theory will be exceptional and hopefully do not irritate the reader. .2 The Constant Classical Field 539 and in any case deserve detailed study.* assuming that this is acceptable to a reader with some familiarity of the basics of the more complicated field theory of electrodynamics.1) ' T h e reader who wants a highly advanced overview of instantons. it is important for a better understanding of the considerations which follow. to keep this case in mind.g. 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). and also to see these in a somewhat broader context (by comparison with higher dimensional cases). We begin therefore with a brief recapitulation of the simple Higgs model which exemplifies this case and exhibits the phenomenon known as "spontaneous symmetry violation" or the Goldstone phenomenon. 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.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. 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>. monopoles. a static (time independent but spacedependent) solution of the classical equations of motion. 22. The important aspect we want to draw attention to here is that of "spontaneous symmetry breaking". modified by going to imaginary time).V(4>). (22. or even a solution to a modified field equation (e. We shall not be concerned so much with the case of a constant first approximation. d^] = d^d^ . t The classical first approximation in the procedure outlined above may be simply a constant (time and space independent) quantity. vortices and kinks. Tong [272]. This means theories with field densities which are therefore infinitedimensional. However.
The EulerLagrange equation is d„ dC d(d.g. for m§ < 0 the potential is a double well potential with two minima at \<f)\ ^ 0.540 CHAPTER 22.2) We ask: Are there classical (i. V ) . 22. (In models of the early universe one often considers V(0) with m\ > 0 at the early stage.S£/<90(x. B. Felsager [91]. p. as illustrated in Fig. of the equation of motion? For these we must have all derivatives of <j) zero. § (22.1 Different potentials for different signs of m§.£). m l + \04>*4> )$ = <). 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).1. then after cooling with a phase transition one considers expectation values 0 7^ 0 corresponding to minima at \<$>\ ^ 0. . Classical Field Configurations where d^ — (—d/d(ct).e. cnumber) solutions <j)(x) = 4>c = const.e. i. 433. For UIQ > 0 the potential V((f>) has a single well with minimum at </> = 0. 22. £)) is bounded from below we must have Ao > 0.3) See e. To insure that the Hamiltonian H = f dxH with Hamiltonian density 7Y(0(x. and hence here [d^ + m20}(l) = Xo(f (22. V[<t>] m 2>o o Fig.
TT = —J. w A iP 3ml 5* "XT" (22. aq> For (b = const. and so ft[<M=m6:(^* 1 + Ao(0*0)2. But for Ao > 0. 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>. 22. The U(l) phase transformation <b — exp(ia)((> leaves both £ and the EulerLagrange equa> tion unaffected. H = —C = V.e. The Hamiltonian density H is defined by the Legendre transform H[<f>. the equation has this invariance. 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.2 when allowed to fall chooses an unpredictable phase parallel to a radius in the (0>i.5) . (22.4) Here (3 is a spontaneously chosen phase like a stick held upright along H in Fig. mg < 0. i.</>2)plane. We can convince ourselves that <fic is the field associated with a state of minimized energy. the only such solution is the case we wish to consider. not the solution x(t)). we have 541 0.2 The Constant Classical Field For Ao > 0. Thus the classical solution 4>c violates the U{\) symmetry of C and of the equation of motion.22.ir]=7r<fiC. whereas the field (bc becomes (a) A oi(0+<x) V~2 + Every new phase defines a different solution. 22. Fig.e.2 The spontaneously chosen phase. TTIQ > 0. with 0 = 0.
(22. (22. the one for (3 = 0. where (f>c = — T](x) = —= [Vl (x) + ilp2 (x)} • (22. For every value of the phase (3 we have a different constant cnumber field configuration 4>c. Clearly these are the field configurations which trace out the circular bottom of the trough of the double well potential.7a) (22.tp2. 22. &H/d(fi = 0) if [m20 + ^>]</> = 0.e.7b) the constant on the left and the coefficient of A on the right of Eq. 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. one obtains the equations (d^ and (d^ + ml + ^ A 0 A 2 V = 0 ( ^ 2 .e.3ml = 2ml = ml + TA 0 A 2 = 0.3). Suppose we choose one such configuration. i. i. V?) (22. we set cf>(x) = <f)c + ri(x). (22. ^ ) o + m 2 + 1A 0 A 2 ) ^1 = A (± A0A2 + m 2 ) + 0 ( ^ 2 . e.2.3) obtained above.6) Inserting this into Eq.e. as indicated in Fig. Classical Field Configurations the density Ti is minimized (i. and we wish to investigate the behaviour of the field cf> in its neighbourhood.7b) In Eq.7a) vanish on account of Eq.2) and separating real and imaginary parts.Then m\ m\ = ml + A0A2 = m2. This is the condition (22. and partly by climbing up the parabolic wall on either side.542 CHAPTER 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). i.e. >° for m 0 < °> . (22.g. in a transverse direction. along the trough of V (which we can call a longitudinal direction).
22. tangential to the classical path. which. Like the Goldstone mode in the above Higgs model it leads to a divergence in the Green's function of the theory.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. has to be removed in order to allow a well defined perturbation procedure (i.3 The spontaneously chosen phase seen from above. There the wave function with an associated vanishing eigenvalue is called a "zero mode". whereas ipi(x) is y ' " " " " " ^ . 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. i. of course. 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. irrespective of the question whether the perturbation series as such converges or not. Thus the Green's function is similar to that in electrodynamics. (22. the existence of individual perturbation contributions). The appropriate Green's function G is the inhomogeneous solution of the the equation [8^ + m20]G(r. 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. The elimination of the .r')S(t . Goldstone's theorem applies to fully relativistic field theories. though not identical phenomenon. 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 £. r'. 22. ^ = h eip / \ ^2 l .t').e. t') = d(r . t. then a massless boson exists. We can see the problem here by looking at Eq. (3=0 Fig.e. In our later discussion of theories with nonconstant classical field configurations we shall encounter a similar.7b).
equates to zero the coefficient of this wouldbedivergence. A. 7711 = — 1 ( 1 6 Minkowski manifold). The Lagrangian density is taken to be C[^. and we choose the metric ?7oo = + 1 . W. MullerKirsten [291]. 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.£) in one spatial dimension. F. In the following we will not be concerned with classical cnumber solutions of the EulerLagrange equations which are simply constants. Lohe [179]. however. instead we shall consider static and timevarying solutions in important models of one spatial dimension. C. N. Behera and A. t Potentials of this type have been discussed by M. We make one more assumption concerning V which. Soliani [237].* We consider the theory (later: quantum theory) of a real. 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. see e. Zimmerschied and H. The principal aspects will always be very similar so that it really pays to study simple models in considerable detail.* Potentials of higher polynomial order in $ than $ 4 can also be considered. 22. spinless field $(x. Later we wish to develop a perturbation series about a classical configuration of $. . so that we are interested to have a parameter *For a collection of many informative papers on solitons etc. zero mass configurations and constraints even before the question of quantization of the fluctuation field T){x) can be considered. To insure that the energy (see below) is positive definite we require V[$] > 0.8) where V[<&] is a selfinteraction or potential of the field (to be specified later).d^} = ~d^d^V[^) = ^2^(^j V[$]. we shall need only at a later stage. (22.3 Soliton Theories in One Spatial Dimension We consider here the two basic soliton models known as <J>4 and sineGordon theories.544 CHAPTER 22.g.t In particular the sextic potential can be considered along parallel lines.e. i. J. *S. Khare [17]. Rebbi and G.
12) we see that they satisfy the Newtonianlike equation <j.e. i. In a quantum theory the field $ is an operator defined on a space of states. at least in the present case.3 Soliton Theories in One Spatial Dimension 545 which serves as the expansion parameter of the series.l]. . For this purpose we assume that V[$] depends on a scaling parameter g such that Vm = V[t>. The other condition which we impose (and study in detail) is that of stability. In fact to begin with we restrict ourselves further and consider cnumber fields <> —• (f) which are static.e. (22. Since $(x. (22. but in such solutions which are subject to (ignoring the dimensional aspect) reasonable physical conditions.g] = \ v m For instance.9) with quartic potential is defined by ^V\g$\ = ^V\g*. (22. ^ ^ * + V'[*]=0.l}. (22. 9 9 The EulerLagrange equation is seen to be (with c = 1) D * + V'[*] = 0. — m$ 1 — cos Im 1 I ^ $ (22.§ Unstable classical configurations are also important and will be considered in later chapters. The first such condition to be imposed is that the energy of the solution of interest must be finite.22. From Eq.t) has an explicit timedependence it is a Heisenbergpicture operator. 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). The field $(x. the socalled ^theory = ^V[g$. the reason for this condition being that we visualize the classical solution as representative of a lump of energy.13b) We are not interested in just any solution of this equation. (22. 1].«{x) = V'(<p). i. Before we consider quantum aspects we study classical cnumber fields which satisfy the EulerLagrange equation.13a) We write such fields <j>(x).10) n*]  m4 r 25 i m2 and the sineGordon theory with cosine potential by ±V[g$} = \v[g$.11) (22.12) where the prime denotes differentiation with respect to $ . such that J > — — — = 0.
17) Since we have to integrate over all space.e. <9$ i.4 The wellknown soliton of $4theory. dC • Too = r?oo£ + =r$ = C + vr$ = H.e.2\dx 2 ' 1 fd& + V[$].7r) where IT is the momentum conjugate to $ defined by dC ? r = . We observe here already that the integrand of the integral in Eq. 22. i.14) where the overdot implies differentiation with respect to time t.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>. Classical Field Configurations The energy JE?[$] of the system is defined to be the spatial integral of the Hamiltonian density 7f($.T = <I>.T^ = 0. the classical solution will therefore have to be such that this integral is finite.16) In the static limit we obtain therefore and write this as Em = H[<t>\ + E(4>) = fdx 1 2 \dx 2 + V(</>) (22. The energymomentum tensor T^ and the conservation law it satisfies are given by the equations dC T^ = n^C + Q^^d^.c = <£z + . (22. (22.546 CHAPTER 22. HI tanh m(x>k) 9 Fig. n = & . d. (22. (22.15) The zerozero component is equivalent to the Hamiltonian density H. .
See J. In the sineGordon theory^ the classical or soliton configuration is (see below) found to be (x) = ±4— tan" 1 with energy 777 . it is not difficult to find classical cnumber solutions 4>c to Eq.13b) for the specific potentials (22.4.3 Soliton Theories in One Spatial Dimension 547 As we shall see below. The energy is correspondingly obtained — as will be shown — as 4 m & 9 The typical shape of the configuration given by Eq. Again XQ is an integration constant.e. 22.x n ). m ( x . — XQ).777. The sineGordon theory differs from the KleinGordon theory in that it possesses invariance under the shift rf> — <f> + 2n". i.20) (22. > .21) This sineGordon configuration is depicted in Fig. 22. (22. < = — tan [e >  ° ] 9 2 rrm/g — Fig. . Rubinstein [240]. O±m(xxo) (22.22.1 .5 The soliton of sineGordon theory. the $*theory.10) and (22.18) is depicted in Fig. in the case of the first. "The name sineGordon is derived from the analogous KleinGordon equation or theory.11). one obtains by straightforward integration (but see also below) the soliton configuration m 4>c{x) = ± — t a n h 777(2. In fact. 4m . (22. around <j> = 0 the sineGordon potential behaves like the KleinGordon potential proportional to 2 4> . 22.5.18) Here XQ is an integration constant. (22.
Classical Field Configurations The solutions (22. or \(<P'? = V(4>)+ const.o) =m7r. Here the constant is zero for V(<p) and <j>' zero at x = ± o o . .548 CHAPTER 22.1 "»JtKxo) 9 1 . <^>(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 .e.10) we obtain f<t>{x) x — XQ i 9 ! 1 = ± / J<j>(x0) dtf> 1 .13b) which we can write d \W? = !"(•». ± 1[tanh.18) and (22. (22.* ( m' m 2 r)] ^2" .20) are obtained from Eq. In the case of the sineGordon theory defined by the potential (22.^^! m ^m <p(x0) m 4>(x) = ±—tanhm(x — XQ). 4m In tan tan 5 l/e±m(*xo)tan^£^)U. and hence 4/71 9<t>{x.9 2 4>{x) = i. Thus J<t>(xa) \/2V(6) JXOJxn U(x0) y/2V{</>) dx = x • XQ. 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. With we obtain g(/>(a. 0(x) = ±4—tan" 1 j e * " 1 ^ " * 0 ) ! .
Then we have to demand that the functional derivative of the energy E(<p) be zero. (22.17) 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) . i. (22. Suppose now that we demand stability in the sense of minimized energy. we obtain these expressions later (cf. 22.17). 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.4 Stability of Classical Configurations The concept of stability can be complex.23) See for instance R. in the onedimensional Ising model. however. However. the others as antikinks.22. The relation (22. Jackiw [141]. .22) and that its second variational derivative be positive semidefinite. the discussion of Bogomol'nyi equations) in a simpler way. the curves rising monotonically from negative to positive values are known as kinks. SE((f>) 5cb(y) 0. This is called more precisely "classical stability J The EulerLagrange equation is obtained by extremizing the action or Lagrangian.4 Stability of Classical Configurations 549 Prom their monotonic behaviour the solutions <f)c in these cases derive their name as "kinks". A state of stability is therefore associated with minimized action and/or energy.21) for the energies can also be obtained from Eq. (22. this does not require the action to be minimized.19). 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. (22. instead of doing this now.e. The expressions (22. for instance.
A theorem on zero modes Let S[U] be the action of some field U defined on Minkowski space. i. we see that the latter has one eigenfunction. its eigenvalues w\ > 0. In fact.e. Classical Field Configurations Since y ^ ±00. the function 4>c(x + a). in particular the invariance under spatial translations. namely (p'c. with eigenvalue w2. The Newtonlike equation (22.25). of course. (22.13b) with classical solution 4>c.550 CHAPTER 22. i.25) Before we begin to study this equation in detail for specific potentials we can make a very important observation on very general grounds. if we apply the generator of spatial translations d/dx to the classical equation. (22. (22. = 0. in fact.13b).13b) retains the invariance although the solution (f>c does not. it is also called a "translational mode". Thus we have to investigate the eigenvalue spectrum of the Schrodingerlike equation d2 + v"{<t>c ipk(x) = wkipk(x) dx2 (22. Thus the vanishing of the first variational derivative yields again Eq. The eigenfunction 4>'c{x) is for this reason called a "zero mode". dx2 + V"{4>) (22. in fact ip0(x) const. and the EulerLagrange equation retains this property. the integrated contribution vanishes. (22. (22.24) For socalled "classical stability" the differential operator of this expression has to be positive semidefinite. Since it arises from the violation of translational invariance by 4>c. The original Lagrangian was.e. if 4>c{x) is a solution. 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. This is. we obtain 4'(x) = v"(4>c ^ or d2 dx2 4>'c(x) = 0.26) Comparing this equation with Eq.28) . written down in Lorentzinvariant form.A.e. Eq.a / 0 is not the same.27) shows that the zero mode results from application of the generator of translations to the classical cnumber field configuration. Assume that S[U] is invariant under the transformations of some symmetry group G with elements g = exp(i\T) € G. a very general phenomenon which can be established as follows. Proceeding to the second variation we obtain at 0 = <fic the relation 52E((j>) 5<j){x)5(t)(y) S(xy). dx (22. i.
33) x—>±oo .27). 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. We see therefore that the occurrence of these zero modes is a very general phenomenon. we are interested in classical cnumber field configurations of finite energy. in fact.31) Inserting this into Eq. Invariance of S[U] under transformations of this group implies S[U} = S[gU]. lim [V(<f>c) + const. We can choose the potential V such that the constant is zero. We return to our considerations of stability. 0 = S'[UC] = S'[gUc}. and that S[UC] is finite (or equivalently the energy). As stated above. i.e. (22. (22.4 Stability of Classical Configurations 551 where T is a generator and A a parameter.20) are such configurations in two particular models. (22.] = 0 or x—>±oo lim V{(/)c) = 0 for const. In the onedimensional case under discussion we can rewrite Eq.22.32) = V'(</>).18) and (22. = 0. Then the vanishing of the first functional variation implies that 5S = 0. we obtain 0 = i^S'[exp(iXT)Uc]\x=o = S"[UC](TUC).29) Suppose now the EulerLagrange equation possesses the classical cnumber solution Uc.30) Differentiating this with respect to A (applying —id/dX) and setting A = 0. Thus TUC is a zero mode. (22.e. (22. it is an expression like (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. (22.13b) as ~d~4> > ' The first integral of this expression is \(tfc? = V(<l>c) + const.17) we obtain oo dx[V(4>c) + const. / oo Thus for E(4>c) to be finite. (22. and we have seen that Eqs. i. this integral has to be finite.]. (22.
V{4>) = 0 at different values of </>). ± l . Thus the classical energy or vacuum configuration is not unique. » In the <fr4theory defined by Eq. (22. .10) we must therefore have that for x — ±oo: 4>c — ±m/g as indicated in Fig. Classical Field Configurations V Fig.552 CHAPTER 22. 22.6. (22. the potentials have degenerate minima (i. For a configuration to have finite energy. 22. . n = 0 .1). 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. oo J—oo y The topological quantum number q is defined by 0. in view of the antisymmetry of eM^.c o and another for x — oo. . .6 The minima of V in $4theory at x — ±oo. Solutions which approach > > the same minimum for +oo and —oo are the constant solutions <f>c = ±m/g in the $ 4 theory. In the case of the 3>4theory we can define a conserved current k^ by (22.11) we must have correspondingly for x — ±oo: » 771 9 2vrn. The charge Q is therefore given by dxk° = / dxe01di(bc = [ ^ ( s ) ] ^ = ±—. .e. ± 2 . In the sineGordon theory > • defined by Eq. This number is defined like a charge in field theory by the spatial integral of the time component of a current.34) jfc" = e^dv(l since d^k^ = 0. the classical field (bc must approach one minimum for x — .35) and it is seen that in this case q can be ± 1 or 0. (22.
cos(27rnA)]. 2g Thus even without varying E with respect to A we can see that E(X(f)c) is finite only for A = ± 1 . 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. and we call this "topological stability.A2)2 + 0 except for A2 = 1. In the case of the sineGordon theory the minima of the potential (22. if it remains unchanged under continuous deformations of the (field) configuration while preserving finiteness of the energy. Such a smooth deformation (in the sense that exp(i%((2))<^c depends smoothly on 6. continuous values. zero is not a meaningful topological number. x—>±oo so that under such deformations the configuration remains in the same topological sector defined by the boundary condition. ± l ± 2 . Again we see that E(\<f>c) is finite only for A = ± 1 . Tfl \ (22. A property is. Then for x — ±oo: > > > V(\4>c) (2 ^ 0 ) ? 4z( 1 . one may ask how topology comes into the picture.11) are given by ( so that for x — ±oo: > — J.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. . .22.7r) does not change a boundary condition lim [\<t>c{x)\ = (/>c(±oo). in general.4 Stability of Classical Configurations 553 Since we call q a topological quantum number. For such configurations we have the energy oo 2 2 dx ^ A ( ^ ) + y ( A ^ / (22.37) for the onedimensional case here. The difference Tfl 4>c(oo) — 4>c(—oo) = 27T—An . termed topological. We can see this as follows. .7T (22. Thus we consider the family {X(pc(x)} of (field) configurations where A is some parameter which assumes real. n = 0 .38) 4 V(\<t>c) >m ~2[l . First. and exp(ix(0))<pc reduces to 4>C when x(#) = 0. .
Classical Field Configurations involves the difference An = rioo — n.40) (22. MiillerKirsten [215].554 CHAPTER 22. the energy is minimized.39) E(cp) = J — 0 0 we can write / dx oo ~1 2(0'(x)) 2 + y(^) (22. Eqs.13b) or ^ ' ( * ) ] 2 = V(</>).33)) that the classical configuration cf)c is the solution of the nonlinear second order differential equation (22. (22. its integral is nonzero only on account of nontrivial boundary conditions.17)) f 00 > 0. (For a broader view we consider briefly in Sec. W.43) **See e. We observe (a) the topological current k^ of Eq.g.428.31) to (22. (d) the topological quantum number iV arises in a completely classical context.41) dxy/2V{$)4>'{x). H. pp. 00 (22. where n±<XJ are the integers n corresponding to the minima of the potential at x = ±00.e. i.5 Bogomol'nyi Equations and Bounds We have seen previously (cf.^ = N. 22.42) Thus the energy has a lower bound given by the right hand side of this inequality. Thus with one space dimension higher there.** (b) the time component k° depends only on <p but not on momenta (again in contrast to the case of Noether currents). J. (22. From this we construct the inequality [</>'{x) = Vm^)}2 F Since (cf. 425 . if oo / dxy/2V{<l>)<f>'{x). the factor A = exp[ix(#)] varies over a complete circle). (22. The inequality is saturated. 00 (22. Eq.34) is conserved independently of the equations of motion (other than Noether currents whose conservation follows from the equations of motion). 22. (22.9 the foregoing arguments in a higher dimensional context. (c) the time component k° is a spatial divergence. .
39). . for instance. 555 (22. Here (V(c/)) > 0. 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. (22. (22. In the sineGordon theory we can consider. (22.n > 1. B.40) is called the "Bogomol'nyi equation" since it was first introduced by Bogomol'nyi.5 Bogomol'nyi Equations and Bounds We observe that (cf.21). 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.44) This first order equation which saturates the inequality (22.22. Bogomol'nyi [27].19).40)) this occurs when 4>'(x) T V2V(4>) = 0. Eq. the set of configurations (fic which interpolate between 0(—oo) = 0 and <f>(oo) = 2nnm/g. 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.^ Clearly it solves the second order equation or (22. 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— . We discuss this in the following Example. n E .
ka —> K. rotate) only in the (y. We can now write down an expression for the energy of such a system of (say) n pendulums. Classical Field Configurations Example 22. Solution: Consider a string along the xaxis as described.e.t) 4>{xn. from the yaxis as shown in Fig. * / dx OO + 2^ V dt 1 K (d<t>\ 2 \7T I + ^9^(1cos 4>) (22. Initially the pendulums are suspended freely in the gravitational field. Construct an expression for the energy of such a classical system and its Lagrangian. rotate) only in the (y.t) <j>(xn.7.556 CHAPTER 22.7 The pendulum analogy. all being identical.g. t)).g.ij\2 Thus the The potential energy due to gravitation is (maximal for <j> = 7r) mgr(l — COS <f>{Xji. Fig. 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. Chapter XI.46) **See e.i)) The continuum limit is obtained by t a k i n g ^ (since xn = na. H. Assuming that the pendulums can move (i. 22. z)plane and only such that their elastic strings remain taut.2 V dt + k[(f>(xn+i. so that oo > u. total energy of the system is with the number of pendulums allowed to become infinite *= £ i ''( "' ) .1: Classical analogy to sine—Gordon theory Consider a string along the xaxis. } n= — oo roc —> / L J .t)\2 +mgr{\ . Goldstein [114].cos</>(x„. . 22. with mass m and connected with short strings to neighbouring bobs. 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. xn+\ — xn = a) 00 a —> dx. Assume that the pendulums can move (i. V dt „ 1 (d4>_ dx • i^gr(l — cos(j>) (22.e.t) 2 V dt The energy of interaction with neighbouring bobs is (recall "fcx 2 /2") k[(f>(xn+i. At equidistant points xn = na along this string we attach strings with pendulum bobs.t) measured e.45) We can therefore write the Lagrangian L=TV• r dx fir P 1 2 .
22. t) of Eq. we see that ipk(x) has to obey the equation 92 dx2 +V"{(t>c{x)) ^ fc (x) = w^k{x).25) which we obtained as the condition of classical stability for eigenvalues w\ > 0. It is instructive to obtain this equation by yet another method. and so would invalidate the procedure which assumes that rj(x. (22. wk would be imaginary and so the factor exp(—iwkt) in rj(x. This number. Inserting (22. t) is the fluctuation (field). each finite energy configuration beginning and ending with a classical vacuum corresponds to an integral number of rotations about the iaxis (i.t): {w With the ansatz r)(x.22.49) would imply an exponential growth in the future t > 0 or in the past t < 0.12) we obtain d2 d2 \ Since (j)c{x) obeys Eqs.25) obtained earlier.47) into (22.t) = <l>c(x) + r}(x.6 The Small Fluctuation Equation 557 Thus if the classical vacuum corresponds to the case when all pendulums are pointing downwards. the topological quantum number. (22. (22.49) exp(iwkt)ipk(x). ' Equating coefficients of exp(—iw k t). (22. is therefore also called "winding numbed. We can understand what stability means in the present case. Hence we set ${x.50) This equation is identical with Eq. Eq. t) = ^ k ~ iL2)71^ + V "^c{x))'q[x. of the continuous curve in Fig.13a) and (22. The equation is called stability equation or equation of small fluctuations. We are interested in studying (field) configurations $ in the neighbourhood of the classical solution cpc.47) where n(x. which represents an expansion in terms of normal modes.t) = ° (22'48) (22.13b) we have for small fluctuations rj(x. 22. If some w\ were negative.t) is a small .e.6 The Small Fluctuation Equation We now return to the Schrodingerlike equation (22.7 linking the pendulum bobs). (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).t). (22.
50) has the general form of a timeindependent onedimensional Schrodinger equation. V"(<b) = 2m2 + 6g2(f>2. t) have to be orthogonal to the zero mode (which is thereby circumvented) so that the Green's function of the expansion procedure exists. we obtain V'^c We have m cos ± 4tan" 3±m(:r—xo) (22.558 CHAPTER 22. we obtain V"(<t>c) .x0) = —2m + 6m 2 (l — sech 2 m(x — XQ)) 6m Am' cosh m{x — XQ) = (22. V"((/)c(x)). For x — ±oo the quan> tity acting as "potential". Classical Field Configurations fluctuation. cos46» = 2cos2(2(9) .A \ Inserting <f)c of Eq. Thus for stability w\ must not be negative. the zero mode. 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. 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£^)*.e. (22. i. This means that the stability is not universal but only local in a certain sense. we shall see later that the fluctuations n(x. i.18). The equation of small fluctuations (22. approaches a constant value which is nonzero in both the <54theory and the sineGordon theory. . and need 9 = tan" 1 j e±m^Xo) 1.20). In fact.52) tan 6 := e±m^xo). Inserting (j)c of (22. If that is necessary.e.2 m 2 + 6m 2 tanh 2 m(x .51) (b) SineGordon theory: Here 1 — cos m m 9 V"(<f>) = m 2 c o s f ^ . In fact.1.
We can therefore construct solutions of definite parity.tan20 1 + tan 2 # Equation 559 1 — e x p { ± 2 m ( x — XQ)} I + exp{±2m(x — XQ)} sinh[=Fm(a. (22.* For such a potential the spectrum can be b o t h discrete and continuous.e.ijj^ respectively.56) *We shall see later that the PoschlTeller equation is a limiting case of the Lame equation. even and odd solutions. ip+. 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. A  *(* + !) cosh 2 z tp(z) = 0.53) = m(x — (22.55) The differential operator is even in z. which satisfy the boundary conditions V>(o) = o.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 .2 (22. ^+(0) d 0. — XQ)] cosh[±m(x — XQ)]: 1 . (22.22. i. . Consider the equation fd2 dz2 . The eigenfunctions of the discrete case vanish at ± o o and are squareintegrable. Those of the other case are complex and periodic at infinity.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. and as a consequence a limiting form of the small fluctuation equation of all basic potentials.6 The Small Fluctuation Using the formula cos 20 we obtain cos 40 1 .
(22. + l2) = 0.4 n ( n .61) Discrete eigenvalues are obtained when the series (22. 2n < Z.57) Then the equation for x{z) 1S (22. (22. i.560 CHAPTER 22.( 2 n .l2 = .e. A ^ An = . when in the even case n ( n .e.Z)2.59) Setting (22.e. i.Z ) + (A + Z2) "n+1 (n + l)(n + §) &n (22. Since for £ — oo: * the function ^>+ is normalizable provided Z > n.I) .58) = (1 + sinh2 z)~l/2x(z).Z ) + ^(A + Z2) = 0.e. and in the odd case n+\)U+\l)+\(\ i. . + \{\ ) 2 ~Q"m 1/2 oo «n+l (n + i ) ( n + l ) ( n + ^ ) ( n + ± .Z ) 2 .60) terminate after a finite number of terms.60) n=0 n=0 oo we find the recursion formulas x+(o = E ^ " > n(nl) = x(o+=l £ £&»£n. 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.( 2 n + l . A * An = . i.
e.62) where even (odd) iV are associated with even (odd) eigenfunctions. the continuum is w > Am2. °2 i. We can see the continuous spectrum of Eq. (22. m i. (22.j < I .. (a) ^theory: the equation Inserting (22. We can now derive the spectra of the stability equation for the potentials (22. 2 .. for N = 0.6 The Small Fluctuation Equation 561 and the function ip. 3m 2 .62) we see that the equation possesses discrete eigenvalues w —2 .63) d + K(2ik  2LO)—V ' dijj' + 1(1 + l)V = 0.22.u2)—^V 2 dio and ip(z) = elkzV{to).e.( 2 . i.51) into (22.54) by setting X = k2 Then 1 . From (22.. (22.50) we obtain with z = m(x — xo) tp(z) = 0.63) we see that the continuum starts at A= 2 2^ = 0.2.\=Kl2 From (22.64) This is the equation of Jacobi polynomials V\a' (u/) with a = —ik of degree I in tanhz.2Z = .is normalizable provided 2(n + . (22. (22. m N<2.1.10) and (22. d w = tanhz.Ny.e..11). We can summarize the results in the statement that the discrete eigenvalues are A = (ZAT) 2 .N<1.1: w2 = 0.65a) _dz2 cosh 2 z l=2. N = 0. w2 = Am2.
%(x) Fig. Eq.562 CHAPTER 22. 22.e. . . also be deduced from d(j)c/dx. of course. Classical Field Configurations (b) SineGordon theory: Inserting (22. so that its derivative is nowhere zero (except at ±oo). In the $ 4 theory we saw that (cf.8 and we see it has the shape of a typical ground state wave function. This property can. Thus either spectrum contains the expected eigenvalue zero.65b) dz2 cosh^J.50) we obtain with z = m(x — XQ) the equation (22. (22. .. Thus 4>'c(x) has no node. It is now particularly interesting to look at their wave functions. d . i. for N = 0: to2 = 0 and the continuum starts at w2 = m2. . The zero mode is depicted in Fig. 22. We know that 4>c is a monotonic function. .66) We see that this is a nonvanishing function for any finite value of x. the zero modes. m d .=1A=^_.8 The zero mode as typical ground state.2 The discrete eigenvalues are now given by w ^l m^ = (lN)2. N<1.53) into (22..27)) . and — as expected — this eigenfunction is even. m m ipo{x) oc —<fic{x) = ——tanhm(x — XQ) = g cosh m(x — XQ) dx g dx (22. Thus it has no node and so represents the eigenfunction of the lowest eigenvalue.
. d . — XQ) OC Hence the conclusion is similar to that in the previous case. ipo{x) oc —<bc(x) ax oc 4 m d _i Q±rn(xXQ) —tan e g dx m D±m(za. . Imposing the condition F we have in the <I>4theory: /•oo dx[^o(x)}2 = 1.= E(<f>c).t a n h x .70) .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.e.3 . . Finally we consider the normalization of the zero modes. (22. 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 . without quantum corrections).68) d{mx) m(xx0) 4 m3 3 ^ g Ml J cosh / i. 1 o rrr tanh x — . (22.67) J —( Mx)= and hence m w0i^ 1 2 4 m ±—tanhm(x — XQ).z 0 )]oo = 8 .oc (22.o) ±4 m g l __ e ±2m(xx 0 ) 1 cosh m(a.e. .6 The Small Fluctuation Equation 563 In the sineGordon theory the zero mode is proportional to . i.22.
This difference may not always be appreciated later in our very analogous treatment of (topological) instantons and (nontopological) bounces. H. i.564 CHAPTER 22. P(</>) = f ddxV{(f)) > 0. for if such configurations exist in a realistic theory with three spatial dimensions. Goddard and D.e.73) n<(>a) = j ' d^V^)2 = a2~d / dd(ax = J' ddXl{V4>(ax)f d 4>{ax) 5ax . *See e.7 Existence of FiniteEnergy Solutions So far we have been considering the case of one space and one time dimensions.g. This question has important consequences. H. The arguments of Derrick were later rephrased by several authors whose line of reasoning we shall be using here.g. however.T(4>). Affleck [5] who uses in his Eq.> 4>a{x) = 4>(ax). *The considerations of this section extend in part beyond quantum mechanics and into field theory.g. it is important to understand their physical implications. One can have different types of scale transformations — see e. Kastrup [145]. Classical Field Configurations 22. J As a first example we consider the expression (22. I. ^For basic aspects of scale transformations see e. Derrick [68].4) <j>(x) —• a<f>(ax). P. I.41) for the energy of the classical configuration but now with respect to a ddimensional position space. (2. They may help. ddx {V4>f + v{4>) (22.* The general scaling argument used for this purpose was first introduced by Derrick t and is therefore referred to as Derrick's theorem. We now wish to inquire whether static finiteenergy configurations could exist in more than one spatial dimension. Under this transformation (the verification is given below) £(</>) ^ E(</>a) = T(cf>a) + P(</>a) = a^T^) since for instance + a" d P(0).71) We now consider the scale transformation (with 'a' some number € R) 4>{x) . we consider E(<P) which we also write as (22. to underline the fundamentally different nature of topological and nontopological finite energy configurations.2d. f G . Olive [113]. (22. A. .72) with T{$) = I' ddx]{V(j))2 > 0.
e.. a=l (22.xd).78) For the energy of a static configuration. In fact. (22. d = 1 minimizes the energy. It is also a virial argument in view of the virial theorem relation T{<j>) = P(<f>). 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. Consider first the Lagrangian density £[&M = \{d^)(d^) . we set 0 dE(<pa) da J a = i (22. (22.e. since d2E(<pa) da2 2(2d)T(<f>). Next we consider more complicated cases.7 Existence of Finite Energy Solutions 565 We saw that the static configuration with a = 1.d)T(<f>) = dP{4>).22.71).73) we see that this implies (2 . we write • = lim J r .x2.77) with V(4>) > 0 and x G Rd and the set of minimum configurations Mo := {4>\V{4>) = 0}. to be finite the field <P ~^ 4>oo £ Mo for x — oo.79) In order to emphasize that (f> — (p^ in any direction.76) we see that E(<f>) can be minimized (i. i. Can the energy also be minimized for more than one spatial dimension? To see this we stationarize E(<f>a) with respect to a..80) . (22.V(<P) (22..\ eMo(22.75) Since both T and P are > 0 and d is integral. x = (x1. the second derivative is positive) only for d = 1.74) From (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. (22.e.
x = 00. These observations are summarized in Table 22. Table 22.e.e. If we go to two spatial dimensions (d = 2). is a connected region. i. —1. therefore. i. 0oo 7^ <>< > in the case of the kink solutions) and /x associated with this a set of disconnected points in space (00. as is the case for the constant solutions. . Thus in order to obtain a topologically nontrivial case we must have a different classical vacuum in association with each point at infinity. of • course.566 CHAPTER 22. i. = —00 : S = — 1 <—> ^oo = 27rn^ The topologically trivial case is given when the classical vacua associated with S1 = + 1 .81) and (since d = 1) S = ± 1 . the easiest way to visualize points at infinity is to take a circle Sl of radius r and let r — 00. that we have a set of disconnected classical vacua (</>oo><^oo. — 1).38). Then each point at infinity is characterized by a different value of the polar angle 9. Classical Field Configurations In the sineGordon theory (d = 1) the equivalent statement is (22.e. (n — n = 0) (B) Topologically trivial mapping x = +00 : 5 = +1 <—> <> o = 27rn^ /o a. are the same.and sineGordon theories we assumed $ to be real. —00. and. in fact. 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 . We see. one set can be mapped into the other. —00 or equivalently S = + 1 . Thus in this case S^1 is a disconnected set consisting of two points.1 manifold Sd~1\d=i <• manifold Mo <—> 0 O = 27rn^ Q <—> 0_oo = 27rn'^ kink limits (nn'/ 0) const. (ptx = <f>oo. S1.1. ( with — = lim <pc(x) 9J *>±oo TYl \ (22.1 (A) Topologically nontrivial mapping x = +00 : S = +1 x = 00 : 5 = . 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).
assuming invariance of C under the transformations of the group SO(3) in internal space. .83) . we would have only two points at infinity. VMx)) with (0(oo) = linvxx) (p(r.e.22. i.e. if we consider a theory in three spatial dimensions and so the unit sphere S2 = {x x{ + xl + xl = 1}. cj)(r. the field must map position space into field space at infinity. Thus. given any (00) = «Koo)e* = <?[<p(xMx)a2]2 567 say with 9 — 9$.82) Now Mo := \ 4>(oc) ^24>2 = a2\. —00 or S = 1 . On the other hand. we have w e ca {Mx)}> and 1 3 1. if cj> were real and position space twodimensional. the one point S = + 1 .e. This can be achieved e. We see. i. i. 00. 9) would depend only on r = x and hence the problem would be effectively onedimensional. therefore. Similarly. i.• M0. Then the entire set of configurations (f)^ = \4>oo\ exp(i#) would correspond to.g. that if the problem is to be topologically nontrivial.1 . * n reach any other <^(oo) by simple phase shifts. (22. 9)) Mo = {</>(oo)</>*(oo)0(oo) = a2}. i. This will be "nontrivial" in the above sense only if 0oo is independent of 9. by allowing <fi to have three components in some "internal" (isospin) space. if <p is real. for instance. to have topologically distinct classical configurations. phase transformations 6Q ^> 9Q + 59.3 (^real) v{4>) = / (22.e. the manifold of classical vacua would have to possess a similar geometry if the theory is to be topologically nontrivial. If position space were onedimensional.7 Existence of Finite Energy Solutions Suppose. Thus. map into.e.e.2. i. $(r) : Sl .
e. The theories are: (a) The Nielsen—Olesen (vortex) model (also called Abelian Higgs model) " with Lagrangian density V(4>) = Ug2 2A 9 2 FM„ = du.Ta} in an 50(3) invariant theory and so on — determines the transformation from one point at infinity to another (on a line. Classical Field Configurations The manifold A4Q is a sphere in a 3dimensional Euclidean space. circle. 11 H.2. (b) The GeorgiGlashow model I! with nonabelian gauge field in 3 spatial dimensions with Lagrangian density (V as under (a)) £[AM. the subsequent discussion requires only the general form of the Lagrangians. Summarizing we see that </>(x) is a mapping from S^T1 to <S^1. The symmetry of the potential (and Lagrangian) — i. In the following Example the use of Derrick's theorem is demonstrated by application to some gauged theories.Av  dvAf i where 0 is a complex scalar field and the spatial dimensions are 2 or 3. Example 22. $ — $ + 27 > • T in sineGordon theory. . it is effectively a surface (and so a continuous set) S2. + i ( D " f t W ) a . invariance under replacements $ — —$ in $ 4 theory. the winding number is n. going once around S%~1 (i. Georgi and S. Olesen [220]. depending on d) and correspondingly (with n complete windings) the transformation of one classical vacuum into another. Any 4> £ MQ can be reached from any other cf> e Mo by application of an element of SO(3). $ — $exp(ia) in a U(l) symmetric theory.a M J . a theory being described as 'gauged' if a gauge field like the electromagnetic field is involved.568 CHAPTER 22. If. if SO (3) nonabelian) are two models defined by the following Lagrangians which we cite here solely for the purpose of being explicit.. . we cover <S^_1 n times. A) = T(4>. In order to preserve lim y ^ i a\ such rotations must tend to the identity at r = oo.e. so that details of these models are irrelevant. i. % . a = 1.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) £(</>.</>] = ~G^Gai. i. through each point at infinity).„ where G r = 9 M £ . the appropriate rotation. Nielsen and P.e. A) + T{A) + P{<j>).e. $ — > > $ exp{i Xlo=i Pa.. B. sphere.3. Glashow [108].V{4>a].2: Derrick's theorem applied to gauged theories Wellknown and important theories involving a gauge field (AM if abelian and A". L.
d /" d d x ' a</)(x') <9x' d</>(Ax) 3x M A 2 /" ^ ( A z ) d<p(Xx) A / A^ d(Ax M ).<^[x) transforms like a vector. T(A) = f ddxV(4>) > 0.86) (22.d)T(4>.dP(4>) = 0. Mx') = E with a scale transformation x — x' = Ax. i. i. >• A„ x ( x ) = AAM(Ax)..Ax) dX Since T((f>.A).87b) We consider explicitly the example of <j> — </>\(x) = <p(\x) for > T ( V(4>):= Jddx(d.e.22. negative. (2 .84) in this way one finds that the quantities involved transform T ( ^ .A) P{4>) 569 j / i F ( « ( P ^ ) t ( P ^ ) „ > 0.84) and investigate the existence of finite energy classical field configurations.d T ( A ) + AdP(</>). (22.87a) C?M„(x) — \2G^{\x). ^ A ) = A 2 .7 Existence of Finite Energy Solutions where T{<j>.d T ( ^ .89) P{4>) are positive semidefinite and not all zero.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'>. it must be stationary with respect to arbitrary field variations and so with respect to the above scale transformations also.d)T(A) . (22. (22. A) + (4 . Solution: Each contribution in (22. A.i T(AX) = X4~dT(A). P(tf>A) = A " d P ( 0 ) If £ is to allow a static finiteenergy solution. J 4 ) + A 4 . = 0. (22. Thus we set dE(<fix. <0 >• <£A(X) = <KAx). l j ddxGijGai] > 0. We now consider scale transformations with A £ R.e. (i)^ 2d T Considering the contributions of (22. 4>Y We have T«(4>A) = J ddx(d„4>x)2 = J dd . A A ) = A2dT(^. we have {^/»)A^ (22. and so B ( ^ A .84) is positive semidefinite and so must be finite by itself. A). (22. > Since T>IJ.T{A). some contribution must be .
(e) For d > 5: In this case it is not possible to stabilize Eq. MiillerKirsten [268]. tt The — now established — microscopic BCS theory of superconductivity was formulated some eight or so years after the macroscopic GinzburgLandau theory. In the following we consider a related theory. Thus such a contribution could counterbalance T{4>.90) and we would have to add to Eq. R.89). (b) For d = 2: A theory with all terms would be a candidate — in fact the NielsenOlesen theory is an example. 22.g. 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. L. (22. if we allow a term T^ 4 ' {<f>) with fourth powers of derivatives. of course. One can thus define a scalar field or wave function ty(x. We can therefore use our understanding of the BCS theory in interpreting the GinzburgLandau theory.89). Skyrme [253].570 CHAPTER 22. We first demonstrate — as a matter of interest — that the NielsenOlesen theory for static equilibrium (i. a = a(T) = a T —T c .** e. (c) For d = 3: All three terms must be present as in the GeorgiGlashow model. We then have a situation of the potential like that described in the context **T. is temperature dependent and changes sign at the critical temperature Tc. . T^{<t>x) = \4dT(</>). Ginzburg and L. This case is. exemplified by soliton and sine—Gordon theories.8 Ginzburg—Landau Vortices We have already referred to the NielsenOlesen theory.89) the contribution (4 — d)T(4>). (22. 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. 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. (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).t) such that  ^  2 describes the density of Cooper pairs. Tchrakian and H. (22. Landau [111]. There are other posssibilities if we permit higher powers of derivatives as in Skyrmion models. (22.92) The expression ^ ( x ) ! is known as the "order parameter. H. An explicit and solvable Skyrmion model in 2 + 1 dimensions is considered in D. D. Equilibrium states of the superconductor are assumed to be described by the timeindependent static wave functions ip(x). a > 0. A) or P{4>) for d = 3. W.e. H. timeindependent) field configurations is equivalent to the GinzburgLandau theory of superconductivity. J.
i. from x < 0) means — as is familiar from electrodynamics — replacing V by V . Returning to the three dimensional case and switching on a magnetic field (applied from outside of the superconducting material. Then the static energy density becomes (cf. is a measure of the distance from the surface of the superconductor to where .94) to check the dimension) 1 = I—a.(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. setting S£. ^(00) = a (22.e.2) £(^. (22.22. The parameter 7 serves to put the minima of the double well at V = 0.8 GinzburgLandau Vortices 571 of the complex Higgs model (cf. assuming there is no superconducting material in this domain). Example 22. Sec.94) the solution is (in the domain 0 < x < 00) i>{x) = ^/^tanh(^x\ . 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.A) = iB 2 + i iXQ + 7 + ^ H 2 + /^4. Clearly the length (see Eq.^ 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. (22.93) J' (22. but for T < Tc it has a double well shape.2): For T > Tc the overall potential has a single well (minimum).e. Variation as before. i.95) This expression can be shown to be equivalent to the integrand in the NielsenOlesen theory. 22.
the order parameter \ip\ attains (approximately) its asymptotic value. 2mi Setting < / > \P\e vp B = ViA.e.M. (22.96) ipH. H. The magnetic field on the other hand.9. yJ—a/[3. J. MiillerKirsten [215].g. exhibits a completely different behaviour — in fact. We can see this as follows.e. V x M = j s . i. i.97) we can rewrite j s as m \ h In macroscopic electrodynamics the magnetic field strength H is (in SI units) defined by H := — B . (22. 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 ).e. Classical Field Configurations superconductor (x>0) Fig. . (22. The density of the Cooper or super current is given by the typical expression of a current. i. 450.9 Behaviour of ip and B in the superconductor. In the Maxwell equation See e.572 no superconductor CHAPTER 22. 22. as indicated in Fig.98) where M is the magnetization resulting from atomic currents. W. 22. h [^DV'V'DV]. p.
i. (22.div grad" and V • B = 0. and V x (V x B) = / J 0 V x j s . Approximating p by its equilibrium value..100) we obtain 2 m ( 0\ The ratio XL/XQ is called the GinzburgLandau parameter. 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. those of the socalled type I and those of type II. Hence 0 = V x H = V x ( — B .8 GinzburgLandau Vortices 573 the quantity j is the density of a current applied from outside. m Here A^ is known as the (London) penetration depth (or length) (in practice this is of the order of 1 0 . Superconductors of type II are characterized by the fact that when the magnetic field is again increased and beyond a first critical value. V x B = [i0V x M = /lois. In the onedimensional case we have B{x) = B(0)ex/XL. These vortices therefore carry magnetic flux which is necessarily quantized . this becomes A B = no V x j 5 .99) Inserting (22. m xi A B . Using the relation "curl curl = grad div .M i. Also dD/dt = 0 in a static situation like the one here.5 to 10~6 cm).e.22. (22.L B = O.e. More and more vortices are formed as the magnetic field is increased further and further until only small superconducting domains remain which then disappear completely beyond a second critical value of the magnetic field.97) this becomes (since curl grad = 0) AB = Mo—Vx m A = /x 0 —B. K i ™*. There are two types of superconductors. This is zero in the present case.
The existence of these vortices has been confirmed experimentally. de Vega and F. 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 — 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 < 1/4.102) In the onedimensional soliton case of Eqs. Finiteness of E requires \<t>\ = \<fioo\ on this circle.9). Classical Field Configurations (see below). 'See the remark in H. where exp(ix(0)) is an arbitrary phase factor. *A.9) ^e^ '>\<poo\. A.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. 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). Abrikosov [2]. 22. i. We can think of (e. .574 CHAPTER 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. Thus the asymptotic field </>(r. in other words.e. we must have that lim V ( H ) = 0.. of two points on a line. De Gennes [67]. i. See also P. (22. In the NielsenOlesen model defined in Example 22. In all the models we considered so far we observed that the energy E contains a contribution P(V) = j ' afxVM).36). (22. Schaposnik [69] after their equation (3. (22. the space consists of two diametrically opposite points on the circle. by the element exp[i%(0)] of the internal group A(s) = e * W : SI+SI. In the two models of Example 22.g) twodimensional space as being bounded by a circle at r = oo. J.e. x—>oo Thus at spatial infinity </» has the value of a zero of the potential. in fact a periodic function of the polar angle with period 27r.37). A. G.101) (p(r.
States of the strip corresponding to winding numbers 0 and 1 are illustrated in Fig. The map satisfying X(x + y) = \{x)\{y) is said to be a homomorphism. defines the orientation of this line element. 22. We can illustrate the simple lowdimensional example under discussion by reference to a strip with Mobius structure. topological group with the usual multiplication as group operation. .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 22.2 imply that such a map from a circle to a circle is associated with an integer which we called the winding number n.X ( 0 = O)].118)).e.10. W. Thus smooth deformations *See D. Then a continuous function of (j> on [0. abelian (i.e. i. Sl := {zeC\\z\ = 1}. 2n] remains continuous if the function assumes the same value at the two endpoints.22. We may define this as « = ^[X(0 = 2 T T ) . commutative). (22. 22. X is called the domain of / and y its range. The two theories we considered above in Example 22. Fig.10 States of winding number 0. Naturally an integer cannot change continuously. (22. The map A : S1 > S1 defined on S1 C R2 by (22. Finkelstein and C.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".102) is called the exponential map of S1 onto S1. 0 < 0 < 2TT. Suppose <j>. The 1sphere is the unit circle in the space of all complex numbers. In mathematical language S1 is therefore a compact. Misner [96].1).1.
finite. i. no hole.g. i. connectivity.) remain the same under homeomorphic transformations and are therefore called topological invariants.105) is described as "pure gauge" since it is determined entirely by the phase x(#). if it has one twist as above or a hole. A field configuration with n ^ 0 therefore cannot be deformed continuously into one with winding number zero.104) so that d2x hJ Since A has direction e#. we required that lim r—>oo w o. In our consideration of the NielsenOlesen theory we also required the gauge field A^ to behave such that the energy is finite. elements in the same class have the same topological characteristic. We observe that since time evolution is continuous. it cannot be deformed continuously into the socalled "classical vacuum" (the latter is a constant configuration in our soliton example).576 CHAPTER 22. The continuous deformation or distortion of a set of points does not alter the topological structure of the set (e. we can write 1 d h J ee rde w At h r^oo . 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.g. r^oo 1 dx(9) n r o0 (22.104) to be finite we must demand that g . one hole. (22. The different kinds of connectivity of these topological spaces (e.i'ldx(0) 9A Thus for (22.105) the corrections falling off faster than 1/r.. it retains this twist or hole under the continuous distortion) and is described as a homeomorphism. The gauge field required in the asymptotic limit (22..e. i. For that reason n is called a "topological invariant?. Homeomorphism subdivides the possible spaces into disjoint equivalence classes (to be explained below)..e. The winding numbers n are such topological invariants. In other . since otherwise the energy integral would behave like f d x f dr dr 2 J r J r lnr.e. Thus two homeomorphic spaces have equivalent topological structures.
Of course. for B / 0 .e. going once around S^. we can go n times around S^.108) map the spatial circle S^. (22. i. the vector potential A. Then we can construct a oneparameter family of maps or transformations which transform </>n (0) continuously into (j>h (6). we have B = V x A = curlgradx = 0. Suppose two such configurations are given by O) W(9) = ei&)V\ n fixed. 9){^ and hence the functions (22. cannot be deformed continuously into one with two loops (n = 2).e. It follows that the field Fij. We can calculate the magnetic flux $ through the region with $ = * rd9A0 = .110) . i. Ffiu. However.105) insure that B ^ O .l]. i. the higher order contributions in (22. An example is the mapping H(t. into S i .9):=t(f>£\9) + (lt)(f)W(9) with te[0. with n = 1.106) J g j r d d g Thus the magnetic flux is quantized.109) A W(0) = e^ W. (22. we can consider transformations which describe continuous transformations of a configuration with fixed n. A field configuration with one loop (or twist).22. (22.<b rd9—4^ = n. cannot be pure gauge everywhere.9 Introduction to Homotopy Classes words. if A = V x . In terms of the phase factor \{x) = eixie\ we can rewrite the relation (22. the number of flux quanta 2irh/g being the winding number n. is also zero at r = oo. Therefore. the gauge field.105) as «A(I)'AWV*<I> \(x) = J*V\ ( =• VxW)) (22. 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".107) The fields <f>(r. However. the group space of (the phase transformations of) U(l) continuously. it is determined by a gradient (l/r)d/dd.e.
9) is an equivalence relation.111) In principle an expression like (22. But this type of combination is excluded sind H(t. and the space of t. (22. Example 22.e.110) could be written down for configurations with different winding numbers. e.9) is called a homotopy (emphasis on the second syllable). We verify these in the following example. n' ^ n. (fin (9) in the above example.e. n. f T h e standard texts on homotopy theory are N. A brief introduction can be found in the very readable book of W. Since the passage of time is continuous. The relation (22. (22. to the internal group space of £7(1).0) = $\6). time cannot make a field jump from one homotopy class (see below) to another. the unit interval I = [0. (b) symmetry and (c) transitivity.111) as a map from the space described by 9. i. Two curves which can be related to each other in this way. H.lQSl^S^ with H(0.112) is an equivalence relation — and as such it is defined by the three properties of (a) identity. 8) = cfinl\9). we have to show that the map satisfies the three properties of (a) identity.113) . i. i. H.3: Homotopy an equivalence relation Verify that the homotopy map H(t. In fact. Hu [134]. Solution: As stated. i.112) are homotopic iff there exists a homotopy connecting (fin (9) and (fin (9). S%.e.1]. (b) symmetry and (c) transitivity. Steenrod [261] and S.9) = <fiW(9) and H(l. Classical Field Configurations which is continuous in the parameter t and is such that H(0. We can consider (22. A superposition of classical configurations with different winding numbers does not minimize the energy. are said to be homotopic (emphasis on the third syllable) to each other and one writes ^(0)^^(9).8) itself must be a configuration with definite winding number.e. the phase factors of (f)(9).T. Dittrich and M. H(1. Reuter [77]. Thus by varying t from 0 to 1 we deform (fin (9) continuously into (fin '(8).578 CHAPTER 22.g. Still on the level of a mathematical text is (in spite of its title) the Argonne National Laboratory Report of G. i. Thomas [270]. it is clear that this is also a homotopy. these two maps (fin from S^. The map H(t. 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.e. — S\ > (22. like <pn (0).9) = ^°\0).
since and tf (1.ffi(O. i<t<l. \ M)\ H!(2t. 0) = ^ (0) and 6(1. . 9) connects <pn (0) with itself continuously. Hence <f>(0^ ~ <p 2 ). 9) = <f>W (0). 9) = H(l.9) = . 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. 0). so that H(t. <^(2). i? is a homotopy connecting </>(°) with <^>(2'.22. (c) We can demonstrate the property of transitivity as follows. We now define the function H ( f 0 ff . which therefore demonstrates that the homotopy relationship is transitive. these groups are then called homotopy groups (cf. (b) Next we check that if 0<°) ~ <p^. Moreover. ' .e. 9) := H(l .t. On the other hand H(0. . 0) = <£{2). In some cases the set of homotopy classes possesses group structure (satisfying the group properties). Hence the set of all such maps breaks up into equivalence classes or homotopy classes.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.e) tf2(2il. <f>W . U(l)). 9) = H(0.e. Then there exist homotopies Hi and Hi connecting <j>(0> with (j>W and t^ 1 ' with </>(2) respectively. below). 0) = 4>W (0). negative or zero). 0). as they are called.* 22. Thus 7Ti[C/(l)] = vrifS'1] corresponds to (i. we see that of the set of these maps some are equivalent and others are not. being the continuous map connecting </>(°) with <j>(1'>.0).0) = </>(0) is continuous there.9) = <f>(0\6) and H(l.9) H(0. Two elements of 7ri [f7(l)] which belong to different classes are homotopically inequivalent. then </>W ~ <t>{0)• We define an H(t. i. Then H(t. Homotopy classes are sometimes also called Chern—Pontryagin classes. at t = 1/2 both i?i and H 2 give c/^1'. so that H is a homotopy connecting <jyl>{0) with <f>^>(9). Returning to the maps \(x) : Si ^ S1^ (i. is such that H(0. (22 1U) At *=2!: But ^ ( ^ ) = ^ i ( M ) = #2(0. Assume <p( ' ~ (j> and <p( ' ~ <> ' /( for continuous maps <^(0'. a class being a set of equivalent maps.0) for for 0<t<. 0) = tf2(l. is isomorphic to) the set of integers Z (positive.10 The Fundamental Homotopy Group 579 (a) Evidently the homotopy H(t.e.
e. .) why 7i"i[S2] = 0. 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. This trivial map into a single point is described as a "degenerate map" and is illustrated in Fig. T (22.11 The trivial n = 0 map allowing shrinkage to a point. Appropriate maps.g. How can we see this? We first sketch the idea.580 CHAPTER 22. and then introduce some of the related mathematics. Classical Field Configurations Thus there is a denumerable infinity of such homotopic classes or sectors. which is to be mapped into another circle Si indicated by dashed lines. to make the result plausible.11 the circle drawn with the continuous line is the circle S%. In the diagrams^ of Fig.11 for the special values of t = 1 and t = 1/2 with Xo(0) = to t(2vr 9) for for 0 < 9 < 7T. 22. but TTI[S2] = 0: S 1 is a circle.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. Fig. p. 52. Rajaraman [235]. Sketch of idea demonstrating that nilS1] = Z. 22. In order to really appreciate how topology comes into the picture here one should understand (e. T < 9 < 2TT. n=0: t=1: t=1/2: i.115) Varying t € [0. S 2 the surface of a sphere in a threedimensional Euclidean space. 22.
phase factor or element of U(l)): A(x) : Xn(x) = em*W = [Xl(x)]n. The maps for these cases are illustrated in Fig.h. is invariant under infinitesimal (continuous) deformations. (22. We make another observation at this point.118) is an infinitesimal (continuous) real function on the circle with (5f)o=o = (Sf)0=2n 1 See S.103)) A(x) > \'{x) where 5f(x) = \{x) + i\(x)5f(x) = A(x) + 6\(x).117) in ±.22.e. 22. i. to the case of winding number zero.117) 2^ d6X(x)\1(x). Coleman [55].e. and hence the winding number. In the cases n = 1. 22.e. For winding number n we have the mapping (i. One can prove that the expression (22. i.10 The Fundamental Homotopy Group 581 n=l: cutting of dashed circle n=2: Fig. 2 the maps are wound around the dashed circle.12.12 The n = 1 and n = 2 maps not allowing shrinkage to a point.116) In fact from this we obtain (see below) 2  n and this means r.117). (22.h. d (22.s. 0 < X(P) < 2vr. .s. It is seen t h a t in these cases the continuous curve can be distorted into t h a t of case (a) only by cutting the dashed circle — thus these cases do not allow a continuous shrinkage to a point.I** dOJn*W±e {8) X 2vr J0 — / 2Wo dO(iri) dO \ d6 l. the transformations^ (the result may be looked at as 8n of expression (22. of (22.
Steenrod [261]. 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. an integer modulo 2) i. whereas 7ri(50(3)) = Z2 (i. In fact. 1 1 See N. Sec.8.582 CHAPTER 22. il**( ^') r2Tr A ( 21 9 2 1) ' de\x2 *(«/W).e. only one homotopy class.e.e. the cyclic group with two elements (since 50(3) is doubly connected). the identity class (the group space S2 is simply connected. Thus there is only the trivial map into a point. there is a theorem which says thatH TTg(Sn) = 0 for q < n. . It is intuitively clear that any circle drawn on the surface of a sphere can be shrunk to a point. 15. 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. 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. i. Classical Field Configurations Under this transformation n changes by 5n Now. Hence i /"27r ^ i Thus we have shown explicitly that the winding number remains unchanged under continuous deformations.
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. In each of the cases considered we obtained the result by solving the appropriate Schrodinger equation. and in particular the exponentially small level splittings of the double well potential. Together with the infinite Euclidean time limit this requires for the calculation of the tunneling contribution to the ground state energy the FaddeevPopov method for the elimination of the zero mode.Chapter 23 P a t h Integrals and Instantons 23. in such a way that the factors appearing in the result are exhibited in a transparent way. Since we have not yet introduced periodic (i.2 Instantons and AntiInstantons A onedimensional field theory is a quantum mechanical problem. We consider such a theory now with action S((p) depending on the real scalar field 583 .e. different from literature.1 Introductory Remarks In Chapter 18 we considered anharmonic oscillators and calculated their eigenenergies. finite Euclidean time) classical configurations (which are nontopological). 23. In the following we employ the path integral in order to derive the same result for the ground state splitting (only that in this chapter!). 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. 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.
* /•tf CHAPTER 23. i. Path Integrals and Instantons S(<f>) / dt \i>2 .2. The Hamiltonian of the simple harmonic oscillator (cf.1 The double well'potential.1.1 — degenerate minima (positions of classical stability or "perturbation theory vacua") at positions 4>{t) = ±m = ± 0 O .1)) and hence E0 = mh/y/2. n = 0. 2 2 4 ™ = ~.2 = 2m 2 (here m is the parameter in the potential V{4>) of Eq. 23. Gildener and A. in particular E. "For comparison with literature.2) Fig. Fig.584 <)>{t).v{4>) <?VV m4 __ 1 . and ^2(0) : m* . (23. 23.e. note that here the mass of the equivalent classical particle is taken to be mo = 1. ! " ^ j = ^ 2 .m ^ + §^24 ' ^2 > a (23J) The potential V(</>) has — cf. (23. Chapter 6) 1 2 . 2mo has as the normalized ground state wave function </>o(?) = <<?0) Thus for mo = 1 we have in our notation here a. Patrascioiu [110] and Chapter 18.
23. and in the above 4> corresponds to x.e. The corresponding wave functions ip±(x) then must be peaked at the extrema of stability — as in the examples illustrated in Fig. this alone does not yield the level splitting._(0) = 0. These boundary conditions imply nonperturbative contributions to the eigenvalues which yield the splitting of the asymptotically degenerate oscillator .g. 23. Fig. we have the Lagrangian L(x. </. must have definite parity as in a physical > situation. This tunneling affects the eigenvalues. The states of the system here must be even or odd under x — ±x. i. as we did in Chapter 18 — but as we saw.2 Instantons and AntiInstantons 585 Classically the position <f> = 0 is a point of instability.x) = x2 V(x).2 — and we can deduce from this the boundary conditions. 23. We can develop a perturbation theory around either of the minima. We know from the shape of the potential that the quantum mechanical energy spectrum is entirely discrete (no scattering). but there is tunneling between the two minima if the central hump is not infinitely high. 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> ~ * ±</>.In mechanics of a particle of mass 1 with ~ position coordinate x(t). e. W(0) = 0.2 The wave functions of the lowest states.
Equation (23. Euclidean time T = it. i.2 V'(<l>) = .e. In the semiclassical path integral procedure that we want to use here.9) .£'(0) = 0. dr We observe that Eq. Path Integrals and Instantons approximations into an even ground state and an odd first excited state.7) so that (23. The transition amplitude for this.5) and iS = dLE d dr[d(d(f)/dT)} i. that given by Eq.8) V(4>) + const.7) is to be solved for 0 with the boundary conditions m 0(r) = — for Tf — oo and 0(r) = > m for n oo.U) = J'D{<f>}exp[iS(<l>)/h] (23. A classical particle starting from rest (0 = 0) at —m/g cannot overcome the central hump and move to +m/g.  . For small h we can attempt a stationary phase method.£j — —oo (we assume the normalization factor or metric to > > be handled as explained in Chapter 21). SE. dr m+™ (23. (23.3) for tf — oo.4) J <bi =—(An rTf (ITLE = I JT. But quantum mechanically it can tunnel through the hump from one well to the other.e.e. d(f> d ' dr2 dr It follows that the equation of motion is now given by 55.3).<h. (23. n) = f with the Euclidean action rTf SE(<i>) = / J TV f V{<f>}exp[SE(<P)/h} (23. we consider the amplitude for transitions between the vacua or minima at 0/^ — ±</>0 = ±rn/g over a large period of time.2.0o.2 m 2 ( l m z (23.7.e. i.586 CHAPTER 23. is finite if we go to imaginary. i.6) d*l dr 2 . (23. We are therefore interested in the amplitude — called Feynman amplitude or kernel as we learned in Chapter 21 — (</>o. classically forbidden domains then become "classically" accessible.7) resembles a Newton equation of motion with reversed sign of the potential as a result of our passage to Euclidean time. Then (0o. dLE = 0. (23.tf\ . r .
Fig. this means the total energy of the particle in Euclidean motion is zero. in the present case (T — To) = m tanh m(r — TO). 23. — (23. Thus we can picture the particle as starting from rest at one minimum of V(</>) or maximum of —V(<p).18). i.2 Instantons and AntiInstantons 587 In fact.e. 23. timeindependent) in the 1 + 1 dimensional <J>4theory which we considered previously.e. is depicted in Fig.3 The instanton path. and hence kinetic energy gained is equal to potential energy lost. from fa — —rn/g.23. and having just enough energy to reach the other extremum at 4>f = m/g where it will again be at rest. 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.e. we see that the equation is the same as the classical equation for a static soliton (i.10) where TO is an integration constant. the classical path. . Thus we know the solution from there which is the kink solution (22. Its trajectory.3. i. If the constant arising in the integration of the Newtonlike equation is chosen to be zero (see below). In the present context of Euclidean time the configuration (f)c is generally described as an "instanton".
Tj) oc exp 4 m3 m^2 (23.10) 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. (23. (23. 'Note that this integral may be evaluated for finite limits r = ± T and presents no problem in the limit T oo.10) is given by (23.13) 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. Using Eq. The classical equation (23. This antiinstanton configuration rises monotonically from right to left as shown in Fig.13).7"/! ). We also observe that the expression is singular for g2 —> 0.TJ T fro.13). The configuration (23.12) dimr) cosh m(T — TO) m tanh x tanh x 00 _ 4m*_ (23.4).588 CHAPTER 23.5).' .14) But this is not yet enough.13) into Eq. we obtain lim f+ OO.r 0 ) ' (23. (23. thus indicating the nonperturbative nature of the expression.7) admits another configuration which is the negative of (23. Inserting the result (23. Tf = oo. We also need to know how we can calculate the difference AE of the energies of the two lowest levels from this expression. Also note that we can reexpress SE in the form 2 SE f0 c (oo) T Joo \ dr j J0 C ( where ^iw = .11) For this quantity to be finite for TJ = —oo.^ . 23. Path Integrals and Instantons The Euclidean action of the instanton solution (23. the constant must be zero. For a nonzero constant we obtain the periodic instantons discussed in Chapters 25 and 26.8) we obtain T f SE dr[2V((j)c) + const.* With this choice and inserting mo = j dr J = m we obtain* SE = m T lm4 2 g2 cosh4 m{r . but with the same Euclidean action (23.].10) and is called the antikink or antiinstanton configuration and obviously (following the arguments of Chapter 22) carries topological charge — 1.
e. 23. The terminology is motivated by analogy with the phenomenology of particles: Thus like a proton has electric charge + 1 (in units of the fundamental charge e) and the antiproton has electric charge —1. as we can see e.TO far apart and T'Q 3> TQ. so correspondingly one defines the instanton or kink with topological charge + 1 and the antiinstanton or antikink with topological charge — 1.4 The antiinstanton. 23.15b) a solution of the classical equation? Is it a topological configuration — if so. 0c(r) = &(T).15b) since we require configurations from vacuum to vacuum. Questions which arise at this point are: Is the nonmonotonic configuration (23. and an antiinstanton localized at (say) TQ.5. .5. from the sketch or from the variation of SE. #(r) = </>C(T .15b) In the case of (23. in the case of (23. / > Looking at the instanton (23.15a) we have a simple superposition.g. 23. (23.15b) has the shape shown in Fig.15a) or # ( r ) = 0(rro)^c(rro)0(rir)e(TTi)^c(r^)0(T2r).g. The sum (23.23.TQ are very far apart their sum differs from an exact solution only by an exponentially small quantity.15b) a mending together of an instanton and an antiinstanton at r = T\ with —oo = To < T\ < T2 — 00 for TQ.TO) + ^ c ( r . (23. 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 (23.10) we can ask ourselves: Does it make sense to consider the sum of an instanton localized at TQ as indicated in Fig. b u t if TQ.Writing ••• + L ••• and performing the variation in the usual way with 5<j) = 0 at r = ± 0 0 . Clearly we are here interested in configurations of type (23. Both configurations comunicate between the two wells of the potential at < = —m/g and (ft = m/g.r0).2 Instantons and AntiInstantons •<l> 589 ^ ^ ^ 0 ^ ^ ^ _ _ ^ Fig.15b) is not exactly a solution.
23. vanishes as T\ —> ±oo (of course.5 Superposition of an instanton and a widely separated antiinstanton. In fact. 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 . of course. T\] with LECIT = 0.r 0 ) At r = Ti the variation Sep is finite but nonzero so that we are left with an exponentially small contribution which.15a) is a configuration which starts from 4> = —i^/g and ends there. Finally.T 0 ) } ] . 5^ In order to answer the second question we observe that (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 . Path Integrals and Instantons '. g cosh m(r — TQ) (23. since (23.2 m ( T o .15a) can be argued to be zero. m/g / o / o ^ 0 1 \ c'\ o\ T2 ' ^^__ m/g Fig.16) . Ignoring exponentially small contributions the Euclidean action of the configuration (23. we see that in the domain [—oo. it also does not exactly stationarize the Euclidean action.15a) is not exactly a classical solution. 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. at oo the variation is <<> = 0).CHAPTER 23.e x p { .
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).
Thus the factor a inserted above has to be taken as 1/2. are summed over all values from 1 to n.64) D (23.TI) K(T2.exp • /_:T JV(T')" dr' = e x p [ .e. the kernel K(r.67) i=l v ) = lim exp V In ( 1 + ' L K(T. we obtain exactly the nonvanishing terms on the right of Eq. (23.n) i=l N lim exp £*(T. [44]. Thus in our case one of the kernels is zero and we have to decide what we do when. .T2) I exp I K(n.lnJV(T)}] N(T) N(T) (23. p.T')dT' (23. X(TI. (23.23..n) o K(TJ. (23. .64).j. Cameron and W. n — oo the expansion becomes > D lim Dn = 1 + n—*oo J K(r1. r ' ) is to be evaluated at r = T ' .63b).TJ) The factorials arise since every determinant of I rows and I columns appears l\ times when i. T. H.7 Y. 2! .n) K(Tj.66) •Dr) .. (23. as in Eq.n) z=l x (23. In the limit S —* 0 . = 1 52 K(n. This determinant can be expanded in powers of K (more precisely in powers of a parameter A which we attach to K). We observe that if we discretize Eq.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.Ti) i=l : exp \ £ K(T. i.64) Inserting the explicit expression for the kernel given in Eq.a { l n i V ( r ) .63a). the value of the determinant is half (or arithmetic mean value) along the diagonal.65) The quantity D is really the Fredholm determinant which is obtained as above by solving the Fredholm integral equation with constant integration limits. Our final step is to relate the VolterraFredholm determinant to the required functional determinant. the kernels to the left and to the right would be associated with step functions 8(T — r') and 9(T' — T ) . 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. Felsager [91]. r i ) + . 192. 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. so that (with T *T) 1/2 •Dz N(T) (23..T1)dT1 + J dn J dr2. where a f 1 See R.64) as above and consider the determinant of the expression. n )d? (23. Martin [43].. whereas the determinant arises only once in the original determinant Dn. ^This is the method explained by B.62) — and with insertion of a discretization dependent factor a still to be explained — we have in the present case D: . Thus D„ = l + 5 ^ K ( r i . 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.68) D above.N{T) Finally we add the following remarks concerning the discretization (23.TI) 0 K(T2.
(23.59)) + f(T)] = — f dz(T) f°° dae^W+f™.^ .J dz(T) J T>{z} J da•Dz hJ dz{T) v{z} I f_A\^+W)i{T)} J H § I"exp [ . Briefly. .69) Incorporating this condition into the functional integral Jo with a delta function 1 = / dz(T)S[z(T) we have (cf.l + £7t) «i (23.Z ( T ).73) .( .3: Implementation of the endpoint constraints Using a Lagrange multiplier a insert into the functional integral (23. and hence may be ignored. I2 and using induction one obtains (cf.72) The endpoint integration dz(T) = dz(rn+i) can be combined with this.608 CHAPTER 23.57)) z(—T) —> 0) ^)/>'£^') = iv(r)/_T/T'A(__l_)2(T0 T JV(T') T . In = dzi dz2 • • • dzn e x p .22)) In = (7T6)"/2 Vn + 1 exp e(n+l) ( Z n + l . N(T') . (23. (23. (23. Evaluating Ii.e—*0 In.— . (21. so that (with V^N(T)/N(n)) 7t = J V{z}^[JT_Tdr[\[z(r)+ia^)' — lim n—*oo. rn+i = T.Z i . Path Integrals and Instantons Example 23.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. 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) + /_ dr — . Eq. n J=.70) and this now requires also integration over the endpoint coordinate z(T). represents a constant translation. In every integration / dzi the contribution €7. Solution: From Eq. With partial integration (in the second step) we obtain (since (cf. Thus 0 = 2 (rf) + f(rf) or 0 = z{T) + f(T). Eq.71) Hence (with h = 1) l = j .59) the endpoint conditions »)(±T) = 0 and evaluate the integral. f(T) = N{T) J^ dr'^^z(r') (23. T + N(T) / T' = T JT dr' N(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).2o) 2 rc^O.0.N(T) . (23. I0 = — f dz(T) f 1>{z} f da T'Z x exp T>r\ «/>5^> + ix(*<T> + '"< T >/• T *>m™ 7V2(T') (23.
4.3 T h e Faddeev—Popov constraint insertion The FaddeevPopov constraint insertion implies the complete removal of the zero mode. D. / — / / We started off with V(T) = ^CnVniT).11). . £n = ~2 dTT](T)7]n(l (23.4 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. / d£o <9r0 5[F(TQ. ^E. 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. (3.23. i. J co § dr Recall that / f ^ doce^ = yfVfp.12).^ 23. 1/2 T W 2 (r)J (23. dxv 0(r) = (f>c(r) + V(r) and set 1 TOO a — const. (23.76) Since the zero mode T]O(T) is proportional to d(f>c{r)/dT with — SE. (21.£Q) — aTo]dTod£o.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 . Eq. Popov [89]. Patrascioiu [110].59) and agrees with that in the literature.e.. a = const." The method consists essentially in replacing the single integration J d£o by a double integration involving a delta function. This result verifies Eq. N. Ignoring the remaining metric factor with the reasoning explained > between Eqs.10) and (21.74) where we reinserted h. 11 L. Faddeev and V. (23.
e.77) The set {r/ n (r)} constitutes a basis of the fluctuation space about the classical configuration. where L = in + cn. i. (23. (23. Path Integrals and Instantons _^ ^ w . (23. demand validity of) the constraint £o = 0.75) enforces the collective coordinate ro to be given by F(ro.J drm(r)Mr + T0) (23.81) the rdependence is integrated out and hence the integral does not provide the function F(TO.83) **The collective coordinate in the present context is also called "instanton time". The number £o is the component of the fluctuation rj(r). / dTm{r)n{T). in the case of the soliton of the static 1 + 1 dimensional theory this is the kink coordinate. i.e.78) <t>ij) = Y. at r = ro. (23. We establish the function ^(To>£o) with the following reasoning. The delta function 8[F(TQ. Thus.76) we set 4>c{r) = ^CnVnir).610 this means that CHAPTER 23.^n{r).g. We now impose (i. In the integral of Eq. (23.79) (23. (23.^O) with dependence on ro which the replacement (23. e.£o)/a. .£Q) — aro] contained in Eq. Then cn = ~2 / dr(pc(T)rjn(T).e. A position ro of this classical configuration is called the collective coordinate** of (the field) 4>{T). this constraint implies that we go to the subspace which is orthogonal to the zero mode. if the set {Vn(T~)} spans the Hilbert space of fluctuation states about 0C (at collective coordinate position r ) . In analogy to (23.81) 2 / dr»7o(r)0c(r) = . We consider the following integral / ( r 0 ) := . in the direction T]O(T) at the point TQ of the parameter r which parametrizes the trajectory of the classical configuration </>c(r). in the direction of the zero mode.75) requires.80) i r or •i /*oo dr7?o(r)0(r) = 0.82) This condition says that (j)(r) is not to possess any component in the direction of T)O(T). i /*oo (23.
91) " F o r more discussion of this point and related aspects see also B. (23.85) PJ W i t h Eq.84) (23.(23.87) l/M) (23.82) we can replace here 7?O(T)0C(T) by 77O(T)T7(T).£O) OTQ = = £ " .4 Field Fluctuations 611 for an infinitesimal translational shift TO of the position of 4>c(r).88) ro0 Then in Eq. Felsager [91].. (23. We evaluate this quantity as follows in the leading approximation. 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) .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 ) . . p. so t h a t with F(TQ. We have AFP = /F(T0.<t>c{r)} .Zo)y/SETQ We now define A FP '• =0 • (23.y/SET0 = 0.£O) F(r0^0) we have defined as + ro) + T?(T)] = / ( T 0 ) + p£o.y/S^ro] (23.90) Now tt 0^C(T + TQ) ro=0 <9T0 dr (23.23.^) ./ drri0(T)<pc(T) + PJ (23. 158. .89) Here the factor App is called the FaddeevPopov determinant. jdrrj0(T)4>c(r) + JdTm(r)(p'c(T) X/SET0.75) 1 = AFP f dT06[{F{rQ.86) = Jdrrjoir^Mr F(TO. (23.
e.93) where AT = TT — 00.92) 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 (23.94) ' ' T h i s is an important result which will be required later in the explicit evaluation of path integrals.e. /(TO) ^/SE~T0} V~SET0 c d£. P J AT (23.90). the "static energy" invariant. in d(j)c(To) + dr/fa). if the coefficient of 770(70) vanishes. 7 Thus there is no fluctuation along the direction of the zero mode in the sum.e. P the relative sign being of no significance. Consider. 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. i.77) between the zero mode d(^ c (r)/dr and the fluctuation vector in its direction.91) into Eq. =~ (23. ~ There is a quick way to obtain the result (23. (23. i.0AFP8 4o H P . Path Integrals and Instantons Thus a shift TQ in the direction of the zero mode (i. tangential to (J)C(T)) leaves this and hence the square of this quantity.93). iytQ 0C(TO + dr0)  4>C{TQ) = dT0^'c(T0) = dr0 7 0 (r 0 ).e. we obtain (remembering that % = pd(j>c/dT/y/S^) AFP = 4 = f dr ( ^ V \/!5E JT \ dr * = yfe.75) now becomes d£0AFP8[f(T0) = = / dr0 / J&T J + p£0 . Inserting (23. fs d(/)c{ro) = We have also drj(T0) = d£0r]o(To) + ^2d£ir)i{To). 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./ dT0AFP. 770(7").612 CHAPTER 23. 0 d r (23. . i. if ^ + d£o = 0. using the relation (23.
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. (23. Since w\ is for finite T given by WQ of Eq. We have to take into account also multiinstanton configurations which almost stationarize the action integral and satisfy the same boundary conditions as the instanton.96) lim .43)).55b).55b) implies I'o w0p m \ ' LOQP '2K irh) m \ ' irh) 4:\/6mp mAT (23.60a) in not involving the zero mode contributions.4 Field 23.98) . 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.96) does not yet allow a comparison with (23.4 Fluctuations T h e single instanton contribution 613 We now return to the multiple integral (23.52a) and replace in this J d£o by the expression (23.A T A F P i " o e x p T—>oo p where I'o n^O.35a) since the dependence on Tf — Ti = 2T of both expressions differs significantly. We observe t h a t the result (23. We can now perform the Gaussian integrations and write the result lim T^oo p 4 m3 exp 3^J (23. so t h a t h = det r)/(n\ppATexp 4 m3' ATAFPdet(^ \ a£ jexp (23. (23.93). we can set I 2TT ^o = l'o\ wlp2'' which with Eq.97) is a constant still to be evaluated. 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. (23.4. We determine I'Q by observing t h a t this differs from IQ of Eq. (23.60a) for I0 and Eq. (23.
they have to introduce the inverse of that factor on the far right of their expression (3. (23.2 m T e x p 2 irgh ' T»oo (23.100) in such a way that it exhibits the crucial factors of which it is composed. This means we are concerned with configurations like* 2n+l d(r . ~ rKo')e{Tx .101) 2T > oo (cf. i.r).a n t i .100) We observe that the dependence on the normalization constant p cancels out. the approximation is then known as the "dilute gas approximation". Patrascioiu [110] make the replacements \QP <> 2g 2 and p?GP <» 2m 2 .5 I n s t a n t o n . In view of their normalization of the normal modes. (3. . 0c M. we drag the factor p2 along for ease of comparison with literature.4>C exp h (23. the result becomes h lim 2 m r ^ ^ 1 e . Gildener and A. (23.4.99) limlATAFF/0f^^eA^exp 4 m3 3 / ^ Inserting the explicit expressions for IQ and App. their Eq.614 CHAPTER 23. Patrascioiu [110]. ^See 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. *For the validity of the calculation the instantons or kinks have to be far apart. T0 < 4l) <C Ti « 42) < • • • «C T 2 n + 1 = T — oo (23.i n s t a n t o n contributions As already mentioned. Although we choose p2 = 1.e.TiMiy+'Mr «>. 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 .32). Thus — the subscript 1 indicating that it is the Feynman amplitude for the oneinstanton contribution — lim where with A T ATAFPI0 W0 SE{.* Finally we rewrite the expression (23.55b)) w0 = v0e VQ 23. Gildener and A.20).102) *For comparison with the parameters of E.
(23.17b) and there perform first the integrations over the intermediate configurations whose endpoints are not fixed.(r) = ^ 2 n + 1 ) ( r ) + 77(r) with r ^ ) « 0.6.T2}((t)2. (3) (23.4>i.23. this means we have to consider the amplitude (23.Tl\(j)i. Consider (with T 0 = T. .36) (/. .r 0 ( 1 ) = .2 T / 3 .105) where we write 'proportional' since only the classical instanton action depends on the endpoint configurations.fo) = Id2 St SE(<Pf.<Pi)2 + • • • .Tl}{4>l.Ti} OO J — OO /OO OO OC OO d(f>2 J — OO LeSE{<l>f.4>l)eSE{4>l.103) For instance for n = 1 one has the configuration <jyc (T) which has the form shown in Fig.106b) SE{<t>l. here (f>2. (23. Taking these configurations into account.l ) i + 1 in the sum of Eq.(l>i) (23. . Proceeding as before we now consider first in analogy with Eq. we can expand the Euclidean actions here involved as follows (the first derivative vanishing at a solution): SE{<j>f. 7 i = . 1 0 6 a ) and SE{(J>I. 23. ( 2 3 .T2\(f)l.<t>ff + h . Since these endpoint configurations.<j>i) . (23.4>I) ld2SE SE(c/>i.Tf\(p2.Ti) = X^/' T /I&> r i> ( 2 n+1) n=0 (23. r0(3) = 2T/3.<t>2)eSE{<t>2.104) Considering the case n — 1 of a kink with one additional kinkantikink pair.102) distinguishes between instanton (rising from left to right) and antiinstanton (rising from right to left).r/l&.r / 3 .H)2 + .4 Field Fluctuations 615 for n = 1. T2 = T/3.<j>i) + 2 d<& 1 r)2 *? k? + (<f>2 . are varied. the total Feynman amplitude is given by (^/. The factor ( . r 0 ( 2 ) = 0.<t>f) + 2 d</>{ SE{4>fi4>i) + ld2SE 2 d(/){ ( h . (n = 0 implies the single instanton solution). T3 = T) d(f>l((pf. 2 .
It follows that — always within our approximations — c (^TflfaTjW SE{<t>f.106b) is to be substituted into the second relation of (23. rn ^2 J2H 9 1m' ±2gcj).e. the dilute gas approximation. The allowed positions of the individual kinks have to be such that their order is maintained. each of the three intermediate kernels contained in Eq. Thus when in each of the three cases the zero mode elimination has been performed with the help of the FaddeevPopov method.e.105) can in other respects be treated in analogy with the single instanton amplitude.12) we have JM~°°) i. in the case of l2n+i therefore ry2n. ±m/g (23.109) The result of these integrations is that the amplitude l2n+i involves 2n intermediate integrations.4>f) d4>\ 4>s72 (23. of course. d2SE{4>) d(j)2 2m. (23.2d2SE ~c 4>i d(f)i e x p SE^f. With our remark after Eq.616 CHAPTER 23. (23.4>i) / d(f>2 e x p {<h . On the assumption that the kinks of the instantons are widely separated. each of which implies a multiplicative factor 7. It has to be remembered. i. i..106a).e.108) where the last expression is obtained as follows. (23.4>i) { 2 " 2d SE .(01 . dT 9 70(oo) V m J sE(4>) = ± m Hence d2SE{<t>) dej)2 .2J. 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.e.107) with (the evaluation is given below) 7T 1/2 7 = IT 1/2 1/2 li&SE/dfi) {d^SE/dcft) 2m J (23.3 9 L 3m 2 and 9SE((/)) ± . the . i.
e. T ] with ] T 3 = 1 A » T = A T = 2T.)(3) 1 o A T 3 Ao lim .Tj)( 3 ) these considerations imply the following result parallel to t h a t of the single instanton calculation (23.^ .112) We now compare this result with expression (23.101): (^/.r/^. 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) .r.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.4 Field Fluctuations 617 replacement (23. Gildener and A.(2) (1) (2) (3) _ ( 2 T ) 3 JT JT JT 3! (23.A 3 r>oo 7 ' 3! exp SE(<J>C) T6 VQ mAT FP10 3 h (23.93) has been applied.2 T / 3 this product becomes that given by E. In the case of (<^f.6 the collective (i) coordinate TQ has to be to the right of TQ .23. then sign reversals TQ i ) .V ^0 SE{4>C X exp mAT lim — e ~mAT Sinh T^oo 7 IjATAppIo vo exp h (23. Am3 2m J rW ( r ( 0 and this followed by setting aj.105) t h a t each of the three amplitudes contributes a factor e x p [ .m A . (3) ° JrW: 2T/3 (2T)3 .110) Finally we have to recall from Eq. a 2 0. and in the case depicted in Fig. we now have to integrate over three collective coordinates. Patrascioiu JT/3 ° ir< >2T/3 1 dr.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 . (23.(*) o ax = 2 T / 3 . 23.a 3 [110].7 3 ^ — .T/0j. i.
by either replacing A T by 2AT or doubling A*.e. — the relation between the parameters is ^LMK = ^I^GP = 4 r " 2 a n < i C\MK ~ \^<3P = 92. Patrascioiu [110]. Wiedemann [4]. We observe one more aspect of our calculation. has been obtained without explicit use of canonical commutation relations (i. . AE.113) This is (in slightly different notation) the result of Gildener and Patrascioiu [110] for consideration of the instanton plus associated pairs.618 and hence AE A. The difference between levels. even without explicit specification of the prefactor. To help the comparison we note that — with indices indicating the authors. MiillerKirsten [30].Q.113). 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. W. Gildener and A. W. J. W. MiillerKirsten [163]. Taking similarly into account the antiinstanton (since the physical level splitting does not distinguish between an instanton and its antiinstanton).113) is very complicated. Finally we realize that the computation of (23. we obtain AE = 2A* and agreement with the result (18. Path Integrals and Instantons 8V2hm5/2 = exp ( 4m3\ . here as 1) is also taken into account. Bose and H. (23. explicit quantization).^r . K. (18.113) for qo = 1: A1E „ .W We also see that the level splitting is a nonperturbative expression (with an essential singularity at g2 = 0). J. (23. 2 V TT [ i ( g „ _ i ) ] ! V 9 J V 3 9 HJ For an elementary discussion see S.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. ^See P. whereas the method of the Schrodinger equation in this case is of the same degree of complication as for the ground state. H. J. Achuthan. We shall see that the path integral applied to excited states is even more complicated in necessitating the evaluation of elliptic integrals. J. Liang and H.182) (with mass = 1 and H = 1) becomes the following which implies twice the result (23. MiillerKirsten and A. GP for those of E. l CHAPTER 23. l~2h = 2qo+2\ = m /4m3Vo/2 / —*exp 2 P 4 m3 \ — . .182) of the Schrodinger equation method provided the difference in the mass of the particle (there taken as 1/2. 23.With this correspondence A\E of Eq. 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 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.
Chapter 24 P a t h Integrals and Bounces on a Line 24.a 2 ) 2 + c^T2. Potentials of this type play a significant role for instance in models of the early inflationary (expanding) universe (as a result of its cooling). Since problems of instability are ubiquitous and occur in all areas of microscopic physics their understanding is of paramount importance. 619 (24. Consider for instance the temperaturedependent potential yr(0) = i A ( 0 2 . however. 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.2) . We shall see that these classical configurations are associated with quantum mechanical instability and hence with complex eigenvalues. i. Previously we considered in particular the quartic or symmetric double well potential for a scalar cp.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). and their interpretation and uses. in a somewhat different formulation.
but implies instability for T < Tc. Thus if we start with the system (like the early universe) at a high temperature and then allow the system to cool.A a 2 ) .620 CHAPTER 24.a2)4> + 2cTz 2 0. its symmetry under .e. a spontaneous symmetry breaking leads to the double well (i.1 The asymmetric double well. T = 2\(<f>2 . A2 _ a — \„2\ u ) ~r u/\y> 2 (24. We see that <fi = 0 minimizes VT(</>) for temperatures T > TC. Path Integrals and Bounces on R1 Here dVT{4>) vanishes for . J 4.3) Since d2VT{4>) 2 d^ we see that ~ ^ ' > (d2VT(<f>)\ V # V # 2 2 A=o 2 2 = 2(cT 2 . 24. at the minimum of the potential the case < = 0 changes to the case 0 ^ 0 ) .2=a cT /\ Thus there is a critical temperature T c := (X/c)l'2a. We can destroy the reflection symmetry of V((f)) (i. = 4(ACT2 cT 2 ).e. > / V[<H 'true'vacuum 'false'vacuum Fig.
24.t In the following we consider first (after some recapitulations) a simple model of a false vacuum. We shall see that: (a) solitons (or equivalently instantons in the onedimensional theory) describe vacuumtovacuum transitions.g. .) .g. L. Cornalba. V(<j> = m/g) = 2e. tf(*) =^ 1 . 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).g. the quantum mechanical vacuum).6) The EulerLagrange equation is given by • $ + */'($) = 0 (metric + . (24. also in contexts of string theory. the vacuum being understood as the classical. <44 2) ' This is now an asymmetric double well potential as in Fig. Guth and S. Since the true vacuum minimizes the potential. a linear term as in w^^'A1^'*0V(<l> = m/g) = 0. Costa and J. False vacua play an important role in models of the inflationary universe* and elsewhere. The Lagrangian density of the theory is given by C = dp*d»*U(*). H. T False vacua and bounces occur.^J >o. .5) We see that one minimum of V now lies higher than the other.1 Introductory Remarks 621 exchanges 4> <* — <P) by adding e. see e. and (c) determine the lowering of the energy to the true quantum mechanical ground state (i. (24. which (b) are saddle point configurations of E (or SE). On the other hand we shall also see that (a) bounces describe vacuumtoinfinity transitions. and (c) determine iQE of E which results from a nonvanishing probability of tunneling away from the classical vacuum only. for instance. the higher minimum is called a "false vacuum".e. S.1 with (24. Pi [126].7) *See e. A.24. Penedones [59] and references cited there. M. (b) they are minimum configurations of the energy E (instantons of the Euclidean action SE).Y. perturbationtheory vacuum). We recapitulate briefly a few points from our earlier detailed treatment of (1 + l)dimensional soliton theory.
so that the EulerLagrange equation becomes the Newtonlike equation <t>"{x) = U'(</>). (24. (24. — XQ) (24. i. We observe that this kink solution is odd in (x — XQ). we see that (24.9) 9 for the doublewell potential given above.10) Inserting (pc and evaluating the integral we obtain E&c) Am6 3ff 2" (24. Path Integrals and Bounces on R1 Static fields given by d$(x. t) = <f>c(x) + Y^ iPk{x)ekt (24.8) This equation possesses the kink solution <t>c{x) m tanhm(a.7) yields the following equation called stability or small fluctuation equation.12) Since <fic(x) solves Eq.e. (24. which is an important point as we shall see. We have seen previously that the same equation is also obtained from the second variational derivative of the energy functional. dx2 + u"(<j>) 5(x y). The energy of this classical configuration is obtained by evaluating the Hamiltonian for (f) replaced by <j>c.14) . Thus dx J —c 1 k'\2 {4>'cy + c/(0 c (24. h " 52E(<j>) 8(j){x)5(j>{y) +u {Mx)) = ip {x) k wlipk(x). i. (24.8). small perturbations around </>c(x) do not grow exponentially with t (the zero eigenvalue requires special attention).622 CHAPTER 24.11) We investigate the classical stability of the configuration by setting in the full t—dependent equation for small ipk: iw $ ( s .13) For small w\ > 0.t) dt 0 are written (f>(x).e.
1) the fluctuation equation becomes 6m