INTRODUCTION TO OUHNTUM MECHANICS
•
Schrddinger Equation and Path Integral
INTRODUCTION TO QUANTUM MECHANICS
Schrodinger Equation and Path Integral
This page is intentionally left blank
*»«*?:#•
m
m
't
Schrodinger Equation and Path Integn
University of Kaiserslautern, Germany
\(P World Scientific
N E W J E R S E Y • L O N D O N • S I N G A P O R E • B E I J I N G • S H A N G H A I • HONG KONG • T A I P E I • CHENNAI
Published by World Scientific Publishing Co. Pte. Ltd. 5 Toh Tuck Link, Singapore 596224 USA office: 27 Warren Street, Suite 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
e. has not always been understood well. that is the Coulomb potential and the harmonic oscillator. With the growing importance of models in statistical mechanics and in field theory. Excluding spin. 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.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. screened Coulomb potentials and maybe singular potentials. one problem being that there are few nontrivial models which permit a deeper insight into their connection. In the first part. in Feynman diagrams in quantum electrodynamics). again point the way: For scattering problems the path integral seems particularly convenient. but also about complex energies that he encounters in a parallel course on nuclear physics. Thus important level splitting formulas for periodic and anharmonic oscillator potentials (i. we recapitulate the origin of quantum mechanics. its mathematical foundations. These basic cases will be dealt with in detail by both methods in this text. even though the expansions are mostly asymptotic. Thus this first part . To what extent the two methods are actually equivalent. 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. be treated to a considerable degree of satisfaction perturbatively. the aforementioned exactly solvable cases. any problem more complicated is frequently dispensed with by referring to cumbersome perturbation methods. like periodic potentials. with degenerate vacua) were first and more easily derived from the Schrodinger equation. 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. With various techniques and deeper studies. However. and it will be seen in the final chapter that potentials with degenerate vacua are not exclusively of general interest. The introduction to quantum mechanics we attempt here could be subdivided into essentially four consecutive parts. in fact. 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. numerous problems could.g. but arise also in recently studied models of large spins. basic postulates and standard applications. whereas for the calculation of discrete eigenvalues the Schrodinger equation. Chapters 1 to 14.
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).e. Thereafter the concepts of instantons. Using perturbation theory. 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. In the second part. 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 deal mostly with applications depending on perturbation theory. This is followed by the important case of the cosine or Mathieu potential for which the perturbation method was originally developed. which is presumably the only such singular case permitting complete solution and was achieved only recently. 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. i. we derive respectively the levelsplitting formula and the imaginary energy part for these cases for arbitrary states. 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. 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.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. Chapters 15 to 20. After a treatment of power potentials. and the behaviour of the eigenvalues is discussed in both weak and strong coupling domains with formation of bands and their asymptotic limits. The earlier Chapter 17 also contains a brief description of a similar treatment of the elliptic or Lame potential. the chapter thereafter deals with Yukawa potentials. and the elliptic or Lame potential — here introduced earlier as a generaliza . the method of matched asymptotic expansions with boundary conditions (the latter providing the socalled nonperturbative effects). The potentials with degenerate minima will be seen to reappear throughout the text. The following Chapter then deals with Schrodinger potentials which represent essentially anharmonic oscillators. periodic instantons. We also consider inverted double wells and calculate with the path integral the imaginary part of the energy (or decay width). and their eigenvalues. The most prominent examples here are the double well potential and its inverted form.
This puts Schrodinger equations with e. In fact. an approximation whose higher . Employing anharmonic oscillator and periodic potentials and reobtaining these in the context of a simple spin model. for instance. 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. The physical behaviour there (in the transition region between quantum and thermal physics) is no longer controlled by the Schrodinger equation. cubic). the Mathieu equation. With a fully systematic perturbation method and with applied boundary conditions. the Schrodinger equation can be solved for practically any potential in complete analogy to wellknown differential equations of mathematical physics. quartic. such as limiting procedures. from our precise reference to unavoidable elliptic integrals taken from Tables). The introduction of collective coordinates of classical configurations and the fluctuations about these leads to constraints. we can make the following observations. this comparison on a transparent level being one of the main aims of this text. 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. anharmonic oscillator potentials on a comparable level with.g. We then illustrate the relevance of this in the method of collective coordinates. for instance. it is. except that these are no longer of hypergeometric type. In addition this part considers in more detail the region near the top of a potential barrier around the configuration there which is known as a sphaleron. The particular solutions and eigenvalues of interest in physics are — as a rule — those which are asymptotic expansions. and whenever available also with the results of WKB calculations.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. 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. however. All results are compared with those obtained by perturbation theory. 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. The application of path integrals to the same problems with the same aims is seen to involve a number of subtle steps. 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.
also for the verification of results.g. Instead a glance at a nearby footnote provides the reader immediately the names of authors. 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. with the source given in the bibliography at the end. Dingle (then University of Western Australia.Q. e. Tchrakian (Dublin) and Jianzu Zhang (Shanghai). whose research into asymptotic expansions laid the ground for detailed explorations into perturbation theory and large order behaviour. an additional motivation was that a sufficient understanding of the more complicated of these problems had been achieved only in recent years. The author has to thank several of his colleagues for their highly devoted collaboration in this latter part of the work over many years. Liang (Taiyuan). as well as applications and illustrations. The line of thinking underlying this text grew out of the author's association with Professor R. W. Whittaker and Watson [283].g.g. This endeavour developed into an unforeseen task leading to periodic instantons and the exploration of quantumclassical transitions. . the references referred to are never cited by mere numbers which have to be identified e. K. at the end of a chapter (after troublesome turning of pages). shows him that each such formula here has been properly looked up). Nonetheless. which is particularly important in the case of elliptic integrals which require a relative ordering of integration limits and parameter domains. Andrews). H. Throughout the text some calculations which require special attention. we also consider at various points of the text comparisons with WKB approximations. other topics have been left out which are usually found in books on quantum mechanics (and can be looked up there). thereafter University of St. Since this comparison was the guideline in writing the text.* H. 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. in particular Professors J.XIX order contributions are difficult to obtain. T. MiillerKirsten *In the running text references are cited like e. Their deep involvement in the attempt described here is evident from the cited bibliography. Whittaker and G. D. Watson [283]. J. For ease of reading. are relegated to separate subsections which — lacking a better name — we refer to as Examples. 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. N. B. As a rule. not the least for permitting a more detailed and hopefully comprehensible presentation here. formulas taken from Tables or elsewhere are referred to by number and/or page number in the source. so that the reader is spared difficult and considerably timeconsuming searches in a source (and besides. The author is deeply indebted to his onetime supervisor Professor R. D. like E. B. Park (Masan).
Planck had arrived at his formula with the assumption of a distribution of a countable number of infinitely many oscillators. Kirchhoff's law in thermodynamics says that in the case of equilibrium. very few explain in this context what he really did in view of involvement with statistical mechanics. the amount of radiation absorbed by a body is equal to the amount the body 1 .1 Origin and Discovery of Quantum Mechanics The observation made by Planck towards the end of 1900.and RayleighJeans laws for the limiting cases of small and large values of the wavelength A (or AT) respectively is generally considered as the discovery of quantum mechanics. We do not enter here into detailed considerations of Planck.Chapter 1 Introduction 1. which involved also thermodynamics and statistical mechanics (in the sense of Boltzmann's statistical interpretation of entropy). 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.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). we want to single out the vital aspect which can be considered as the discovery of quantum mechanics. Thermal radiation (with wavelengths A ~ 10~ 5 to 10 . A "perfectly black body" is defined to be one that absorbs all (thermal) radiation incident on it. that the formula he had established for the energy distribution of electromagnetic black body radiation was in agreement with the experimentally confirmed Wien. Although practically every book on quantum mechanics refers at the beginning to Planck's discovery. Instead.
1) ' y Here u(v. energy per unit volume) of the radiation (i. and then measuring the increase in temperature of the heat bath. v + dv in equilibrium with the black body at temperature T.626 x 10 . the formula (to be explained) u(u.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. How did Planck arrive at the expression (1. i. Black bodies as good absorbers are therefore also good emitters. The parameters k and h are the constants of Boltzmann and Planck: k = 1.e. radiators. T) were known and tested experimentally. In Eq. Let us look at the final result of Planck. Introduction emits. 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. and the other in the region of large A (or AT).T) = dv3eC2U/T.e.1) c is the velocity of light with c = u\.2 CHAPTER 1.e.38 x 1(T 23 J K'1. u(u. of the photons or photon gas) in the cavity with both possible directions of polarization (hence the factor "2") in the frequency domain v. kT kXT (1. h = 6.3 4 J s.T) = 2*?£(?)kT. (1. 1. T)du is the mean energy density (i.e. A being the wavelength of the radiation. It was found that one expression agreed well with observations in the region of small A (or AT).1 Absorption in a cavity. i.l ) where x = ^ = ^ . These expressions are: (1) Wien's law. (1. y J c 3 \ex . Fig.2) and the ..
C3 being constants. When Planck had found this expression. Indeed.T) = av e6"/Ti' where a and b are constants.1.2 Distributing quanta (dots) among oscillators (boxes). in the first place Planck had tried to find an expression linking both.T) = 2^C3T.1) in regions of a.1 Origin and Discovery of Quantum Mechanics (2) RayleighJeans law: u(i>. and thus over a discrete number of admissible states. we obtain: u(i/. To this end he considered Boltzmann's formula S — klnW for the entropy S. every oscillator corresponding to an eigenmode or eigenvibration or standing wave in the cavity and with mean energy U. he searched for a derivation. Here W is a number which determines the distribution of the energy among a discrete number of objects.T) 2^^kT. which are indistinguishable) among the N indistinguishable oscillators at . 1. but — here the discreteness appears properly — only in elements (quanta) e. (1.3) are contained in Eq. (1. (x large). and he had succeeded in finding such an expression of the form u(v. so that W represents the number of possible ways of distributing the number P := NU/e of energyquanta ("photons". We see. 3 (1.3) Ci. This is the point. "small" (i. exp(x) ~ 1+x) and "large" (exp(—x) < 1). C2. Planck now imagined a number T of oscillators V or iV oscillating degrees of freedom.2) and (1. that the formulas (1. discretization begins to enter. where the Fig. T) u(i/. e xhv.1) as approximations. Moreover Planck assumed that these oscillators do not absorb or emit energy continuously.e.47TZ/ 2 {x small). Considering Eq. .
6) Fig. = 1/T). Then W is given by w = (N With the help of Stirling's formula* {N + piy. N * oo.1) u = vmrri (L5) as the mean energy emitted or absorbed by an oscillator (corresponding to the classical expression of 2 x kT/2.e. Agreement with Eq. M.2. i. v + dv. e.3 Comparing the polarization modes with those of a 2dimensional oscillator.1)!P! (1.7) *See e. S. u(i>. Oberhettinger [181]. by multiplying U with the number nvdv of modes or oscillators per unit volume with frequency v in the interval v.3. e = his. p. p. formula 8. there not called Stirling's formula. one obtains (cf.g.4 CHAPTER 1. . We now obtain the energy density of the radiation. Introduction temperature T. 1. W.2) requires that e ex is. We visualize the iV oscillators as boxes separated by N — 1 walls. Magnus and F. Ryzhik [122]. and the second law of thermodynamics ((dS/dU)v Example 1. as in most other Tables. . with riydu — 2 x —w—dv. 1. (1. 940.4) IniV! ~ JVlniViV + O(0). as for small values of e). with the quanta represented schematically by dots as indicated in Fig.g.e. Gradshteyn and I.T)dv. The Stirling formula or approximation will appear frequently in later chapters. U{T) being the average energy emitted by one oscillator. I. i. (1.343(2). (1. h = const.
In terms of A we have u(X.e..2. so that the derivative of u implies (x as in Eq. so that (lvL\A 0 I I = rr. rii = 1.7) for instance. Then L^j = nrii. (1.3 (as for ideal conductors).UJ = 2KV. d ™4*±\IL> — —dv = nvdv dv dv _8 3 \ c / 14 8 2 4TTV2 = 3 dv 83 ^ ^ = ^ ^ ' as claimed in Eq.. 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. '''From the equation I \ JW .3.1. where^ 2 [2KUY . as in electrodynamics.UJ2/C2 + K? = 0.T) has a maximum which follows from du/dX = 0 (with c = vX).965 and xmin = 0. (1. as indicated in Fig. (1. v + dv.i = 1.V 2 ) E = 0.1..8) This is Planck's formula (1.2... nvdv per unit volume. We observe that u(v. 2 2 2 r2 L K — 7T n .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.2.1).. so that . . L for i = 1. We obtain therefore u^T) = Unv = 2^fJ^—i.1)) The solutions of this equation are ^max = 4. i.7). is given by . KT = I J . The number with frequency v in the interval v.3.3. We obtain the expression (1. . The number of possible modes (states) is equal to the volume of the spherical octant (where n^ > 0) in the space of n^.T)dX = ^ehc/*kT_idX. dj\l dM .
n = 0.10). which did not arise in Planck's consideration of 1900. ra = 0. 2 . However. and from which the constant h can be determined from the known value of k. Lorentz and Planck that Eq. A. . We are not dealing with the linear harmonic oscillator familiar from mechanics here. that it is easier to consider first the case of T = 0. which can assume the discrete energies en — nhv. we then have at T = 0 independent oscillators. Thus we have a rather complicated system here. We therefore examine such contradictions next. i.965K = Const.e.—v <L9) We observe that for T — 0 (i. that of an oscillation system at absolute temperature T ^ 0. (1. the behaviour of the system at zero absolute temperature. such a procedure leads to contradictions.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=—. One expects. which can not be eliminated without a different approach. This is Wien's displacement law.10) Thus here the socalled zero point energy appears. Introduction he 4. . which had also been known before Planck's discovery. Since temperature originates through contact with other oscillators. x — oo) the mean energy vanishes (0 < U < * > oo). If an oscillator with thermal weight or occupation probability exp(—nx) can assume only discrete energies en = nhu.8) could be derived much more easily in the context of statistical mechanics. .1. that we arrive at quantum mechanics simply by discretizing the energy and thus by postulating — following Planck — for the harmonic oscillator the expression (1. Later it was realized by H. 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 .e.6 The first value yields AmaxT = CHAPTER 1. but one can expect an analogy. of course. One might suppose now.l.
1)! .1 Solution: Inserting W into Boltzmann's formula and using In TV! ~ A In T — TV. so that 1 f ds\ T — \dUjy . for 2 degrees of freedom.i n .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. kT/2.1 ' u~ e . with Heisenberg's discovery of the uncertainty relation.2 Contradicting Discretization: Uncertainties The farreaching consequences of Planck's quantization hypothesis were recognized only later.. .kT This means U is then the classical expression resulting from the mean kinetic energy per degree of freedom.1)! . We "This is what was effectively done before 1925 in Bohr's and Sommerfeld's atomic models and is today referred to as "old quantum theory".k dU 1+U\ f ln(l T + U\  U U k ( e .+ 1 e \U e u = exp(e/fcT) which for e/kT — 0 becomes > .= . {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) .In .2 Contradicting Discretization: Uncertainties 7 Example 1.1.e. 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. d . 1. we obtain T V S =fc[ln(TV+ P . 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: 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. around 1926.ln(TV .
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
since .24 CHAPTER 2. which can be observed directly at time t. i. The following properties can be verified: *See e. chapter VII.P. Compared with an arbitrary function f(qi.1) as d . chapter VIII. whereas the velocity requires observations of space coordinates at different times.4) as the generalization of Eqs. (2. All functions u(qi. Rather. M.. the entire timedependence of observables u(qi. One now defines as (nonrelativistic) Poisson bracket the expression^ With this definition we can rewrite Eq. the equation of motion of the observable u.g. in analogy with Eqs.1). A.Pi) is contained implicitly in the canonical variables q^ and pi.4) contains as special cases the Hamilton Eqs. this expression is simply a functional determinant. It suggests itself therefore to consider more closely the properties of the symbols (2. Dirac [75]. 6t>0 5t Real quantities which are directly observable are called observables. Goldstein [114] remarks at the end of his chapter VIII. S »(«.) = £ [wm + WiK) = £ [WiWi . Goldstein [114]. the standard reference for the application of Poisson brackets is the book of P. (2. H.2) as This equation is. qi(t + 6t) . du \ x^/dudH du dH\ . (2. Pi has to be considered as an independent quantity.qi(t) qi = hm f . dH In this Hamilton formalism it is wrong to consider the momentum pi as mqi.n n.1). . A system consisting of several mass points is therefore described by a number of such variables.t).^ ^ j . one obtains the Hamilton equations* • OH . Hamiltonian Mechanics is wellknown.p% are therefore again observables.3). We can therefore consider Eq.1).2) If we have only one degree of freedom (i = 1). which all together describe the state of the system. (2. (2. (2. x^fdu. The total time derivative of u can therefore be rewritten with the help of Eqs. It was only with the development of quantum mechanics by Heisenberg and Dirac that Poisson brackets gained widespread interest in modern physics. as mass times velocity. One can verify readily that Eq. (2.e. . ^As H.Pi. (2.pi) of qi.
5d) = {A*. B}. (2.5a) (2.5d) is useful in calculations. If. we expand A and B in powers of qi and pi and apply the above rules until only the fundamental brackets remain.2.2 The Hamilton Formalism (1) Antisymmetry: {A.C}.3) of the Poisson bracket. where A and B are arbitrary observables. These are {Qi. which combines the Hamilton equations.5c) The first three properties are readily seen to hold. (2.A}. {C. it is irrelevant whether we write {A.5b) (3) complex conjugation (note: observables are real. we obtain the fundamental Poisson brackets. B2}. that the very general Eq.B*}.B}C or C{A. (2.B}* (4) product formation: {A. that of the linear harmonic oscillator. the Poisson bracket {A.4).6). Bx} + a2{A. Later we shall consider noncommuting quantities. {pi. Property (2.6) We can now show. for example. then the ordering is taken as in (2. in other words without any reference to the original definition (2.B}. B}} = 0.B} = {B. {A.B}C + B{A. we wish to evaluate {A. Since Eqs. like here. but could be multiplied by a complex number): {A.5d) above. The original definition of the Poisson bracket will not . can be solved solely with the help of the properties of Poisson brackets and the fundamental Poisson brackets (2. (2.5e) = {A. (2. C}} + {B. A}} + {C. B} is completely evaluated. i. a i S i + a2B2} = ax{A.e. If we evaluate the Poisson brackets for qi.Pk} = 5ik.Pi.Pk} = 0.BC} (5) Jacobi identity: {A. 25 (2. (2.6) give the values of these. (2) linearity: {A. {B. As an example we consider a case we shall encounter again and again.Qk} = 0. As long as we are concerned with commuting quantities. {qi.
and Pi are ordinary real number variables and that H(q.qot2 . In the evaluation one should also note that the fact that g.p.8b) p=q.7) q = [q.H}.H}.4) we have for u = q.10) In classical mechanics one studies also canonical transformations.p. Then we obtain: (2.p}p + p{q. "4' = q. These are transformations qi—>Qi = Qi(q. Since constants are also irrelevant in this context..p\) P. and so q = q.11a) 'q' = q. Hamiltonian Mechanics be used at all. According to Eq.yPot3 + •••• (2. or q(t) = qo+ Pot . Pi—> Pi = Pi(q. q = {q..12) . and P={p. (2. (2. from which we infer that q(t) — qocost + posint.8a) and use the properties of the Poisson bracket and Eqs.t).p) is an ordinary function is also irrelevant. (2.9) Similarly we obtain from Eq. (2. q + q = o. we consider as Hamiltonian the function H(q. (2.26 CHAPTER 2.7) into (2.9) and (2.t). (2.10) we deduce q = p = q..6).p) = ±(p2 + q2). From Eqs.8b) We insert (2.p.8a) (2. (2.P2} + {q. (2.\(p2 + q2)} = l({q.11b) (2..q2}) = = 2i{q.
. which transforms qi. . {qi. that (dropping the subscripts therefore) {Pi.Qk} = 0.e. which means that a Hamilton function K(Q.13) We write the reversal of the transformation (2.3) we can now express the Poisson bracket {^4. i.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.B}Q>P. this transformation is described precisely by the equations of motion but we shall not consider this in more detail here.P One can then show that {A. and Example 2. {Qi. The proof requires the invariance of the fundamental Poisson brackets under canonical transformations.e.P.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).Pi back to their constant initial values. {qi. for which Hamilton's equations hold. (2. i. _ax p__dK_ (2. those at a time t — 0.e.* Hence in Example 2. P) exists. chapter VIII.e.15a) °raS {AiB}Q. (2.P provided the transformation q. verify that the Poisson bracket of two observables A and B is invariant.2 The Hamilton Formalism 27 for which the new coordinates are also canonical. {QhPk} = Sik.e. i.3 deals with the relativistic extension. as {A. In classical mechanics we learned yet another important aspect of the Hamilton formalism: We can inquire about that particular canonical transformation. Goldstein [114].pk} = 5ik. (2.B}q. Of course.Pk} = 0. Example 2. Eq.P is canonical in the sense defined above. Pi^Pi=Pi(Q.p = {A. *H.P.14) With the help of the definition (2. {PhPk} = 0.qk} = 0.2.t). (2.15a). i.t).2 below contains a further illustration of the use of Poisson brackets.12) Qi—>qi = qi{Q. Example 2.p <> Q.1 we verify only Eq.15a). (2. i.B}q. B} of two observables A and B in terms of either set of canonical variables.
P^pr. The relativistic Poisson bracket (subscript r) therefore becomes du dF dq dp du dF dp dq du &F _ du dF ' {u.15a) r„ ™ v .p = J2 d~Fk Replacing in the above A by P and B by A. we obtain as claimed by Eq.15b) dA {Qk.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).p^ "Pk.p^v dP.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. 236.F}r ~di d~E ~ d~E~8t Consider F = H(q.s F 5 7 r ={A.E(t).Qj}q.28 CHAPTER 2. we have —Et. § E x a m p l e 2.Pj}q.3: Relativistic Poisson brackets By extending qi.p^— + {A. OQk Pk}q. dA {Qk.A}q. 1 „ H = p + m0gq and solve the canonical equations with Poisson brackets for initial conditions q(0) = qo. we obtain dA {Qk.B}q. § See P. Qk}q.pc). Solution: Relativistically we have to treat space and time on an equal footing. {A. Mittelstaedt [197].B}QIP.q/c) and (E. p. (2.p = dB dA dB \ rA 1 {A'B^ E b r s r . Thus we extend q and p to spacetime vectors (t. V V °Qk dPk dPk dQk J E x a m p l e 2. (2.pt to fourvectors (in a (1 + 3)dimensional space) define relativistic Poisson brackets.A}q. Replacing here A by Q and B by A. . their product Et — qp being relativistically invariant.p) .+ dQ.( 9A = {Pk. Thus whenever q and p are multiplied.p(0) = poSolution: The solution can be looked up in the literature. we obtain analogously dA dQk' Inserting both of these results into the first equation. 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 „ „•.
dudH s l u .2. H(q.Po.po) = Gi(q. du q\ P = —. p= p(qo..1 Liouville Equation.. since H is expressed as a function of q and p.16) For example. g oc AqAp. where dr is the difference of proper time given by dudH 1 du dE(t) du — = 1 du ..E(t)}r = Hence at Relativistically we really should have clu/dr. du .3 2. (2. 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.17a). in the case of the linear oscillator with energy E given by Eq. but partial derivatives of F do not vanish.e. (1. and the space spanned by the entire set of canonical coordinates is described as its phase space.dt y/l  qZ/c2' 2.p) . . if qo.p) to G\{q. rrr . i. . Probabilities Single particle consideration We continue to consider classical mechanics in which the canonical coordinates qi.._.p) around some point qo. c dt 2 2 w^ du du dr . dE . it is the equations of motion which lead from Go(q.p).Pi are the coordinates of some mass point m.t). Of course.t0.14) and area A of the phase space ellipse given by Eq. But now we consider a system whose phase space coordinates are not known precisely. and at time t > to in a domain G\(q.3 Liouville Equation.e. to . Since Hamilton's equations give a continuous map of one domain onto another.Po. This probability is evidently proportional to AqAp. We consider first the a priori weighting or a priori probability. 2TTE UJ and hence g oc — .po is a point on Go.y.p).' ±=j1V c \dtj' u^.)•'=(*)**£. so that Go(qo. numerically zero. I . (1. (*.t). also boundary points of one domain are mapped into boundary domains of the other. We distinguish in the following between two kinds of probabilities. g. of course. and E as a function of t). Probabilities 29 (This is..p).po. one obtains Gi with q = q((lo. Instead we assume a case in which we know only that the system is located in some particular domain of phase space. Then .to. Let us assume that at some initial time to the system may be found in a domain Go(q.3. we have A = (p apaq — J . i.
Hamiltonian Mechanics If g depended on time t it would be dynamical and would involve known information about the particle. this has the same value at a time to. dt dq dq dp dp d2H dqdp d2H = 0. Example 2. which means. ? « + •sin 2 6>/' 21E 2IEsin2e' in spherical polar coordinates. as at a time t'0 ^ to • Solution: We consider dln(A9AP)rf(A<Z) ' .4.)curve for constant E and $ is — as may be seen by comparison with Eqs.15) — an ellipse of area § dpgdp. Je=0 . Pi 2 / \. (1. dpdq and similarly dt = — Ap./. Solution: Integrating over the angles we have =2TT fe=7r 2frEsin. the (pg. Example 2. J0=0 Hence g oc 8n2IdE.eded<t> = 8TT2IE.14) t o (1. and hence g oc %ir2IdE. Hence from the difference: — .5: A priori weighting of a molecule If the rotational energy of a diatomic molecule with moment of inertia / is 1 / 2 .1): d .p^. 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.30 CHAPTER 2. = 2TrIEsm6. dt dq and with with Hamilton's equations (2.5 thereafter provides an illustration of the a priori weighting expressed in terms of energy E. Show that the total volume of phase space covered for constant E is 8n2IE. as is demonstrated by Liouville's theorem in Example 2.— = — Aq. dq Aq dq 2 dq Aq dq 2 qH to the right and q to the left. Thus g must be independent of t.4: Liouville's theorem Show that A q A p is independent of time i. — dp A A — In(AqAp)s = — + — = Example 2.. in view of this independence it can be expressed in terms of the conserved energy E.
by jr = P(9»P.Thus we assume a large number of identical sytems.e.p.*). whose positions in phase space are characterized by points.q + dq. (2.p + dp. (2.2 E n s e m b l e consideration 31 We now assume a large number of identical systems — the entire collection is called an ensemble — all of whose initial locations are possible locations of our system in the neighbourhood of the point qo.20) .e. i.1 The system moving from domain Go to domain G\.3.t)dqdp = N. We consider the totality of these systems which is described by a density of points p (number dn of points per infinitesimal volume) in phase space. 2. (2.p.3 Liouville Equation. dn — p(q.t).t)dqdp= 1. it is suggestive to introduce a factor 2KK with every pair dpdq without. i. (2. d d 1 P = JJdqidpi. Since W has the dimension of a reciprocal action.p. Thus dn is that number of systems which at time t are contained in the domain q.19) Thus W is the probability to find the system at time t at q. leaving the basis of classical mechanics! Hence we set / «w)^ =.p. however.Po.17) F ^GT^^ 0 Fig. Probabilities 2.p.2. The total number of systems N is obtained by integrating over the whole of phase space. W=p(q.18) With a suitable normalization we can write this / W(q.
p(q. that in our consideration no additional points are created or destroyed. 2. t)dqdp p(q. p + dp G P O q qndq Fig. The number of points at time t + dt in domain G.p.p / .p) and vp(q.t)dp( — I . how the system moves about in phase space. we take into account.3 and establish an equation for the change of the number of points or systems in G in the time interval dt. we consider the domain G in Fig.32 CHAPTER 2.p. t + dt)dqdp.p. t + dt)dqdp — p(q. Hamiltonian Mechanics We can consider 2irh as a unit of area in (here the (1 + l)dimensional) phase space. is equal to the number in G at time t plus the number that went into G in the time interval dt minus the number that left G in the time interval dt.p. i.a t dt + q+dq.p. In order to derive this equation.t)dp{ jt Q. The equation of motion for n or W is the socalled Liouville equation.3 The region G.2 The ensemble in phase space. — if vq(q. 2. In doing this. i.e.p) denote the velocities in directions q and p — p{q.P dt p(q + dq.e. 2. We are now interested in how n ov W changes in time. i. * q Fig.e.
p + dp.p(q. t)vq(q + dq.t)vq{q.20) and (2.t)vq(q.4) shows that p and u satisfy very similar equations.t)vq(q + dq. put differently.p.P *' % = iH>P} ( 2 .p.p. (2.p.p. Comparison of Eq.t)vp(q.t) dt {H(q.P.p) . dH Mq.19).p. (2.p).p + dp. = K(9. so that dt ~~dq\P Hence dp) + dp\P ~dq)~ ~~dq~~dp~ + dp~~dq~ ~ { .t + dt) . Probabilities and thus p(q.p + dp) dp = or  However.p(q + dq.21 ) This is the Liouville equation which describes the motion of the ensemble or. (2. (2.t)^ = 1.22) The generalization to n degrees of freedom is evident: The volume element of phase space is . .P)]. 33 Dividing both sides by dqdpdt this becomes p(q.W(q. . .p)=p = 3H —.p) dq p(q.21) with Eq.21) we can also write dW(q.t)vp(q.P)]K(Q.3 Liouville Equation.2.t)vp(q. p.p(q. p.p(q.p.p)dtdp +p(q.p) . .p.p)dtdp .t)vp(q.t) dt p(q.t)} with JW(q. (2.p + dp)dtdq.p) = q = g^.p)dtdq . vp(q.t + dt)dqdp — p(q.p.p. With Eqs. the probable motion of the system under consideration. t)dqdp = p(q.p(q + dq.
(2.4 Expectation Values of Observables Let u = u(q. Equation (2.4). The first and most immediate possibility is — as indicated .p.p)W(q. (.p). and hence that equal phase space volumes contain the same number of systems. Hamiltonian Mechanics where is the probability for the system to be at time t in the volume q. We deduce from the Liouville equation the important consequence that ^ M = 0. i. We define as expectation value of u(q.p (cf.1 the area Go is equal to the area G\. Thus in Fig.t)(^J. (2.4).that the density or probability W(q. p+dp.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. p. for i<«>!/«<*p>"w>(^)". (2.po are the initial values of q. t) depends explicitly on time t (if determined at a fixed . 2. that dtj\y v if qo. (2.e.24) since the total derivative is made up of precisely the partial derivatives contained in Eq. Example 2. since no systems are created or destroyed. i. q+dq. and this means — since these systems are contained in a finite part V of phase space — that dt We have in particular. With Eq.26) we described the time variation of the observable u(q.p) be an observable. 2.24).34 CHAPTER 2.p) the following expression: (u)=Ju(q.e.p.
. (2. 1 (2.po.0) = u0{qo.i). where we used Eq.p) assumes certain values) that depends explicitly on time. and hence u(q.t).Po) at time t = 0.W(q. We verify this claim as follows.f(qo.Po. ie. 8.po.p.30) p = f(qo.p) = uo{qo. of course.p0)(^^J. at t = 0.4 Expectation Values.p.t). so that Q = g(qo.t)W0(q0.Po.24): W(q.28) = Ju(q. we can also employ a more complicated consideration. Observables 35 point in phase space).t).t).po. We expect.p. W is the density in the neighbourhood of a given point in phase space and has an implicit dependence on time t.Po.0) = W0(q0.p. Sec.p). Goldstein [114].f(q0.33) po = f(q. (2.t). we can express these in terms of their initial values qo.p){H(q.8.^ Solving the equations of motion for q. (2.p.t). (2. Reversing Eq.t) = = W(g(q0. that (u) = (u)0.p.p.t). (2.p) = u(q. (2.27) becomes !<»>  /«(*P>!"W>(^)" (2. With these expressions we obtain for the expectation value (U)Q: (u)o = Ju0(qo. (2.Po.2.t)}(^y. Then Eq. and the time variation d(u)/dt is attributed to the fact that it is this probability (that u(q. However. (2.t).32) In this expression the time t is contained explicitly in the observable u(q.29). since W oc p is constant in time according to Eq.po.po.e.Po.0) =u(g(qo. Thus we can write.29) The distribution of the canonical variables is given by W(q.t).p0. (2.34) S e e also H.22).t).31) i.t) W(q0. we have Qo = g(q.
i.35).38) we obtain £<«)„ = .P.p).t).t) v{g(<j(q.P.t)\—^\ = (u). (240) .31)) W0(q0. (2.30): duo(qQ.p.po.t) 0H_ 0 q =  {{H(q.25) into Eq.35) Inserting these expressions into no we obtain uo(qo.f(g(q.PO. (2.t).P0. (2.p.M 27Th i . (2.p.t) d t _ (du\ \ d<i dq / P=f(qo.p.p.t).t) (2.t).t). (2.t).t) .t) dp M du\ ° q 'P=f(qo. Inserting Eqs.32).t).t). (2. Taking now the total time derivative of (it)o.t). (cf. d dt {U)0 = dt. (2.37) and (2.29) q = g(g(q.28).P)}).o.p)W{q.Po.f(q. Hamiltonian Mechanics so that on the other hand with Eq.p.0)=u(q." V " " ™0' > • • » « » 0 .p).p.36) = (2.t).po)(~ .p. (2.36 CHAPTER 2.Po.t) M f 9u\ V °P / P=f(q0.39) Substituting this into Eq.p)})q=g(QO.37) as had to be shown.po.t).e. (2.30) u0(g(q.0) u(q.u{q.) Wo(qo.t). (2.p0) = W(q.f(q.t).f(q.36) and (2.Po. / ^fdq0dpc . .u(q.(dq0dp0\n We deal with the partial derivative with the help of Eq. (2.\„.(2.) f / u (q0. we obtain («)o = / u{q.t) dB__ (du\ °P \°PS P=f(qo.p).30) Moreover.Po.p.=9(. p = f(<j(q.t)J(q.p.t).t) = (2 30) (2. Eq.P0.po. Eq.f({H(q. [ du0{q0. we obtain an expression which is different from that in Eq.f(q.POit)i p=f(qo.32).t)W "(q0.p.Po)(^^)n.
This timedependence is described by the Liouville equation dW(q. u}W instead of u{H.34) and use (2. . lim ——uW — 0.44) iji^ioo dpi (which is reasonable since the density vanishes at infinity).43) ——uW = 0. However.4 Expectation Values.42) the relation d .p. (2. (2.28).28) and (2. u}W + u{H.t)}.. (2. In the first case.28). x r „. (2.25) and (2.p).t) dqdp\n 2irhJ (2. The phasespace integral of a Poisson bracket like I (2. described as "Schrodinger picture".41) This expression contains {H. dt <«)o v f u(q..2. Observables 37 Here we perform the transformation (2.p){H(q. uW} = {H.31).W(q. / d q d p \ n 2irh J (2.p)) is attributed to the probability W(q.W(q. 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. .p). and the time variation of (u(q.p).p.41) and (2.W} in Eq. from the properties of the Poisson bracket.p. Consider {H.p.t) of u assuming certain values q and p. ^ „.p)}W(q. we obtain zero after partial integration of I and hence from Eqs.t)} ..(2. we obtain {H.45) the Liouville equation.p). W}.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. the observable u is treated as a function u(q. Alternatively we could deduce from Eqs.u(q.t) dt {H(q.45) in agreement with Eq.42) wxtf)" ~dHd(uW) dqi dpi vanishes under certain conditions.p. so that ~d~t <«>0 = J{H(q.
.t) dt __ ~ (2J30) ( 4) du0(qo.5 Extension beyond Classical Mechanics With the above considerations we achieved a general formulation of classical mechanics.p)." and the explicit timedependence is transferred into the correspondingly transformed observables Uo(qo. which describe the state of a system. " This means: Only in the Heisenberg picture dW/dt = 0. called u Heisenberg picture". the probability of the initial values Wo(qo.po.t) dt du(q. The equation of motion is then that of an observable. which are described as "Schrodinger picture" and "Heisenberg picture"— all this on a purely classical level but with the use of canonical transformations. does not have to be commutative. i.p)}.4)) ^ = {«(.38 CHAPTER 2. (c) A multiplication of the quantities among themselves is defined.t).po are constant initial values — we have du0(q0. (2. This formulation deals with observables u representing physical quantities.p)}. 2.po) is assumed. .46) We thus also recognize the connection between the Liouville equation. (b) As usual a muliplication by a complex number is defined.po. which.4) of an observable on the other. Thus we arrive at a more general theory if we define u and W with the following properties: (a) An addition is defined between the quantities. " T h e author learned this approach from lectures of H. and probabilities W. but does satisfy the usual associative and distributive laws. The time dependence of the expectation values can be dealt with in two different ways. for which the axioms of a commutative group apply. (2.** 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.F(«.rt. Koppe [152] at the university of Munich around 1964. (cf. Hamiltonian Mechanics In the other case.p0. and the equation of motion (2.e. the reason being that — since qo. Eq. as the equation of motion of an ensemble or of a probability distribution on the one hand.H(q.p) It = {u(q. as we observed.
Eq. and to be able to correlate these with the above classical considerations. px > in—. therefore.5 Extension beyond Classical Mechanics 39 Quantities satisfying these properties define a linear algebra.p] = l.2.26)) (u) = Tr(wW). B] := AB — BA in the following way: 'i {q. can be deduced from the following characteristic properties of a trace. Introducing it here assumes.q}=0. In Chapter 9 we attempt a corresponding approach for classical systems with .) = TV (vu). The quantity corresponding to W in quantum mechanics is the socalled "statistical operator'''. (d) Tr(ut.49) With these considerations we have reviewed aspects of classical particle mechanics in as close an approach to quantum mechanics as seems possible.48) In matrix theory the symbol "trace" has a welldefined meaning. y:\p. We shall then interpret as "canonical quantization" the procedure which allocates to each of the fundamental Poisson brackets (2. Moreover.p} = 0 • — ^ ~[q. (2. which all apply to (2. so that we consider this next. (c) TV (u*u) > 0. it is helpful to introduce already at this stage some additional terminology.q} = 0 {p. if u ^ 0.48): (a) TV u* = (Tr «)* = TV^. ox In view of our later considerations. That this is the case.p} = l {q. for a phasespace integral we define the word or symbol "trace". (2.p)(^y. (2. also written "TV". ~[q. Thus we can write the expectation value of an observable u (cf. also called "density matrix". (b) TV(cm + j3u) = oTV u + /3Tr v.p]=0(247) One verifies readily that the commutator relations are satisfied by the differential operator representations qx > x. that its use here implies the essential properties it has in matrix theory.6) a socalled "commutator''' [A. by Traceu:=Ju(q.
. However.40 CHAPTER 2. Thus we can now proceed to prepare the ground for the extension of classical mechanics into an operator formulation. excepting the Poisson brackets. electrodynamics. since gauge fixing (i. the Poisson brackets require modification to Dirac brackets. it will be shown that in electrodynamics. i. a constraint) has to be taken into account.e. and. obtain corresponding results — as one would envisage in view of the expected particlewave duality in quantum mechanics. Hamiltonian Mechanics a wavelike nature. These aspects will be considered in Chapter 27.e.
i. = {ipi} is called a linear vector space on the set of numbers I 6 {C}. of measurable quantities. we can define the Hilbert space as the space of states of a physical system. if the elements ipi of M satisfy the usual axioms of addition and K 41 . quantum mechanics: The Hilbert space as the space of state vectors representing the states of a physical system. i. A set M. and selfadjoint operators in this space as representatives of observables.e. with the canonical commutation relations or Heisenberg algebra defining the basic product relations. 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.2 Hilbert Spaces We first recapitulate some fundamental concepts of linear algebra and begin with the axioms defining a linear vector space.1 Introductory Remarks In Chapter 2 we investigated the algebraic structure of classical Hamiltonian mechanics. We also introduce in this chapter the concepts of linear functionals and distributions so that we can make free use of the delta distribution and similar objects in later chapters. which turn out to be those of the theory today known as quantum mechanics. We found that the Poisson algebra permits extensions to noncnumber formulations.Chapter 3 Mathematical Foundations of Q u a n t u m Mechanics 3. 3. Building upon this.e. In this chapter we therefore introduce important basic mathematical concepts of this noncnumber mechanics.
x M. Mathematical Foundations of Quantum Mechanics multiplication by complex numbers.2) Vectors ip\.. if numbers eti. so that n 5 > ^ = 0. i=l (3. 2 . f > 0 if V T ^ O .tp2.^1) = (01. . In each case n linearly independent vectors are said to form a basis. n.ipn are said to be linearly independent. ip2 of this space can be associated with a complex number (V'l.aiV>i + a 2 ^ 2 ) = ai(ip. . are linearly dependent. . the vectors il)i.ip2).^) called inner product. ( n (3.i = 1.ip2. (0 : null element and with complex numbers a and j3: a(tpi + tpj) = a{Wi) lipi = = < M). . 2 . (aptyi. . ^ i = 0 if ^ = 0.5a) (3. . .  . (a. • • • . E (3. (ipi.02)* (hermiticity). (3 5c) + a2{ip.42 CHAPTER 3. —»• IK. (ipi + ipj) + ijjk = ipi + (ipj + ipk). if every vector ip E M can be associated with numbers Q . If n + 1 elements ^ € M.3) If all on = 0. fa.ipn are said to be linearly dependent. n is called the dimension of . if any two elements ip\./?eK).M is said to be a metric vector space or a preHilbert space.M. and n is the smallest such number. such that n ^ = YJCi^i.1) aipi + aipj.4) The vector space . . . with the properties (a* G IK): (^2. i. i = 1. (V>.~ip2) '• M.5b) .e. . (3. A + tpj = tpj + i'i. (3.^i) . n exist (not all zero).
(3. ^ 2 )  . meaning oneandahalffold linearity). (310) (^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. for ip\.i>2\\2 = 2'i/'i2 + 2'i/'22 (parallelogram equation).tp2 £ M. (3..2. tp2 £ M: 1(^1.9b) = 0 (Pythagoras theorem).*h)=\\AH\(38) (3. for arbitrary A and ip2 ^ 0: (V>i + AV>2.3.9e) 11^11=1 We restrict ourselves here to some remarks on the verification of these wellknown relations.2 Hilbert Spaces 43 where the asterix * means complex conjugation. if Wi.9d)   V i   = sup  ( ^ i .9c) HV'i + tp2\\2 + \\1p1 . antilinearity in the first component (also described as sesquilinearity. ^ ) + A 2 (V2.^ 2 ) =   ^   2 + 2 ^ i .7) In addition the following relations hold in a metric space M. WA+Ml2 = ll^il 2 + ll^2 2 . The first two properties imply ( a i ^ i + a2ip2.^) = ai*(V'i.9a) we start from if) = ipi + \if>2 6 M. In order to verify Eq.V') +"2*(V.e. ^ i ) + A * ( ^ .^1) 2 h 11 0 <%«e + M .^2)1 < H^ill • 11^211 (Schwarz inequality).is defined by d(ipi. »2 2 . The distance between two vectors if>i. \\1p1 + ^211 < HV'ill + IIV^II (triangle inequality). ^ i ) + A ( ^ i . (3. (3. The norm of the vector ip (preHilbert space norm) is defined as H:=(^)1/2.rh) (3.9a) (3.V. linearity in the second component.V>i + A</>2) > 0 .)> (36) i. which we can write 0 < ( ^ i .
so that   ^ l + ^ 2   <   ^ l   + ^2. o ^ V i = o.10): ll^+^H2 = < <   ^  2 + V22 + 2K(Vl. be the set of all column vectors v\ V = (Vi) = I 2 V with complex numbers Vi.* Two vectors i f i . V 2 ) = 0 .^1)1 Vi 2 + ^2 2 + 2   ^   . (3. . ll^ill + llVdl.11). i=l * Not all the wave functions we consider in the following and in later chapters are automatically normalized to 1.9e). (3.V2) IIV'l2 + V'22 + 21(^2.9d).13) the vector is said to be normalized. (3. for which CO 2 := V^l^l 2 < 00. Examples of metric vector spaces: (1) Let M.12) We omit here the verification of the remaining relations (3. a £ K : ll^ill HVill 11^1 +^211 M If for a vector tp e M: > = < = 0. In verifying the triangle inequality we use this result (3. We also omit the verification of the following properties of the norm of a vector tpi € M with ^ € X . hence verification in each case is necessary.44 or CHAPTER 3.9c). ^ e M are said to be orthogonal if (</>!. HIHI (3. beginning with A = 1 in the second line of Eq. Mathematical Foundations of Quantum Mechanics 1(^2.  ^ l l = (HlMI +   « 2 . (3. (3. (3.9a).\\ih\\.^)1 <IIV>il.ii) thus verifying Eq.
g)= [ /(x)^(x)d3x. Eq.e. The Schwarz inequality is then oo oo \M\<* ^2\Vi\ ^2\Wj\2. In order to avoid this difficulty. for which the scalar product. (3.e. which satisfies relations (3. and one defines addition and multiplication by complex numbers with respect to these classes. (2) Let M.5a).5b). are combined to an equivalence class [/] (with space L2). •'^ \ 0 otherwise. = 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. With the scalar product (f.2 Hilbert Spaces Then we define v + w := (v{) + (wi) := (vi + Wi). But this applies also in the case of any function which is nonzero only on a set of measure zero. / ) = 0. 2 and so on. i. (3. with inner product oo 45 (v.[$]):= / JM /(x)* 5 (x)d 3 *. which differ solely on a set of measure zero. (3.3. all squareintegrable functions / which are "almost everywhere equal".5c). etc. Elements of the classes are then called representatives of these classes. Then L2 is the space of all these equivalence classes. Js the space C? is not yet a metric vector space although for (cf.5c)) / ( x ) = 0 = > ( / .w) :=J2v*Wi. i. . */ x } = / /o for x = 0. is defined by ([/].
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.
e.4b) (4. if (u\v) — 0. (2) (u\u) > 0.9a): \{u\v)\2 < (u\u)(v\v). if x\u) = 0 V kets \u) E V. 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. (3) (u\u) = 0. It follows that we can now write the Schwarzian inequality (3.5c)).62 CHAPTER 4. or (4. Hermitian Operators A linear operator A in the space of ketvectors V acts such that A\u) = \v) EV . Dirac's Ket. \v) E V are said to be orthogonal.5) 4. The dimension of V is the same as the dimension of V. i. Two ketvectors \u). (4. the norm squared.3 Linear Operators. which we write (u\.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). As in our earlier considerations we demand for (v\u) the following properties: (1) (u\v) = (v\u)*. Also: (xi = (X2I) if (xi\u) = (X2\u) for all \u) E V. that the vector \u) E V is associated with a vector in V.\u) = (u\0*.4a) \v) = J\(m<% and (4.(t.and BraFormalism Furthermore we have: ( x  = 0. This expression is linear with respect to \u) and antilinear with respect to \v). V \u) (in the case of £) with infinite norm!). Because of the antilinearity the conjugate bravector associated with the ketvector u) = Ail> + A22> is (v\ = Xl(l\ + X*2(2\. if \u) = 0 (see (3.
e. defines the unit or identity operator. AB^BA. Hermitian Operators 63 with A = 0 if A\u) — 0 for every vector \u) £ V.. i. Furthermore linearity applies. The inverse of A is written A . the space Vi <S> V2 has the dimension . with n)(1) = Aiw)( 1 )+A 2 k) ( 1 ) .. BA = 1.e. Then one says: (771 results from the action of A on (x. if \v) — A\u) i. The operator A. Let \u)(l> be vectors of the space Vi and similarly ?j)(2) the vectors of the space V2.3 Linear Operators.\u) = ((X\A)\u) = (X\(A\u)) = (x\A\u). Then the following rules apply: A + B = B + A.. In this way the operations of operators A. Two operators A and B are said to be inverse to each other. on bravectors are similar to those on ketvectors. Furthermore. i. as we can see as follows. Then (x(^4it)) is a linear functional of \u) £ V (since A is linear).4. This product is commutative. denned as a linear operator on V. if AB = 1. also acts on the dual space V. we have \u^u^) = X1\v^u^) + \2\w^u^) and so on. The inverse of a product is (AB)'1 =B1A~1. ltt) = \u). and one writes: (V\ = (X\A.e. This does not always exist. B. 1 and \u) = B\v). let (771 6 V be the bravector defined by this functional. The product of such vectors is written: \u^u^) = \u)^\u)^. Also A = B it for every vector \u) £ V : (u\A\u) = (u\B\u). The product vectors l?^ 1 )^ 2 )) span a new vector space. Finally we define the tensor. It follows that (r. Let the following bravector be given: (x\ £ V.or Kronecker. If Vi has the dimension A/i. Multiplication by unity. the space Vi <S> V2.or direct product of vector spaces.
Dirac's Ket. since AWAW\UW)\UW) = \vWuW) = AWAW\UW)\UW).and BraFormalism an operator in the Mi A/2. in the product space. If (v\ = (u\A. AW\uW)\uW) = \vWUW). i. then \v) is a linear function of \u). i.7d) . here designated with the same symbol.e.64 CHAPTER 4. we obtain the conjugate relation <tAt«> = (u\A\t)* (4. by replacing in the above the bravector (t by <tflt) (AB)* = B*A\ (cA)* = c*A\ (A + 5) t = A* + B\ (4.7a) (4. Every such operator is then also associated with an operator. Assuming now that \v) G V and (u\A G V are conjugate vectors as explained above.e. the operator A^ is called hermitian conjugate or adjoint operator of A (cf. Next we construct the following scalar products: (t\v) = (t\Ai\u). Since we demanded the scalar products to have the property (t\v) = (v\ty. we write \v) = A*\u).7C) («)(u)t = \v){u\. Every operator A^> commutes with every operator A(2\ i. 3.g. \t) G V.7b) (4. [A^. (v\t) = (u\A\t). i. Examples of operator relations: (1) (AB\u)(v\Cy = C^\v)(u\B^Al (4. Let A^> be an operator in the space Vi and A^ space V2.6) for all \u).3). also Sec. AW\U)W = \V)W.e.e.A^] = 0. and (v\ = (u\A. As a consequence we arrive at the following properties (e.
but only if these commutators commute. so that (n^4n) is real. it follows that (u\A\u) = a(u\u). or (u\A\u)* — {u\A^\u) = (u\A\u). hermitian The product of two hermitian operators is not necessarily hermitian. An important theorem is the following. as is also (u\u). It follows that a is real. if H is hermitian and (u\H\u) > 0 for all ketvectors \u). Verification: Since A — A^ and A\u) = a\u). i.3 Linear Operators. .K\.e. The commutator of two hermitian operators is antihermitian: [H. H^K] The separation of HK into hermitian and antihermitian parts is therefore HK=±{HK + KH)+l[H. An hermitian operator A has the properties (a) that all its eigenvalues are real. K] = 0. [H. Analogous considerations apply in the case of bravectors (u. Every operator A can be written as the sum of two such parts: A=\(A + A*) + \(Atf) anti—hermitian . Hermitian Operators (2) The conjugate bravector of AB\u)(v\C\w) is {w\C^\v){u\B]A]. if H = H' and K = K\ then (HK)* = HK only if tfrf = KH.e.4. so that (a)(a)t = a)(a.K]^ = = [HKKH]* KHHK = = K^H^ [H.K}. Obviously the quantity \a)(a\ is an hermitian operator. 65 The linear operator H is called hermitian. The operator H is said to be positive definite. i. for we had (\u){v\)i = \v){u\. and (b) that the eigenvalues with respect to \u) are equal to those with respect to (u\ and vice versa. if H — H' and antihermitian.
e.66 CHAPTER 4. Dirac's Ket.e. A continuous spectrum can immediately be included by passing to the extended Hilbert space. Then every vector \u) £ "K can be written \u) — its) + \us*) with \us) e S. (u\u') = 5{v . The above theorem remains unchanged.e.and BraFormalism Moreover.u').) = 0. since a ^ b. i. so that (u\Ps = (us\. (u\Ps = {us\. Assuming A\u)=a\u). a ^ b. i. {us\us*) = 0. The entire set of ketvectors spans the extended Hilbert space. it follows from A\u) — a\u) that (u\A = a(u\ or the other way round.9b) (4. Ps2 = Ps(4. and let S* be its complement. \u8*) E S*. Ps\ua. i. so that (v\u) = 0. we have 0 = (uArt) — (u.Au) = a(v\u) — b(v\u) — (a — b)(v\u). since a is real.9a) . This follows from observing that (u\Ps\v) = (u\vs) = (us\v3) = (us\v). as we can see as follows. (4.8) The projection operator has the following important properties: Ps is hermitian. i. (n\n')=5nnl. Next we introduce the very useful concept of projection operators. If these form a complete system. The operator Ps which projects onto S is defined by the properties: Ps\u) = \us) = Ps\us). which includes vectors of infinite norm. Orthogonality: Two eigenvectors of some hermitian operator A belonging to different eigenvalues are orthogonal. (v\A = b(v\. then the hermitian operator defined on this space is an observable. but the continuum vectors are normalized to a delta function.e. Let S be a subspace of the Hilbert space !K. ( n  z / ) = 0 . Ps is idempotent.
. iV) is a set of orthonormalized states.1. . \u) arbitrary. Ps2 = PsThe only eigenvalues of Ps are : 0 and 1. i. (1 — Ps) is projector onto S*.9d) Thus the vectors P\u). i. p 2 . Set \u) = \ua) + \ua*) with (a\ua*) = 0. i. 67 (4. The projection \ua) of an arbitrary vector \u) onto this subspace is ita) = a)(aw). so that (Ps2 ~ Ps)\u) = 0. Then 0 = (P2P)\p) and hence p = 0.3 Linear Operators. P\p) =p\p). so that \ua) = (a\u)\a) = \a)(a\u). If for these (by construction) the projector P2 onto a subspace S2 is P2= J*I f2\odm. (1 — P)\u) are orthogonal. Let p be an eigenvalue of the projection operator P.e. Example: The vector \a) normalized to 1 with (a\a) — 1 spans a 1dimensional subspace. if the set of vectors 1).p = 0. Let a continuum state be written £). (4. = (p2p)\p). Obviously N PN = y]ln)(re n=l is the projection operator onto the A^dimensional subspace SN. as can be seen as follows. The quantity \a)(a\ is called elementary projector. Hermitian Operators This property follows from the observation that Ps2\u) = Ps\us) = \us) = Ps\u).e. Then (a\u) = (au a ) + (a\ua*) = (a\ua) and so {a\u) — {a\ua) and hence {a\u} = c(a\a) = c.e. where £ is a continuous index. Very similarly we can construct operators which project onto the subspace spanned by the continuum states.9c) This property can be seen as follows. 2). c = (a\u).e.4. \ua) = c\a). Ps is projector onto S. .. i..
114b).6. The eigenvalues Aj then form a discrete sequence with associated eigenvectors \ui) € !K.68 CHAPTER 4.114c).e. Coulomb potential. are linearly independent). See e. the operator P is correspondingly P2\u)= f2\0m\u). Degeneracy will be discussed at various points in this text.4 Observables Operators which play a particular role in quantum mechanics are those called observables.6.r)(ui. Let us assume that A is an hermitian operator with a completely discrete spectrum (as for instance in the case of the harmonic oscillator). 4. In the case of the infinitedimensional Hilbert space with a countable number of orthonormalized basis vectors.g. which are orthogonal to each other (i.. 8. (11. in this case the spectrum also has a continuous part. that one and the same eigenvalue Aj is associated with several eigenfunctions. Dime's Ket. ' The projector onto the subspace with eigenvalue Aj can then be written Pi = ^2\ui. which we introduced earlier. i.. u2).1 and 11. be a system of basis vectors in this space. Let \u\). one says the degree of degeneracy is r .and BraFormalism and hence '6 Numerous properties of the projection operators Pi are selfevident. see Eq. and Eq. If there are r such vectors. 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.. Sec. However. Examples 8. (11. P = £n)(n.r\ "An example is provided by the case of the hydrogen atom. In general it is possible.e. Observables are representatives of measurable quantities. .
the projection of this operator is the entire space. the relation (4.e.11) i.r \ui. is always implied).r\ = l. ^ A i P i = A = ^Aiui.r)(ui. i i (4. (4.12) . or also as subdivision of unity or of the unit operator. The uniqueness of the expression (4. PA = y£pi i = Yt\ui.10) This expression is known as completeness relation or closure relation. in the case of degeneracy with (ui.r)(ui. Applied to an arbitrary vector \u) G "K Eq. Together with the orthogonality condition. i.e.10).r \(ui. It follows that the norm squared is given by (u\u) = {u\PA\u) = ^2(u\ui.4.10) gives the linear combination it) = ^ i.r)(ui. we obtain i i i i i.r (4. and the eigenvectors \ui. 69 i If A is an observable and if the spectrum is purely discrete.4 Observables The dimension of this subspace is that of the degree of degeneracy. The operators Pi are linearly independent.r.r>(ui.10) follows from the fact that i Applying the operator A to the projector (4. If Aj ^ Aj'. then PiPi' = 0. the operator A is completely determined by specification of the eigenvalues A. which we do not enter into here.r\u).e. Let us set {A .e.r'} = Sii>6rr>. i.r\u) i.\i)Pi = 0.r) (the convergence. and APi = XiPi. (4.10) expresses the completeness of the orthonormal system.r\u)\2.r\ui/.r =^ i. i.
for instance.e. If the spectrum of an observable A consists of discrete as well as continuous parts. now however.v'). The ketvectors \uv) are orthonormalized to a delta function. (note: {uv\A\uvi) is the matrix representation of A) A\uv) = \(v)\uv). 1/2)) PA = 22 \ui)(u*\ + / dv\uv){uv\ = 1. for instance exponential functions.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). (3.17) previously. Then we denote by \uv) the eigenvector of A whose eigenvalue is \{v). Let v be the continuous parameter which characterizes the continuum. (4.14) . as explained earlier. (4. the eigenvector \ui): f(A)\ui) = f{\i)\ui). we have the corresponding generalizations. > rV2 + / dis\(uu\u)f \uj)(ui\Xi + / \(v)\uv){uv\dv. Then f(A) = f(A)PA =^/(AOk)^! + / f(\{v))\uv)(uv\dv. i.70 CHAPTER 4. In a similar way we can handle functions f(A) of an observable A. Dime's Ket. i.and BraFormalism This is the expression called Parseval equation which we encountered with Eq.e. (u„zv) = 8{v . written in Dirac's notation. One defines as action of f(A) on. 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.
form a complete set of commuting observables. (4. Finaly we recapitulate the following theorem from linear algebra: Two observables A and B commute. i. if (a) they all commute pairwise. 4.. [A. In the following we consider more generally ketvectors \ip) as representatives of the physical states. as a formal orthonormality condition. We shall return to this theorem. First of all we can rewrite the integral representation of the delta function. For many practical purposes the use of the Fourier transform is unavoidable. which is discrete in the case of discrete energies and continuous in the case of scattering states (with continuous energies). \k)€Fk.and bravectors. with energy E.17) The Fourier transform provides the transition from one representation and basis to the other.16) represent subdi .B.e. for which the completeness relation is f dk\k)(k\=l.e.C. and (b) if they possess a uniquely determined system of basis vectors. x) € Fx.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. (4. here presented without proof.. i. their commutator vanishes.58) in terms of ket.. Therefore we want to reexpress the Fourier transform (3.15) and (4. again later. (4. where the symbol ip is already indicative of the Schrodinger wave function ij.15) where the vectors {\x)} are to be a complete set of basis vectors of a linear vector space in its position space representation Fx.59).4. Extending this we can say: A sequence of observables A. (3. Since both expressions (4.5 Representation Spaces and Basis Vectors We began by considering ketvectors \u) in which u is a parameter of the (energy) spectrum..16) Correspondingly we also have a complete set of basis vectors {\k)} of an associated vector space F^. In the onedimensional case we write {x\x') = S(xx'). Eq. B]. In the differential operator representation the problem to establish such a relation is known as the SturmLiouville problem. / dx\x){x\ = 1. which all commute with one another. if and only if they possess at least one common system of basis vectors.
(4. Rather \x) and \k) serve as basis vectors in the representation spaces Fx. which are representatives of the states of our physical system. 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). this expression has to be identified with — / dkeikxeikx\ i. i.e. il>{x) := (x\i/}) : "K * C. (4.Fk.59) or Eq. (4.19) Comparison with the orthonormalized system of trigonometric functions (4.x') = (x\x') = (x\t\x') = (x\ f dk\k)(k\x').22) All of these expressions can readily be transcribed into cases with a higher dimensional position space.1a).space representation. The vectors \x) and \k) G F are not to be confused with the vectors u) or \ijj) G !K. V 27T (k\x') = .20) The representation of the corresponding bravector (^1 G IK in the position space Fx is correspondingly written (il>\x) = {x\ipy.e. Obviously we obtain this by inserting a complete system of basis vectors of Fk.18) According to Eq. (4. called wave function. i. . (x\k) = ^=eikx. we can rewrite Eq.15): 5(x .L e " * * ' = {x'\k)\ V27T (4.21) provides the ketvector \ij)) in the A. The Fourier representation ij){x) = )= V 27T J Ieikx^{k)dk (4. shows that these expressions are the corresponding continuum functions (the continuous parameter k replaces the discrete index n).e.and BraFormalism visions of the unit operator.72 CHAPTER 4. ${k) := (k\if>). (4.2a) etc. Dime's Ket. (3. <a#) = [ dk{x\k)(k\ip).
the calculation of which for specific cases is the subject of Chapter 7. 1.1 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. We mentioned earlier (in Sec. which is the quantum mechanical analogue of the classical probability density p(q. Schrodinger [244]. This will then also clarify the role of p as an operator in the quantum mechanical analogy with the Liouville equation.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.Chapter 5 Schrodinger Equation and Liouville Equation 5. with the complementary system or *E. 73 .e.p.t). if we do not take into account the complementary part of the universe. 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. We postulate the (timedependent) Schrodinger equation* (with the Hamiltonian as the timedevelopment operator) and obtain the quantum mechanical analogue of the Liouville equation.2 The Density Matrix We shall establish a matrix p. i. Finally we introduce briefly the canonical distribution of statistical mechanics and show that with this the density matrix can be calculated like the Green's function of a Schrodinger equation (inverse temperature replacing time). 5.
b. but also of the states of the complementary set of the universe. p" ( C C * ) t = CTC* = p. (52) (5. \a)..j ij (54) Plj = (i\p\j)^YCiCJ*a (55) Here p is an hermitian matrix.. where i represents collectively all quantum numbers of the state of the system under consideration. These are the quantum mechanical states which are also referred to as microstates. we have (a\b) = 5ab.. we have to deal with a socalled "mixed state". (a\i)=0 = {j\b). i. 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).e. i eH ®^' (5J) We assume that the states are orthonormal and in their respective subspaces also complete. which in Dirac's notation we can write as = J]C a »i) a. are called "pure states".3b) The quantum mechanical expectation value of an observable A.b... \a).i.i. called the density matrix. The actual and real state of a system is then a superposition of two types of states.. that is. The states \i) £ "K.^ Since p is hermitian. i.. T . a (5. of an operator that acts only on the states of the system we are interested in. the matrix can be diagonalized by a transition to a new set of states ^The hermiticity can be demonstrated as follows. \i). as well as 2~] «)(* = 1 in the subspaceof pure states i (i\j) = 5ij. is then given by (A) := {MA\^) = Yl U\{b\A\a)\i)CaiCbj* a.3a) and N J  a ) ( o  = l in the complementary subspace. The states \a) £"K'on the other hand are the corresponding states of the complementary part of the universe.74 CHAPTER 5.e.j = where J2 (J\A\i)5baCatCbj* = X>'l4*>/0« a. where T means transpose: p = CTC*. The difficulty we have to face is the impossibility to specify precisely the stationary state of our limited system. Schrodinger Equation and Liouville Equation the measuring instruments.
as we discussed previously.) = YJc t . (5. i. in a specific representation. Thus the real system is in the pure state \i) with probability uii/ J2i Ui.7) p = ^2\ii)(i. The operator Pi = \i)(i\ projects a state \<p) G "K onto \i).e. The ensemble can be represented as a set of points in phase space.e. In this sense we have p(x'.ji(x / )i*(x). which today is a fundamental term in statistical mechanics. Gibbs introduced the concept of ensemble. then (cui/ '^Zi^i)Z is the number of ensemble elements in the pure state i).j(x'i)(ia.10) and is therefore.3 with Pi^YtCaiCaj*. . (5.. (see below) eM.x) := (x'\p\x) = yju.5..\p\i")(i"\ = Y.3a). Thus the pure states \i) form in the subspace of the space of states which is of interest to us.e. (5. a complete orthonormal system with properties (5. quantum states. in the position state representation or xrepresentation. as we saw earlier.2) and (5. If the ensemble corresponding to the system under consideration consists of Z elements (i.9) is diagonal. can be achieved with the help of the completeness relation of the vectors \i): \i') = J2\i)(i\i') i. This transition to a new set of basis vectors in which the matrix becomes diagonal.g. which are all subjected to the same macroscopic boundary and subsidiary conditions. P ^ ) = i)(#) = (#)i) (5.4) we had the expression (A) = Y/Pij(j\A\i) i. e. seen macroscopically of Z copies of the system).* We can specify the state of our system. i i (511) In Eq. This projection operator represents a quantum mechanical probability as compared with the numbers ui{ which represent classical probabilities. but occupy in general different microscopic.. where x represents collectively the entire set of position coordinates of the (particle) system. a projection operator.6) (i'\p\i") = Wi'Si'vi = real.2 The Density Matrix 75 \i') 6 "K. a (512) *In his reformulation of the statistical mechanics developed by Boltzmann.e. This ensemble is interpreted as a large set of identical systems. or (5. i.^'\i'^i'\> (58) since then (f\p\j") = ^Ui'ij'li^ii'lj") v = ^Ui'Si'fSi'j" v = UjiSj/j. In the following we write simply \i) instead of \i').
) := (mm = (#') W> = l(#')2 > 0.4).17) Hence the sum of the classical probabilities is 1.16) (5. as claimed by Eq. Then the expectation value of the observable A is (cf. Schrodinger Equation and Liouville Equation We now define the operator p by the relation (cf. (5. (5.13) {A) = ^(j\A\i)(i\p\j) ij = 52{j\Ap\j).76 CHAPTER 5. (5. In particular for A = 1.5) above) (i\p\j)=Pij. We can also show that Ui > 0. To this end we set A —>• Ai' := \i')(i'\ (no summation over i'). Thus Tr(j>ii)<i)=l. Hence Wj/ > 0. (5.14) which we can write (A) = Tr(pA).14) we obtain (Ai') = ^2(j\i')(i'\i)(i\p\j) ij = ^Sji'Su'Pij i. Eq. (5. 3 (5. i t (5.18).4)) (5. (A. (5. 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. On the other hand. Eq. (5.15) 5^0'l5^w«l*X*b') j i = 1 » or J^wi(it> = ^ W i = l. according to Eq. we have Tr(p) = l.3 = pvv = uv.
We now consider the latter expression.t) = {x\i).e. so that p{x. thus enters these states. t) p= i The expression (5.§ {\x)} is a complete system of basis vectors in the position representation space Fx with dx\x){x\=l. represents a quantum mechanical probability.e. but we see here. and in the other states with probability zero.24) s (x\x') = 5{x . one says. .3 The Probability Density p(x. Equation (5. (5. the only remaining one is 1.3 The Probability Density 77 We see therefore that p defines the mixed states. The effect of the interaction of the system under consideration with the surroundings. defines the statistical operator. i.23) Instead of i(x.8). i. (5.5. in all other cases the state of the system is a mixed state. We observed above.t) we now write tp(x.t) = \^{x. i.t).e.t) and instead of (x\Pi\x) we now write p(x. not in a mixture of states.22) In the following (x\i) is always to be understood as (x\i(t)}. since they do not take into account the (interaction with the) rest of the universe. If all ui but one are nought.21) p=\i){i\=Pi.t)\2. by representing the vector \i) by the wave function i(x. in which the numbers Wj represent classical probabilities and the operator \i)(i\ quantum mechanical probabilities or weights. For the case n we have TJ=J (520) (5.20) expresses that the system can be found with probability 1 in the state \i). Then {x\Pi\x) = (x\i)(i\x) = \{x\i)\2. the system is in a pure state. that these are actually not sufficiently general. We see this more clearly by going to the position space representation. that — different from u>i — P. (5. which is contained in the coefficients Cai. (Question: Is the state of the universe a pure state?) 5. Thus we have a system in a pure state. In the following we are mostly concerned with considerations of systems in pure states.x'). ^2ui\i){i\.
(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. This differentiation requires the time derivative of a state vector \i) E "K. Hence we postulate: The time development of a state \j) E "K is given by the equation ih^\j)=H\j). which is chosen in such a way that the desired analogy is obtained. i. 2. Equation (5. with respect to time. can be found with probability 1 in the space R1. (5. We shall convince ourselves later that the postulated equation is sensible (in fact. but the linear operator states are moving — even if the system is observably stationary). (5. t h e equation of motion with states represented in a system for which the q's or here x's are diagonal).8). t).e.3. (5. The generalization to a threedimensional position space is selfevident: JK 1 / p(x.27) where H is the Hamilton operator of the system.78 CHAPTER 5.4 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. Schrodinger Equation and Liouville Equation The relation (5.27) is known as the Schrodinger equation. To this end we have to differentiate the density matrix p. the expression (5. it is not "derived").25) J Thus the particle whose state is described by \i). the Schrodinger equation is always postulated in some way.e. (5. and hence the integral over all space dxp(x.28) Differentiating Eq. and the timeindependent equation should correspondingly be called Schrodinger wave equation (i. We therefore postulate first an equation for the time dependence of a state vector.8) and multiplying by ih.t) = \rp(x.22) can also be written (xi) 2 = (i\x)(x\i). or the wave function tp(x. . We now want to obtain the quantum mechanical analogue. (5.t) = / dx{i\x)(x\i) = (i\i) = 1.t)\2.26) 5.^ The equation which is the conjugate of Eq.27) is iHJ\ jt= m .
e. (5. (5.34) . (5. the relation ih^ = [H.4 Schrodinger Equation and Liouville Equation 79 Inserting here Eqs. we observe that here on the left hand side we have the total time derivative.e. (2.28) we obtain (see also Eq.32) with its classical counterpart (2. Eq.p}.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). M\/n\ (530) where in statistical equilibrium dt £^">01=03 (5.Xi)\j). (5. The correspondence to the classical case is obtained with the substitution h J< > <5'33> and dp/dt = 0 (cf. p also contains "quantum mechanical probabilities" (i. With Eqs. The reason for this is that in addition to the "classical probabilities".e. both uji and i)(i). (3. The Schrodinger picture and the alternative Heisenberg picture will later be considered separately. We can now consider the Schrodinger equation (5.21).37)) (5. (5.27) and (5. dp/dt ^ 0.28) the states are considered as timedependent. i. (cf.27). such a formulation is described as Schrodinger picture.5.24)). i. Eq. not the partial derivative. We thus obtain the operator form of the Liouville equation. (5.p} + ihJ2^\J)(J\.31) below) 3 j 3 = y £u>J{HPjpjH)+thyE^m\ dt 3 = [H.32) Comparing Eq.27) in the position space representation: ih(x\\j) = (x\H\j) = {x\H(pi.
**R.e.. we write V>(x. a (5. For a mixture of states \i) of the system under consideration caused by the rest of the universe we have Ui ^ 1. which we shall need later. With the help of Eq.Xi){x\En) = En(x\En). (5.35) can be written'! ip(x. 0) can be expanded in terms of a complete set of eigenvectors \En) of the Hamilton operator H. with respect to time t. . to) = exp[—iH(t — to)/H\. (5. but with respect to the parameter /3 = 1/kT. since the timedependence cancels out (the exponential functions involve the same operator H but with signs reversed. 0) or \il))t = \il>)t=o exp (5. In expression (5.8) for p. P.4.t) = Hld . the state \i) is still timedependent.0): where H(ih—. Schrodinger Equation and Liouville Equation or with (x\j) — ip(x.t) = exp ip(x.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. in Sec. i. we shall describe as "time development ator given by U(t. namely that the density matrix satisfies an equation analogous to the timedependent Schrodinger equation — not. t ) .4.36) in in The initial wave function or timeindependent wave function ^(x.** In view of the close connection between time..and temperaturedependent Green's functions. however. it is plausible to refer to this equation at this stage. n (537) This equation is described as the timeindependent Schrodinger equation. Feynman [94]. (5.80 CHAPTER 5.. T meaning temperature.x 1 ^ ( x . this is an application of the BakerCampbellHausdorff formula which we deal with in "Later.t): > ih—</>(x. „ / ith—. in p = ^2i^i\i}(i\. appearing in the Boltzmann distribution. or H\En) = En\En). 10.35) Here we can look at the Hamilton operator as a "time generator3''. operator" the exponentiated oper .e.36) however. we can replace these states by corresponding timeindependent states. since the solution of Eq.38) (x\j)t=o = Yt(x\E^(En\J)t=o. 5. i.
Hence we have (with (Ei\Ej) = 5^) in what may be called the energy representation p = YJ"i\Ei)(Ei\. Eq.40). (5. or ^ = ^ e 0Ei _m.^ n tne position space representation Eq.1). (543) Since H(f>i(x) = Ei<f>i(x).5.44) r (TrePH)' (5.45) since on the one hand •PEi i i i and on the other hand .9) with En > nhu) L0icxePE\ so that Yliui becomes = f3 = l/kT.4 Schrodinger Equation and Liouville Equation 81 Example 5. (5. the expression _Zie^\Ei)(E>\ also as (see below) e0H (5.40) 1. we obtain p(x.x'(3) = Si e ~^W*V).39) therefore (x\p\x') = J2"i(x\^)(Ei\x'). (5. i.e. x') = ^ i LUiMxWix'). i ( 5 . (1.42) Inserting here Eq.x')^p(x.41 ) or with 4>i(x) := (x\Ei) this is p(x. we can rewrite p.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. i (5.
/?).x. This solution is the Green's function or kernel K(x.50) Differentiating this equation with respect to j3.(3).tt tt For a more extensive discussion see R.x'. Tr(pN) = / dxpN(x. (5. we obtain 9pN gpP) = SijEie^ = EiPNij(J3).46) In the energy representation this expression is pNij(P) with PNij(P) = 6ij.52) Hence this expectation value can now be evaluated with a knowledge of pN.47) with respect to (3. Schrodinger Equation and Liouville Equation so that p is given by Eq.x'.45) and (5. we obtain = HxPN{x. (7. PN(0) := (EilpNiP^Ej) = (Ei\ePH\Ej) =e"^^ (5. We now rewrite the factor exp(—(3H) in Eq. Example 5. (5. (5. [B. Wilcox [284]. the following relation holds: exp(A) exp(S) = exp ( A + B + i [A.51) where the subscript x indicates that H acts on x of PN(X.48) = 1. whose calculation for specific cases is a topic of Chapter 7 (see Eq.45) without normalization as PN(J3) := e~pH. (5. (5.47) (5. Differentiating Eq. A}} + • Solution: The relation can be verified by expansion of the left hand side.4> = ^ g ^ . (5.49) In the position or configuration space representation Eq.45). B] + L [A. With Eqs. .1: Baker—Campbell—Hausdorff formula Verify that if A and B are operators. [A.x'.X'. (5.82 CHAPTER 5. (5. (5.51) is seen to be very similar to the Schrodinger equation (5.46) is pN(x. M./3).L [B. B}} .35). (3). Equation (5.51). (5.(3) := (x\pN(P)\x') = (x\e~0H\x').46) we can write the expectation value of an observable <^Tr<„.The quantity pjv is obtained from the solution of the Schrodingerlike equation (5.55)).
" Quite apart from this basic significance of the harmonic oscillator. S. Kim and M. its associated space of state vectors also provides the best illustration of a properly separable Hilbert space.1 Introductory Remarks In the following we consider in detail the quantization of the linear harmonic oscillator. can — in fact — be quantized in quite a few different ways. we shall see on the way.Chapter 6 Q u a n t u m Mechanics of t h e Harmonic Oscillator 6. like the Coulomb potential. that the oscillator. 83 . However. 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). * Comment of Y. This is a fundamental topic since the harmonic oscillator.51) to (8. Noz [149]. 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..53)). Since the spectrum of the harmonic oscillator is entirely discrete. 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. (8. E..the present form of quantum mechanics is largely a physics of harmonic oscillators. 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. The importance of the harmonic oscillator can also be seen from comments in the literature such as:* ".
hence we do not use a distinguishing notation. This can be done in a number of ways. linear harmonic oscillator the Hamilton operator is given by the expression XJ 1 2 . which is — in the first place — representationindependent.MvT« 7=r)^(VTvir>' + («> 'Note that we assume it is obvious from the context whether q a n d p are operators or cnumbers. To this end we introduce first of all the dimensionless quantities rriQUi fl q. so that there is no ambiguity due to commutation.p]=ih (6. of which the only nontrivial one here is [q. This equation would then have to be solved as a second order differential equation. in the present case + [EmooJ2x2)ip = 0. p^ih—.p.to = 1 reduces to the parabolic cylinder equation. h = l. which is called Weyl ordering. we can also proceed in a different way. 1 2 2 The quantization of this oscillator is achieved with the three postulated canonical quantization conditions. the boundary conditions on the solutions ip(x) would be square integrability over the entire domain of x from —oo to oo. Also observe that H does not contain terms like pq or qp. 2mo dx \ 2 2 (6. Quantum Mechanics of the Harmonic Oscillator 6.1) for the hermitian operators q. However. For instance we can go to the special configuration or position space representation given by q^x.2) which for mo = 1/2.2 The OneDimensional Linear Oscillator In the case of the onedimensional. ox and then to the timeindependent Schrodinger equation Hip(x) = Eip(x). When such a term arises one has to resort to some definition.84 CHAPTER 6.^ The first problem is to determine the eigenvalues and eigenfunctions of H.' ^mgLOh by setting and *~. . like taking half and half.
11) Thus the eigenvalues are real and nonnegative.B}C + B[A. A^] = A\ (6. .14b) [A* A.8) The eigenstates of H are therefore essentially those of N := ^ f A (6. (6. and [AB. i.6. A*} = A^A.A^] = 1.9) We observe first that if \a) is a normalized eigenvector of N with eigenvalue a. + 1). (6. C] + [A.2 The OneDimensional Linear Oscillator 85 (p. i\/2 Inserting these expressions for p and q into H we obtain H = hiu{A^A + AA*) = hi* (A^A + ^ . for ^ A  a ) = a\a).14a) (6. We now use the relation (3. if A*A\a) = a\a).12) in order to obtain the following expressions: [tfA.5) Reexpressing q and p in terms of A.14a) we deduce for an eigenvector \a) of A^A.13) ( ^ A ) i t = A\A^A From Eq.10) then a = (a\A^A\a) = \\A\a)\\2>0.6) and p = y/motoh AAi =^. (6. q are hermitian).e. A\ we obtain h A + A^ rriQLU y^ (6. i. (6.1 ) . With the help of Eq.1) we obtain immediately the commutation relation [A. Prom these relations we obtain ( ^ = ^ . (6. A] = [A\ A]A = A. (6.34d): [A.C]. (6. (6. C] = A[B.BC] = [A.7) (6.e. C}B.
 a) is eigenvector of A^A with eigenvalue (a + 1). we must have (a) Let \an):= Ana)^0.2A)\a) = A(Aa . unless A^\a) = 0.l)\a) = A(a . In this case we have (A*A)A2\a) (6 (6.. If we continue like this and consider the vectors An\a) ^ 0 for all n.14b) : (AiA)A*\a) = A\A^A + l)a) = A\a + 1) a) = (a + l)A^\a).17) (6.11) this is not possible. (6.20) .l)A\a).l)A\a) = A(A^AA . since this equation implies that the eigenvalues cannot be negative.A)\a) A{AAU .2)A a). unless A\a) = 0. but (b) An\n>\ llA An+1a)=0.e. (6.18) = '= a ) A(A^A . A 2 a) is eigenvector of A^A with eigenvalue (a — 2).86 the relation CHAPTER 6. Quantum Mechanics of the Harmonic Oscillator (A*A)A\a) = A{A*A . This would mean that for sufficiently large values of n the eigenvalue would be negative. Next we consider the vector A 2 a).21) . in view of Eq. (6. (6. However.l)\a) = (a .2A)\a) 2 (a . unless A 2 a) = 0. or Ata) = v ^ T T .19) i. Thus for a certain value of n > 0. The norm of A la) is Aa) 2 = {a\A^A\a) = a ( a  a ) = a. we find that A n a) is eigenvector of A^A with eigenvalue (ct — n). Similarly we obtain from Eq.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. (6. (6.16) This means. or \\A\a)\\ = Va.
or a = n > 0. i.26) <0A4 t AA t + A4 f 0) = (Q\2(A]A + 1)0) = 2.. But A 2 A t 0) = A(AA^)\0) = A(A^A + l)0) = 0. Moreover.24) This is a very important relation which we can use as definition of the state vector 0).18) we obtain for a = n: p t  n )   2 = n + l. we obtain an = \\A\an)\\2(6=) An+l\a) \\An\a) (6. that an = 0. HAU+IO)!!2 = = (O\AAAWI\O) = {O\A{A^A + I)A^\O) for all n. I „ .22) With relation (b) of Eq (6. In the following manipulations we always use the commutator (6. (6. \ I 2 2 x 87 Replacing in Eq. v. (6..20) that A0) = 0.5) to shift operators A to the right which then acting on 0) give zero. From relation (6. For a — n = 0 we deduce from (b) of (6.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) = 1.6.20) we obtain from this that the right hand side vanishes.n ) = " A»a>=» ' = A«a)F = 1 .28) .17) a by (a — n). The state 0) is called ground state or vacuum state. (6. \ 9 \ I / II 4 T ) .23) Hence the eigenvalues of the operator iV := A^A are nonnegative integers. (6.20) we have for a = n = 1: A2\1) = 0.27) According to (b) of (6. so that (a(A n )tA n \a) __ \\An\a)\\2 ( a .I n U . (6.25) (6. (6.' " /.2 The OneDimensional Linear Oscillator be a normalized eigenvector of A^A with eigenvalue (a — n).e.
The states \n) thus defined are orthonormal. But using Eqs. (0^ 2 (At) 2 0) = 2.29) Hence we obtain and in view of Eq. we have 2> = ^=dAi\0).(6.11). According to Eq.Al] = 2A^A^ + (A^)2 x 1 = 3(A^)2. A^A* + A*[A.e. as we can see as follows. (6.30) (arbitrary phase factors which are not excluded by the normalization have been put equal to 1). l>ocAt0).32) (6.33) .5) and again (6. (A*)2} = [A. (A^)3] = [A. i. and in view of Eq. (6.29).11) we have [A. (6. In general we have \n) = ^(A^n\0) (6. (At)2]A+ + (A*)2[A.(A*)n] = n(Ai)n\ (6. Similarly we find 2)oc^Ut0).88 and CHAPTER 6. 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. (6.^ ( O l A ^ n o ) . and in general [A.32) we have (nm) = .31) (6.26) the equality l) = A t 0). A^} = 2A\ and [A.34) (6. (6.
(6.1 ) . With Eq.34) and (6.J \n) = hw (n + i J \n).32): rf\n) = . . In view of the properties (6. we obtain (n\m) = 5nm We also deduce from Eq.34) A\n) = ±=A{Al)n\0) .35) the operators A^ and A are called respectively raising and lowering operators or also shift operators. l .e.38) (6.41) The contribution hu/2 is called zero point energy. n = 0 . i.40) (6.1) (6. whose number is given by the integer eigenvalue of the number operator N = A^A. however. (6. chosen in close analogy to the postulated "second quantization relations" of field theory. . In view of the same properties A is also called annihilation operator and A^ creation operator of socalled u quasiparticlesv. (6. ldnm = n\5nm.37) n^)711}^) = v n  n . The terminology is.6. the eigenvalues of the Hamiltonian H are En = hw\n+\+ i ) . •(» (6. (6. 2 .33). we do not have creation or annihilation of any real particles as in field theory (hence the word "quasiparticle"). . . in quantum mechanics. . (6. .1) = n\n).'n\ and = v n ![ ( A t ) M i= + (6.8) we obtain therefore H\n) = hu (A^A + ].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 .1 0).35) Inserting this result into Eq.1 (A + ) m .L ( A t ) n + 1  0 ) = VnTT\n Vn! and with Eq. 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 . .36) + 1) (6. Here.39) A^A\n) = y/nA^\n .
.... 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. is defined by projection of operators on eigenfunctions of energy.39): {n\A. (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 .6) and (6. A^ from the relations (6. (6.44) yfn. n = 0. and — in fact — we have used this already previously (see e. The representation of the Hamiltonian in this energy representation is {n\H\n') = En5nn. (6. En = hu(n + ). i. also called Heisenberg representation.43) = = y/n(n'\n — 1) = y/n5n^ni.e. (6.39)).g. Quantum Mechanics of the Harmonic Oscillator 6. Eq. Vn + 1.7) the energy representation of the operators q and p.1. ( 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. (5.3 The Energy Representation of the Oscillator The energy representation.38) and (6. other elements zero.45) Correspondingly we obtain with Eqs. other elements zero.2.46) v^ .90 CHAPTER 6.42) We can deduce the energy representation of the operators A..
Correspondingly the position space representation is given by the wave function <t>n(x) := (x\n). Eq.e.47) LP 0) = 0 .48) Applying from the left the bravector (x\ and remembering that (x\p\<f>) = we obtain ih—(x\(j)}. (6.4 The Position Space Representation 91 It is an instructive excercise to check by direct calculation that Eqs. (6. (6. Hence ( * Recall J™ dxe™ * ? 2 2 2 1/4 W \ i/4 \ em^x*/2h_ (g 5 Q ) = yfln/w .4 The Configuration Space Representation We saw that the eigenstates are given by Eq. (6.1) and (6. i.5) are also satisfied as matrix equations. 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. 2 . A\Q) = 0.32). mow / (6. The ground state wave function (/>o(x) is defined by Eq. by (cf.24).6.3)) ^ (q 2n \ + (6. ™ou(x 2h \ + J _ d ] m = 0 (649) mQUJ dx CemouJx2l2h. 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 .e. c/>n(x) := (x\n) = .32) to apply the appropriate number of creation operators A^ to the vacuum state 0).54) (6 55) i/4 rzri7r 1 / 4 vo. i. (6. Vn! Now {XlA (6. (6.L ( x  ( A t ) »  0 ) .l ) n f f „ ( 0 .l ) > n ( x ) . 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.51) \l 2h {X q \ m0uP)\l 2h \X m0u. it suffices according to Eq.0 = ( . ) e"^2.2 n n! The functions Hn(£) are obviously polynomials of the nth degree. These polynomials are real for real arguments and have the symmetry property # n ( . they are called Hermite polynomials.92 CHAPTER 6. so that <t>n(x) = ( .56) < ^ = We have as an operator identity the relation . (6.
57b) • Generalizing this result and inserting it into Eq.58b) by W. gain as an operator relation we have de = een± = een da ee/2±_^ee/2 d£ 1 u ^ d? dt. (6.6. Magnus and F.54) we find that (observe that there is no factor 2 in the arguments of the exponentials!)§ Hn{i) = ( .It follows that 0en(Pe/2^Lez 2 n\P?i2 — c2 d d£ ? = 2f. . d£" (6.en_ dt. Oberhettinger [181]. (6. — = n+ d ~^ i 2 + d 2 de) ' 1+e ~ (6. ( 2 0 4 .2)m = (±t)m for some function VKO. 93 .58a) ' T h i s definition — apart from the usual one — is also cited e.g. e.12(20 . (2036(20.e. 81. 2£.l ) n e ^ In particular we have #o(0 #i(0 #2(0 #3(0 tf4(0 = = = = = 1. p.4 The Position Space Representation i. (202"2.12.
(6. 81.59c) The recurrence relations satisfied by these Hermite polynomials (the second will be derived later in Example 6.^ + n) He n(0 = 0 (6.He'n{£). £.3) are (as given in the cited literature and with a prime meaning derivative) Hen+1(0 Hen(0 He'n(0 = ZHenifi) . £ 4 .60a) and I d£ Hen{Z)Hem{Z)e?l2 = n! V & „ m  (6. £2l. (6. (6. Quantum Mechanics of the Harmonic Oscillator The differential equation of the Hermite polynomials Hn(£) defined in this way is ( ^ 2 ^ + 2n )ff"(O=0.(e^/2) or Hn(0 = 2n/2Hen(V2ti). (6.60b) J—< "See for instance W.6 £ 2 + 3. (6.94 CHAPTER 6. 80.58c) 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. £33£. Oberhettinger [181].59b) = (l)^2/2^. pp. Mathematicians often define Hermite polynomials Hen(^) as^ Hen(0 Then He0(O Hex{0 He2(0 He3(H) fle4(£) = = = = = 1. = ffen+itO+ntfeniCO. Magnus and F. . = nHe^iO.59a) The differential equation obeyed by the Hermite polynomials Hen(£) defined in this way is (^2 .59d) The normalization of Hermite polynomials is given by the relations: oo / d£ Hn(OHm(Oe~e oo = 2 n n\^5nm.
s) as a Taylor series. Thus the generating function can be written F(Z.4 The Position Space Representation 95 The relation between Hermite polynomials and parabolic cylinder functions Dn(0 — which we use frequently here — is given by" Dn(0 = = Dn(V2£) = ( .s) = J2 n=0 (6.63) The meaning is that if we expand F(£.s) = eM?(Zs)2}(6.^ = e^/42~n/2Hn(aV2).g. F(S. p. 2(£  s)£2& s=0 s=0 ~ ds = m sf2]e?= •««) : S=0 2(2e2l) = (2e)22 = ^2(0.l ) ^ e^HeniO 2 / 4  . W.. (6.62) Joo The following function is generally described as generating function of Hermite polynomials Hn{£): F(S. 93. and so on.6.s)=F(£. Magnus and F..64) n\ See e. Oberhettinger [181].61) e?l*2nl2Hn{Z). Their normalization is therefore given by OO /"OO / OO J —OO z (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. .
Furthermore we observe t h a t the wave function of the nth state has n zeros in finite domains of the variable x. T h e comparative behaviour of t h e wave functions of the three lowest states of the harmonic oscillator is shown in Fig. in order to estimate the lowest energy of the harmonic oscillator. i. mow' Apx = ^m0Luh/2. ground) state is fko/2. 6.96 CHAPTER 6.1 The comparative behaviour of the lowest three wave functions of the harmonic oscillator. In classical mechanics the energy of the oscillator can be zero. We see t h a t the lowest state is symmetric. for x(t) = 0 and p(t) = mox(t) — 0. E0 ~ —{Apx)2 + IrriQ l mQuj2{Ax)2 moLoh/2 2TUQ 1 moU!2h/2 2 TTIQU = Ku).e. 2 . i.e. 6. W i t h Apx = moAvx we have = moLoAx Ax and thus h/2 Apx h/2 mQuAx Ax I h/2 . Quantum Mechanics of the Harmonic Oscillator Fig. In fact we can use the uncertainty relation AxApx ~ H/2. But this would mean t h a t b o t h x{t) and p{t) would assume simultaneously the "sharp" values zero. has parity + 1 .1.e. and we see t h a t even and odd states alternate. We also see from the eigenvalue t h a t the energy of the lowest (i.
back to 0. setting q = 1/z.4 The Position Space Representation 97 The zero point energy plays an important role in atomic oscillators.1 we demonstrate how the eigenvalues may be obtained by the method of "poles at infinity".: 27 T mov2a2. The analogous Example 6. original path q .1/a O l/a Rez all at infinity Fig.plane Imz z . to 0. and is directly observable. like twoatomic molecules. It is clear that the uncertainties are not due to deficiencies of measuring instruments (we are concerned here with a system in the pure state 0)).1: Eigenvalues obtained by contour integration Use the corrected BohrSommerfeldWilson quantization condition (cf. to —a.6. pdq. 14.e. in corrected version instead with n* = n + 1/2. According to "old quantum theory" with integer n. In the plane of complex q the integral has no poles for finite values of q.. 6.plane pole at z=0 ^ . and with potential V(q) = 2K2mov2q2 and total energy E = p2/2mo + V(q). Thus. This observability can be taken as direct proof of the uncertainty relation.2 contains only the answers since the evaluation of the integrals proceeds as in Example 6. which is the complete orbit. Finally we point out that the Hamiltonian of the harmonic oscillator can be diagonalized in many different ways.4) to obtain by contour integration the eigenvalues of the harmonic oscillator. classical orbit i. mass mo. /classical orbit Here q goes from 0 to a. E . Example 6. 27rmoi/ <f> V'a 2 — q2dq = n*h. we obtain the integral . But it has a pole at infinity.2 The original path and the transformed path at infinity. The kinetic energy is always positive or zero. In Example 6. n*h. Sec. Solution: Consider the simple harmonic oscillator of natural frequency of vibration u.1.
Quantum Mechanics of the Harmonic Oscillator This integral has a triple pole at z = 0 (i.1.3. 1 . a J \d Solution: The solution proceeds as in Example 6.) hv. 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).V62 a2). £ = n*/w = ( n + . / " ^ .( 6 . are the most basic and exactly solvable cases in quantum mechanics..c2 = (I + 1/2) 2 (natural units. After removal of a Gaussian or W K B exponential. (m . Example 6. but here we leave this . .2TTI a ^ = . i.114a)). 2 .98 CHAPTER 6.a) m = 2 .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. mo = 1/2) the integral I2 can be checked to give the eigenvalues for the Coulomb potential Ze2/q (2 turning points). 6.5. In the following we investigate in more detail the important differential equation of the harmonic oscillator.2: Contour integration along a classical orbit Setting q = 1/z. and obtain with this all properties of the solution. in q at infinity). 6. Eq.It follows that 1 = s2 i ^V»a*ai ~2f z=0 2wi r m '„2 = 7ria 2 La*Va 2 2 2 . since these two cases. verify that II = d> dq yi92 aq + ft 2TTI 9 /y/F^I 9z V a + bz z=o .65) f ^Va2z2 .e.e.2b = Ze2. z = 0 Vi 2Kmov(Ka2) = n*h = E/v.e.5 The Harmonic Oscillator Equation In the above we considered mainly the energy representation of the harmonic oscillator and encountered the Hermite polynomials. (11.1. hence we consider the following differential equation in which n = 0 . for instance — without deriving all properties of the solutions. We can evaluate the integral with the help of the Cauchy formula (the superscript meaning derivative) Jc Hence we obtain (z . i. . « qdq + I2 = f dq . . E = Z2e4/4(l + n+ l ) 2 (cf. harmonic oscillator and Coulomb interaction. we start with a Laplace transform ansatz for the remaining part of the solution.1)! (6. In Chapter 11 we perform similar calculations for the radial equation of the Coulomb potential — as in Example 11.5 . With a2 = E.
Eq. .6. We demand that C(p) := — \ps(p)e~pt] = 0. Substituting these expressions into Eq. Thus — ignoring details — / dpf(p)erpx is the Fourier transform of / ( p ) .66) First we have to remove the £2term. / dpf(p)e~px the Laplace transform of f(p). W. 80.67) the latter being the differential equation of Hermite polynomials Hen{t) for integral values of n. (this determines the limits). Hence we set (with the chosen sign in the exponential) m=y{t)et2l* with *0. since this can give rise to exponentially increasing behaviour (at most a factor t can be tolerated). fp{ps{p)) = {v2 + n)s{p).5 The Harmonic Oscillator Equation latter property open for determination later: 99 ~dP + 1 1 2 V> = 0 .68) § = J p2s(p)e*dp. Oberhettinger [181].67) we obtain \ps(p)e pt] + I P2s(p) + —(ps(p)) + ns{p) e~ptdp = 0. (11. Then (with partial integration) *dt = ~*/Ps(p)e.g.69) **See e. and / dpf(p)pn the Mellin transform of / ( p ) .g. (cf. (6.** We make the Laplace transform ansatz^ (with limits to be decided later) y(t) = [ s(p)e~ptdp. Magnus and F.111)) the Mellin transform of e~p is n!. This has to be true for all values of t.t*ij& + ny{t) = O. so that the integrand of the remaining integral vanishes.i ^  (ps(p)) P n P Integrating this expression. i. (6. .p t dp = \ps(p)ePt} . (6.J ±(ps(p))e*dp. E. p.e. we obtain P2/2 ln(ps(p)) = — p — nlnp or s(p) <x Tl+l P (6. (6.
Eq. 6.67)) M) * e"* /4y(t) ~ et +2/ 2 /4 But this is not permissible.100 and hence CHAPTER 6. since otherwise (with p = \p\ exp(i8)) we will get part of the solution from 0 to infinity. Hence a possible path of integration > is from — oo to oo. This can be attained if p is allowed to be large (t can be of either sign). when p = —t (which is possible. Thus the only solution left which does not involve large values of \p\ is that with § Q for n an integer. For singlevaluedness. To see this. i. n must then be an integer. then for large t we have y(t) ~ exp(i 2 /2).3 The path when n is not an integer. Therefore we take ptp2/2 y(t) .3. of course).70) = 0 P. so that y(t) ~ exp(i 2 /2). The value of the exponential at this point is ~ exp(£ 2 /2).71) . (3) We can choose a contour once around p = 0 in the complex pplane.. We now investigate possibilities to satisfy the boundary condition C(p) = 0. = / • Im p Rep n not an integer Fig. we can choose a path from zero to infinity (the condition will vanish at p = 0). so that p cannot go to infinity.n+l between the two limits of integration. as indicated in Fig.e. The quantum mechanical wave function is then (cf. We can show that if \p\ is allowed to be large. Quantum Mechanics of the Harmonic Oscillator ptp2/2 » ( « ) ptp2/2 dp with C{p) = (6. (1) This will be satisfied if \p\ — ±oo. y(t) oc 2vri J pn x dp for n integral and ept p 2 = coefficient of pn in ~^ (6. This exponential assumes a maximum value when pt \—p2 = minimal. (2) If n < 0. (6. 6. consider the exponential in the integrand.
. this becomes jfj. . Therefore the coefficient of (pt)n in exp(—pt — p2/2) is (—l) n /n\.3. The exponential function here is described as the generating function of the Hermite polynomials Hen(t). The Hermite polynomial is therefore given by H ^{t){^—f—^r~dp.if n is odd.6. 2 ..72) We now obtain the polynomial form with Cauchy's residue theorem as Hen(t) = ( . In y(t) above. if n is even and v = 1. .P 2 / 2 00 C—tY ( .. We can also obtain the derivative form of this Hermite polynomial. the term of highest degree in t in the exponential corresponds to taking all p's from exp(— pt).„y2 v l { ' ^ r (^) \ 2 ' where v = 0. n—v even ^ 2 ' Hence we obtain the polynomial of degree n Hen(t) = nl £ see below ( _ 2 ) ( n . (676) .l ) n n ! x coefficient of pn~v in V LJ_L e p 2 /2 *^ v\ u=0.l ) n n ! x coefficient of pn in = = (~l)nn\ e~ ~ x coefficient of pn in V l z ^ e . 4 . We have 27T! Setting q — p + t. pt p2/2 (6.0) p p»+i v ' Hen(t) = (l)^ e * 2 /2 / 2vri /(9=t) dg " ^(gi)"+i e ?2/2 = e i 2 / 2 ()ne"i2/2.5 The Harmonic Oscillator Equation 101 with Cauchy's residue theorem. The Hermite polynomial Hen (t) is now defined such that the coefficient of tn is unity. . .5.
( n .102 CHAPTER 6. Solution: We have to evaluate °° 2 . n — l)Heni(t). (6.3: Recurrence relation of Hermite polynomials Show that tHen(t) = (n. (n. Hen(t).it) _ (n + l)Hen+1{t) (l)""1^1)! (—l)"+ 1 (n + 1)! ' tHe„(t)=nHen1{t) + Hen+1(t). Pptp 2 / 00 /2 dp.70). as in the normalization of asymptotic expansions of Mathieu functions).5 is a further application of the method for the evaluation of integrals that occur frequently in practice (e. for the evaluation of expectation values. we obtain tHe„{t) (l)"n! ~ i. Example 6. in radiation problems. Starting from Heo(t) = 1. Hei(t) Henj. and elsewhere. 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. We leave the consideration of the recurrence relation of Hermite polynomials (of paramount importance in the perturbation method of Sec.n + l ) f f e n ^ i ( £ ) + (n. (6.75) as Hermite polynomials. Quantum Mechanics of the Harmonic Oscillator which we recognize as agreeing with Eq.7 that we use throughout this text) and their orthogonality and normalization to Examples 6.4.3 and 6.e. 8.)} 2 /2dt = ePde(P 2 W)/2^^} J —CO .g. 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).77) = t.1 '2dt using Hen(t) = (l)n27 r . one can construct the tower of polynomials Example 6.4: Orthonormality of Hermite polynomials Establish the orthogonality and normalization of Hermite polynomials.n— 1) = n. Reinterpreting the remaining integrals with Eq. n + l) = l. (6. Hen(t)Hem(t)et ™ It follows that y ' (27Ti)2 fp=0Jq=0 P "V+l^+l J ^ ^ The integral with respect to t is a Gauss integral^ and yields ePde(p 2 +12)/2 f°° J —OO e{t+(p+<. Example 6.59a).
jQ) n n!I ^—' ^Q fi\(n . zero otherwise (6.n)\ i s an+2rij. Hence the normalized wave function of Eq.5 The Harmonic and hence Oscillator Equation 103 / = (!)" But (2iri)2 { J Jp=0 J q0„ pn+ i0m+l P I e d d P 1 1 / epq p" <k rrdq = coefficient of qm in epq = — 27ri / qm+1 m Therefore (IT ^n+m ^ n l I Jp=o ' v d P ' 27U pnm+1 =2ni if Ti=?7i. 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. g) is the product V^Tre" ^oM!(n/i)!(n + 2rM)! (6. (6.^ + s 2 ) / 2 f°° 2 e(P+9+«)*~t /"GO 2 /2di 2 J — oo o = e " /2eP9e"(p+9) / e(t+p+9+a!) /2d4 J —oo ^/^Tga /2e9Qep(q+«) ^ gl^ The coefficient of p n in / ( p . the coefficient of qn+2rn the coefficient of p"qn+2r m exp(.5 nm .(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). (6. and 0 otherwise \/2n(—l)n+mn! if n = TO.82) ^ .q)./(n __ 2r — /i)!. so that /2 Q2(n+rrf in / ( p .q) = = e .80) f(p. g) is n! Next.78) \/27rn!.66) is \Zn\V2w Example 6.6.5: Generalized power integrals with Hermite polynomials Establish the following general formula for Hermite polynomials: / Jo t2sHen(t)Hen+2r{t)et ir n\(2sy.
e. n + 3)(n + 3.l ) ( n .i / ) ! ( 2 r + i/)! > .r .e.r . the relation (n. i.u)\ 2s<—" v\(n .t • 2l/2dt _ J : Q2 » a2(n+r^) ~ f ° ° f f „ t2/2 „ !ZoHen(t)e2 Wdt .A o?(.83) = (n + 2r)!e a / 2 V „ „.(t)e«2/2dt Now. n + 2){n + 2. Quantum Mechanics of the Harmonic Oscillator It now follows t h a t — in later equations with v := (n — /i).v)\(2r + v)\ > Q (2s)!(n + 2r)! AT 2~ s r 2V v\(2r + u)\{n . Prom Example 6. Hence t4 He„ (t) = = Hen+i Hen+4{t) (t) + [(n + 3) + (n + 2) + (n + 1) + n] H e n + 2 (t) + • • • + 2(2n + 3)Hen+2(t) + . /i = (n — v) _ / r o o e .3 we know that upgoing coefficients are 1 and downgoing coefficients (n.^ ) ! ( n + 2 r .ra + 3)(n + 3.' . K is the coefficient of a / ( 2 s ) ! in the expression J ..— — ^ .78). n. n)(n. by n!(7r/2) 1 / 2 . n + 2) + (n.^ / 2 * .e '' ^ o^M ! ( n .e. i.A i ) ! (6.' '2dt= / Jo t4 Dn(t)Dn+2(t)dt = V2^(2n + 3)(n + 2)!. i / ! ( n .r+v) a2s coefficient of in (n + 2r)\ V (2s)! h . . we obtain the result (6. A « 2 ( r '+ I/ ) 2 coefficient of (2s)! in (n + 2r)\e ' ^ . n + 2)]ife„ + 2 (t) + ••• . n .n + 2) + (n. so that now with integration limits from 0 to oo K = coefficient of a2s (2s)! in J a = 2.l . n + l ) ( n + 1. n + 2)(n + 2. n + l ) ( n + 1. the coefficient of T 2s »{sr u)\ As we observed. The result then follows with half the normalization and orthogonality integrals (6.2s _oo ( a 2 / 2 ) i ." ) a2^) r coefficient of (2s)! in f^ (s . It follows that the coefficient of a 2 s / ( 2 s ) ! in this expression J is the quantity K given by f^ot2sge„(t)Jfen+2t. n + l)(n + l . n + 2) + ( n .79).104 CHAPTER 6. n — 1) are given by the first index. . n + l ) ( n + 1. n + 2)(n + 2.° i i f e T 1 ( t ) i f e „ + 2 r ( t ) e . i. 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 + l ) ( n + 1. 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.Y^ i[ ^0vKnv)\C2r + ») cc2s A (n + 2r)! a 2 ^ .77). n + 4)_ffe n +4(t) + [(ra.v)\(s .n+ l)(n + 1.v)\ ^ Multiplying both sides by the value of the normalization integral.( n + Jr). n ) ( n .
which is a point of unstable equilibrium for a classical particle. of a differential equation of second order. With reference to this case.1) be the timeindependent Schrodinger equation which has to be solved.Chapter 7 Green's Functions 7. in Chapter 21 (with a different calculation).1 Introductory Remarks In this chapter we consider Green's functions of Schrodinger equations in both timedependent and timeindependent forms.x'). The other example of the oscillator enables us to evaluate corresponding expectation values of observables in the canonical distribution introduced in Chapter 5. and those which are timeindependent.x'. which we write here K(x. Let H^ = E^> (7. and is thus an important point to be noted here.t). we consider the sojourn time of a quantum mechanical particle at a point.2 Timedependent and Timeindependent Cases In the case of Green's functions we distinguish between those which are timedependent. We consider first timeindependent Green's functions. This example also offers a convenient context to introduce the inverted oscillator potential. which we write G{x. which is normally not discussed. in Feynman's path integral as the kernel or free particle propagator. The first of these cases will reappear later. 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. more generally. These are the Green's functions of the timeindependent Schrodinger equation or. 7. We 105 .
although the explicit calculations can be much more involved. appropriately decreasing behaviour at infinity).x') by recalling the completeness relation.g. Green's Functions H = H0 + Hi. of the kinetic part p2/2mo and some part of the potential. . (7.106 assume a Hamiltonian of the form CHAPTER 7. the relation X') = 5{x . The boundary conditions which G(x. We can obtain an expression for G(x. e. aq2.EW)xG{x. (7. i.5) M = Z*W=EW i l for Ei0) E * (7 6)  This holds since *For simplicity we consider the onedimensional case. (7. i.3) i which has the specific representation ^2{x\i)(i\x') i = 5(x . x') has to satisfy correspond to those of * (to insure e.g. We can readily see that G(x. Eq.x') can be written t G (7. a perturbation part like (3q4.e. square integrability.x') is defined by the equation* (Ho . for instance. cf. and Hi is some other contribution. (7.2) where Ho consists. since the generalization to higher dimensions is selfevident.x') or ^2^(x)^*(x') i = S(x .13).x'). The timeindependent Green's function G(x. where £ ( 0 ) = lim E. e.4) where the functions ipi(x) are solutions of HQij)i{x) = E^iix).x'). 'The case of £(°) equal to some Ei is considered later.e.g.
.t).10) The contribution on the right hand side is the socalled perturbation contribution. We see that the initial condition (7.7. which consists of the perturbation part Hi of the Hamiltonian and the corresponding correction E . We can see the significance of the timeindependent Green's function for the complete problem as follows.E^ We can check this by considering (H0 .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.x'. For the solution V we can immediately write down the homogeneous integral equation (again to be verified below) *(a.t) = ^Eie^ihM^i(x') i = Y^eE^ihUx)^(x').11) = = 3) = fdx'(H0 . The timedependent Green's function K(x.EM)X*(X) ( .x'.7) with the initial condition K(x. We rewrite the complete timeindependent Schrodinger equation H^ = E^ as (#0£(0))* = (££(0) #/)*.Hi{x'))^(x'). (7. x'\ t) (7.9) = i H0YieE^i%(x)^(x') = H0K(x. x'){E . (7.£(0) . i (7.8) This Green's function can be seen to be given by the following expansion in terms of a complete set of states (see verification below) K(x.x'.E^ of the eigenvalue.8) is satisfied as a consequence of the completeness relation of the wave functions ij)i{x).E<®)xG(x.E® (EE{0) Hi{x))m{x).t) since ih^K(x.t) (K for "kernel") is defined as solution of the equation ih—K(x.(*'))*fr') Hi{x'))^(x') f dx'5{x . . .x')(E .x. (JL x'\ t) = H0K(x. (7.#.0) = S(xx').) = f dx'G(x.x'){E .
e.Ej ffj(x'))*(a:') is a vector in TL which is orthogonal to ipj. then at best such a contribution would be something like cipi(x). Maharana.. if we assume that E^ = Ej. integral equations are more difficult to solve than the corresponding differential equations. it is possible to solve the integral equation by an iterative perturbation procedure (see later: Born approximation).x'.15) ''This problem and its circumvention can be formulated as a theorem. Examples in various contexts have been investigated by L. But if we assume that E^ ^ Ei for all i. In general. Wiedemann [183]. 2nd ed.11) are called (homogeneous or inhomogeneous) Fredholm integral equations.108 CHAPTER 7.2 (not contained in the first edition!). Green's Functions Can we add on the right hand side of Eq. (7.12) Equations of the form of Eq. H. Sec. (7. see E. however.14) We can now see the significance of the timedependent Green's function K(x. But then the Green's function (7. Instead of Eq. Booss and D. The timedependent Schrodinger equation ih\><l>)t = H\il>)t (7.t) for the complete problem as follows. In cases where an inhomogeneous contribution is given. (7. This difficulty* can be circumvented by demanding from the beginning that 0= f dx'tf(x'){EEjHi(x?))V{x'). Merzbacher [194]. In this case the perturbation is restricted to that subspace of the Hilbert space which is orthogonal to ipj. MiillerKirsten and A. D.11) with a system of linear equations of the form Vi = Mijyj. i. . This will be different. 17. the function ipi(x) would not be a solution of and it is not possible to add an inhomogeneous contribution. W.6) we would have (7. (7. see B. J. Bleecker [33]. All such methods of solution are based on the analogy of Eq. that (E . The theorem is also known as Fredholm alternative.6) is not defined. (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).
2 for the computation of the sojourn time. *>0. (6.5. Comparison of Eq.21) dx'K(x.7.17) is the initial condition of \4>)t. (7.x'.x'.20) is the timedependent Green's function. E.t) = J2 fdx/^n(x)^*n(x')^(x'..e.0)eEnt/in.t) = where if(x^'. t)\2. <a#) t = ]T n J [dx'(x\En)(En\x'){x'\^)t=0eE^\ (7.16) As usual. for the oscillator potential.g.22) According to Eq. (5.g. \En)(En\ip)t=oeE^\ n (7. 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.' which obviously satisfies the initial condition K(x. (7. in different formulations.18) or.2 Green's Functions: Timedependent and Timeindependent permits stationary states \En) defined by \1>)t = \En)eE^ih.19) i/>(x.t)il>(x'. 2nd ed.t) describes the evolution of the wave function from its initial value ^(a^O).x'. (7. 109 (7. t > 0. (7.158. . H\En) = En\En).t) is known.17) as initial condition is obviously W)t = Y. 7. x'. Eq. x'.20) the Green's function K(x.0). We can write this expression also as§ ip(x. (7.The solution of Eq. Merzbacher [194].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.Q) = 6(xx').t) = ^ e £ " f / « V n W < ( ^ ) . (7. 1 See e. t) with the Note that when K(x.15) which contains (7. p.51) shows that we obtain a very analogous expression for the density matrix PN(P) a s for the Green's function K(x.7) with Eq. this relation provides the probability density \ip(x. e. i. (7. We use this in Sec.
between (with E^0' = E) G(x. 7.E . With Cauchy's residue theorem we obtain I(t) := J2eEnt/ihMxWn(x'Mt) n = K(x.ie (7 . (7. Inserting for GE+ie the expression above.i /f e «/.«^«2 W .110 CHAPTER 7.x') = GE(x.x') and K(x.t) K(x. x')6(t) (7.x'.23) along the contour C in the plane of complex E as shown in Fig.1 The contour of integration.e. we obtain then also the corresponding density matrix. we obtain / W : = . i. We therefore consider the following integral with e > 0: I(t) := i J ^eEt/*hGE^e(x. 7. As a consequence of the above considerations one wants to know the connection between the timedependent and the timeindependent Green's functions. 24 )  ReE Fig. Jc 27T „ hn . t > 0. x'.t).25) in agreement with the timedependent Green's function . Green's Functions difference that (5 = 1/kT plays the role of it.1.x'. t) = J2 e £ " f M i ( x ) < ( x ' ) . = J2 ^„(s)C(*') En — E We see that G = GE depends on E. In the following we shall derive the Green's function for the case of the harmonic oscillator.
X>. i.x'.7. (7.0) = 5(xx'). (7. and identifying coefficients of the same powers of t on both sides. which is moving in one space dimension.31) The constant A has to be chosen such that K(x.x1) t B(xx')2/t B(xx')2/t 2AB 4AB2(xx')21 ~¥l 2+ W 2 (7. 2mQK h h2 ih(AB) = —{4B2A). A and B being constants.x'.28) B(xx')2 t2 DB(xx') 2 /t (7. B m0 n _ 2ih 2iK (7.t) t h2 = .e.30) into Eq. It is clear that it is nontrivial to solve an equation like Eq. we obtain *I£I = h2 —(2AB). . K(x.30) Inserting Eqs. (7.t) = ^eB^X In this case we have dK ~dt and dK dx 2 8K dx2 A A 2t3/2 A tV2 '^'K (7. / dxK(x.7).— d2 —2K(x. In this case the Green's function is the solution of the equation d hK(x.x'.3 The Green's Function of a Free Particle 111 7. v ' 2m 0 v h i.26) This is the case of a free particle with mass mo.e.27).t).27) An equation of this type — called of the type of a diffusion equation — can be solved with an ansatz.29) tJJ 2 2B(x .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>. Thus we try the ansatz.0) = 1. (7. We therefore consider first the simplest case with Hn = P 2mo h2 a2 2mo dx2 (7.29) and (7.
38) . (7. (7. p.H We can insert the expression for K(x.x'.. 158. (7. (7."737) \27rii/i in agreement with Eq. Green's Functions For parameter values such that the following integral exists.t) into Eq. 1st ed.(7.33) V Lirvnt Can we demonstrate that this expression can also be obtained from Eq.oo 27T (7.20) and can then obtain ip(x. j3 = i(xx').x't) = Y.36) 2mo Then — provided that the parameters assume values such that the intergral exists — K{x.0)oce"Qx2+ifcox. n For a free particle moving in the onedimensional domain \x\ < L — oo.36) ^ W Q ^ J emo(ii') 2 = —eP2lia 27T dke~^k^2^2 J„00 /2«tS .t)= f^em2t/2m0iheik(x~X')_ J 27T (735b) We set a = i—.21). (7.t) — for instance for a wave packet given at time t = 0 of the form ^(x. 2m0 (7. from K{x. Merzbacher [194]. Mx)^M^) „ n so that J =^ . (7. i. 32) It follows that x'.x'.zEnt.t) = — / dkeak2+Pk 2vr J.ihMxWn{x'). *>0.35a) K(xy. t) = /jJo_ e mo(««') 2 /««. "See the excercise in E.e.34) K(x.33).112 CHAPTER 7. v27r En^^. we have This is 1 provided A=JB= ^ (7 . we > have to make the replacements V ^ fdk.
42) .8). .d r. . mQLu . .t) + ~x K(x. x'.x'. —KK(X..7.33). (21.e. i. / ) = ~^K(x.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. (7.e. .x usn ot i.e. in this domain *Recall that S(x) = ± f eikxdk.. . z .x'. 7.x'..X . / ) (7.43) 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.t) 1 + m0u2x2K(x. I (7.40) becomes 2 .7) is the solution of d ih—K(x.t) = h2 2 d2 TW . —ih—K(x.39) In this case the timedependent Green's function K of Eq. (7. (7. the relation* 6(x) = o<y(os). x'.t).37) will later be obtained by a different method — see Eq.t). 5{ax) = ± J eikaxdk = ArS(x). (7. We consider the onedimensional harmonic oscillator with Hamilton operator HQ. §jK(x.x'. . (7. i.40) Then Eq. 1 2mo u>n ox n with the initial condition (7. i. HQ = ^—p2 ZrriQ + \m0uj2q2.3 The Green's Function of the Harmonic Oscillator 113 The result (7. x'.——K(x.0) = 6(xx') at / = 0./ ) + fK(x.t) We now set h2 d2 = .x'. K(x.e. We rewrite this initial condition in terms of £ and use for a = const.25) — in the context of Feynman's path integral method.
48).47) We insert this ansatz for K into Eq.e.51) See R. (7. it • is suggestive to attempt for K the following ansatz^ m.48) ocexp[{a(/)C 2 + 6(/)e + c(/)}] (7.c o t h 2 / .x'. we must have /o = 0. Identifying coefficients on both sides. p. that is. (7.e.b2. so that for / — 0: > a^—.45) in this sense. (/ = 2a.46) If we interpret Eq. (7. a = .^ . Integrating Eq. W ) .t) become similar to expression (7.4a&£ + 2a .42) and obtain the equation (with a1 = da/df etc. and we expect K(x.50c) (7. .Z'. f i.. P. c(0)^. Feynman [94]. as the limiting case / — 0.45) in accordance with (7.) a'£2 + b'i + c' = (1 . (7.114 CHAPTER 7.49) a=icoth2(//J0 ) w 2 "' 2tanh2(//0) * To ensure that the expression (7.50a) we obtain (7.b2.4a 2 )£ 2 . 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.4 a 2 . 50. i.47) becomes (7.50b) (7.33).50a) (7.f) with a(0)»y. Green's Functions to the particle behaves almost like a free particle. we obttain a' = l . (7. (7. b' = Aab.
as we shall see in the following.47) we obtain ^ c o t h 2 / + ^ + ^ c ° t h 2 / (7.53) or. / —————rexp y 27rmsin(u. (7. x . . B independent of / .c(f) K = 5 =exp Vsinh 2 / into Eq. Felsager [91].t) = . (7. c(0) = £' / 4 / . i.45). Finally Eq. 7.e. if we return to x. (5. we must have A = ('. (7.50b) Kf) = —r^77) smh 2 / A independent of / .t) to obtain this element (x.In B.* With this result we have another important quantity at our disposal. and to ensure that we obtain the prefactor of Eq. Comparing the Eqs.0).e.5. i. 115 To ensure that in accordance with Eq. with A.2. as one can verify.i) (7. (7.x'. (7. x'.33) for the Green's > function of a free particle.50c) yields for c(/). 174.£) _  {(x2 + x /z ) cos tot — 2xx'} 2msm(o. x') of the density matrix p^ (with respect to the canonical distribution with (5 = 1/kT): pN(x. K(x.55) 'For an alternative derivation and further discussion see also B.48) 6(0) = — £'/2f.51) for the density matrix PN(X.54) For t —• 0 this expression goes over into the expression (7. 2TT^ Inserting a(f). in particular for the derivation of the sojourn time in Sec. we can use K(x. (7. In order to satisfy Eq. c(/) = i ln(sinh 2/) + ^ 2 c o t h 2 / .48).40) of the timedependent Green's function with Eq.7.b(f). t.27). (7. p.3 The Green's Function of the Harmonic Oscillator Correspondingly we obtain from integration of Eq. we must have (besides A = —£') = /m 0 u.x'./3) = J IJIQUI 2irhsmh(hLu/kT) 2hSinh(hw/kT) \{X +X j C x exp ° S n kT lXX J (7.X'. (7.
p. P.. 52. [ [ 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. 11 For T ~+ 0: u>0 .. )x Inserting into Eq.55). eP E i = i + y. Feynman [94].p)dx W ) = —^r = —p .52)) the expectation value of an observable A in the canonical distribution (i. Eq. ^—• Irpiv J pN(x.x.e.x.56a) the expression (7. <2 (7. we obtain § < Z> = ^ c o t h ^ ^ ° — . R.x.uij>0 >• 0. Eq.116 CHAPTER 7.52): Tr (PNq2) Jx2pN(x. Thus the system is in a mixed state and the expectation value "Cf.p)dx Thus for (q2) = Tr(pq2) = ^2i(i\pq2\i) we obtain: 2 (7. . (5. at temperature T): {A) = Tl{pA) = ^ > . (7. 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"}. we skip the algebra here. For instance we have with Eq.» l.56a) {q2) = = Y.56b) w v ' 2mQu} 2kT 2m 0 u ' What is the meaning of this expression? At temperature T the fraction (cf.40)) 1 epEi 1 Ui = y. = jdxp(x.e.P) 2.eft >E*) E (7 57) ' of the number of systems of the ensemble occupies the quantum mechanical state i. Green's Functions With this expression we can evaluate (cf. (5. (5.
J (w — ie)< e (n+l/2)Et e i(n+l/2)u. and we replace to by LO — ie.7. (6. /2N 2~7T: = eE°/iht(t>Q(x)(t)0(x') for £ > 0. setting a = mooj/h. If we consider the system in the pure state \i). the expression for (q2) would be:'I (fti = J(i\x)x2{x\i)dx j(i\x){x\i)dx (Q2)o = h 2m. of K(x. t) = J2 eEnt/ihct>n{x)<t>n(x')For t > 0 and En = (n + l/2)frui the factor exp(Ent/ih) exp is (7.59a) E 0 = ^fiw.QUJ (7. We assume t > 0 and t —> oo.3 The Green's Function of the Harmonic Oscillator 117 (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.t) w—>UJ—ie to—n t—>oo moco/irh ^ ' piuit g—iwt exp m ^\(x 0 ^ .54).54) is K(x. 2 + x'2)le^2xx' to—>w—it t—>co 2 . t) of Eq. 1 mo^ exp lmow / 2 2 ft X (7. This is the first and hence dominant term of the expression (7. In this case the Green's function K(x. <*2>o dxx a •K j e IT I da r dxeax *\1/2 da a J _ 1 _ _ h 2a 2mou j —( .e.21). which means in the oscillator state \i) with eigenenergy fku(i + 1/2). i.t "With the normalized ground state wave function of the harmonic oscillator given by Eq. x'.56b) is that with respect to this mixed state (whose cause is the finite temperature T).x'.59b) n+ l \ ftwt here 2hf exp _ i ( n + . x'.50) we obtain. (7. We return to the Green's function (7. e > 0.
(7.5.e. However quantum mechanically in view of the uncertainties in position and momentum. 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.62) . In Example 7. which are planes.x'. is defined as vv = . to infinity) the contribution with n — 0 dominates.1 we estimate T semiclassically. lie on surfaces.e.. The wavelength A is given by The phase velocity v<p. a particle placed at the maximum of the inverted oscillator potential (which is classically a position of unstable equilibrium) will stay there indefinitely. i. the particle will stay there only for a finite length of time T.118 CHAPTER 7.61) The word "plane" implies that the points of constant phase <p := k • r — uit at t = const. Considered classically.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)]. 7. k = k. Green's Functions For t large (i.59a). and with this we estimate the sojourn time T. 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. In a very analogous manner we obtain the solution of the equation for the density matrix.(3) = J2^EnMx)K(x') n „ ^ ° e^MxWo^') (760) 7. (7. i.e. of dpN as pN(x. the velocity of planes of equal phase.5 The Inverted Harmonic Oscillator We encountered the inverted harmonic oscillator already in some examples. (7. as in Eq.
e.<p(k"))}dk'dk".7. or dco' da The centre of mass determines the particular phase. i.65) Let f(k') = \f(k')\eia and <p := k'x . to the function ip given by oo / fik'y^'x^dk'. The wave packet describes a wave of limited extent. /OO / / / oo /*oo \f{k')\\f(k")\ei^W^k"»dk!dk" l/(fe. i. e . the fundamental postulate on matter waves in Chapter 2): E = hu.64) where / ( k ' ) differs substantially from zero only near k' = k. A wave packet is defined as a superposition of plane waves with almost equal wave vectors k. i. xt— + — = 0.Jt + a. One defines as centre of mass of the wave packet that value of x for which d(p — = 0. t) is the spatial Fourier transform of this Gauss distribution — as we know.Q ( k . for which \ip\ assumes its largest value: OO . (7.4 The Inverted Harmonic Oscillator 119 Every frequency u belongs to a definite (particle) energy E (cf. e. u = w(k). then ^(r. a > 0 .)ll/(fc")l[cos(¥'(fe.k ') 2 . essentially again a Gauss curve. If we assume for / ( k ' ) a Gauss distribution. ' t) dk'. We now ask: How and under what conditions does the time variation of the function ^(r. ^ ( r . oo (7.e. duo' da . (7.63) The relation u = w(k) is known as dispersion or dispersion law. / oo J—oo .)¥'(*:")) +ism(<p(k') . i ) = /*/(k')e i(k '" r '' . t) describe the motion of a classical particle? For reasons of simplicity we restrict ourselves here to the onedimensional case.e.
(7. (7. a > 0. The threedimensional generalization is evidently du . = — (768) It is instructive to consider at this point the following examples. 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 = . Green's Functions This expression assumes its maximal real value when <p(k') = ip(k") = const. i.120 CHAPTER 7.^ g=r a rad. Example 7. p = hk the group velocity is equal to the particle velocity v..— g(k).fc w> vv = ~a^ = S r a d P E «i7 Sd u . 9 cku dk dhw dhk dE dp dE dp' We can also argue the other way round and say: By identifying vg = v.e. Solution: The function to be calculated is the integral dxeax OO ezkx. For E = hio. .67) The centre of mass moves with uniform velocity vg called group velocity of the waves exp[i(kx — cut)]. .69) In general one uses the theory of functions for the evaluation of this integral. however. as claimed for =0. In the present case.1: Fourier transform of a Gauss function Calculate the Fourier transform of the spatial Gauss function e . we obtain the deBroglie relation p — hk. dE v 9 = . Thus the expression g{k) is solution of the following first order differential equation g'(k) + Aff(fc) = 0. we can use a simpler method.
Afc = 81n2. .3: The uncertainty relation for Gaussian wave packets For the specific Gauss wave packet of Example 7. XZ (e > 0).. (7. g . (7.0 0 7T 7 0 7T e From Eq.72) Solution: From Eqs.2: Representations of the delta distribution Use Eqs. with partial fraction decomposition.a) + 6{x .4 The Inverted Harmonic Oscillator Simple integration yields g(k) = c e " f c 2 / 4 a .73) we obtain the requested representation of the delta distribution with 1 f°° 1 € Six) = lim — / dke~e^eikx = lim .. Since /•c 121 9(0) = J—00 dxe J_ we obtain S(fc) = ^ e " f c 2 / 4 a ( 7 . Example 7. MiillerKirsten [215].71) (7. \a — b\ (7. Appendix A.69) and (7. J...69) and (7.1 verify the uncertainty relation Aa. where c = 9(0) is a constant.74) Example 7. —±=ek*/ia2 / 4 " = lim 1 — — lim ^e* = The second important example can be verified by immediate integration.7. 2 f + .70) to verify the following representations of the delta distribution: X2/€2 5(x) = lim ?—=e ^ o eV7r and 5{x) = \\m^r^.70) we obtain 6(h) = lim — f°° dxeax2eikx = . or see H. (7. W. (7. (7. (7.73) 27T 7 .7 °) With the help of this example we can obtain some useful representations of the delta function or distribution as in the next example.43) one can verify the following important relation* 5[{x a)(xb)] = T—!— [S(x .75) ' E .&)] for a + b.r „• With the help of Eq. We have — 1 / * ° ° dke~eWeikx =  I / " 0 0 dke~tk cos kx =  1 .. (7.
77b) (7. is Afe = 2 V 2 1 n 2  (7. i.e. and hence implies large values of Ak. Physically .r)1/2a Re f(x) Ax Fig.g. The wave function corresponds to the function f(x) in the above considerations (e. (7. where g(k) = max<?(fc)/2.77a) g(k) = e .e. Green's Functions 1/(27i)1/2a — 1/2(2. A simple calculation yields A i = 2v/2~In2a. Thus a sharp maximum of the function f{x). (7.t) is a measure of the probability to find the particle at time t at the position x.2 The Gaussian curve. when Ax is very small. the width Ax is a measure of the uncertainty of the probability. requires according to Eq. The breadth A/c of the curve g(k) around k = kg.CHAPTER 7.70) the Fourier transform of f(x) is (7. According to Eq.77a) small values of a.76) 2ira The uncertainty Ax is defined to be the width of the curve at half the height of the maximum.78) It follows that the product of the uncertainties is AxAk = 8 In 2. where  / ( x )  = m a x  / ( x )  / 2 . i. 7. In quantum mechanics the square of the modulus of the wave function ip(x.* o ) 2 / 2 . 7. Thus a slim maximum of the curve of /(a.« 2 ( * . Solution: In Fig. at time t = 0).) leads to a very broad maximum of the Fourier transform \g(k)\.2 we sketch the behaviour of the Gaussian function p—x 11a ~—ikox (7.
«> = / ^ C ' W . (7.79) According to Eq. (7. many supposedly elementary problems request the calculation of the sojourn time of a quantum system near a classically unstable equilibrium configuration. Barton recalls that W. For reasons > of simplicity in the following we set in addition t mo = l.. Drastic idealizations are then required in order to reformulate the classical situation into that of a quantum system.54) these parameters appear in the combinations moui/h (dimension: l e n g t h .54) with the substitution ui — iu.2 A particle's sojourn t i m e T at t h e m a x i m u m The very instructive topic of this subsection has been explored in detail in a paper by Barton. In the reverse case a sharp localization of the particle in momentum space implies a correspondingly large uncertainty of its spatial coordinate. In the same limit the momentum uncertainty Afc grows beyond all bounds.2 ) and U)t. The result (7. E. the less precise is the determination of its associated momentum. O ) . nr = S(x). (7.".u> = l. the wave packet has its centre of mass at the origin and has the following Gaussian shape as the pure initial state: iP{x^) = ^L=ex2l2b\ (7. (7.76) ex 2 /2a2 ^2na a>0 um 1/0*01 = lim a^O .x'. In the limit a — 0 we obtain from Eqs. <7*» We assume that initially. As G. . 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.* We obtain the Green's function K{0 for the inverted harmonic oscillator from Eq. Thus G.5.t) = V 27Ti s i n h t exp 2 sinh t {(x2 + x'2)cosht2xx'} .81) *G. Estimate the maximum length of time compatible with quantum limitations before the pencil falls over. i.e. The function g(k) is therefore described as momentum space representation of the wave function f(x).7. (7. Barton [15]. 7.h = 1. so that Ki0(x. at time t = 0. ^In Eq.20) the wave packet at time t > 0 is obtained from that at time t = 0 with the relation «. This means the particle is localized at x = 0.. Barton mentions at the beginning of his paper.71) > and (7.75) implies therefore: The more precise the coordinate of a microscopic particle is determined.4 The Inverted Harmonic Oscillator 123 p = hk is the canonical momentum associated with x. which means all values of the momentum are equally probable.
85) = —r= o2 7 2 sinh 2i. 9 (7.83) sinh 2 t + (i/2b2) sinh 2i 2 sinh £ 2A 2 sinh 2 1 (7.82) i/f(x. so t h a t (this defining C(t) and the imaginary part ip on the right) {B2 + icotht) sinh 2 1 + (i/26 2 ) sinh 2t ( l / 6 ) s i n h 1 .81) and (7.84) dx' oo tvirl/2b With some algebra one can show that (B2 2y and A2 sinh 2 t f7 s ^ sinn / + icotht) = .79) into (7.t) = J oo dx' Setting A2:=±we can rewrite ip as ex b2 and AB sinh t' (7.t) = A\/2msmh 2TT 2 /2C2+i<px2 tVir^b In t h e further calculations it is convenient to set t a n 29 = b2cotht Then ~A 4 and VA + iB := ReiB (7.(Ax' + Bxf \/27risinh + (B2 + ^cothi)a. (7.t) = J oo P .2 (7.80) we obtain „/2 oo ex P 2 s i n h t { ( : C + x')^ht2xx'} \/27usmh . (7. . ' 2\Aswht % cosh t (7.(i/2) sinh 2t 2 2 2C 2 (i) + *c^.86) Evaluating the Gaussian integral of Eq.88) .87) <9 2 „9 2 sinh21 i? = A + B f / ^ d x e " ^ / 2 " 2 = v^F<*. + cosh 2 1.84) we obtain* ip(x.cosh t sinh t' Wisll2b ^ (7.83) ip(x. = ... Green's Functions Inserting (7.124 CHAPTER 7. A .
C2(t) (7. _2 1 exp C 2 (i) = l + 2sinh 2 i.t)\' 2 exp[x2/b2] rrVaft In the following we choose 6 = 1 . t) the phase ip drops out and one has (x. since dy Jo the derivative dT_ ~dT For b = 1 we have f°° dt= / (7.90) with the expected limiting behaviour \im \ilj(x. We have therefore .94) / dxf(x)=g'(y)f(g(y)).86) has C2(t) := b2 cosh 2 t + 72 sinh 2 1. (7.t)\2 Ji == 71 / V " Jo dO ? 6' (7.89) = exp[x2/C2(t)} 7rV2C(t) (7.7. •&(t)=//c(t) = / or.£ 2 (7.t)\' 2 125 (7.95) ..91) C(t)' (7. In taking the modulus of ip(x.4 The Inverted Harmonic Oscillator The real part of Eq.t)= where Co(t) / dx\^(x.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. The probability for the particle to be at time t still within the distance I away from the origin is plausibly given by Q(l.93) where —{dQ/dt)5t is the probability that the particle leaves the vicinity of the origin in the interval St around t (partial integration leads from the second expression back to the first).
From this we obtain C(t) C'(t) Z2 ^(r?2_.126 We set I C(t)' so that dT ~d~l Now.98) We insert this into Eq. V and (cosh t = y 1 + sinh2 t) 1 r CHAPTER 7 Green's Functions dr} dt C'lt) ' C(t)' (7. 136.96) °° l Jr.99) .=o V(i v W + v2) oV "dT J L 1 f A x 2 __H_I r'=s/21 Adx'.=0 V^ C'(t) I Jt=0 V^ (7. Dwight [81].2)(p_7?2)(7. + 0(e'2) (7.. we obtain dT _ 1 ~dJ~l 'See for instance H. p. B. (7.97) sinh 2 1 = . £ 1 dr\ 2Z %_ rl 2 2 yfx Jr..(4 2 .= \ / l + 2sinh 2 t.1 ) 2 \n J ^"•""ff^^V^K?.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 /2^7.97) and obtain dT ~dl or with £ = ?y/7: 2 2 d^e. C(t) = .
. B. Such a field arises in the spectrum of states in string field theories and is there called a "tachyon". 7. this is T ~ ]n(ly/mou>/h). U) (7. (7.3 How stable is the particle on top of the eggshell in quantum mechanics? E x a m p l e 7.7.101) A more detailed evaluation of the constant can be found in the paper of Barton cited above.g.102) 'As a matter of interest we add that in the theory of a free scalar field x. Fig. 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.4 The Inverted Harmonic Oscillator 127 for I sufficiently large. See e. 238. Explain the parameters entering the calculation. T h e following Example 7. 2 2 2mo mow x The classical equation of motion is therefore' it — uj x = 0 (7.100) Reintroducing the dimensional parameters we had set equal to 1. p.4: The sojourn time calculated semiclassically A tiny ball of mass mo is placed at the apex of an upright egg. It follows t h a t T~lnZ. this equation is the equation of motion of this scalar field with negative masssquared. The equation there describes the classical "rolling down" of this tachyon. 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. also motivated by this paper.4. Zwiebach [294]. is instructive in revealing the basic quantum mechanics involved in the finiteness of the sojourn time.
102)) p(0) = ^/(p 2 >. the smallest macroscopic length. so that AxAp= In our semiclassical consideration we set therefore x(0) = y/{x*). the expectation value of x is the integral with an odd integrand) <x) = 0 = <p).128 with x2 = LO2X2 + const. (7. i. .e. pifi) have to be replaced by the positions of the spatial and momentum maxima of a wave packet. Thus we set I = A cosh LOT + B sinh coT ~ (A + B)e"T. x(0) = A. (x 2 ) h 2mou 2mou> and so l^J^ULl). We ask: At what time T > 0 does the ball reach the point with horizontal coordinate x = I. so that e. (p2):=mW(x2). Thus 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 . Eq. Green's Functions x = Aco sinh cot + Bco cosh cot. v A x A p = \ (x2)(p2) v = m0u(x2).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. i. and for simplicity we take (cf. / 7 ^ 2 =2J(x }. but instead are subject to an uncertainty relation of the form AxAp > h. and solution x = A cosh Lot + B sinh cot.5)) Ax = <J(x2){x)2. (7. Quantum mechanically x(0). For a symmetric state like the usual ground state of the harmonic oscillator we have (since the wave function is an even function. CHAPTER 7. so that raoto _ We assume now that at time t = 0 the ball is placed at the point x(fi) with momentum p(0).e. Ap= v /(p 2 )<p)2. (10. For h —> 0 the time T —• oo in agreement with our classical expectation. 2 x(0) ^ where A i .p(0) cannot be determined with arbitrary precision. 2V ' Let A and Bio be the values of x and x at time t = 0. A p are defined by the mean square deviations given by (see Eq.g. ^(x2)(p2). 2 10 x(0) = B u .
(i)+/^(2) + ". but this is no longer the case for an anharmonic oscillator potential like ax2+(3x4. 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.E^ is already a bigg problem.1 Introductory Remarks The Schrodinger equation can be solved exactly (i. 129 . In the case of the harmonic oscillator potential ax2 the Schrodinger equation can be solved exactly with ease.e.Chapter 8 TimeIndependent Perturbation Theory 8. one would set tt E = = ^(°)+/ty. in closed form) only for very few potentials. It follows that in general one depends on some approximation procedure which is usually described as a perturbation method. For mathematical purists the question of convergence of the series is an even bigger challenge. In general perturbation series do not converge.e. Frequently even the calculation of the next to leading contributions ip^. i. 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.2h2 cos 2 a # = 0.
.g. 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.2 However. E.* i. . It will be seen later (e. 581. ^For further details. In the following we explain the difference between convergent and asymptotic series.g.• + (ir~ X X* 1' 2' r?l Xn + Rn(x).. N. the expansions in ascending powers of the parameter h2 can indeed be shown to have a definite radius of convergence. Whittaker and G. 0.2) n—»oo n It is interesting to compare the behaviour of the series (8.g. e.3) *See e. p. explore the latter in somewhat more detail and finally consider methods for deriving perturbation solutions of the Schrodinger equation. the expansions i> = v (0) + /*V (1) + /*V 2 ) + • • •. The series n=0 of the exponential function is well known to converge absolutely for all real and complex values of x. TimeIndependent Perturbation Theory In the case of this equation.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. W. (8. perturbation theory and path integral methods — that the lightminded way in which perturbation theory is sometimes discarded is not justified.+ . E = E^ + h2E^+hAE^ 71 72 7 + . concerning also the uniform convergence of the exponential series. the expansions in descending powers of h2. at the end of Chapter 26) — and by comparing the results of WKB. Meixner and F .g. 8. (8.. Cll E = a_ 2 n + a_i/i + a0 + — + —r \ ..e.= 0 < 1. J. This convergence can be shown with D'Alembert's ratio test since''' lim . T. which is known as the Mathieu equation. Watson [283]. see e.130 CHAPTER 8.. Schafke [193].1) and its terms with the behaviour of the following series and its terms: f(x) = 1 . h hl are in the strict mathematical sense divergent.
We observe in the first place that since lim n—>oo co (8. / ( I ) = 1 — 1! + 2! — 3! H • However.0.g.g..33333 . T.e.4938.l n ( l + ^ ) n=l L ^ ' (8. From the Weierstrass product which defines this function* one obtains the series _ M*l)! = 7(*l) + E [ ^ . It is evident that the larger the value of \x\. .0..5) Cf. We can see this as follows. .3) with the convergent series (8. .2 Comparison of Asymptotic Series with Convergent Series 131 where Rn(x) is the remainder sum. However. . .22222 . E. This type of behaviour is characteristic of the terms of an asymptotic expansion as we shall see in more detail in Chapter 20. Whittaker and G..1). after reaching 6/27 = 0. and we see that the moduli of successive terms first decrease and then.222. p... we have ^ ' ~ = 3 + 9 ~ 2 7 + 8 1 ~ 2 4 3 + 7 2 9 ~ 2187 + " ' ' 1 . Comparing the asymptotic series (8..22222 .3) can still be used to obtain almost correct values of f(x) for x sufficiently large..001 + 0.376 .. for x = 3. i.4) for every value of x. /(1000) = 1 .0.3) diverges for every value of x. + 0.0..8. the series (8. N. Normally a divergent series is characterized by an ever increasing behaviour of its terms as in the case of x = 1.2. the partial sum of terms up to and including the least term yields a reasonably precise value of the function at that point with an error of the order of the first term of the remainder. in spite of this the series (8.000 000 06 + • • • .. +0. + •••. . Thus e. 235. begin to increase again. e. The theory of asymptotic expansions claims that if the expansion is truncated at the least term.9876.0..29629.000 002 00 . we observe that the individual terms of the asymptotic series have the form of a factorial divided by the power of some parameter which is large. and in fact increase indefinitely.. + 0. As a second example we consider two series expansions of the gamma function or factorial T(z) — {z — 1)!. Watson [283]. the better the approximation obtained. for larger values of x.
In many respects Messiah aims at more rigour in his arguments. Messiah [195].5772157— Here the series on the right is an absolutely and uniformly convergent series of the analytic function. p. I. This is not appreciated by mathematical purists. in fact. E. 236.1 between Eqs. Vol. the leading term of an asymptotic expansion.5) are practically unknown.J In2. p. Schiff [243].e.1. 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. meaning convergent. . it is sensible to make the assumption. Watson [283]. 11 L. at a later point in his treatise Messiah admits that the expansions are mostly § Cf. Thus one can say that the vast majority of expansions of this type in physics is asymptotic and not convergent. 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. § However. 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. T. It is interesting to observe the comments of various authors on this point. 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. whereas applications of the convergent series (8. Sec. N. 152. for ln(z — 1)! one can also obtain the Stirling series ln(zl)! = ln(2vr)5 z+ (z. p. 371.5). although this has not been investigated except for a few simple problem^. Applications of Stirling's series can be found in all areas of the physical sciences (particularly in statistical problems). the dominant approximation is. ip) can be represented by rapidly converging power series in e. In fact. **A. TimeIndependent Perturbation Theory where 7 = 0.132 CHAPTER 8. (16. On the other hand Schiff" says:" We assume that these two series (for \I/ and E) are analytic for e between zero and one. Whittaker and G." This is a very clear statement which does not try to pretend that a perturbation expansion would have to be convergent.41) and (16. This latter observation hints already at the importance of asymptotic series in applications. Merzbacher [194]. that E and \ijj) (i. ^E." These "rapidly converging power series" in physical contexts are most likely very rare cases. 16.
which have been studied in great detail. 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. possess asymptotic expansions which have for a long time been important a n d accepted standard results of mathematics.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.ft somewhat afraid t o say so himself a n d therefore with reference t o investigations of T . we can rewrite Eq.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. Vol.11) A . Dingle [70].= / e~t2dt.^=e~x2 / e^^dt. there is no need for such bias.. (8. p. II.. solutions of second order differential equations.7) This function has been considered in detail in t h e book of Dingle. Messiah [195]. 198 (German edition). In fact.10) Vn Changing t h e variable of integration to u = t2x2. (8. (8.* and we follow some of t h e considerations given there. Replacing t h e exponential in Eq.. *R. . 8." Of these three authors t h e first.8. We know from books on Special Functions t h a t all of these functions.7) as / e'* dte* dt = 1 . ft Jx du = 2tdt. It is therefore suggestive to write iy poo <t>(x) = 1 . (8. B. (8. Kato: "Indeed the perturbation expansion is in most cases an asymptotic expansion .2 Comparison of Asymptotic Series with Convergent Series 133 asymptotic.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. 2 (8. He says.2. Merzbacher.
7T (ni)!sin{7r(n+^)} ( „ . 2 9 f00 / Jo du e" 2(u + x2)^ u <£(*) = 1 T h e binomial expansion . 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. Using the reflection formula (*!)!(*)! = for z = n + i . « \~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. (8.15) < 1 we can insert this expansion into Eq.12) VK Jo n V oo 7 n=o x) 2 2*Qn\(±n)\\x2) (8.e.' we have 2 nH 7T 7T (8.e. (8. since (—4)! = y/n.16) . du.i)[(_!)«' Then.^ X e U ~ ( 1 + * \ 2 .14) sin7T2. we obtain 1+ i.^e~x \Ar i.13) presupposes t h a t < 1.12). TimeIndependent Perturbation Theory <P(x) = 1 .134 we obtain CHAPTER 8.
and in fact so slowly.l)! = / Jo Then 2 cHnxdt. (8. the asymptotic expansion of the error function is. . We now see how the asymptotic expansion (8. We can actually understand the deeper reason for this in practice. (8.e.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. i. /•oo T(n) = (n . that a large number of terms has to be summed in order to obtain a reasonable approximation of the quantity concerned.19) originates.15) into (8.12) ignoring the fact that the latter is valid (i. ' arg. (8. much more useful than the convergent expansion.19) implies either that we ignore the remainder or the correction term given by the integral in Eq.18) or that we insert the binomial expansion (8. convergent) only in the restricted domain u < x2. In an asymptotic expansion the first few terms successively decrease in magnitude. The expansion (8. Frequently "=" is written instead of "~". Whichever way we look at the result. i.19) approximates <f){x) the better the larger x is. 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. In the foregoing discussion of the error function we assumed that x was real and positive. argx = 0. We now want to relax this condition and allow for phases argx ^ 0.8.e.—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^. whereas in a convergent expansion the terms first increase in magnitude. As in the case of the gamma function.17) 0(X) = i. (8.e. in general.19) where "~" means asymptotically equal to which in turn means that the right hand side of Eq. It is fairly clear that the above considerations remain unaffected for argx < 7T.r<7r.
(8.7) and obtain 4>(iy) = —= j e~l dt = = / es2ds.e. however. (8. we set x = iy in Eq.26) or else return . 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. =dv. thus with \v/y2\ < 1.20) where we set t = is.vds.^ f V e ^ . dv = 2sds = . Proceeding along lines similar to those above we write <f>{iy) = ^=ey2 I* e^^ds.2 y V . V7T Jo Changing the variable of integration to v (8.22) we obtain 2 4>{iy) = .^ = . We can therefore rewrite the integral as (j)(iy) i 2 fy2 y —=e / e~v 1 . The situation becomes critical. when argx = 7r/2.24) We can read off the binomial expansion of the factor in the integrand from Eq. (8. we can try to proceed with Eq.23) Throughout the range of integration 0 < v < y2 the integrand is real.136 CHAPTER 8. (8. i.21) = y2s2. We therefore consider this case in detail. VK Jo V^ Jo (8. TimeIndependent Perturbation Theory since for this range of phases the decreasing nature of the exponential is maintained. (8.15).
28) ignores the integral contribution of Eq. (8. (8.2 Comparison of Asymptotic Series with Convergent Series to Eq. (8.29) where 5t means "real parf. we obtain 4>(iy) = ^ey2TZ f°° e~v (1 . ey is much larger than something of order 1/y.^ ^ p .28) n=0 We observe that this expansion differs significantly from (8.28) we observe that the expansion (8.28) because then the correction term oc f°£ becomes smaller and smaller. However. i. for  arg x\ < 7r/2 and for arg x = IT/2 the error function 4>(x) possesses different asymptotic expansions.e. (8. i (8. (8. We have therefore <j>(iy) = ^e^TL fV e~v ( l . and the larger y. (8. „2 \v J v^(ni)! y(nj)l ^—' y2n ie"2 /00 ^ ieir /•" W7r /„2 y ^—' _v^{n\)\(v n! .27) Then .26) we can write* 137 v* Jv2 ie « „2 OO / 1 n=0 M ~k n! . for large values of y.. the better the approximation expressed by Eq.^] * dv. We know that for v < y2 the integrand is real.19).24). — (8.8. Eq. If we proceed with Eq.2 yir (8.27) and (8.27). (8.24).17): J£° e~vvndv = n\. e ^ ( i y ) ^ ^ .30) yvn f Jo V y J Cf. For v > y2 the second integral is seen to be f imaginary and thus drops out in taking the real part. 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. and so the integral is real. which will be studied in more detail below. Returning to Eqs. .^ ) * dv JO V y /»00 /»' yV^ 1 i Vsfv 2 ey n oo J0 Jy' 1%) 'dv. i. (8.e. y y 2 ( n _i)! y arg(* = iy) = i * . Suppose now we consider Eq.
. R. (8.33) 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. y. (8.. e xZ ^ (n — £)! „ ^ ) . (8. i.2 OO rz=0 V . Looking at Eqs.31) and (8.* We observe that this Stokes discontinuity is a property of the asymptotic expansion but not of the function itself. (8. With a similar type of reasoning one can show that the same expansion. (8. .8)) 4>(x) = <f>(x) we see that _x2 OO / ] \  n=0 for 7r/2 < I arg x\ < 3TT/2. This result is identical with that obtained previously. 7r = . Chapter 1.33) we see that the Stokes discontinuities occur at those phases for which these expansions (i.28). Dingle [70].25) and integrate from 0 to oo. For the sake of completeness of the above example we continue the phase to  arg x\ > IT/2.31).30) can be written </>(iy) = ^ey2 fV ev(l^) yVn Jo V y J "dv = ^!rV"E^(4)V (8.e.138 CHAPTER 8. since (8. *Cf. and • higher order terms of the associated series all have the same sign. (8. Eq.3!) which is (8. TimeIndependent Perturbation Theory If we insert here the expansion (8.26).e. The corresponding phase lines are called Stokes rays. We have therefore found that XTT n=0 ^ ' and . is obtained for <f>(x) with arg x = —ir/2." — E ^ ^ forarg. Since (cf. B. we use the expansion outside its circle of convergence and hence obtain a divergent expansion.33)) would have • a maximally increasing exponential e+x .
36) has an exponentially decreasing solution and an exponentially increasing solution for zero phase of x. for arg x = 7r/2. Magnus and F. 1/1 ipv{x) v1 vl (±ix).e. pp. e. R. (6.34) the contributions 1.13. B. Oberhettinger [181]. W. See also Tables of Special Functions. . i. in the sum of the right hand sides of Eqs. we have e z2/2 _ 2 e(x Rxj+2ixRxI)/2_ The exponential is maximally increasing for XR = 0. Later we shall make extensive use of parabolic cylinder functions Du(x) which are solutions of the equation dx2 + v+ :X y =o .2. i. 91. though. (8.g. These rules. and so may not be universally applicable.2). Chapter 1. —1 cancel out.35) 8. p.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. Equation (8.2 Stokes discontinuities of oscillator functions Dingle' has formulated rules which permit one to continue an asymptotic series across a Stokes discontinuity without the necessity of a separate calculation of the asymptotic series in the new domain. Dingle [70]. Thus if we set x = XR + ixj.37) Cf. 9 . which can be written (for v not restricted to integral values) D^(x) DM(x) f xu(f>u(x). 5.e. We also observe that the expansion for  arg x\ = ir/2 is half the sum of the expansions on either side of the Stokes ray.32) and (8. apply predominantly to asymptotic series of the solutions of second order differential equations.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.rgx<n/2 argx>7r/2 _ X1T •<—' v — ( n=0 £ (ni)! X2)n ' oo (8. '"Cf.8. Eq. and 4>(x) argx=7r/2 = 2 r2 ^ + <f>(x) a.
Thus D{J\x) (i. i.ei<v+VxvHv{x\ argx = £ . Hence Dil){x) = xv(j>v(x).38)) has a Stokes ray at Dl '(X) the arg x = 7r/2.x 2 oo «*)  — E^fi=0 v . D^(x) = = xv4>v{x)Jr\^l'1^'Kl'1\A~v~X^^) xv4>v{x) + \a. as can most easily be seen in the case of the Mathieu potential. > > The equation (8. We consider Dv (x). (838) We observe that one solution follows from the other by making the replacements^ x — ±ix. B. = 7r/2. Eq. Chapters I (in particular p. 8. x~l/~1ipu(x). At arg x = 7r/2 the exponential factor becomes an increasing exponential and late terms of the series have the same sign. Dingle [70]. i. Chapter 17. v — — f — 1.7. Dingle [70].39) onto the Stokes ray and from there into the neighbouring domain is determined by the following rules which will not be established here:" • Dingle's rule (1): On reaching arg a. the series of D ^ x ) develops an additive contribution (the discontinuity) which is IT/2 out of phase with xu4>v{x) and proportional to the associated function.e. 0 < argz <  (8. TimeIndependent Perturbation Theory i.e. ' S u c h symmetries are extensively exploited in the perturbation method of Sec. 1 1 Cf. (8. The continuation of (8. (14). (8. Chapter 1.140 where^ CHAPTER 8. . ti asymptotic series (8. B.40) § Cf. 9) and XXI.36) is invariant under these replacements. R.39) (as in the example of the error function). We also observe that the late (large i) terms of the series in (8. R.37).e.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. with a real proportionality factor.
sm(iri/) = —a/3/2 or a/? = 2sm(irv).8.e. = ir.e. Applying rule (2) we obtain D$\x) = U™ + ^iap) (xyMx) + a{x)vlMx) (843) for IT < arg a. i.e.41) the Stokes multiplier a is. (8.40). beyond arg x — 7r/2.41) xuct)v{x) + a{x)~u~lipu{x).42) T ^(eiwu i. Since Dv (x) is real when x is real. 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. where in the first line on the right hand side the second term contains the real proportionality constant a/2. we must have at arg x = 7 in the dominant factor on the right of Eq. The value of Du (x) on the other side of the Stokes ray. 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). is given by • Dingle's rule (2): One half of the discontinuity appears on reaching the Stokes ray. for xj = 0) and therefore possesses a Stokes discontinuity there.41) to arg a. another half of the discontinuity appears on leaving the Stokes ray on the other side.2 Comparison of Asymptotic Series with Convergent Series 141 for argz = 7r/2. In (8. On reaching ir the part containing ^ ( x ) . with \x\ = (—x) for arg a.44) + ^iap) = 0 .42) el™ + ±iafi\ {x)v<t>v{x) + a{x)vl^u{x). Thus for arg x > ir/2 we have D$\x) = xp(t>u{x) + = 2]iaei<v+^x1/l^v{x) ^ < arga. (8. (8. e +fx2 _ 2 2 e±(x Rx I+2ixRxI) is maximally exponentially increasing (i. < 7T. (8. as yet an unknown real constant. < 37r/2.e. = ir. i. We therefore proceed to continue (8. 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.
l)0(v .1) (8.51) .45) is therefore given by a{v)a(v . i. argx = vr.38) = cos7ru(x)u(f)1/(x) + (—x)ve~*x COS7T^— —r a(u)(x)l'1^u(x) (2i .48) We compare this with the reflection formula (8. (8.1). (8.45) that af3 = 2sin(7r^)) D^\x) = (8.1) = 2sin7ri/. (8.14).47) Equation (8. V i=0 \U0 v 2w . We can see that the equation is satisfied by (3{u) = ±a(u . (8.1)! TT.42)) for argx = ir (in the second line of Eq.e. or sin7r(2: + 1) sin irz' W ( = 7 ^ ) ! = 2sil"r2Comparing this with Eq.e.e.1) = ±a(y).50a ) (aMb) »W = f "M=(=^)i In order to decide which case is relevant we observe from Eq. Eq. i. a(u)P(v) = 2sin(7r^) = 2sin7r(i/ . Dl \X) is (cf. (3{v .42) we insert from Eq. (8.e. (8.48) we can set <8 49) ' ^ ) = {_V\)V PM = ^f> or ( 8 . TimeIndependent Perturbation Theory Now a and (3 can still be functions of v. . (8. i. i. Hence the left hand side must have the same property. (z)\(ziy. 8.v .42) that a multiplies ipv(x).46) For a given function a(y) this equation determines /3(u).—^ ON • — ^ > + ^ ^ L .142 CHAPTER 8.45) Thus the right hand side of this equation remains unchanged if v is replaced by —v — 1. a{y)P{y) = a(u .1) (8.
2.50a) because this enables us to define normalizable parabolic cylinder functions. (8.7r). In a similar way we can examine Dv (x) and its Stokes discontinuities (at argx = 0..36)) in the domain 0 < arg x < IT. —f..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.3 Derivation of Asymptotic Series from Differential Equations 143 We choose (8. ' ^ (i/l)!z! ^ . ' (X) (cf. Then (in the second step using (—u — l)\v\ = — ir/sm(Trv)) DW(x) ~ c o s ^ ^ * f ^ .*"ei°> f ( ^ _  _ L _ = (!)««(. the error .35)..53) We have therefore obtained in a natural way the quantization of the harmonic oscillator. Chapter I. (8. i.36).A / ± sin^Cx)"M* 2 V ^ .) (8..8. (8.7).e. (8. (8. It should be noted that in Eq. (2inl)! (n1)! n! {n2i)\ So far we have been considering the asymptotic series expansion of Di.)(. For details we refer again to the book of Dingle.1 ) ! 4~L i\{2x2Y i=0 V2^(x).52) (argx = 7r).e. i.1 ) ( .53) the factor (—n — 1)! is to be understood in association with (2i — n — 1)! in the numerator. u 1 y> {2i + v)\ 1 n 0.„ _ix2^(2ivl)l e 2 > —. However.2^1 v y 7T i=0 ' (2i + v)\ z!(2x ) v (8. We observe that for v — n = 0. the second contribution vanishes and we obtain £>«(*) .e. Eq..I / . (2x2)1 = COSTTU(X) v . which vanish at x = ±oo. Proceeding along similar lines we can continue the expansion into the domain (2) ir < arg x < 2ir. i. 8.e^ . solutions of Eq.1..
^ + 2 .36).54) we consider again the important equation of the harmonic oscillator.36)) 0 where + x2Q(x)y = 0. (8.60) into (8.54) (8. Instead of dealing with Eq (8.144 CHAPTER 8. (8.. \ + U (8. Eq. previously we had ^.60) 2 x x l Inserting (8.. we obtain equations from which the coefficients So. x°. S2. Solving these equations we obtain = = 0. cf. x. how series of the type (8. (8. Eq. + — + ^ + • • • . 0. TimeIndependent Perturbation Theory function can also be obtained as the solution of a second order differential equation. We write Eq. Thus 2S 0 Si 5o + 25 0 5 2 + ^ + ^ and so on.37).59) and equating to zero the coefficients of powers of x2.56) ~l^(8 57) Q{x) =  We then set y = es^ so that the equation becomes (8. i.38) can be obtained.e. and demonstrate. % = {. (8.^ + 1 ) . i. S\.55) The large x asymptotic expansion can also be obtained from the differential equation. ^ = 0 dxA dx with the boundary conditions lim 4>(x) = ± 1 . 5„=4 Sl=o. (8. x—>±oo (8. • • • can be determined.58) Since x2 is the highest power of x appearing in x2Q(x) we set S(x) = ]S0x2 + SlX + S2 In a.36) in the form (coefficient of — x2 here chosen to be ~..e.
(8. 2 ) + V2(l)! lt1! 2 1 — v 3x\ 1 2 2' 2 where iF\(a. Oberhettinger [181].37) with (8.+ X M W = Q. are asymptotic series in a variable x. 4~l~2 ' i i _i.91.62) 4 \ 2 XiFi 2'2 .8. therefore.19). Whittaker and Watson [283]) is now somewhat elaborate (although in principle the same as before). The derivation of the asymptotic series from integral representations (cf. Magnus and F.61) (the dominant terms of (8.36) have correspondingly e±x / 2 ) . i. » 1. function. a z a a ( + 1) z2 which has the unit circle as its circle of convergence. . p. T h e parabolic cylinder function normally written Dv(x) is frequently defined via the Whittaker function WK^{x) which is a solution of Whittaker's equation d2W dx2 Thus* Du(x) = 24 + 1 K 1 + T+ .b.38).38) in a variable for the solutions of approximated differential equations.63) **W.3 Derivation Then of Asymptotic Series from Differential Equations 145 eS(x) = 2 e±^ xS2 1+ 0 (8. \x\ ^> \v\ and (a) for  a r g x  < D„{x) 3/4TT: e 4 x x i/(i/l) 2x2  v(y\){V2){yS) 2Ax2  . n 22e"T JziL (8.e. 2 X 2Wi. etc. T h e expansions (8. We.z) is a confluent hypergeometric T. In particular we shall need the asymptotic expansions of parabolic cylinder functions in various domains. In the following we are mostly concerned with asymptotic series in some parameter. do not enter here into their derivation and simply quote the result: W i t h \x\ .35) and (8. / i N . and it is clear t h a t higher order terms follow accordingly. Nonetheless we shall need asymptotic expansions like (8. (8.37).
whereas e~x '4 increases exponentially.1 a2 an a0 + — + ^ + . Watson [283].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.4x2 (8.+ — + ••• of a function f(x) for a given range of a r g x .4x 2 27F (*/!)!  e^e^x""1 1 + (v + l)(i/ + 2) 2x 2  (i/ + !)(»/ + 2)(i/ + 3)(»/ + 4) 2. T.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. Thus there is no contradiction^ between (a) and (b). 349. . N. 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 3.0. 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) 2.146 CHAPTER 8. 4 (8.4 Formal Definition of Asymptotic Expansions Finally we introduce the formal definition of an asymptotic expansion. Whittaker and G.67) Cf. 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. If Sn(x) is the sum of the first n + 1 terms of the divergent series expansion . E.66) 7 =e(^) 2 i(3x)2ei(^)(?Jx) decreases exponentially. top of p.
thereby (unknowingly) discarding the vast majority of expansions which have been used in applications to physics. first published in 1902. (8.68) is (effectively) simply a statement of the observations made at the beginning.69) and the consequent nonunique definition of the expansion.Sn{x))  = oo n—>oo (8. the divergent nature of asymptotic expansions together with a vagueness of definition (which is avoidable as explained by Dingle) frequently tempt mathematically prejudiced purists to turn away from asymptotic expansions. in particular with regard to possible exponentially small contributions to the expansion for which e.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). It may be noted that Whittaker and Watson's internationally aclaimed text on "Modern Analysis". . The equation to be solved is the equation H<S> = EI/J with H = H0 + eV.g. A critical discussion of the definition can be found in the book by Dingle [70]. (8. (8.e. In fact. i.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.68) The definition(8.8.73) Vn + q/#> + e2T/i2) + • • • . we set E^En * > ^ n = En°1+eEnV = + e2EnV + . lim xneW x—>oo = 0 (8. Chapter I. The definition is due to Poincare (1866). which is assumed to be small. 8. possesses a whole chapter on asymptotic expansions.72) For the eigenvalue E near En and the eigenfunction \l/ near ipn we assume power expansions in terms of the parameter e.5 RayleighSchrodinger Perturbation although Theory 147 lim \xn [f(x) .
e. 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.As a consequence the state vl/) is a vector in this space.77) and obtain {Ho ~ 4 0 ) ) E "i^i = (Ein] ~ W n .4 0 ) M = / ^C(4 1} . .n = 40)V>n. With the help of the orthonormality condition (8.w » .75) HoW + vW = E^+E^+E^. i. We obtain then the following set of equations: Hrj.72) we obtain oo roo dxifc{H0 OO / ~4 0 ) ) E • / °i ^ = / J— OO d x < i E n ] ~ V )^n. ( ^ . i. so that we choose / • dxrpM^O.i4 0) )V4 1} = {EM . (8.77) The states \tpn) of the unperturbed problem span the Hilbert space"K.74) (8. i. TimeIndependent Perturbation Theory We insert these expansions into Eq.e. The second equation can be rewritten as {Ho . °° i J or 0 = EWJdx{rnVil>n). (8.V)i.148 CHAPTER 8. (879) We multiply this equation from the left by ip^ and integrate over x. (8. ^ ) = 0.n. (8.78) into Eq. J f°° dxrn E(^ 0) . Ho^+V^n = EW^+E^n.70) and equate on both sides the coefficients of the same powers of e. (8.e. as are also the states Wn )j Wn )i — We therefore write these vectors as superpositions of the basis vectors provided by the unperturbed problem.76) The first equation is that of the unperturbed problem. We insert the ansatz (8.
8.84) from the left by ip^ and integrating over the entire domain of x we obtain coo J —( oo />oo / oo dxi.m ^ n and integrate: oo /•oo / oo dxrmY. the procedure forbids degeneracy and we can consider here only nondegenerate eigenstates. (8. (8. (8.a\^(E:S0) „• 4 0 ) )^ = / J—oo dxrl?m(E$V)1>n. It is clear that we can proceed similarly in the calculation of the higher order contributions to the expansions of En and \I/n.79). S ' ti^+Q^)• We observe that Eq. We therefore consider Eq. we return to Eq. (8. and hence It follows that to the first order in the parameter e: En = 4 ° ) + e ( ^ . (8. the coefficients a\ .e.^(EW ~ V ) ^ + E& / J—oo dx^miln . (il)i.4 0 ) M 2 ) = ( 4 1 } .82) *> = ^ + E .5 RayleighSchrodinger Perturbation and so Theory 149 E& = {n\V\n) = Wn.76) which we rewrite as (H0 . i. i. with En .Vil>n) (880) In order to obtain ipn . multplying Eq. We multiply the equation by ipm. In order to appreciate the structure of these expansions it is instructive to go one step further and to derive the next order contribution.V)^ Setting + 42tyn.e.i ^ 0. (883) Thus the procedure forbids the equality of any E\ '.84) (885) ^2) = £ a S 2 ) ^ (excluding i = n for reasons discussed above).Viprt e (8. ^ n ) + 0(e 2 ).83) makes sense only if Ef] ^ EnV for all i + n.
~Ej E E. FVn) + e 2 ^ ( V > n .VV . ^ n ) j&iift + ^n .V4>n){lpn.86) E^ iy^n JJn J J i ' i^n For m ^ n w e obtain from Eq.V^n) .79) •/—oo oo / Using (8.V'>pTl (E^Eny V'fc (8. (8. ipn ) = 0 i.89) • • • where we guessed the term of 0(e 3 ).^(8.£i°») 2 Hence +^£o (£«? .2 (V> m . TimeIndependent Perturbation Theory Setting m = n we obtain with (^„.V^j)(^j.(1) (i&0) .B i 0 ) ) ( £ f .4 0) ^+ e 2 E (lpk.Vlpn] M _ P (0K 2 ^r .88) (Q) Inserting Eq. (8.81) we obtain (2) _ _ {lt>m.^j)(V .e.85) together with (8. Inserting the expression (8.Vlpn)(lpn. j. (8.87) (i>k.Vlpn)E.86) into the expression for En we obtain En = 4 ° ) + e ( ^ n . En0) (8. n) £f) (8.^ ) + $& w £ (^fc.Vlpn) E^ . V ^ i ) X (Q) (Vi.88) into xjjT we obtain similarly * r i + f ^ i^n (lpi.150 CHAPTER 8.90) + £ ( 4 0) 4 0) )(4 0) ^ 0) ) ^ +' .^w .
5 RayleighSchrodinger Perturbation Theory 151 The expressions (8.89).Vlpn) = y^Ci(lPk. Merzbacher[194]).91) The coefficients Cj follow from the recurrence relations.n(x) = J2crtn+ii (8...8. 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.n+i = Ckn(2) We can now rewrite ipn ' as / (2) _ V"^ c ^ n knc0 ~ 2j ~ TJfi)0 ) k^n (4 ^) 2 n(0).90) can be found in just about any book on quantum mechanics. These recurrence relations allow one to reexpress expressions like Xm^n as linear contributions of ipi with constant coefficients.90) is the term with only one summation J2 m the contribution of 0(e 2 ).e. (8. in the form i Adopting this procedure for the perturbing potential V(x) we can set V(x)i.lpn+i) = ^ Cj5k. An ugly feature of the expansion (8. We then obtain (lpk.7 is based on this procedure. These functions generally obey one or more recurrence relations which are effectively equivalent to the differential equation in the sense of difference equations. 8. + l^t /„(0) V~^ c kjCjn &(Efi»Ei )(EfE&»)\ 0) „(0)w. The functions ipn(x) are eigenfunctions of some secondorder differential equation. .g. sometimes expressed in terms of projection operators which one can construct from the states \ipm) (see e. i. One would prefer to lump it together with the other contribution.
.e. m = I.93b) The lowest order eigenfunctions belonging to En. In order to deal with the problem of degeneracy we consider the simplest case. degeneracy. Proceeding as before we obtain Ho(cniln + Ci21pm) = En0) (cU1pn + C121pm) = EnVn.e.97a) and (8.We now insert Eqs. I — 1 . Equations (8. (8.93a). (8. however. are allowed to be equal since otherwise the expansion becomes undefined. (8.96) (Ho . For example the magnetic states enumerated by the magnetic quantum number m.95b) H0(c2l^n and + c22ipm) = En°\c21^n + c22^m) (8. (8. (8.94a) into Eq. —I. . Em can now be linear combinations of ipni'ipm. that of double degeneracy En = Em . Equality of eigenvalues. . i.93b).ipn).97b) are . + c22^m).96) are effectively the known equations of ipnitpm and so yield no new information.94b) with (ipn. i.97a) (8.70).4 0 ) ) ^ } = (Em] . .6 Degenerate Perturbation Theory It is clear from the expressions obtained above for En and tyn that no two eigenvalues En '. (8.4 0) )</4 1} = (En1] ~ V)(cu^n (Ho . i. Em = 4 ° ) +eE$ + e2E$ + . (8. in a central force problem all have the same energy unless the system is placed in a magnetic field.94a) * m = c 2 i ^ n + c22tpm + e ^ + e ^ + ••• (8. . is not uncommon..ipm) = 0 = (^m.Hence we write * n = CllV'n + CisV'm + € ^ + d 2 ^ 2 +••• .V)(c2l^n + ci 2 ^ TO ). (H0 + eV)^n and (HQ + eV)Vm = Emym.e.. Equations (8. (8.95a) (8. (8.94a).152 CHAPTER 8. so that En = 4 ° ) + eE™ + e2EP + ••• .93a) (8. TimeIndependent Perturbation Theory 8. all n.97b) with ipn y£ ipm.
^ ) = ( V .98a) Setting I = n. Vl/.(Vn.ipm ) the equations (iPn.98b) + Culpm) + cuV^m). Then (^. (8. H0lP$) = C22[E$ ~ (Vm.ipn n) = 0 = (ipm.Vil>m)] V^m).J#»a&> = 0j since ii4i = En (degeneracy). V^n).m.H0^) . we obtain with (ipn.EJ®(lPn.C 21 (^ m . t f ) = a k 1 } 4 0 ) " Eg>aU = 0. Vlf>m)] .y>m.^ ( ^ . But (8.C22(Vn.8. (^m. and ( ^ m . c2iV^n + c22V7pm).H0xl>$) . (^n.99) can now be rewritten as (frvvA \1"'Vr\ )(cn)=w(cn) (siooa) .ci 2 (V„.99) and ( ^ ^ O ^ ) " ^ ^ . (8.(^04°))41)) = EM^CuTpn (ipl.(lPn. H0lpV) . cn(ipm. Wm.6 Degenerate Perturbation Theory 153 inhomogeneous equations. Vi/>n)] .£ < « = ag)^) .^ ) ( ^ m . ^ o ^ ) ) = 0. Equations (8.V^n).H0il>W) = cn[EW .cnVipn and tyi. . Similarly Wn./>£>) = C2l[E£> . V^n)\ . ^ ) ) = CU[EW . We now multiply these equations from the left by ip* and integrate. £ aW^°Vi) .m).
It follows that ip and <j> are linearly dependent. It is then possible that a special perturbation. In this particular case therefore nondegenerate perturbation theory can be used (cf. (8. However. (V>V</>V>) = const..e.E)tp = 0. 9} ip.5). 4. there is degeneracy in the case of the hydrogen atom with Coulomb potential (Chapter 11) which is based on threedimensional spherical polar coordinates r. Solution: We consider the Schrodinger equation with potential V in the simplified form i>" + (V .e.100a) determine the coefficients Cn. Example 8. i> 4> Hence there is no degeneracy.^Vm) 0.6.Vl/>m) J \ C22 J m \ C22 J V ' These matrix equations have nontrivial solutions if and only if (lf>n. ip = c<p. Then / > i>" 2— = V ip Hence 0 = V ' V . Example 11. If one wants to perform a perturbation calculation in such a case.101) This secular equation determines the first order correction En • The equations (8.Vlpn) CHAPTER 8. d>" E=^. dx i. just as there is no degeneracy in the case of the onedimensional harmonic oscillator (Chapter 6). i. cannot be degenerate because there is only one quantum number corresponding to one definite energy and hence cannot be degenerate. like that providing the Stark effect (proportional to rcosO). i.154 and V (^m. \nip = ln(j> + c o n s t .E™ (Vn.^ ' V = — (•>!>'<t><t>'i>).e.Similarly Eqs. Let ip and < be two eigenfunctions belonging to the same eigenvalue E. . the secular determinant. An example which emphasizes the significance of the spatial dimensionality in this context is treated in Example 8.100b) determine C2i.C22. as we saw.Vrl>n) . splits the problem into two independent onedimensional systems. the degenerate perturbation theory must be used.3).1: Do discrete spectra permit degeneracy? Show that in the case of a onedimensioanl Schrodinger equation with discrete point spectrum there is no degeneracy. At x = ±00 the constant is zero. TimeIndependent Perturbation Theory (l/>m.1. Nonetheless we shall see that the Schrodinger equation of the hydrogenlike problem can also be separated in some other orthogonal coordinates. .ci2. (8. 11. in particular in parabolic coordinates (see Sec.This completes the derivation of the first order perturbation corrections if an unperturbed eigenvalue is doubly degenerate. if)' (/>' — = —. A onedimensional system.
Dingle [72] and H. In the following we describe such a procedure as first applied to the strong coupling case of the cosine potential or Mathieu equation.104) are chosen such that the equation Dcf) + (E0 . The subdivisions (8.103). Dingle and H.102) can be written D<f> + (EQ .8. V = V0 + v. also possible in areas not of immediate relevance here.105) (8. T For instance spheroidal functions can be studied with this method. W. and the solutions then have to be matched in their regions of overlap. B. with some restrictions on the function V.e)<t>. the considerations given below would be valid even if D contained only the derivative of the first order. We assume that V can be expressed as the sum of two terms. Muller [210].103) such that if E is written correspondingly E = E0 + e.7 DingleMuller Perturbation Method We have seen above that although the procedure of calculating higher order contributions of perturbation expansions is in principle straightforward.t Since in general a solution is valid only in a limited region of the variable. (8. and such that one can even obtain a recurrence relation for the coefficients of the expansion. J.102) where D is a second order differential operator and E a parameter. . the method has to be applied in each domain separately. It is therefore natural to inquire about a procedure which permits one to generate successive orders of a perturbation expansion in a systematic way. H. See also R. one quickly arrives at clumsy expressions which do not reveal much about the general structure of a higher order perturbation term. (8. W. B.e. anharmonic oscillator potentials and screened Coulomb potentials.7 DingleMuller Perturbation Method 155 8. of course. (8.V)(p = 0.V0)<f> = 0 (8. in particular in applications to periodic potentials. Eq. Applications of the method are.* In various versions this method is used throughout this book.V0)<t> = (v. Muller [73].106) (8. We consider an equation of the form Dcj) + (E . J. cf. W. It is immaterial here whether D does or does not contain a first order derivative — in fact. a potential. J. i. (8. Muller [216]. Thus the method is a systematic method of matched asymptotic expansions.104) "This method was developed in R.
m + s)4>m+s s (8.e ) ^ = 2 ( m .111) s then (see below) 0(1 = E (m.V0)<f>m+a = 0.109) D(j)m+s + (Em . We write the coefficients "(m.V0)<f>^ . v = {D+EmVo)<t>m. the function 4>(0) = <t>m (8.102) with E ~ Em.108) with coefficients (m. Consider (D + Em= (D + E m V0)(4>W + ^(1)) = (D + Em . where m is an additional integral or nearintegral parameter (depending on boundary conditions).Vo)(j)m+s = (Em . s an integer: (ve)(pm = 'Y^{fn. Then if ]e < \E\ and \v . the equation Dcf>m+S + (Em+S .156 CHAPTER 8. since D<f>m + (Em .+s ' (8. (8.E0\ (the latter in a restricted domain).m + s)" in order to relate them to "steps" from m to m + s.m + .110) If we write the next contribution <f>W to <p(°> in the form m+st (8.e\ < \V0 .) This result is obtained as follows.VQ) ^ s c s4>m+s = 22 s C s(Em ~ Em+s)<fim+s = ( u . Considering (frm+s we have.113) . (8. and so v (8.Em+s)(/)m+s.107) represents a first approximation to the solution (f> of Eq. TimeIndependent Perturbation Theory is exactly solvable and has the eigenvalues Em and eigenfunctions (f>m. 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. m + s) which are determined by the recursion formulae.V0)4>m = 0. m + s)0m+s.
£ ^ T^o (^m ~ Em+s){Em — Em+r) ( m ' m + s ) ( m + g ' m + r ) $m+r (8. . (8.m) + > ^ .Em+S) = 0 (second approximation). (8.TP \(T? .m)4>m in (8. (8.T? S The first expansion determines the solution 4> (apart from an overall normalization constant).e.m) E J^ ( m ~ Em+S) ' • fa.^ (Em .m + s)(m + s. Thus (m. and the second expansion is an equation from which e and hence the eigenvalue E is determined (i.m + r)(rn + r.m + s)(m + s. The factors (Em — Em+S) in the denominators result from the right hand side of Eq.8.m + r) m ^ (8 118) r^O ' ~~ Em+s)\Em ~ Em+r) together with n _ / x y^(m. .114) To insure that 0 = ^(°) + f^ ) H is a solution of Eq. V ^ V ^ {rn. Eq.m + s) 157 (B\IA\ s^O: 1 cs = J m _ ^m+s . We observe that the coeffcients of expansions (8.110) and are therefore related to the inverse of a .118). /01_.Q (m. (8.119) follow a definite pattern and can therefore be constructed in a systematic way.102) to order (1) of the perturbation (v — e). .m) — 0 (first approximation).116) m ^ v (m.115) Repeating this procedure with cf)^1' instead of 4>^ we obtain the next contribution 0(2) =YY together with (m.m + s)(m + s.117) „ . .. .7 DingleMiiller Perturbation Method and so (m.113) must vanish. (8. the coefficient of the term (m.m + s)(m + s.m) — —.119) is effectively the secular equation). (8. We then find altogether ^  l^m  Em+S) + y^y^ SJ.m) 1^1^ IW .n\ ^119) + . (8.„N This procedure can be repeated indefinitely.
§ H. Other applications can be found in the literature.7 1 9 2 + 32g + 2976) 3•215g2 +32Z 2 (252g 2 .9) + 4096J4] + 0 ( l / g 3 ) .120) n4 . for instance.423) + Z(2720 9 3 . W. as will be seen in later chapters. has been treated by R. A further conspicuous difference between this method and the usual RayleighSchrodinger perturbation method is that the first approximation of the expansion.* Example 8.2 therefore illustrates the use of this perturbation method in the derivation of the eigenvalues of the Gauss potential. . This means the solution is not normalized. (8.1)( 2 + 64J3] — . Kaushal [146]. The result is the following expansion.38). Imposing normalization later one obtains the normalization constant as an asymptotic expansion. this is — so to speak — the price paid for obtaining a clear systematics which is essential. S. 2 k2 + g2 = ga(2l + q)^[3(q2 „3 + l) + 4(3q~l)l + 8l2] 32ag r .2).[q(llq2 + 1) + 2(33g 2 . 1 . n = 0 .I2q + 64) + 256Z 3 (41 g .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. for the investigation of the behaviour of the late terms of the perturbation series. One should note that in this perturbation theory every order in the perturbation parameter also contains contributions of higher orders. (8.[ 4 ( 8 5 g 4 + 2q2 . and is typical of a large number of cases. 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. in which q = in + 3.e. (f>m in the above. does not occur in any of the later contributions. . . The eigenvalue expansion obtained is an asymptotic expansion. TimeIndependent Perturbation Theory Green's function. . An additional aspect of the method is the full exploitation of the symmetries of the differential equation.158 CHAPTER 8. The systematics of the coefficients suggests that one can construct the coefficients of an arbitrary order of this perturbation theory. as already mentioned after Eq. The following Example 8. Miiller [211]. This can in fact be done and will be dealt with later.6q + 1)1 + 24(5? . 2 . for instance. ^The potential A r 2 / ( 1 + gr2). i.e. as will be seen later. An explicit application of the method (not including matching) seems suitable at this point. since it will be used extensively in later chapters. i.
1). .V ! 4 I z S *! V2 ipq(z)._ „!2 ^ ( " « 2 r r 2 > . (8. 159 •g'g'a'V U E Here we change the independent variable to z = y/2gar.b. 1 . 1 1(1 + 1) 1 . (.z2 is a normalizable function if a = —n for n = 0 . (8.l)ip(a . multiply the equation by onehalf and set h = —a/g. . . 2 The solution XQ(S) = $ ( a . „2 J„2JL.• 0 ( ° ) = ipq(z) = zl+1ez2/4<S>(a..122) where A is of order 1/g.b. We then set V>o(2)=2i+1ez2/4X0(2) Now the function XQ{Z) satisfies the confluent hypergeometric equation (cf.7 DingleMuller Perturbation Method Solution: We expand the exponential and rewrite the Schrodinger equation in the form rfV (ir 2 . i/> with Dq = d2 .6 and 16. <«»> In the limit g —• oo we can neglect the right hand side and write the solution ijj —* I/IQ. Therefore in the general case we may set k2 + g2 = ga(2l + q) .a + l)if>(a + 1) + (a. The recurrence relation for the functions ip(a) follows from that for confluent hypergeometric functions: z2f(a) = (a.121). 2 . \2 . Then a?ip dz2 fk2+g2 2ga l(l + l) 1 z*)i> ±(T'(^*and S=z2.2) where 1/ . Sees. S) is a confluent hypergeometric function. We now insert this equation into Eq. 11. (8.2 '(l + l) .^z2) = i>(a) leaves unaccounted for on the right hand side of Eq. 3\ 2 V ' 2. .2a2 A.123) The first approximation (note the last expression as convenient notation) ^ . a . (fc2 + <?2) 4ga and b = I \—. The solution i. a)ip(a) + (a. . b.0(z) = zl+1ez2^fa.123) the contribution fl<°> Ah + . . Setting q = in + 3 implies k2 + g2 ga(2l + q).8. Then the equation is Dqi> = Ah+y 2 4 ^ 1 1 » fc i\ .
a + j)i/>(a + m). As a check on the usefulness of the perturbation method employed here. a + r) = 0 for r > m.l . TimeIndependent Perturbation Theory (g .a + 2]i P . In fact. a]i + h fa. a + i) = 0 for i ^ 0. one can write down recursion relations for the coeffcients of powers of hl as will be shown later in a simpler context. a ] i + 0(/i3) [a.Z. a + 111 r [a.a — 211 . where [a.a]1=0. a) = b . . o + r) + S m _ i ( a .b = (g + 3) . a) = 1.2. a . a ] i + L ' ^ J 1 [ a + l . ( 5 m ( a . a + i)(a + r. The coefficients S m ( a .a]2+L ^ J1[a + 2 . 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. So (a.^ A + J7^S2(a'a)' K a +J]i+l = 4 / i + 2)!' S ''+ 2 ( a ' a + j) for i and j not zero simultaneously.e. a ] x + L _ 2 J 1 [ « .120). Hence the next order contribution to tp^ becomes oo i+2 r [a a . If we now evaluate the various coefficients in the last expansion we obtain the result (8. a + r) with the boundary conditions Sn(a. (a. of course. (a. a + r) + S m _ i ( a . _ J [a .a + j}i+1^(a + j).160 where (a.j^0 ' + jU+1 J ^(a + j) with h[a. This equation shows that a term fiip(a + j) on the right hand side of Eq.a]i = . . a + 1) = a = In general CHAPTER 8.2a = I + q. „ . one could now use the RayleighSchrodinger method and rederive the result (8.a]i It is clear that the various contributions can be obtained from a consideration of "allowed steps". fa.3). (8. a + r) satisfy the following recurrence relation: Smi(a. i. The above remainder i j ' 0 ' can now be rewritten as oo i 1 R(0)=J2h + i+2 i+2 ]T (i+2) [a.a—lli. a + r + l)(a + r + l . a + r — l)(a + r — l .S m (a.120). 0 = h[a.i ^(i)=£V+i i=0 J2 j = (i+2). when j = 0.1) = a . Now we make the following important observation that Dqip(a + j) = jip{a + j). the latter determining to that order the eigenvalue. a + r) = i 2 \ m ?n z J 0(a) = J^ j= — m S m ( a .123) and so in R(°> can be taken care of by adding to the previous approximation the contribution fitp{a + j)/j except. fa.
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. light. although not exactly straightforward.e. the generalization to many particles. 9. as we shall see in Chapter 27. Regrettably this has to be left out here in view of lack of space. and we shall then see that these lead to analogous results as in Chapter 2. We therefore begin with the appropriate classical considerations with a view to generalization. except for the canonical algebra which has to be formulated with Dirac brackets replacing Poisson brackets.Chapter 9 The Density Matrix and Polarization Phenomena 9.2 Reconsideration of Electrodynamics In this chapter we consider some aspects of classical electrodynamics with a view to their generalization in the direction of quantum mechanics. i. 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. 161 . and this means electrodynamics. 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. proceeds along similar lines.
(9.1) Fig. i. 9. (9.4) (9.oei(krwt> which satisfies the transversality or Lorentz condition k • A = 0. t) = A 0J R cos(cji) + A 0 / sin(u.t) = 3ftA.1 The vector potential ellipse. When Ao/? is perpendicular to Ao/ and Ao# = Ao/ the ellipse becomes a circle and the field is said to be circularly polarized. This is the equation of an ellipse as indicated in Fig.5) which is described as right polarization as compared with left polarization for which (A0R x A 0 /) • k < 0. A 0 = A0R + iA0I. 9.i). When AQR is parallel to Ao/ the ellipse becomes a straight line and the wave is said to be linearly polarized.3) (9. (9.e.2) (9.162 CHAPTER 9. t) = A(XR cos(k • r — cut) — Ao/ sin(k • r — cot) and at r = 0: A(0. It follows therefore that A(r. The Density Matrix and Polarization Phenomena potential A(r.6) . In general the constant Ao is complex. t) given by A(r. In the latter case one has the possibility of {A0R x A 0 /) • k > 0.1.
/ = 4 0 ) * 4 0 ) + 4 0 ) * 4 0 ) = (A(°\ A<°>) = A(°)2. determined with the help of a polarizer.9) The intensity is characterized by the fact that it can be looked at as the "expectation value" (1) of the unit matrix 12x2.A i = 0 = k . i. however.e. (9.A2 is the sum of the moduli. (AW.8). not. the direction of polarization.A«).AW). the intensity of a wave. Rae [234]. From the vectors A(°).e.2) the vector potential A is completely determined by its two (say (x. In classical electrodynamics.9. We set therefore A 0 = Ai + A 2 .2 *i*4 0) (98) with coefficients F^. See for instance A. (9.A(°)). Observable quantities can only be those which are invariant under translations and rotations. (AW. ie. I=<l2x2>:=E40)*l*40)i. A2. represented by a filter "F".y)) components A i . (9. 17. i. the vector potential A. for instance.k (910) We give the initial wave the label "0" and subject this to an experiment.e. . AW = £ fc=l. we recall. which leads to an outgoing wave which we give the label "1".A« we can construct only the following scalar products with this invariance: (A(°\A(°)).t in which k and ui remain unchanged.g. Then their connection is given by Eq.11) 'This means we consider the polarization. p.2 Reconsideration of Electrodynamics 163 As a consequence of the Lorentz condition (9. with k . (9. the connection must be linear. The intensity 7 of a light wave with amplitude coefficients A\' .A 2 . as e.7) We consider now measurements with polarized light. 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. i. the fields E and B are observables. a calcite crystal or a film of nitrocellulose. (A(°). Here we are interested in observable properties of light waves — like.
(F)=Tr(Fp).13): j2^FikAk = Y^AiF*kAi = Y.15) in order to rewrite Eq.A*FikAk. We use definition (9.k l f c 4 ° \ (9. it must be real and hence (F) = <F>*.k (9. i. i. (9. (9.k i (9.15) This matrix is hermitian since (Pife)t = (AiAtf = (A*Akf = (AkA*f = {pkif = (Pik).14) We now introduce a matrix p — called density matrix — which is defined by the relation pik := AiA%. But we can convince ourselves that we achieve exactly the same as with our earlier consideration of observables.17) .e. We write the observable quantity as (F):=Y.e. if (F) is an observable quantity. (9.A*kF£A* = E and hence F = Fl A p i tkAk. the connection between the initial polarization and the final polarization.A( 1 )) = ^ ^ F i. To this end we observe that as a consequence of Eq. Thus (F) = J2FikPki = ^(Fp)a. i.164 CHAPTER 9. The Density Matrix and Polarization Phenomena Only the second last of these combinations describes the experiment. (9.12) i.e.13) This consideration of translation and rotation invariant quantities is somewhat outside the framework of our earlier arguments.13) the quantity F must be an hermitian matrix and thus representative of an observable. Thus. (9. so that (F) can be looked at as the expectation value of an observable F. (9. pki = AkA*. Then according to Eq.16) i.13). (A(°).
The socalled pure case.e. In the present case of a polarized beam of light we have two possible. b = 0 and o = 0. i. A2 = b as </*) .* For a lightwave travelling in the direction of z. "incoherently. i. i.21a) follows from Eq. is in general unpolarized.15) with A\ = a. But light. 1 The classical polarization vector or wave vector A corresponds in the quantum mechanical case to the wave function ip. we have (9 19) ' In this case there is no preferential direction.18) The vector A describes a particular ray of light. b = 1 we obtain the expressions (9. (9.(AAt) = ( £ £ ) . 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. that of a pure state. <»^> (9. is given by a single polarized wave.d "»=(o ! ) " In the case of an admixture with equal portions of 50%. The density matrix for the pure state (9.e.( J o ) . .e. mutually orthogonal polarization directions x and y described by the matrices (see also below) * .21b) This state is still a pure state.19).22) For a — 1. 165 (9.2 Reconsideration of Electrodynamics In particular we have / = Tr(p). This admixture is described by introducing matrices p. like that of a bulb. (9. it is a statistical admixture of light rays whose polarization vectors are uniformly distributed over all directions.9. For a 45° polarized wave we have 1 .
(9.8) A(1)=E^40)> k R = R(<*).g. Equation (9. The Density Matrix and Polarization Phenomena P45° = I 1 For a 135° polarized wave we have a = and l M . in which the intensity I = Tr(p) is not conserved (e.Z^ 3. Then according 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. (9.166 and CHAPTER 9. b = —=. (9.15) Pik — A i A k . 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. (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".23) 1 . Pi35° = \ \ 2 ? )• (9. as a consequence of absorption).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 = . We now inquire about the the way p changes in the process of the transmission of the lightwave through an apparatus.™ V 0 n kmAm / ^^jPim^mk' . We set in analogy to Eq.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. l l\ " i °\ ) ' ^925^ 9.
31) (9. the observable is transformed.31). (9.t) in Chapter 2 in what we described as the Schrodinger picture there. We can interpret the result either as (F) ( 1 ) = Tr(F{Rp(0)R}}) or as (F){1)=Tr({RtFR}p<0)). (9. (9. in other words without creation or annihilation of systems.4 The Liouville Equation 167 pW = RpMRl (9.30) 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. i. The corresponding condition here is that no absorption of the wave takes place.34) . 9. The Liouville equation describes the motion of an equal number of systems in phase space elements of the same size. in the state functional) in analogy to the time dependence of W(p.32) (9. (9.17) (F) (1 ) = Ti{FpM) = = Tr(tiiFRpW) F' = RtFR. that the initial intensity Jo is equal to the final state intensity I\.29) where (i.e. that Io = Tr( / 0 (°))=Tr(pW) = / 1 . the observable F remaining unchanged. i.9. (9.33) i.pW).30) Ti{FRp{®I$) = Tr(F. In the case of Eq.4 The Liouville Equation It is natural to go one step further and to derive the equation analogous to the Liouville equation.e.e. In the case of Eq.28) (9. on the other hand.27) The measurement of an observable F in the new state with superscript " 1 " then implies according to Eq. the dependence on a is contained in F' = R^(a)FR{a). (9. here this contains the dependence on a. and p(0' remains unchanged.e.q.
36) (9. = Tr(R^Rp^).33)) we obtain correspondingly in the Heisenberg picture the equation ^ = +i[H.40) Equations (9. For an infinitesimal variation of the state of polarization we have R{a) = 1 + Mda. (9.38) H = Hl R(a)p(°\a0)R\a) (liHda)p(0)(a0)(± + iHda) pM(ao)i[H. In Chapter 27 we shall see that the Poisson brackets here are actually Dirac brackets because the gauge fixing condition (9. CHAPTER 9.40) can be considered as quasiequations of motion.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 2.39) and (9.32) in the Schrodinger picture that p^iao + da) = = = or ( w i t h p(°> —> p) (9. The Density Matrix and Polarization Phenomena Tr(p(°)) = Tr(p^) so that = Tr(Rp^R^) RlR = 1. (9.39) If the dependence on a is not contained in p^1' but in F' (cf. (9.35) we have 1 = R*R = (1 + M W ) ( 1 + Mda) ~ 1 + (M* + M)da.p(°\ao)]da.e. j£ = i[H.M . i.21) and (2. (9.37) (9. For p ( 1 ) ( a ) = p ( ° ) ( a 0 + da) follows from Eq.F}.46) suggests to define Poisson brackets of matrices by the correspondence {A.p].168 i. . In view of Eq.B}:=i[A. the matrix M is antiHermitian: Mf = .e. (9. Eq. (9. (9.2) has to be taken into account. We set therefore M^:=iH. (2.B]. Comparison of these equations with Eqs.35) Thus the matrix R has to be unitary.
10.1) This expectation value remains unchanged if \u) is replaced by e%a\u) with a real.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.1 Introductory Remarks In this chapter we are concerned with the formulation of quantum mechanics in general. the Schrodinger. and consider timedependent perturbation theory. Heisenberg and interaction pictures. Thus the expectation value with respect to the vector describing the state in the Hilbert space does not depend on this phase factor. Thus in particular we establish the uncertainty relation for observables in general.Chapter 10 Quantum Theory: The General Formalism 10. 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). (10. we establish their Heisenberg equations of motion.
which clearly does not hold in the case of operators.2.Qj] = 0. we must have (u\u) = 1.2) (F(A)) = {u\u) <"lf.4) *Even in those cases where this is possible the operator correspondence may not be unique.Pi are called fundamental observables.y. 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.170 CHAPTER 10. \pi. z and for definiteness here — which will not be maintained throughout — the "hat" denotes that the quantity is an operator.g.* a conspicuous exception being for instance a spin system. y. Quantum Theory: The General Formalism expectation value of the unit vector or identity operator is 1. Such a correspondence with classical mechanics is not always possible.2. The first step in the investigation of a quantum mechanical system is to specify its dynamical variables A. For these the following rules of canonical quantization are postulated: Postulate: The operators q~i. . (10.B] = ih. where i = 1. we assume this now.q). however. and assume that A = A(p. (10. The phase space coordinates qi. B We first establish the following theorem: If A and B are observables and hence hermitian operators obeying the commutation relation [A. [qi. Here. 10.3 represent the three Cartesian coordinate directions of x. E. Pi • Pi. q).Pj] = ihSlj. 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.1 U n c e r t a i n t y relation for observables A.Pi are postulated to obey the following fundamental commutation relations (again leaving out "hats"): [Qi. classically qiPi = PiQi. Naturally one can also write (10.3) These fundamental commutators determine the commutators of arbitrary observables A = A(p.Pj] = 0.
11) (A2) = (A2) . (10. (10.10) .0)1/2 = AA..12) where \u) 6 "K is a ketvector representing the state of the system.12) the Schwarz inequality (cf. (3..2A(A)}  {A)2)1'2 (10.9a). C = A.7b) = = [A.(A)2 = (AA)2. Then [A.2 States and Observables then their uncertainties AA. 111 (10. (10.7a) (B) = 0.7b).8) and..6) B:=B(B). B}= ih.. (A{A))(B{B)){B{B))(A(A)) (10.2(A)2 .(A) = 0.2A(A) + (A)2. Since A = A— (A).9) ((A2) + (A)2 . (10. Hence (AA)2{AB)2 = (A2)(B2) = (u\A2\u)(u\B2\u).10. Thus AA = AA = ( i 2 ) V 2 .B\u) G IK as vectors) we obtain (for an application see Example 10. (10. so that (A) = (A) . using Eq.B] = ABBA > ^h.13) AB = AB = (B2)1/2. = \\A\u)\\2\\B\u)\\2 (10. (10. AB defined by the mean square deviation AC=((C2)(C)2)1/2.1) (AA)2{AB)2 = > (u\A2\u)(u\B2\u} \{u\AB\u)\2. (10. Eqs. (4.5) are subject to the following inequality called uncertainty relation: AAAB In proving the relation we first set A:=A(A).5)) (for A\u).{A)2)1'2 {{A2) . we have A2 = A2 .(A)2)1'2 (1 °= a) ((A2 + (A)2 .B. Applying to the expression (10. Ai = = = ((A2) .
R= (AB + BA).16) I = h. A\u)=cB\u).4). Zi (10. The equaltosign applies in the case when (a) it applies in the Schwarz inequality.e. We can therefore rewrite Eq. (10. (10. (10. and with Eq. (10. From this we obtain (uiJBu) 2 = i22 + / 2 . since A and B are hermitian. .14) Hence (u\AB\u) = /^(AB where. (10. Zi Zi (10. Quantum Theory: The General Formalism Next we separate the product AB into its hermitian and antihermitian components^ and use Eq. and (b) when (AB + BA) = 0.172 CHAPTER 10.17) as had to be shown.18) >R2 + I2. (AB + BA) This means (AB + BA) is real. Zi + BA)\ + ^ih. AAAB>I=h.15) as (u\AB\u) = R + iI. for a complex constant c. and hence for condition (b) 0 = = tSee Sec.15) = = (u\AB + BA\u) = (u\(AB + BA)^\u)* (u\(AB + BA)\u)*.19) (u\AB + BA\u) = (u\AB\u) + (u\BA\u) c*(u\BB\u) + c(u\BB\u) = (c* + c)(u\B2\u). (10. (10. The condition (a) is satisfied when.13) (AA)2{AB)2 i. Then AB = \(AB + BA) + \(ABZ Z BA) (1 = 8) \{AB + BA) + \ih.3. 4.
Separating the product into symmetric and antisymmetric parts. (Lz)2^0. q) = ih. Example 10.20) or.3 OneDimensional Systems We now consider onedimensional systems with a classical analogue as described earlier. such that Lz is "sharp".) 2 > = x. i. (ALZ)2(ALX)2 > \h2 = (Ly)2. (10.7a).z. {u\LiLj\u) = R + iQ. (Ly)2=0. h2(Lz)2. implying {Lx)=0. + LjLi\u). (Ly)=0. ALZ = 0 (these are states \lm) with Lz\lm) it follows that (Lx)2 = 0. i.e. (Lx±iLy) ^ 0.1: Angular momentum and uncertainties Show that in the case of angular momentum operators Li. m\lm)). What does the relation imply for states u) = \lm). The observables like angular momentum are functions of p and q. in view of Eq. (A{A))\u)=i(Zc)(B(B))\u).y. i. i. we have LiLj = (LiLj and ^ = y(u\LiLj so that \(u\LiLj\u)\2 Hence {ALx)2{ALy)2 > l + LjLi) + [Li. Thus in this case Lx. 9ftc = 0. For states \u). (10.10.e. ALX 10. the following uncertainty h2{Lk)2.i relation holds: (ALi) 2 (AL.e. = 5R2 + G 2 > 9 2 = hi2(Lk)2.13) to operators Li — Li.e. = 0. implying that the vector it) has to satisfy the following equation: A\u) = i($Sc)B\u) (10. we obtain {ALi)2{ALi)2 > \{u\LiLj\u)\2. for which one component of L is "sharp"? Solution: Applying Eq.3 OneDimensional Systems 173 Thus c = — c*. 9 = h{u\eiikLk\u). ALy = (u\L±\u) ^ 0.Lj] = (LiLj + LjLi) + ih£ijkLk. For these we have in the onedimensional case the relation [p. Ly are not "sharp". (ALy)2{ALz)2 > ~h2(Lx)2. .
p] = ih we obtain therefore [q.21b) now follow with the assumption that A(q.174 CHAPTER 10.A(q. Quantum Theory: The General Formalism Since this says that p and q do not commute (as operators). We demonstrate that (a) its eigenvectors have infinite norm (in the sense of a delta function). as remarked earlier.p)] = . i.BC] = [A.e. A — A(p.A(q. Q2] = [p.p) can be expanded as a power series in q and p. Similarly we obtain \p.21a) Equations (10.p)]=ih^^and \p. for q\x) = x\x) we have p\X)=\x)(th^y Now let A be an observable. (b) its spectrum is necessarily continuous. We obtain these equations as follows. (10.21a) and (10.21b) (10. This is why the operator p requires a representation in the space of eigenvectors of q. (3. Eq.pn]=ih^(pn). Using [q. Next we have a closer look at the operator q. they do not have a common system of eigenvectors. i.e. We have (cf. . (10.B]C + B[A.p2] = [q.23) = ihg(q2) (10.q).p] = ^ihp ~ Proceeding in this way we obtain [q.p}p + p[q.22) d ih—(p2).C\. and (c) the spectrum is not degenerate.qn] = ih^(qn).34d)) [A. q]q + # > v] = ~2ihq and \p. Then (see below) we can derive for A the following commutator equations [q.i h ^ ^ .
3 OneDimensional Systems 6K(x) 175 Fig.1 how the delta function arises in the limit K —> oo. we obtain [q.27) (10.21a) A by U. Replacing in Eq. 10.1 The delta function for K — oo.f^ ^ J—oo dke ikx —ikx' (10.25) It is shown in Fig. U] = ih^= aU. We thus see explicitly t h a t the vectors \x) have infinite norm (in the sense of delta function normalization).24) lim SK(x — x') — S(x — x'). For the cases (b) and (c) we consider the operator U(a — pipa/h (10.e.26) it Here p is an observable. (10. We have (x\x') = = where f^ J—oo dk(x\k)(k\x'} (4 = 9 ) ±. = eipa/h = Uia)'1 = U{a). 10. . and a a cnumber. 6K(xx') = — fK dkeik^x'^ 2TT JK sin K(X K{X — x') — x') (10.e. i.28) i.10. > We begin with (a). (10. U is unitary. Since p = p\ follows t h a t U\a) so t h a t U(a)U\a) = 1. an operator.
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. cf. (10.29) This means: U\x) is eigenvector of q with eigenvalue x + a. 10. Quantum Theory: The General Formalism qU = Uq + aU = U(q + a).176 i. ^Observe that {iji\A{p.1 T h e translation operator U(a) We saw above.q)U^U\ilj). or qU\x) = U{q + a)\x) = (x + a)U\x). we obtain qU\x) = U(q + a)\x) = {x + a)U\x). In an analogous way by defining the unitary operator U(b) = eiqblh (q operator. which means that the spectrum of this operator is a continuum.31) .3.e.e.30) (10. i. CHAPTER 10. Eq. we can see that also p possesses only a continuous spectrum (from —00 to 00). Evidently every eigenvalue has only one eigenvector. that with U(a) = e~ipa/h = U.00).29). Since this holds for any arbitrary value of a in (—00. the spectrum is not degenerate. (10. this means: With a unitary transformation (which leaves the matrix elements unaffected * and hence the physically observable quantities) one can pass to any arbitrary eigenvalue in the domain (—00. or qU(a)\x') = (x' + a)U(a)\x'). (10. Thus all these values belong to the spectrum of the operator q. b a cnumber). The eigenvectors have the same norm as \x): (x\U*U\x') = (x\x') = S(xx'). q any other observable F(p. q) can be constructed representing a dynamical variable.q)\i>) = (i>\U^UA(p. WU = 1.00).
a).hm or „'i„i™" s> i ™' ~/ (x'\p\x")\ = 8'{x' x"). \c{a.35) (10.x')\ = l.e.32) we see that \x' + a)ocU(a)\x').x") U{x'\p\x" % pz. i.x" .39) .34) {x"\U\a)U{a)\x') c*(a. Comparing Eqs. (10. In the expression of Eq. since q\x') — x'\x').37) With the help of the latter expression we can calculate. for instance.38) h 6(x' .x') = = i.e) (x \p\x ) = .e.36) i. (x'\p\x"). (10.e.33) (10. (10.3 OneDimensional Systems 177 Moreover. \x') = U(x')\0).x") . we have q\x' + a) = (x' + a)\x' + a). (10.10. (10.x" . i.x')5(x" = (x" + a\c*c\x' + a) x')./ (10.x" . U(a)\x') = c\x' + a). It also follows that (x'\U(a)\x") = (x'\x" + a) = S(x' . (10.5(x> .x")c(a.e. the operator U(a) acts to shift the value x' by the amount a.37) we set (with e infinitesimal) U(e) ~ 1 Then (x'\U(e)\x") = 6(x' . We choose the phase such that U(a)\0) = \a). It follows that U(a)\x') = = U(a)U(x')\0) = U(a + x')\0) \x' + a). it follows that 5{x" . (10. The operator is therefore called translation operator.31) and (10. i. offdiagonal elelemts of the {x}representation of the momentum operator.32) Since U is unitary.e) ~ 5{x' .e.
Quantum Theory: The General Formalism We mention in passing that.178 CHAPTER 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. it follows that 5'(x) is an odd function of x. .4i) We assume first of all.° ° d(p) (ip) e"« = S'(x).to)m0)). 5.e. Let \UE) be an eigenvector of the Hamilton operator H with eigenvalue E.4. See Eqs.43) (10.42) ^Observe that 6'(x) = £ f ^ dpipe** = ± / .40) 10.x") = (x'\p\x")*.45) (10.4 Equations of Motion Let \tp(to)} be the state vector at time to of a completely isolated system (i.e. (10.to) is called the time development operator. 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). (10. (io.§ It follows that (x"\p\x') = d'(x' This expresses that p is hermitian. so that U{t.f dpipeipx. (5.to) exists which is such that^ m)) = u(t. We introduce another postulate: Postulate: The time dependence of the vector \uE(t)) is given by the relation \uE{t)) = eiH^t0^h\uB{tQ)). In this case the classical Hamilton function does not depend explicitly on t. i. a conservative system. The operator U(t. 6'{x) = !. since 5(x) = . .35). This is in conformity with our earlier postulate in Sec.tQ) = eiH(tt<Mh.44) (10. H\uE(to))=E\uE(to)).L fdpeipx.t0) = HU(t.27) and (5. to) with respect to t we obtain ihjtU(t.t0). i.e. Differentiating U(t.
*l)^(*l)> = I7(*2. Hence U(t.48)). But obviously (cf.t0)]1 Let us set for small values of e: U(t + e.t0)U(t0. In the immediate considerations it is not necessary to restrict ourselves to systems with classical analogues. (10.to) = 1.t)U(t.t0) [U(t + e.t) = 1. Then with U(t + e.t0) or ih—U{t. (10.z / dt'H(t')U(t'. (10. t0) = H(t)U{t.47) e.48): [/(Mo) = 1 . (10.45) has very general validity. (10.44)) we have U(t. The operator H is d efined by this relation.t0) = U(t + we have U(t t) = lim ^ + Mo)^(Mo) = n (1046) = U(t.to) can be expressed as an integral as the solution of Eq.t)l]U(t.t 0 )^(*0)> U(t2. = U(t0.t0)\^(t0)).to).49) .e.t) = l^eH{t).4 Equations of Motion 179 We now demonstrate that Eq. i.10. t0) = E/(t 2 .to) (differentiating we obtain immediately Eq. t0) (10.tiM*i. (10. Ufa.48) with U(to.t) and therefore [U(t.t) = l. (10.41) we obtain ^(t 2 )> = = J7(t2.t). From Eq. Eq.*l)^(*l. Evidently U(t.
to) can therefore be interpreted as a product of a sequence of infinitesimal unitary transformations.51) Since </. Since the only measurable quantities are moduli of scalar products. (io. Then \Mt)) = u(t. the Schrodinger equation ju(t. t)\t/>(t)) =U~ and since H is hermitian. (5. it follows that U(t + dt.5o) ihjtm)) = Hm)). The operator U(t. 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.e. (10.30). and these remain unchanged under unitary transformations. Now let \tp(t)) be the state of the sytem at time t. Eq.53) .41). the instant at which a measurement is performed on the system. whereas the physical quantities are represented by observables in "K which do not depend explicitly on t. In the following the subscripts S and H indicate Schrodinger and Heisenberg picture quantities. >(*)> = i. i. the predictions of different descriptions are identical.t0) m0)). 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.180 CHAPTER 10. 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. Another such description is the Heisenberg picture.e. (10.t) = l % ^Hdt) \il>(t)) Hdt (10.52) is the operator of an infinitesimal unitary transformation (recall that U = 1 + ieF satisfies the unitarity condition UU^ = 1 provided F — F^).t0)\ii>s{to)) (10.(£ + dt)) = U(t + dt. requires the identification H(t) = H= Hamilton operator. The eigenvectors of these observables are also constant in time. Comparison with the equation postulated earlier.
in order to obtain the Heisenberg picture (description).54) Observables transform correspondingly with a similarity transformation. i. t0).56) U^[As. we have to take into account the contribution dAs/dt. In the Heisenberg picture it is HH = U]HU.With explicit time dependence.58) HHAH (U*ASU)(U^HU) [AH.48).HH]. (10. (10. as a rule.H]U = = = U]ASHU AHHH — U^HASU .55) and using Eq.i). It does not always have to be the case that As does not contain an explicit time dependence.59) This equation is called Heisenberg equation of motion.55) Thus AH is explicitly timedependent.t0)\ips(t)) = \ips(to)) = constant in time. = so that ihjtAH or ihjtAH = = [AH. a scalar product) remains unchanged: AH(t) = U\t. we consider only cases without explicit time dependence of As. .e. even if As is not timedependent. (10. (10. for instance in cases of nonequilibrium or if part of a system disappears through absorption. i.4 Equations of Motion and \^H) = U\t. subject to a continuous change. Here H is the Hamilton operator in the Schrodinger picture. Differentiating Eq.e. but here. As(q.10. if ^ = 0. 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. {AH. (10. we obtain ih^AH at = = U^HAsU + ihU^^U at + U^AsHU (10.57) and rf[As. so that a matrix element (i.p) in the Schrodinger picture dAs/dt = 0.e. For As(q. Thus.p. However the associated observables are timedependent. t0)AsU(t.HH].H]U + iW]^U.HH]+ihU^U. 181 (10.(tf HU)(rfASU) (10.
In our treatment of representations. A. as in the transition from the configuration or position space representation to the momentum space representation with the help of Fourier transforms. In many cases the Schrodinger picture is more amenable for explicit calculations.13) to derive the uncertainty relation for energy and time. since in the transition from one to the other the commutator remains unchanged. in the case of the fundamental dynamical variables qi. We end with a word of caution. In the Heisenberg picture essential properties of a system can frequently be recognized more easily in relation to their counterparts in classical mechanics — in both cases the time development of a system is given by the time dependence of the dynamical variables. In the treatment of the pictures we used unitary transformations of operators (and vectors). [CH. Finally we add a comment concerning conserved quantities.HH}=0. provided the Poisson brackets in the latter are replaced by commutators. We see that formally one obtains Hamilton's equations of classical mechanics.H] = 0. In particular.2: Energy and time uncertainty relation" Use the Heisenberg equation of motion and the general relation (10. This holds also in the case of the Schrodinger picture.Pi): and where the expressions on the right follow from Eqs. 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.e. . An observable CH is called a conserved quantity.61) Thus conserved quantities commute with the Hamilton operator. We also observe that the momentum pi is conserved when \pi. if ffCH = 0.Pi we have in the Heisenberg picture (replacing in the above AH by qi. (10.21b). Orfanopoulos [224].21a) and (10.182 CHAPTER 10. i. we used in essence unitary transformations of matrices. Quantum Theory: The General Formalism Wave mechanics is now seen to be the formulation of quantum mechanics in the Schrodinger picture. (10. Example 10. For an extensive discussion see B. Solution: We have for an observable A (A) = (i>H\AH\TpH).
i. .66) This means.p) we then have 0=^M) = U[A.68) To put it The eigenvalues a of A (i. The probability density P ( r ) =  ^ ( r )  2 = (^r>(i#> is independent of t. (A). and TA O C (10.10. This means that for a physical variable A(q. the "centre of mass" of the statistical distribution of A.H}\f)\2. (10. we can replace ([AH. this is practically the same as at time t for times t with \t — t'\ < T.e. i. We construct the following vectors using Eq. H] = 0.15): (AAf(AH)2 Using Eq.)\2 ( = \{i. a position or momentum observation is independent of t. i.64) = 0.65) and rA AA d{A)/dt' . H\tl>) = {H(H))\f).\HAM\2) \(nA. in a different form: The eigenvalues of an observable are called "good" if [A. (10. (10.67) 1 d(A)/dt is oo and AE = 0. if we set AH = AE we obtain TAAE (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 ) . of A\<t>) = a<£)) are then called "good quantum numbers". Let \ip) = \"4>H) represent the state of the system. since \IPH) is timeindependent.HH]) Hence we obtain by ih(A). dt in. 2 d{A)/dt " and. at ^±AH>h.62) The Hamiltonian H does not depend on time.e.Such a characteristic time interval can be defined for any dynamical variable.63) so that with Eqs.13) and (10. then if a measurement is made at time t'.e. is displaced by A A in the interval A T = TA.e.62) and dAs/dt > \WAH\i. (10.4 Equations of Motion Then.H]) = 0.7a). (10. AW = {A(A»V>. 183 dV ' \ dt >"•"»» + ( £ ) • (10. > (10. If r is the smallest such characteristic interval of time. (10.2 > h.
70) we obtain: oo / oo TPEO (x. t) = ^(x) exp(—iEt/h) describes states called bound states of unlimited lifetime.. On the basis of the uncertainty relation AEAt > h these must have an uncertainty AE in their energy spectrum.72) = V'EoW 7/2h + 0i(EE )/h  oo . i. Energy and decay length of the radiated aparticles are characteristic properties of naturally radioactive nuclei. the wave function ^ ( x . states with sharp energy (AE = 0).71) From Eq. This means that in the case of decaying particles. as is observed. (10.69) and (10.H ) ( £ . Quantum Theory: The General Formalism 10. the wave function ipEo(x. (10.5 States of Finite Lifetime When E is real and P(x) is independent of t and J dxP(x) = 1.£ ^ + n/2)^ o ( x ) e{^1/2+i(EE )}t/h 0 o =0 .69): oo poo p / oo *PEo(x. then fE{E') = S(EE') and integration with respect to E' yields again t()E(x.(x)fEo(E')e«EE'Wh 2irh f dE'^E. First we obtain from Eq. for instance. (x)/ E o (E')5(E . have around EQ the form ipEofct) = i.Eo(x)eiEot/he'^2h9(t). Recall for instance radioactive decay.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. (10.e. This means. t)e iEt h ' dt = V>£0 (x) / JO <fte{7/2+i(i5£b)}t/ft (10.184 CHAPTER 10. The wave function of such a state with energy EQ and lifetime r = h/'y > 0 could.)exp(iEt/h)). But the number of radioactive (thus excited) nuclei decreases exponentially. States with finite lifetime (At = r ^ oo) are those whose probability density falls off after a certain length of time called "lifetime".E1) 2irhfEo{E)4>E(x). the particle density diminishes in the course of time (t > 0).tyEt/hdt = = = / J—oo dt J dEJrl>E. (10.69) (Observe that if ip consists of only one state with energy E in the domain of integration of E'.70) we can determine the function fEo{E) integration with respect to t. /oo (10.70) by From Eqs.t) — ipE(x. i ) = J dE'^El^)eiEltlhfEo{E').
toM{to)).e. 7 > 0. (10. which means that a nucleous with discrete energy EQ does not possesses some other admissable level close to EQ. A state is discrete if its immediate neighbourhood (in energy) does not contain some other state. (C/(0) unitary). ihU(t.g. the formula says that JEQ{E) possesses a simple pole at E = Eo — 27/2. (10.e. of course. where U is the solution of Eq. to) be an approximate but unitary solution of this equation. every unitary transformation of states and operators defines a possible "picture"). This description contains effectively parts of the other two pictures and is particularly useful when Eq. (10. whose real part EQ specifies the energy of the state and and whose imaginary part 7/2 specifies the lifetime r = h/j.10.75) (10.105) and (20. ^(*o))> then \xp(t)) for t > t0 follows from Eq.6 The Interaction Picture One more motion picture of quantum mechanical systems is in use. i. Eqs.6 The Interaction Picture Prom the last two equations we deduce for E close to ipE0(x).11). i.41). i.73) precisely by assuming with fEo{E)^8{EEQ) a "smearing" of states around the energy EQ (with uncertainty AE). We see therefore that the states with lifetime r < 00 and TAE > h belong to the continuum of the spectrum. and is called "interaction picture" or "Dime picture". We saw previously that if the state of a system is known at time to.76) **For specific applications see e. (14. Let U^°\t. VE(X) ^ ^ (£ > = dbfift+T/* (la73) This result is known as the BreitWigner formula. They are called "resonance states". . 10.e. (10.e. The state with lifetime r — h/'y is not a discrete state. £/ = £/W(Mo)^'(Mo). that** EQ. We obtained the result (10. Considered as a function of E.to) = HU(t. i.45) can be solved only approximately (in actual fact.45).t0).e. 185 i. (10. W)) = U(t.74) Thus the solution of this equation is the main problem.
81) But [/(°) is unitary.78) (10. H(°Ht) = ih^uW.83) ih^U^ ishermitian. i. with Eq.t0) 0 = l. Now let H^(t).e.82) (10. Quantum Theory: The General Formalism ih^(uWu') i.77) C/(o)t itiuWfLu^HUMu'ih^U' at or i.e.e. (10.e.e. i. i. = HU<®U'. In addition we have (10.78) at with the initial condition c/(0)t f ^ ( 0 ) _ ^ ^ 1 V \ at J U\t0.186 Then CHAPTER 10. (10.e. varies only slowly with t). [/' has only a weak ^dependence (i.h^U> = at ( io. with Hamiltonian H = H^ + H'. be defined by the relation H(°)rj(0) _ i h ^ l = 0 (exact)> n (0) = eiH(o)t/aj (1080) i. dt so that #(0)t ( t ) It follows that H{0\t) = dt = H<®{t).e. (10. For a "good" approximation U « [ / ' ' we have HU^ihjtU^^0. .79) ^'~°i. £/(°)£/(°)t = 1.
U<®\Mt)) = \Mt)). H = H{®{t) + H'.80).86) at = U^HU^U'. 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^. We now multiply this equation from the right by U' and insert the first expression on the right (of Eq. U^H'U^U' (10. (10.78) at (10 85) (10. lfe(*)}. (10.91) Since the Schrodinger equation is given by ihjt\^s{t)) = H\1>s{t)) = ( # ( 0 ) + H'Ms(t)).85) multiplied by U') on the right of Eq.88) and as interaction picture version of an observable A the quantity Ar(t) = tf(°>t AgU^. and obtain from Eq. (10. (10.e.U^HU^U' H'fi'.10. i.89) where ^s(i)) and As are state and observable in the Schrodinger picture. (10. We also set H^U^H'UW. (10. (10.87) = We define correspondingly as interaction picture state iM*)> = t/ (0).92) . (10. Then ihjt\Mt)) = ih~u^\Mt)) = u(0)ihi^W and H'^it)) = U^H'U^U^\^s(t)) + ih(^) \Mt)) (10.85) = where we used Eq.6 The Interaction Picture 187 We use now the subdivision of H into unperturbed part and interaction. (10. (10.90) = U^H'\^s(t)).78).84) U^ where H' is also hermitian.
near zero and \tpi(t)) is almost constant in time. (10. (10.e.89)^0.e.55)) at + imm<Ms_u(o) at dt _umH(. i.93) Thus the vector ipi(t)) satisfies an equation like the Schrodinger equation.188 CHAPTER 10. we differentiate with respect to time the operator defined by Eq.94) Then we have (normally with dAs/dt dt (io=8o) — 0.H\0)}. (10. ^ (0) l^(*)>. (with ihjtAI(t) = [AI.60a) or (10. i.o)AU(o) at +uWAsHWuW UWHWU^UWASUW + iww^uw dt +U^ASU^U^H^U^ (10. Quantum Theory: The General Formalism we obtain by starting from the right hand side of Eq.94) _Hf)Al 1 + lhdAl+AlHf).e. a perturbation so that H ~ H(°)).e. ot i. A^t) = U^AsUi0). H'j = UWH'UW. i. (10.92) ^°)^V/(t)) + ^ ( ^ )  V ' / W > = ^ (0) ^ (0) l^(*)> + ^ .80) (and multiplying by t/(°)t) ih~\Mt)) = u<MH'uM\Mt)).e. Since H'j is assumed to be small (i. (10. we have i. comments after Eq. the vector varies with time (different from the Heisenberg picture).90) and replacing the left hand side of this equation by the right hand side of Eq. (10.89). so that with Eq. dAs/dt=0) .60b) with iff0) instead of H.e. cf. (10. (10. In order to obtain the equation of motion of an interaction picture observable Ai. (note the difference between H' and H'j) ihjt\Mt)) = H'AMt)).95) Thus in the interaction picture the physical quantities A are represented by timedependent observables which satisfy a type of Heisenberg equation (10.
Since .94) Hi = uW(t. (10.102) We assume that before the perturbation H' is switched on at time to.87) and (10.e. where according to (10. i. (10. the system is in the Schrodinger eigenstate \ips ) of iJ 0 with energy Em. We have the Hamiltonian H = H0 + H'. (10. (10.t0)H'uW(t. ^ \<Pn){Vn\ = 1n (10.98).97) The equation to solve is Eq.101) Iteration of this inhomogeneous integral equation yields the socalled Neumann series. i. We obtain the operator U^ from Eq.7 TimeDependent Perturbation Theory We consider timedependent perturbation theory as an application of the interaction picture. ih±\Mt))=Hi{t)\Mt)). (10. i.93). \Mt)) = +' \Mto))^fdt'Hi(t')\Mto)) ft J dt' f dt"Hi{t')Hi{t")\^I{tQ)) + (10.10.80) as E/(°)(t.e.t0).7 TimeDependent Perturbation Theory 189 10. H' = H'(t). Integration with respect to t yields \Mt)) = \Mto)) ~ \ \ dt'Hi{t')\^j{t')).i0) = e<Ho(tto)/ft. {ipn\<Pm) = Snm. H0\ipn) = En\(fn).100) (10.99) (1098) Instead of the original Schrodinger equation we now solve Eq. (10. We assume again that the spectrum and the eigenvectors of Ho are known.e.96) in which the perturbation part H'(t) depends explicitly on time and HQ replaces H^0' in the previous discussion.
(10. The corresponding amplitude or appropriate matrix element is n ( 4 ° V s ) = 5> m (i) n (v4 0) V4 0) >™ = m fl "W ( 10 . (10. we can write l 4 0 ) ) m = elHot/hWm) =e . assuming that this supplies a sufficiently good approximation: IMQ) = \fm) lz\ n dt'HW)]^).190 CHAPTER 10. the time dependence of \ips } can be separated as for stationary states. (10. (10.88) and obtain > n ^ V s ) = {<Pn\eiHot/h\ll>s) = {<Pn\eiHot/hUM\Mt)) = <¥>*(*)>• (10. i. This probability is the modulus squared of the projection of the Schrodinger state \ips) o n t o \ips )„.103) The actual state \ipg) of the system is solution of ih\^s) = H\^s). 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'. (10.103) for m — n and use Eq. Quantum Theory: The General Formalism HQ is independent of time. We express this state \tps) as a superposition of the states \ips ) n of #o with coefficients which as a result of the timedependent perturbation depend on time.^ V ™ ) . (10.105) In the interaction picture the iJostate m at time t is \Mt)) = eiHot/hWP)m = \<pm). (10.106) We insert this expression into Eq.104 ) Here we insert Eq.102) and truncate the series after the first perturbation contribution.e. Hence we set ^> = £°«(*)l4 0 ) >n.107) Jto With this we obtain from Eq.105) rt n Jto h ' Jt00 t rt  x  l h ' Jto t0 (10.108) .
110) With this we obtain from Eq. < KO?t . (10. (10.Em)t] _ sin2 at aH {±{EnEm)}H Consider the function St{a) :1 sin2 at ir a2t t 7T for a —• 0. » — < (10.Em)t' (<Pn\H'(t')\pm) = Wnm(t).Em)t} \{tfn\H'\ipv h 2h(En .111) Different from Eq.109) 10. A further example is provided by adecay.109) the transition probability Wmn(t) = jp Jo . (10. (10. where the momenta of the aparticles form a continuum. but that otherwise the perturbation is timeindependent. This is the case.Jr(En—Em)t _ ^ (ipn\H'\(pm) {En — E„ sm{±(En .25) we here have the positive quantity (note t in the denominator) s i n 2 { ^ ( £ n . for instance. Thus we set H'(t) = H'0(t). We assume that the perturbation is switched on at some time to.En (10. One reason why we begin with these simplifying assumptions is also to obtain reasonably simple expressions. if in the scattering of a particle off some target its momentum p changes to p'. t o = 0.8 Transitions into the Continuum In many practical applications one is interested in transitions into a continuum. or if by emission of a photon with continuously variable momentum an atom passes from one state into another.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^.112) for a^0.
.ii5) This expression has a well defined value unequal to zero only if the energies En belong to the continuum.Em)t} r i 7 TTT. We assume t h a t the matrix elements for transitions into this infinitesimal element can be taken . 10.— = o (En 2h En 2h5{En .io.192 CHAPTER 10. we would have En — Em ^ 0. though sufficiently small. if the energies belonged to the discrete spectrum. Fig.2. In the case of transitions into the continuum we have to consider the transition probability into the interval dEn at En.2 The delta function for £ — oo. t—>oo Hence also (shifting t to the other side) sin2{^(ffn.Em). > 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. (10. so t h a t Wnm would be zero. 10.113) {±JEnEm)¥ or n m irtSt En — ^ 2h 1 s i n 2 { ^ ( £ n . Quantum Theory: The General Formalism T h e behaviour of this function or distribution is illustrated in Fig.114) It follows t h a t for large but finite times t Wmn(t) 2vr ~ — t5(En Em)\(vn\H'\tpm)\2. and no other state would be nearby. Otherwise.E7m)t} = (10. Hence lim St(a) = 5(a).
113) this is* hm . F Jo . if p(En) varies only weakly with En. for instance from a potential. Let p(En)dEn be the number of states in the interval dEn.117) provides the outgoing particle current (see Example 10. sin 2 (tx) ax i—( * According to F.116) The transition rate V is the transition probability per unit time. If we are concerned with pure scattering. Fermi gave this formula the name "golden rule" . Fermi [92].113).115) and (10. '"Recall also the integral J ^ ° dec sin 2 x/x2 = TT. p. (10. i.t The result makes sense as long as the uncertainty AE in the energy satisfies for finite but large times t the relation . 266. Solution: For test functions <p{x) € S(R) the delta distribution is defined by the functional I &(x)4>(x)dx = <j>{0). (10. t We choose a test function which falls off at infinity faster than any inverse power of x: <Kx) = e . so that initial and final states belong to the continuum.4).10. 75. We evaluate the integral 0(0) = 1.W . Example 10. the transition probability into this set of states is J2Wmn(t) = fdEnP(En)Wmn(t) n •* = p(Em)2lTl{iPnlfliPm)l2t. then Eq. (10. Schwabl [246]. 142.118) where 8e is the separation of states around En which in the case of the continuum goes to zero. Then.r Z x 1 ^ ) = 4>{Q). pp.* In view of the usefulness of Eq.3: Application of test functions to formula (10. (10.113) Use the method of test functions to verify in the sense of distribution theory the result (10. in the article of W.117) in many applications.8 Transitions into the Continuum 193 as equal.e.117) were derived by Pauli in 1928./ t>oc TZ J_00 dx . _ 2irh AE > > Se. In the case of the functional (10. . f See E. Pauli [227]. r = J E ^ W =p(£m)y(</>n#W2n (10117) The formulas (10.
.947. with n .„I.194 CHAPTER 10. with v the velocity of the particles.( ^ V V • < ih 2 _ i k n . . Then p(Ek)dEk Since Ek — h2k2/2[i. The ingoing particle flux is therefore v / L 3 . x .. Ryzhik. ^o n . M. 2 T / ^ M.e " 1 * 1 = 2 lim — . . p. . 77^ + . 491. and is obtained as ih. L3/2 ¥>m(x) : L 3 /2 The periodic boundary conditions ipn(xi) = fn(xi ± L/2) imply kiL/2 = ±nj7r. * ^ = . . Gradshteyn and I. .<p). = ( — ) dk= 2ir (—  2irJ k2dkdfl. p = 1 we obtain I = .117) for H' = V(~x) the differential cross section da _ / n V dH ~~ \2Trh) f dxV(x)e where K = k m — k n . K = k 0 .^ V v 4 ) = ^ T 3 e * k ° *^ko)e*° * = — v . V +(ab)2 a _x 2p6 7 m —^ '. 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. Hence the result is 0(0) as expected.? 2TT / L \3k2dkdUn 1 /V(x). Solution: We set ¥>n(x) : so that {<Pn\H'\<pm) = ^3 / d x e i K ' x V ( x ) . 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). Then ^ .( 1 / 4 ) ln(l + 4t 2 ) + t tan _ 1 (2<).a 2 ' i that for a = b = t . E x a m p l e 10.. dEk = h2kdk//j.b2 4 p 2 + (a + b)2 2 b _j 2pa V 1 fc Urn . and hence .4: Differential cross section from the "golden rule" Obtain from the "golden rule" (10. Jin § m = See I..tan p 2 + a 2 . S. . .k. . . 2 . E = h2k2/2fi.ln(l + 4t 2 ) + — t a n _ 1 ( 2 t ) = — tan 4t7T tTV T J< (oo) = 1. ./ cfc a n p 2 +fc 2 . Quantum Theory: The General Formalism with the help of Tables of Integrals. = 0 . § We have r Jo \p\x dx sin ax sin bx  V. iknx . 1 . L \3k2dkdUn It follows that T. . 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. formula 3.
Thus we set ^(x. Schiff [243]. We expand the wave function V( x . (10.122b) S e e also L. pp. (10.x) = S(EE').121) . 193.122a) (10.x)^(E'. specifically in the case of the Coulomb potential. 148.t)+ / n a{E. .69). 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/>).t).9 General TimeDependent Method For a better understanding of the "golden rule" derived above. (10.t) = ^On(t)^(x. (21.^ We start from the Hamiltonian H = HQ + H'. 1 rfx^°)*(x)^)(x) = 8mn. For a possible application of this result.^°>(E.10. 10. t) = (xl^) therefore in terms of the eigenfunctions of Ho with timedependent coefficients.t)^°\E^t)dE (10.9 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. by which we mean without leaving the Schrodinger picture.120) In the case of Eq. (10.119) only the time dependence of HQ can be separated from the state \ip) in the form of the exponential factor of a stationary state.119) where H' = H'(t) is a timedependent perturbation term and Ho\ipn) = En\ipn). we present another derivation using the usual method of timedependent perturbation theory. Thus we assume that the eigenfunctions (fn = (x</?n) of Ho are known./continuum = <San(£)</4°)(x. see the discussion after Eq.
since f rJ>(°>(E..x).120). = e^EE'^h5(E Thus with Eq.122b).125) where an(t) — dan(t)/dt.E') = 5{E . . (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.. .x. Next we use the orthogonality relations .„„.129) . ^0)(E.. (10. V>(x.127) i n c a g e Qf c o n t i n u o u g £. We multiply this equation from the left by </?£(x) and integrate over all space. H'kn = J d x ^ ( x ) # V n ( x ) = (<pk\H'\<pn).„. Then J d^*k(X)[ihSdn(t)cpn(X)eiEt/h = J d^*k{x)[San{t){En + San(t)En<pn{*)eiEntlh] (10.122a) and (10.127) we obtain from Eq. The second term on the left of this equation and the first term on the right cancel out. dnptMrnW = { ^ _ En) (10. (10. (10.e. into dip (10. ..126) + H')<pn^)eiE^h}.123) We now insert these expressions into Eq.. t) = San(t)<pn(x)eiE^h.196 CHAPTER 10. Quantum Theory: The General Formalism The discrete and continuous eigenfunctions of HQ are Vi 0) (x.t) = eiEtlhy(E.iJ5 »*/Vn(x).x.E'). (10.i) = e. . / *. (10. i. f 6kn in case of discrete tp's.x. Observe that these conditions do not contradict Eqs..t)rJ><®(E'. (10.t)dx.128) .126) ihak{t)eiEktlh where = San{t)eiEntlhH'kw (10.
they are fixed by the initial conditions. SH'kn(ty^E^l^\t) (10.e. We obtain ih{ak°\t) + \ak1\t) + \2ak2\t) = SXH'kn(t)e^.133) The numbers ak = const.e. since it is equivalent to the latter. i.131) and insert this into Eq. (10. We restrict ourselves here to the two lowest order equations. (10. In the first place we replace in Eq.130) along with the parameter A in front of H'kn.10.132) These equations can be integrated successively.9 General TimeDependent Perturbation We can rewrite the equation as Theory 197 iMk{t) = SH'knan(t)e^EkE^h. In order to solve Eq. Comparing coefficients of the same powers of A on both sides. (10. 0. SH'kn(t)e^~E^ha^(t). (10. Now we set ak(t) = ak0){t) + \a£\t) + X2a[2\t) + ••• (10. We observe that the amplitude ak of a definite eigenfunction tpk has effectively replaced the wave function ip. to perturbation theory in the lowest order.130) This result should be compared with the Schrodinger equation (10. determine the state of the system at time t — 0. and we set ak°}=5km or ak0) = 8(EkEm).^ [a^(t) E h + } + \a£\t) + A2a£>(t) + • • • ] .130) we develop a perturbation theory. Here we assume (our assumption above): At time t = 0 all coefficients ak are zero except one and only one.120). 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. we obtain the equations: iha[°\t) iha£\t) iha[s+1\t) = = = ••• . Thus first we have ak0) = const.134) . i. (10.130) H'kn by \H'kn and at the end we allow A to tend to 1. let's say am / 0. (10.
Emy in agreement with Eq. (10.136) we see that both "amplitudes" are Fourier transforms and look similar. The amplitude / B o r n is a Fourier transform in spatial coordinates.e. (10. What is the physical significance of Eq.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.135) We put the additional constant of integration equal to zero. an expression of the following type AS?) = " 2 ^ 2 / e j ( k . We shall see that in this approximation the scattering amplitude is effectively the Fourier transform of the potential.136) Comparing Eqs. i. i. ifm{kl) = f J —oo (10.ml2sin2{^frn (En .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. Quantum Theory: The General Formalism depending on whether Em belongs to the discrete part of the spectrum of Ho or to its continuum. (10. In this case we obtain from Eq.132). Equation (10. H'^y^"^^/hd£.137) . 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. (10. (10.198 CHAPTER 10.e. ) ' X V(x')dx'.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. whereas ak is a Fourier transform in time (a distinction which is easy to remember!). Integrating the second of the equations (10. (10. we obtain it*™ = f J—oo SH'kn{t')e^E^'lh5mndt'. ip) in the context of timeindependent perturbation theory.111). ipm) into the state k.135) and (10. so that ihakl\t = oo) = 0.
( 1L1 ) We let pi. meaning the inclusion of the creation and annihilation of field quanta).Chapter 11 The Coulomb Interaction 11.2. Z<i = Z) y(r) = _ M ^ .y. 199 (1L2) . 11. Both cases therefore also serve in many applications as the basic unperturbed problem of a perturbation theory. The electrostatic interaction of the particles is the Coulomb potential (later we take Z\ = 1. Angular Momentum In considering the Coulomb potential in a realistic context we have to consider three space dimensions.2 Separation of Variables. The Coulomb interaction on the other hand is of specific significance for its relevance in atomic and nuclear physics.i be respectively the momentum. r=  r .m.Ti. such as the electromagnetic field (the quantization in these contexts is often called "second quantization".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. 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. Let r : = (x.z) be the vector separating two particles which are otherwise characterized by indices 1 and 2 with electric charges Z\e and Z^e. The Hamilton operator is then given by w= jt + ji_^M. the position coordinate and the mass of particle i with i — 1.
(H4) \mi m. (11. z) one verifies that j9_ = dxi_ _d_ 8X ~~ OX dXl dx2. . But now P2 2mo = "ip/pi 2 \mi m p2\2 m2J  = m0/pi2 2 \m\ p22 m2.4)._d_ _ _d_ dX dx2 ~ dXl d dx2 nifi^i [ ' etc.200 CHAPTER 11. Similarly d_ _dx^_d_ dx dx dx\ dx2__d_ _m^_d_ dx dx2 mi dx\ mo_9_ m 2 dx2 etc. so that as expected according to Eq. so that as expected we obtain the expression for p in Eq. /'Pi p = m0 P2 A v. Solving Eqs. (11. P = pi + p 2 . r} representation they have the differential operator forms Setting R = (X. r = rir2 mi + m 2 (11.2/ mi+m2 The mass mo is usually described as the reduced mass. y. 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 ' . mi + m 2 iTi'2 r2 = R ^ —r.U7&. where m 0 = mim 2 . mi + m 2 va\ (11.3) for the individual position coordinates we obtain ri=R+ •^ —r. p are such that as operators in the position space representation or {R. Y.4) P = p i + p 2 . The Coulomb Interaction We introduce centre of mass and relative coordinates R and r by setting R = m i r i + m2r2 • .5) The momenta P .3) and centre of mass momentum P and the relative velocity p/mo by setting r» . 2Pip2 m\m2 pip2 mi + m 2 . (11. Z) and r = (x.
r2) to (R. 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).16) Although we write ri. with M := mi + vn^R = ^ P.+ ^(r) 2(mi + m 2 ) 2m 0 (H. The motion of the centre of mass is that of a particle of mass M = m\ + 1712 in uniform motion along a straight line.10) One can convince oneself that the transformation from (ri. (i.13) * and = M ' 1V1 P = 0. 2(mi + m<i) 2mo 2m\ 2m. P on the one hand and those in r. etc. (11.15) In accordance with Eq. Canonical quantization of the classical theory implies that we pass from the Cartesian variables rj. The canonical quantization must always be performed in Cartesian coordinates.2 Separation of Variables. pj.8) and so W =777 v + 7 T .pj. Pj (which in the following we write again ri.pj]=iMij.j = x. [Ri. (11. (11.Pj).pj. (1114) m0 We observe that the equations of motion in R.Ri.Ri.= ^.12) (11. p on the other are completely separated.y. Hence Hamilton's equations apply and we obtain. Ri.e.il) ""Op' i.+ P.+ ^2.10) we set H = Hcm + H. (11. p = W. Pj over to operators ?i.z). we mean n = rx s x. r = —. . ' d and 'AW OH'ATJ P =  ^ dR <9# <9r ' (ii. Angular so that Momentum 201 ^ ^ .11. r) is a canonical transformation.2 It follows that p2 2 P2 n2 n2 n2 (H.Pj} = ih6ij. for which we postulate the commutator relations* [ri.9) ft = ftcm + ^ = centre of mass energy + energy of relative motion. (11.
22) If. (11. 2mo (11. i.18) In the position space representation the time independent Schrodinger equation is therefore 2M' A fl ) + h2 2mo A r + V(r) * ( R .23) y — r sin 9 sin tp. (11.Q 2M' + V(r). (11. ip: x z = r sin 9 cos ip.r) = *(R)V(r). In these coordinates we have /\rip 1 d r2 dr \ tdil> dr + r2 sin 9 09 \ S m 1 d_ .24) . The Coulomb Interaction nc H = 2m.17) (11. where Hcm$(R) = and Hip(r) = with ^total — ER (11.e. V(r) = V(r). (11. 0 < r < oo.r).21b) + E. df\ 39 J + 1 d2i/j r2 sin2 9 dip2" (11. Thus the twoparticle problem is practically reduced to an effective oneparticle problem. 9. it is reasonable to go to spherical coordinates r.20) r2MAR_ $(R) = ER*(R).202 CHAPTER 11. —n < tp < ir.21a) L*+v{r)] ip(r) = Ei/j(r) (11. = rcos9. to that of the Schrodinger wave equation h2 Ar + V(r) tp(r) = Eip(r).19) For stationary states with energy E the equation possesses a complete system of eigensolutions * which can be written tt(R. 0 < 9 < ir. r ) = J E t o t a i*(R. as in the case of the Coulomb potential.
2 Separation of Variables. = 0._0di})\ x We now demonstrate that 1. otherwise.27) We can now write (see verification below) p V = h2Ar^ where 2.2. the operator —ihd/dr is not hermitian! The operator pr obviously satisfies the relation [r. We have I2 = (r x p) • (r x p).2. j . with tijk — + 1 .24)) 2 + M\ d dr2 J = tfL^(r^A\ r2 dr\ dr J d2ip (11. .32) (11.5r = .j.26) However. 2 1 9 / 1 9 . that the above expression follows from l:=rxp=ift(rxV).3.3.28) .2. n ( .k a clockwise permutation of 1. (l + .i ' .33) i.1 it is shown that pr is hermitian provided lim n/>(r) = 0.12 — I2.29) . Angular Momentum 203 We can rewrite this in a different form by defining the operator * : =«. is the operator of the relative angular momentum. (11. i—>0 (1L25) (11. i. k = 1.31) . (11.j.A fc2! d ( d^ ^ = _tfU2dA r\dr and (compare with Eq. _ h2 \ . i( i.)In Example 11. if i. = —1. = (p2 + ^ V (r + 0). (11. where (r x p)j = ^eijkrjpk.11.e. k an anticlockwise permutation of 1. (11. (11.3.pr]=ih.
ih)pr and x 2 2 9 9 2 2 Q O (11.27) we obtain r{prr)pr = r(rpr .j.36) It follows t h a t (note the part defined as —5) l2 = = ( r 2 p 2 . (11.40) (11.38) + y* + z .42) (11.k ~ rjpkrkPj) (11. (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. The Coulomb Interaction = &jj'&kk' — Sjk'Skj'.41) o = r pr.44) .ihr • p ) . (11. This means that l2 = so that 9 9 1 (11.39) + ih). Eq. * ' S Since r =xz we have (cf. (11.k Using Eq.204 It follows t h a t 2_^£ijktij'k' i CHAPTER 11.(r • p ) 2 + 3ih(r • p) . ih6jkrjpk (11.25)) r • p = —ihr • V = rpr + ih and hence 8 = (r • p)(r • p . r— = r • — or or (11. (11.ih5jk)pk (11.37) .43) r2(p2p2).15) we have fjPkTjPk = fj(rjPk and TjPkTkPj = = rjPkipjrk rjPj(rkPk + ihSjk) = TjPkPjTk + ~ ihSkk) + ihrjPj.ih) — rpr(rpr With Eq.ihr • p r2p2(rp)2 < + ih(rp).35) j. (11.
e. <p))r2drdns % J r or r<t>(r. 6.l2} = 0. i. (pr4>. or J .2.e.e.26).6. 11. Pri/Oi V0. r := r : lim rip{r) = o i . <p)*rip(r} 0..2 Separation of Variables.[ i J With partial integration with respect to r this becomes (<l>.30) yields l2 = I2 = sin sine— BinO^r 09 + d^2 (11. h f l d —(r<l>*(r.11.Q ' 2mor2 with the condition (11.d_ r dr the operator pr is hermitian. 0. 6.<p) (11.45) oe Thus in Eq. (11.ip) = (<£.<p))ril>(r.e. Angular Momentum 205 Comparison of Eqs.e. <p)r2dfls = . <p)* .1 Separation of variables We deduce from the expressions of H and l 2 that [H. The Schrodinger equation of the relative motion can therefore be written V1 V 2m. reR3. pr = .! /'). /'(r. 8.Pril>) J / r= dQs rcj>(r. 9. <p) d °° .<p)*?(ri.30) I2 is to be identified with the square of the angular momentum operator. 9 € [0. prip) = = I <j>{r.47) Example 11. (11.d..<p)drdna " r r=0 J \ll§r~(r4'(r.2TT]. r—>0 il>(r.— (riP(r.6l. Then {<j>.6d0dip. H 1 2mo pi + K) +V(r).e lim rip(r) = 0.1: Proof that pr is hermitian Show that if Vv lipe H = C2(R3).(r. ip 6 X>Pr C 7i. (11.46) (11.28) and (11.e'iP^} . i.48) .[ 4>(r. ¥ p)r 2 drdn s = (p r 0. S o l u t i o n : Set d£2s = sin.<p))drdns.<P) = Etl>(r. <p)*Pri>(r.7r].<p G [0.
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. they are incompatible variables.l2] = 0.e.51) (11.xpz] = [ypz.49) By cyclic permutation of x. z]px + py[z.53) Either of these operators is the hermitian conjugate of the other.50) Thus the components of 1 do not commute pairwise. (11. These operators here play a role analogous to that of the quasiparticle operators . z we have altogether [lx.ly) are simultaneously "sharp" determinable variables and hence have a common system of basis functions.l2} and correspondingly [lx.e. 3 . so that [lz. /z> *xJ = l <i\}ylx + tx'j/Jj [h. [ly. (11.q2 = V. (11. and this implies in the terminology we used earlier that their determination cannot be "sharp" simultaneously. Example 11.lx] = ihly. The Coulomb Interaction i. We can now proceed as in the case of the harmonic oscillator and determine their eigenfunctions. [lz.lz (or l . g i = x.pz]x = ih(xpy . xpz] = y\pz.e.lx or l . for any two of the components there is no common complete system of eigenfunctions. 2 .2: Verify that [h. y.ly]=ihlz. (11. we have [h. However. zpx] + [zpy. ?3 = z). H and l 2 possess a common system of basis eigenvectors.ypx) = ihlz. ly\ — —tfi(lylx + Ixlyji [hi h\ = 0. zpx .52) This means that l . We therefore first search for a complete system of eigenfunctions of l 2 . Since 1 = r x p . i. In order to determine the eigenvectors of the operators l 2 and lz one defines l± = lx±ily.lz]=ihlx. 2 2 2 = 0. ly] = [yPz ~ zpy. i.206 CHAPTER 11.qj]=iheijkqk S o l u t i o n : This can be verified in analogy to the preceding equations (i = l . [lyA = 0. 11.
58) {l0 : max.lh)\l) = {l0 . and with Eq.54) l 2 = \{l+l~ + 11+) + ll = l+l. i + ] = [l2.54) On the other hand we obtain from Eqs.60) Similarly for l+ with the help of Eq. (11.61) 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. (11. With the help of Eq. A^ in the case of the linear harmonic oscillator.54): lzl\l) = {llz .Z_] = [ l 2 . assuming a finite dimensional representation space.r) := (Z_)rZ) so that lz\l .55) One now defines certain ketvectors as eigenvectors of lz which span the space "K\ of the appropriate states.h~lz(1156) (11. In the present context the operators are called shift operators for reasons which will be seen below. (11. [i+M = 2%iz. since lz is hermitian. Hence we must have Z+I0=0.).i+] = m+.51). in particular one defines \l) € CKj as eigenvector of lz with the largest eigenvalue lo.11. (11. l\l) is eigenvector of lz with eigenvalue (IQ — l)h.r) (11 = {l02)h{l)2\l) (11. Similarly we have lg{l)2\l) and more generally lz(l. (11.i] = m.62) 'At this stage it is not yet decided whether IQ is an integer or not! But we know that IQ is real.3 Representation of Rotation Group 207 A. (n.54): izi+\i) = (i+iz + i+h)\i) = (lo + i)ta+\i). (11. [ig.e. (11.+ ll.Y\l) = (lQr)h(lY\l). (11.57) i. (11.52): 0 = [ l 2 . We write or define: / . (11. Therefore (lo + l)h cannot be an eigenvalue of lz.59) = 9) h(l0 r)\lr). y .l)hl\l).49) we obtain the following relations: [iz. (11. .
e.e. (11. (11.\lr) terminates at some number (let us say) r = n (n a positive integer). [12. (11.65) We assumed that "Ki is finite dimensional so that the sequence of eigenvectors of lz. The Coulomb Interaction (l+l.n) (1 = 60) {l)n+1\l) = 0. \l)..60) = Vli\l) = l0(lQ + l)h2li\l) i0(io + l0(l0 + i)h2(i.56) CHAPTER 11.e.h)\ln) {(lQnf(lQ.l2\l) = l0(l0 + l)h2l\l).\ll).54) (n = 57) (11. IQ(IQ 1 0 (11..55)) we have 12Z_Z> = l.Z_] = 0.y\i) l)h2\lr).. Eq. i.+ 11+) + l2z\l) = i+i.56) that l2/n) = (11 6) (l+l+l2lzh)\ln) (lll. i.64) we obtain on multiplication by Z_: lVl\l) and more generally l2(f)PJ> (11.n)}h2\ln). (11. (11. (11.. (11.64) i. (11.n).63) + l ) ^ 2 .60) and (11. i.67) i (11 60) = But according to Eqs.65) for r = n: l2\l n) = l2(l)n\l) = lo(lo + l)h2\l . l\l) is also eigenvector of l 2 with eigenvalue Zo^o+l)^ 2 .208 Thus for l 2 we obtain: V\l) ' = 11. l\l .From Eq.66) It then follows from the second of relations (11.e.+ i2z (11+ + 2izh) + 1 \ \i) = (izh + ti)\i) i0(i0 + i)h2\i). (cf. Since l 2 commutes Hence Z) is eigenvector of l 2 with eigenvalue with l_.68) .
p.lz transform the vectors \l).m) : Mo).m). The basis vectors \l)..m)\l. The phases are chosen such that§ l+\l.70) According to Eq... y (lo . (11.m). (IQ — l)h.2.m) = mh\l.m) = = mh.63) and (11. (11.e...11.{lo . We write these* for integral values of IQ: \l. Messiah [195]. we skip this here. In the following we consider primarily the case of integral values of IQ.m) = = = = y/lo(lo + 1) — m(m +l)\l. The normalization factor in Eq. Satchler [38]. The operators l+.l. (11. i..65) these are eigenvectors of l with eigenvalues Zo(^o + l)h2. *D.. (11..m) l\l. The dimension of the representation space is therefore n + 1 = 21Q + 1. lo = l0n.m) — mh\l.m + 1)(Z0 + m)\l.n) 2 ..71) can be established in analogy to the method used in the case of the harmonic oscillator.. \l — n) are eigenvectors of lz with eigenvalues loh. . lQ = .m) (l. where lz\l.m±l\l±\l. (H71) These expressions should be compared with the corresponding ones in the case of the harmonic oscillator. R. M. lz\l.m .. According to Eqs.n .m) = IQ(IO + l)h2\l. A.. Vol.Z 0 ).m\lz\l. y/(l0±m + l)(l0^m)h. We therefore write the 2/o + 1 orthonormal vectors: \l.n)} = (l0 .l_. lz) are then: (l. Brink and G.n){l0 .1). (11..69) Prom this it follows that IQ must be either half integral or integral.l0): l2Z.m / + l)h . II. m + l)h.m) with — lo<m<lo2 (11. § See e...m y/lo(lo + 1) — (m — l)m\l.g. —loh.70) they satisfy the eigenvalue equations (with m = l0Jo l.l)h \/(lo + m + l)(l0 .l)h. 17.. As in the case of the harmonic oscillator we introduce normalized basis vectors. \l — n) into one other.  U o ..67) we obtain (observe IQ(IQ 209 + 1) = —lo(—lo — 1)): *o(*o + 1) = Wo .1).m). p. 24. The nonvanishing matrix elements of 1 = (l+.3 Representation of Rotation Group so that with Eq.
—lo + 1 . ip) = e(9)$((p). ip. We . (11. ih d_ dip' lx ± ily = ±he^ d I ^ ± zcotfl^ ..73) 1 d d sm0 06 \Sm9dd) + (11. as this is sometimes described.77) $((p) = m2<b(y).yp. 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. In these coordinates we obtain: h l± so that h2 = xpv .75) the variables contained in the expression (11.<p). where IQ = 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.210 Then.<p) = i)n2Y^{e^).e. . We now perform the analogous procedure for the angular momentum or. ..72) In view of the spherical symmetry it is reasonable to use spherical polar coordinates r. for which the eigenfunctions \ip) are given by l2 l0(l0 + i)h2\ip). with i2Y£{e.. mhY£(9. i.76) d2 (11. CHAPTER 11.74) We choose the eigenfunctions of l 2 and lz as basis vectors of the ("irreducible") representation. .With the ansatz Y(9. as above. 11. l_\l. $ oc eim*. /o. d2 sin2 9 dp2 _' (11. for the simple "rotor".l0)=0. The Coulomb Interaction Z+Mo> = 0.2.. (11. (11. and m = —IQ.1.<p) = i0{io + lzY£(0. as remarked earlier. 9. .74) can be separated: 1 d sin 9 36 2 1 sm6— 89 j — sin 0(9) = h(l0 + 1)0(9).
78) and the completeness relation oo I ' YrVMYTVrf) = °^~' ± J2 Y7n*(f}.77) and o these are determinable by the requirement of uniqueness of the original wave function and its square integrability.ypx = tfrK(11. y2° = \ (3cos 2 01) 2 V V 4vr V 16?r 3 W—.11. (11.=o m=l sin 6 The connection of these hyperspherical functions with the associated Legendre functions P™. We mention in passing that in spherical coordinates d lz = xpy . Uniqueness implies immediately that m has to be integral. (11. so that im(ip±2ir) The functions Ylm(8. m > 0.81) 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.V>)Yir'(8.79) . (11. and that their phases are such that Yt0(0.(5cos 03cos0).t»'\Y7n((l'. (11.^ = *(n . (11.<p) Jo = 6mm'5w. This is the case if one demands that l 2 and lz possess a common complete.* ? ^ .80) y? Furthermore Y°l = V^fcosO).82) .i».<p) = (iy 47r(Z + m)! In particular for m = 0: 1/2 PI11 (cos 6)eimi?. / 3 / 5 J — casO. The functions Y™ satisfy the orthonormality condition dfilimT' = / Jo < W sm8d9Yr(9. orthonormalized system of eigenfunctions.n>).0) is real and positive. is given by (2Z + l ) ( Z .4 Angular Representation of Angular Momentum 211 observe that Z and m define the eigenvalues of Eqs.\ = W .76) and (11.r o ) ! YT{e. These functions are defined in such a way that they are normalized to unity on the unit sphere.(p) are called spherical harmonics.
It is instructive to pursue the appropriate arguments.90) .1.83) Their connection with the Legendre functions Pi (x) is given by the relation pr{x) with the differential equation =£^ Pi{x) (iL84) (1 .g.89) Inserting this expansion into Eq.e. The Coulomb Interaction The number I is referred to as the "orbital quantum number" and m as "magnetic quantum numbed.212 CHAPTER 11.x2)Pf . If we did not know yet that I = IQ = an integer > 0.1(1 + l)]xK+i = 0.2..88) y = ^aiXi+K.e.86) y = 0. 22 (11. (11. exp(imip).88) we obtain for every value of x ^2 ai(K + 0( K + * .e.2(m + l ) x i f + [1(1 + 1) .Y^O»[(K + i)(/e + i ~ 1) + 2(re + i) .m(m + l ) ] i f = 0. (11.e. i.2xy' + 1(1 + \)y = 0. the designation s. we would now have to search for those values of I for which the solutions are square integrable. of Ytm.d.... oo (11.x2)P[' . respectively..h.x2)y" .87) we obtain the equation (1 . Setting y = (lx2)m>2Pr(x) m . is also in use. (11.. The associated Legendre functions or spherical functions PJn(x) satisfy the differential equation (1 .p. Instead of the values I = 0. (11. i. (11. i. (1 . f. For the solution we use the ansatz of a power series.2xP[ + 1(1 + l)Pi = 0. Consider the equation for Pi(x).x2)y" .2xy' + 1(1 + 1) 1x The number m must be an integer already for reasons of uniqueness of the wave function.m = 0. in the replacement ip —> if + 2K.l)xK+i~2 i . i. i=0 (re lowest power > 0).85) (11.
l 2 and lz are the solutions of the Schrodinger equation of the form (11. For the solution in accordance with (11. ai(re + !)« = 0. Margenau and G. M.2 + 2mo dr 2mor V{T)E Viir) = 0 (11.26)) yi(r = 0) = 0. p.1.29)) 2 fe2l 92 Pr = & r 7Tororz We set (11. they have to be polynomials (recall the orthogonalization procedure of E.1 we obtain by comparing coefficients QOK{K 213 — 1) = 0. these are of little interest to us here. Eq.95) Vl = rxiThe hermiticity condition for pr then requires that (cf. . We have therefore (multiplying Eq.94) from the left by r) h2 d2 h2 2 + H! + \)IL. but also those for a different series for a. Comparing the coefficients of xK+3 we obtain aj+2(K + j + 2){K + j + 1) = OJ[(K + J)(K + j + 1) .E Xl(r) =0 (11. l(l + l)h2 _2mo + 2m. (11.46)) p.78) to be square integrable. Schmidt). This is precisely the case.92) This is a recurrence relation. (11. when the series terminates after a finite number of terms. Murphy [190]. first line of Eq. We see that this happens precisely when I is a positive or negative integer or zero.1(1 + 1)]. (11. (11.11. > 1.Qr2 with (cf.93) where Xl(r) is the solution of the radial equation (cf.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. 93. (11.*" 11. Eq. However.96) + V(r) . . (11.94) ^For further details see H.5 The Radial Schrodinger Equation for Hydrogenlike Atoms For i = 0.91) from which we deduce that K = 0. For \x\ < 1 we can obtain from this the coefficients of a convergent series.
The wave function of the relative motion is the solution ^(r) of the equation A 2mo Here •4>{r)=Ylm{eM and Vl(r) (11.1)). He + . for instance. which behaves like 1/r1 in approaching r = 0. r m0 = meMp me + Mv (11. L i + + .e. (11. an analogous consideration applies to hydrogenlike atoms like. In the first place we investigate the behaviour of yi in the neighbourhood of the origin. The Coulomb Interaction (11.100) with (11.1.e.101) 2 r ^ We set  6 = ( _ e ) 1 / a = ( = (11. s = 1 + 1.102) .99) e ip(r) = Eip(r). (11. Jo Jo We thus have a problem similar to that of a particle of mass m in the domain (0.98) 2mo e2 _ 1(1 + 1) yi = o e+ 2 y'l + h r r2 2m0E e = (11. Obviously the irregular solution has to be rejected. (11. and substitute this into Eq. Equating the coefficient of 1/r2 to zero. Z\ = Z2 = 1 in Eq.1)+1(1+ 1) = 0.97) / r2dr\Xi(r)\2 = / \yi(r)\2dr. nor the condition of normalizability for l ^ 0. since it satisfies neither the condition yi(0) = 0. i. This applies to the Coulomb potential but also to Coulomblike potentials like screened Coulomb potentials or the Yukawa potential. In order to verify this behaviour we set <VflV>r> yi(r) = rs(l + air + a2r2 \ ). The equation also has an irregular solution.96). We now consider the hydrogen atom with V(r) — —e2/r (i.214 with CHAPTER 11.96) can be shown to possess a regular solution there which behaves like rl+1{\ + 0(r)) in approaching r = 0. oo) of a onedimensional space. and others. We assume that V(r) has at r = 0 at most a singularity of the form of 1/r. we obtain the socalled "indicial equation!'' s(s . In these cases the behaviour of yi near the origin is dominated by the centrifugal term and thus Eq.
d2 x—j 1 + (2l + 2x) ax d j dx (/ + 1 . We can see from Eq. behaves there like xl+1) extended to an exponentially decreasing branch at infinity. We set therefore Vi X J. .g.529 x 1 0 " 1 0 meters) 1 V/2 e2f hc\ m0c^1/2 2E 2moE J (11.a * = 0. (11.e. (11.6.104) W i t h these substitutions we can rewrite Eq.6 The Discrete Spectrum of the Coulomb Potential 215 For E > 0 the solutions oscillate in the region r — oo. Magnus and F. (11. W.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.100) as^ d2 dx2 1(1 + 1) x v___\ 4 2/1 = 0. hypergeometric (11.v) vi(x) = 0. (11.+if>x/2 vi(x). p. i. see e.11. (11. T This equation is known as Whittaker's equation.88. i.105) t h a t for x —• oo there is a solution behaving like exp(—x/2). Inserting the ansatz into Eq. for E < 0 they > decrease exponentially.6 11.105) 11. (11.106) W i t h this ansatz we separate from the solution the behaviour around the origin as well as t h a t at infinity.e. Oberhettinger [181].103) °y moe2 ( = 0.107) equation. We set x — We set as Bohr radius* a = and 1 v = rea rriQe ft V (47Fe 2KT.105) we obtain the equation for V[(x). (11.108) This equation has the form of a confluent x$" + (6 .x ) $ ' . It remains to determine the function vi(x) for a complete determination of the solution. *The factor 47reo appears in SI units.e.
and for both cases.(0_i)!(6 + n l)!n!' (11. The function T(z) is defined by the integral /co T(z) := / eHzldt.e. 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. The Coulomb Interaction The regular solution of this equation is the confluent hypergeometric series (also called Kummer series) 1 . more precisely. and hence the *Some writers. However. (2) r ( z ) r ( l . called dupZication formula. In general the function $ is an infinite series and behaves at x — oo like » x{l+l+u)ex_ Obviously such a series is useless in our case. (5)r(n + l ) = n ! .! ) ! = imlFI' c a l led inversion or reflection formula.112) The gamma function is a nonvanishing analytic. ^ ( 2 ^ ) ! = 2 2z z!(z .111) Important properties of the function T(z) are: (1) zF(z) = T(z + l). (11. it is convenient to use only the factorial notation. (11.216 CHAPTER 11. are very strict in reserving the factorial exclusively for positive integers and the gamma function for all other cases. the factorial for an arbitrary argument.106). (4) r ( i ) = V5F. with many of these around in some calculation. meromorphic function. 02zl / i\ (3) r(2z) = — ^ r ( z ) r ( z + . (z)!(z .1/2)!. .J.z) = ^f^. particularly pure mathematicians.z(zl)\ = z\.109) In our case we can write the series The expression* r(x + l) = x\ is the gamma function. + ' ' ~ 6 1 ! + 6 ( 6 + l ) 2! + Z. since the exponential function (generated by the infinite series) would destroy the asymptotic behaviour of the wave function which we separated off with Eq. i. (11.
With (cf. Thus the infinite series must break off somewhere.» • » • » .110) as follows: T(l + 1 + p . .e. if the series terminates for certain values of u. . the orbital or azimuthal quantum number I can assume the values 0. which is thus seen to be a consequence of demanding the square integrability of the wave function. 1=0 (11. . Eqs. d = 1. This is achieved. The quantum number n' is called radial quantum number. 1 .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. This number is equal to the number of finite zeros of the radial part of the wave function (excepting the origin).a " \ 2moE we obtain the energy spectrum of the discrete states with angular momentum I ie t m0e2 / 1 N1/2 * =  ( s ) ' ^ . which are also described as its nodes. = n = Z + l + n'.V))T(v . 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 . For every value En.) 'Observe that with property (2) of the gamma function we can reexpress the ratio of gamma functions in Eq. (11. One defines as the principal quantum number the number n. n' = 0 . i. ..v) T(l + lv) _ ~ K sin{7rQ/ .114c) (Recall that for an arithmetic series a + (a + d) + • • • + (a + (n — l)d) = na + n{n — l)<i/2..113) This is a quantization condition. and so to the number of different values of m. for z.I) r(vl p)sm{n(i/ I .1.6 The Discrete Spectrum of the Coulomb Potential 217 latter would not be square integrable. i. (11.1) + n = n(n .11.2. in our case: a = 0.. i. . every n. (11.e.e.. must be a polynomial. • • • • <»•»*> It should be noted that the magnetic quantum number m does not appear in this expression.104)) 1 n =v = K.101) to (11.1) + n = n2. 2 .
ne. nd. hence s). The above treatment of the hydrogen atom has obvious shortcomings: (a) The hydrogen atom was treated nonrelativistically. (c) The nucleus was treated as pointlike and not as something with structure.p. . / = 1.. ' O n e describes as Lyman series the final state with n = 1. 11.l = 2 to n = 2.'.§ One also writes n2l+1l. . .218 CHAPTER 11. nf .1.1 = 0 (called principal series. hence d). from an initial state n > 2. One describes as Balmer series the final state n = 2. . The spectrum of the hydrogen atom thus has the form shown schematically in Fig. The following spectroscopic description of states is customary: nl —> ns.\ I = 0 or 1.d.. I = 0. and hence the fine structure of the energy levels is excluded.1 Hydrogen atom energy levels. where the superscript 21 + 1 at the top left of I gives the degree of degeneracy. and from n.1 = 1 to n = 2. 0.. .'. hence p). stand for I = where n denotes the principal quantum number and s. (b) The spin of the electron was not taken into account.. 11..4.. from n. np.1 = 1 (called diffuse series. 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.1 = 0 (called sharp series. from an initial state n > 2.. I = 0 to n = 2.1. 2 .. respectively. relativistic corrections are of the order of v2/c2 ~ 0(En/moc2) as we illustrate in Example 11.
Unfortunately the definition mostly used in mathematical literature differs. Apart from factors.117) The function *Lk(z) is a polynomial of degree r. With the help of Eq. The mass of the hydrogen atom is given by massif = Mp + m e + E < Mp + me. Gradshteyn and I.2l { + 2. Magnus and F. 844. Vol.414(3). . • (11. 84.6. M.118a) where Lk. . Thus Messiah [195] defines the original or stem Laguerre polynomial as : >Lr(z) = *L°r(z) = e'^(e'zr). k + l. W.115) ' =0 (ri . For clarity and later use we denote the function used there by its normal form L^. p. however. We mention here three different definitions used in the literature. Each can be recognized by reference to the generating function.110) is usually reexpressed in terms of orthogonal polynomials known as associated Laguerre polynomials. p. so that one always has to check the definition. l .116) Laguerre p o l y n o m i a l s : Various definitions in use! The polynomial obtained by terminating the hypergeometric series (11. We distinguish between different definitions in use. S. (11. Unfortunately.1. (11.z) = (r + k)\ Lkr(z). the polynomial contained in y\ of Eq. • ( .11. it is given by^ *Lkr(z) = ^ ^ f 1 *ir. I.x). in fact. • fc.= t . a + 2:*) = t v . and I.{z) is the associated Laguerre polynomial as normally defined and mostly written Lr (z).r = 0 .2 (11. (u.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.115) is effectively the associated Laguerre polynomial normally written Lk(z). (11.p)\(2l + 1 + p)\ The spectrum (11. 2 . . Oberhettinger [181].114b) yields the binding energy of the states.2. for instance. see. 3 . "1'f„+11" P] „. Appendix B.r f . 11. formula 7.6 The Discrete Spectrum of the Coulomb Potential The associated radial wave function is yi(x) 219 = xl+1ex/2<f>{n'. several different definitions are in use. . . Ryzhik [122]. Messiah [195].
(11. N.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.(z)*L TJz)dz 3 = KP + kV}Opqit Jo (11. . p.119) dz dz In particular one has (we cite these for later comparison) LQ(Z) = 1. The reader is there warned that '•'•care must be taken in reading the literature to ensure that the particular convention being followed is understood'. I. Ll(z) = 33z + ±z2. pp. The Coulomb Interaction The differential equation of the associated Laguerre polynomials is (same for *Lk{z)) d2 .122) "See A. Messiah [195].120) ' r=0 It will be seen that the orthonormality condition is" z ezJc*Tk. It is said that this is the definition preferred in applied mathematics.2.220 CHAPTER 11. **I. r r Z r=0 v tr k k (11. L1(z) = l . which we write ^Lk. ^ F o r this definition and the following equations we refer to I. N.3): e2t/(lt) (1 . .** The associated Laguerre polynomial here. Sneddon [255].164.121) In Example 11. .3 we show in a typical calculation how this normalization condition can be obtained with the help of the generating function. and Ll(z) = l. L2(z) = l2z + z2..z .„M„ z"*L. Appendix B. L\(z) = 2z. Example 11.d z—^z + {k + l~z)—+r Lkr{z) = 0.„\*Kk.. 163. Sneddon [255]. is defined by^ dk (—'\\k(r^\2 (11. 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. There at some points n + 1 is misprinted n + 1.1. 162 . For our purposes here it is convenient to refer to yet another definition of the associated Laguerre polynomials. Vol.
Jfpd>2s{p))eptdp.4z + . (11.119) derive the generating function of associated Laguerre polynomials (11. and *L\(z) = .4 + 2z.107) implies k = 11 + 1 and r = n + I.3: Generating function of Laguerre polynomials With a Laplace transform ansatz for the solution of a second order differential equation like Eq. The limits will be determined later.. (11. ^L2(z) = 2 .1 8 + 18z 3z2.124b) Comparing Eqs.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 ... and obtain with this the normalization condition (11.8): (11. N d rk 0. ^L\(z) = .118a) and (11.122) we see that tLkr+k(z) = (l)k*Lkr(z) = (l)k(k + r)\Lkr(z). with y'{i) = — / ps(p)e~ptdp ty'(t) = t f ps(p)eptdpp2s(p)ept s(p)e Pt J ^(ps(p))eptdp.120)) it)k& zt/{lt) ~*P p=k fc (11. (11.121). ..% ' ( * ) + by{t) = 0 with ansatz y{t) = I s(p)e~ptdp. (11. (11. Solution: We consider the following equation of the form of Eq. (11. and partial integration.124c) Example 11. Example 11. '"''Comparison with Eq.119): ty"(t) + (a + 1 . The generating function of the associated Laguerre polynomials so defined is (observe (—t)k as compared with (11.z .124a) (i*) pi The orthonormality condition is (cf.120).11. Then.1 . *L\(z) = .123) For the purpose of clarity and comparison with the polynomials of the other definition it may help to see that here in particular % ( « ) = !. i L1(z) = l .
Hence one defines now as associated Laguerre polynomial L£ and as the generating function g(p) of these (Cauchy's residue theorem requires the coefficient of 1/p): L b(t) •= — <£ 27T.) 2 J J p 6 + i ( l ." b + 1 e~ptdp = coefficient of pb in g(p) := (p + \)a+be~pt. We demand (to be considered later for integration contours) the following condition. s bK> (lq)°+i (11. Then the preceding two equations (inserted into the original differential equation) leave us with I eptdp d {p(p + l)s(p)} . since for n > b : (b — n)\ —• oo.p ) o + i 7 t = 0 . Then dp = dq/(lq)2.119) with a and b as there.p ) « + i q c + i ( i .222 CHAPTER 11. Case (1) with p —• oo implies y(t) ~ / dppa~1e~pt ~ t~a.p + 1 = 1/(1 . this is the rejected irregular solution. (2) from —1 to oo.] from the above equation: C(p) := p(p + l)s(p)ept = 0.124d) In order to obtain the orthogonality and normalization of the associated Laguerre polynomials. 2TT_ Jq=0qb+l(lq)a+l 1 . case (3) similarly. (3) around p = — 1 if a + b is a negative integer.. we insert this contour integral representation of the polynomial in the following integral and obtain __?(_) = P da <* . The Coulomb Interaction since d(p2s)/dp = ps + pd(ps)/dp. (11. the only possibility is the last. 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.118b)) Ll{t) = = coefficient of pb in ^ (p + l ) a + t > n=o £ ( ~Pf" "" coefficient of pb~n in (p + ! ) « + ' ^ = /g <*>" (a + 6)! n! —(.li I u {p(p+l)s(p)} : p(p + l)s(p) dp a + 1 b p+ 1 p(p + 1) a + b+l p+ 1 b p Integrating the equation we obtain s(p) and hence y(t): a+b+l^b ln[p(p + l)s(p)] = In ip+ir+o+'p (p+l)a+b f (p+l)a+b The condition C(p) = 0 now implies for possible contours of integration in the plane of complex p: C(p) = p(p + l)s(p)e'pt („ _i_ na+b+i = ^ ' e~pt = 0. (11. p° Possible paths satisfying this condition are: (1) If b < 0 from 0 to oo. agreeing with Eq. which removes [. t > 0. Case (2) implies the exponentially increasing solution ~ e + t ./ ( l — <?) —— = G(t). Hence the expression in square brackets vanishes.( a + l ) p s ( p ) + 6s(p) dp 0. with requirement of a finite solution. . j p = (P 0 "V?. n! (6 — n)!(a + n)! An alternative integral representation is obtained by setting p = q/(l — q) or q = p/(p + 1). p and one has (as polynomial of degree 6.^ . Thus in the case of Eq.q) and P~ t ( ..a = 21 + 2. It follows that 1 d i I _i. and (4) around p = 0 if b > 0 and integral.
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. /•OO / Lg(t)£?(t)taefctt: (a + c ) ! .(x). 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. The wave function of the state (n. I.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.111). where Fnl(x) = yi(x) (11. (11.6.127) . and 0 if cj=b.—i. Clearly only the second form is sensible. is introduced for dimensional reasons. (11. and obtain r taet[P/(iP)+q/(ig)+i]dt Jt=0 With this the integral I becomes = . .125a) and the factor a~ 3 ' 2 with a = Bohr radius.—: <f h_nJ_[ dp = (a + c)\ T^ if c = o. Jo (11. Eq.?). 11. so that I = (c + a)! 1 f . m) is 3 2 &nlm = a~ / NnlFnl (—^Y^O.112).3 T h e eigenfunctions According to the previous discussion we obtain n2 orthogonal eigenfunctions in association with the eigenvalue En.c (al)\ d(col)! c(c + a)\ c!a! with use of the inversion formula (11.11.
2Z . and for the explicit derivation of the second case below Example 11. 11. This behaviour is indicated in Fig. 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 + /„.3fc + 2).8):* OD  <n!>3 (nk)\ (nfc) (n\)3 i£ n (3) = _ . a (r)is = ~a.v (6n2  6nk + A:2 + 6n . = 0) we have ) r /u =.128) These expressions permit us to obtain the mean or expectation values (1/r) and (r).2.3(2Z + 1) + 2} 2 {2(n + 0 . For the ground state (n = 1.n+l na l2 ° ^ 2 na n+L+lV) (2n + 2/ .! ] + (21 + l ) 2 .129) n2a' Analogously we have roo j = p2 I r n \ 2J ° J0 'V^yr nl _ na J^° xF^(x)x2dx _ na f^SJn+K 3 ) \na) na {6(n + Q2 . 2 (11. thus on the average the electron is the farther away from the proton or nucleus the larger n is. (11..1 + 1) (11.130) We see that the mean value of r is the larger the larger n is.224 CHAPTER 11. The Coulomb Interaction One finds in particular (see also Example 11.131) . + .(l) n+l.(2Z + 1) + 1} = ^[Zn21(1 + 1)].6(n + Z)[(2Z + ! ) . (11.7.
133) for large r.J n 1s 2s 3s *r Fig. v ^ V2n + 1 (11. \f{r ) ~ n a.l ) n ] =  [ 2 n 2 + n] an[ n + (11. = l a nl n N'nll rFnl. Sneddon [255].( n . i. In the Kepler problem of a particle of mass mo in Classical Mechanics with the Newton . we have and tp becomes particularly simple. S.2 1 A r = V (^2) . or (r ) ~ n a 2 2 ^ [ 3 n 2 .6 The Discrete Spectrum of the Coulomb Potential r. In this case (r) = = For (r 2 ) one obtains < r 2 ) = n 2 ( n + ^ ) ( n + l)a 2 .134) *These are taken from I.20.11. / ! J — . 11. pp.2 I 3/2 Kl r 225 \2 1 R„.(r) = x W 2 n + 1 =. This corresponds to E ~ — e2/2n2a like classically with circular radius n 2 o.j . In the case in which I assumes its maximal value n — 1.e.132) (11. Problem 5. 173/174.2 Hydrogen atom wave functions. with n +1 replaced by n and 21 + 1 by k..^ We obtain therefore for the quadratic deviation (r) .
Thus.4 the relativistic equation in cylindrical coordinates. r. determine the energy E by evaluating the BohrSommerfeldWilson integral of "old quantum theory". the states Ytl are predominantly concentrated in the xyplane). in which periods are obtained from the BohrSommerfeldWilson quantization condition.r. However. i. for these values of I it is not distributed with uniform probability over the spherical surface. relativistic corrections are of the order of v2/c2 ~ 0(En/moc2).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). Thus we have to evaluate the following gravitational potential written V(r) = —mofi/r. Since 6 is a cyclic coordinate.— > where ampler is the length of the semimajor axis of the elliptic orbit. In order to obtain an impression of relativistic corrections in a simple context.226 CHAPTER 11. = ngh. pJ = —— and j> Pidqi = nth. The above treatment of the hydrogen atom has obvious shortcomings: (a) The hydrogen atom was treated nonrelativistically. the electron remains practically localized within a spherical surface of radius (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 . identifying yj (r2) with aKeplen w e see that T K ep l e r = ^(r2)^ = ^(n2af/2 = *^a3'*n*. . but rather like classically in a plane (i. Solution: We have H = yjm%c4 + pIn cylindrical coordinates. \/v vT 7 v^ See also Example 14.3. This means.e. p2 = p2 + — pg • eV. we have pg = const. The Coulomb Interaction or TT = ^ = (11. and separating this in cylindrical coordinates r. 6. we consider in Example 11.135) This expression becomes very small for large values of n.6.e. i = 0. (c) The nucleus was treated as pointlike and not as something with structure. Example 11.
6. The deeper reason for this are symmetry properties of the problem. z=^rj).3. Schiff [243].136a) (11. (and subtract moc2 to We then obtain with a = e2/hc ~ 1/137.(p . 86.* See L. where ip = 0.2. (11.d r Kmlc2 + T U^\±AnlK2^Ur 2 2 \ c / r * + * ! + $*—». 2 2 Therefore the energy is no longer a function of n = nr + ng. In the case of degeneracy with respect to the magnetic quantum number m (as in the present case) one can find linear combinations of hyperspherical functions Yj^(9.T). r\ = r ( l — cos#) = r — z.V .11. Thus E2 2EeV(r) e2V2(r) nrh = a> prdqr = ilc2 . which correspond to a new choice of polar axis. ..4 Hydrogenlike a t o m s in parabolic c o o r d i n a t e s The Schrodinger equation of a hydrogenlike atom can also be separated in parabolic coordinates £. as also indicated in Fig. y=y/%qsanp.9. Whenever there is degeneracy. p. where the constants are identified by comparison.136b) £ = r ( l + cos#) = r + z. where r = ^/x1 + y2 + z2. which are related to Cartesian coordinates in the following way. ( v c + i ) . y — 0: V ^ c o s tp.6 The Discrete Spectrum of the Coulomb Potential 227 integral which we achieve with the "method of poles at infinity" explained in Example 16. but depends very slightly on ng alone which gives rise to the "fine structure" of hydrogen lines. It follows that (with h = 2nK) Z2e c h?ni 2 2ixnrfr = — 27r nehtjl  Ze2E >^m2c2E2/c constant'. 11. (p). <p = <p. 11. it is possible to separate the Schrodinger equation in other coordinates. ip are the spherical coordinates. the "fine structure obtain the ordinary nonrelativistic energy) Z2ei c2h2[nr + ne^jl 1/2 E — mac + 1 Z2ei/c2h2n2] 1/2 VTIQC 1+ •+ ng^la Z /r.
137) One now sets the total wave function ip = £(£)!N"(r7)$(</?) and divides the equation by ip. In terms of the coordinates (11. and r\ = const.228 CHAPTER 11.136b) the Schrodinger equation of relative motion (11. 2 . 11. = 0. . (11. one obtains the following two equations: £ ) + TJIQE A m2n  « ' m = o. Then 1 d2$ izimip $ oc e $cV = —m where the separation constant m2 must be such t h a t m = 0 . d_ dirt) + TUQE fm0Z±Z2e2 2h2 V ^2 A V m (ri'N) Arj1 2n Since these equations are alike. a separation constant.3 — h2 ( 4 r d (rd^\ d_f df\~ + 1 d2xl>\ £?7 dip2 2Z1Z2e< Z+V •ip = Eip. for uniqueness of the wave function. Multiplying the remaining equation by (£ + r /)/4 and setting the £part equal to —A. so t h a t the variables can be separated. We rewrite the first of these equations with the help of the relation M (w ^  m + e) .22) becomes — as we show in Example 11. . . with zaxis as axis of rotation. 1 . The Coulomb Interaction Fig. there is really only one basic equation to solve here.3 Parabolas £ = const.
we set (since E is negative for the discrete spectrum we consider) ITIQE =*"• H J?P ^( <9p: 1 / 2 i m0ZiZ2e2 A = n". .114b). Adding now / . In the spherical polar case we had (cf. Eq. . (11.106) and (11. p = /?£ and cr = /^r?.' / 2 p m / 2 L 2 Z + i = m ( / 9 ) ) Analogously we have N oc eCT/2<7m/2L™ (a). . . We leave the calculation of the degree of degeneracy to Example 11. (11. . 1 . ds2 = dx2 + dy2 + dz2. . m0Z1Z2e2 ^ 2 = n (say). as well as the Laplacian A = V and the Schrodinger equation. i. " . .o. T h e n by comparison with the spherical polar separation the answer is as in Eqs. from Cartesian to parabolic coordinates.i n +n = n i + n 2 + m + l = we obtain the energy eigenvalues given by E = n i = n' \(m + 1) = 0 . ph h2p2 _ h2 m2Z2Z2eA hn 4 2 _ ~~ m0Z2Z2e4 2ft2 n2 ~ 2m 0 ~ 2m 0 This agrees with our earlier formulas like (11. E x a m p l e 11.5: The Schrodinger equation in parabolic coordinates Perform the transformation of the metric. 2 . 1 . n 2 = n" . T h e n the first of the two equations becomes 1 £) n' m2 — 1 4 p + 4p" ( P 1 / 2 £ ) .6.e.11. 2 . .105)) Hence we write I = (m — l ) / 2 .± ( m + 1) = 0 .125b) (dividing by p 1 / 2 ) £ oc e .6 The Discrete Spectrum Then of the Coulomb Potential 229 A 22 (£ 1 / 2 £) + moE 2h2 A m2 — 1 +£  4£ 2 (cT1/2£) = 0. . . To obtain a closer analogy with the separation in the spherical polar case. .
> . all cross terms cancel in the sum of 7 the squares. + l 4 {^^)drl f i + v \ . \— < / — cos ipdr] — \/£r] sin ipdtp. Eq. Solution: For the wave functions ip = £(£)J\f(r])&(ip).230 CHAPTER 11. dx = d{ y £77 cos ip) = .E E n l „ .2d? dr)}. We then sum all possibilities and select the coefficient of ojn~1 in the result. We use the method of selector variables. .2 = n — 1. etc.e. {g^Y. 712 and m which give n. for m / . Thus dz2 = .e. i. . (11. 2 _ = (9id02 + (gvdvy + /„ . z = (£ — rf)/2.g<p 9 \ .2)d/2. and one obtains 2 = _ 1 4{ £ ) * (t + 7 )\ j.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. 0 there are two. n the double sum ]T]2 is TI — m. Magnus and F. (11.l)a + (n .l ) ( n . W. / — cos ipdl. Here we have (see preceding text) 711+77. the total number of wave functions we have is the number n with m = 0 plus twice the number with m ^ 0 starting from m = 1. last chapter. The Coulomb Interaction Solution: We begin with z of Eq. and so dz = (d£ — drj)/2. " . / J„^2 .?e 9£ J 4 t d \ g^gv 9 \ dri \ gv dr] J t d \ g^gr. are elements of length. 'Recall that the sum of an arithmetic progression is given by a + (a + d) + (a + 2d) + • • • + {a + (n . In the case m ^ O w e have n\ + 77.2)d} = (n. Beware! For m = 0 there is only one mstate.2 . Thus we want to obtain the coefficient of wn~1 in £2 . when m = 0 = 77. and attach a factor to to each term introduced in the wave function. 2 + ^dV . (11. N2 where g^dS.g. for instance. as t h e two signs of t h e exponential indicate.1) ~ .. The general formula for the Laplacian is§ ^ follows that 1 [9£\ A = ' 9 \ gr. (n _ !) n = n2 §See e. . Similarly we proceed with dx and dy.tx2 . i.2 = n — m — 1 and therefore the number of the coefficient of u}n'm1 . </. = e±im^e("+^'2{par'2L^{p)L^ (a). 2 2V? V^ One now performs the square of this and obtains similarly dy2.* 71 — 1 n—1 n— 1 n n+ 2 E ( n ~ m ) = n +2 E n _ 2 E m= n + 2n(n .[df + dr]2 .136a).114c).. n 2 (1UJ)2 The coefficient of a .1 in this expansion is seen to be n.2) and summing over m. Thus.e. see Eq. 9 \ 9<p\ gv 9f\ _ 1 a2 a / a\ a / ay £77 dip2 E x a m p l e 11. i. Observing that the highest possible value of m is n — 1 (i.e. We consider first the case m = 0. Oberhettinger [181]. Since the lines of constant £ are orthogonal to lines of constant 7 (they cut perpendicularly).136a). we have to find the number of values of the quantum numbers n\.
Apart from the power of the leading factorial (which results from the definition of the associated Laguerre polynomial in the generating function).9) one finds that _ = %2 jm+2 {n2+m)\ (ni + rre)! +1 : > in. However. 2(ni +712 + TO + 1) Therefore the change in energy of the state to first order is given by.r.™(p)] 2 By contour integration (see Examples 11.8 and 11. fc — m. 8. { 6 ni2 + 6 n i ( m + l) + (m + l ) ( m + 2 ) } . if we expressed the perturbation in spherical polar coordinates. with RayleighSchrodinger perturbation theory). .1 (cf. Z2 —> e.128) if one makes there the substitutions n —> n\ (or 712) + m. where F is the electric field. (11. i. with the connection » (11. i. .eF / nh2 \ {v)=3(n1n2)^—lr2j Thus in this case nondegenerate perturbation theory could be used as discussed in Example 8.127) and setting n = r + k. .e.124c).6 The Discrete Spectrum of the Coulomb Potential Example 11.6). we would have to use degenerate perturbation theory and obtain the same result.128).a). L ni\ ny. (11. with z = r cos 9. we can relate the expressions Iq to those of Eq.•* n (l) = / Jo It follows that . 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..drid<fi oc (p + o)dpda. eF{z). and determine the average value of v for a given state. which will then give the change in energy of that state to the first order (i.e. Thus e. with Z\ —> Ze.g. {p — a) — = dxe~*xkvLkn (z)] 2 . with Eq. Solution: Using formulas and results of above we have to determine (v) = eF{z) = leF(e.e.—r~n = 3(mn 2 ).) = ^eF(p . Sec.11. Ikr+k. ( ! ^ ± ^ ) l { 6 n 2 + 6 n i ( m + 1 ) + ( m + 1 ) ( ? n + 2)} ny. . In fact.r+k(l) = / Jo e~*xk[(k + r)\Lk(x)f = [(/= + r)\flk. We obtain the volume element dV in parabolic coordinates from the above results as dV = g£gvgvd£dr)dip = — (£ + rj)d£.(p)L™ (<7)]2(p + a)dpda rev where JJ = / Jo dpe~ "p"[I. s 0/ . = j (2ni+m+l). I. With the help of the wave functions we derived above.7: Stark effect in hydrogenlike atoms 231 Consider a perturbation v = eFz. (11.{ n i {2ni + TO + 1} + {ni j=± n 2 } 6(nin2)(ni+n2+TO+1) ^±n2} ^7—. these expressions agree with those of Eq. .
0 ° e *[l+P/(lP)+9/(l<j)] t m+l di) t  P  9 l [ ( l . tt„:_ (1 ..124d) of the alternative integral representation obtained in Example 11.128).8: Laguerre linear expectation value integral Evaluate the following integral (which is different from the normalization integral of Example 11.3).l)!r! X We now evaluate the integral / as the coefficient of pnqn _ ( m + l)! [ ( l.b.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 .9 we consider integrals with an arbitrary power of t. (11.a. and is a special case of Example 11.9 )« Solution: Using the generating function for both associated Laguerre polynomials. the integral arising here becomes with Eq.9: Laguerre expectation value integrals Evaluate the following integral with the help of the generating function of Laguerre polynomials: /•oo rSc(i):= Jo dte[Ll{t)Lac{t)tae\ i.1)! n\ This verifies the expression I™t in Example 11. (11. 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. E x a m p l e 11.( 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)!. dt = ds/a. In Example 11.g )(l. ) (l_pg)(m+2) Now the coefficient of qn in (1 — g)(l — p<j) .9 ) ] m +h 1 yJo 0 ' 1 —pP 1 q 119 lpq _ (l9)(lp) = a With (see Eq.n+2 in the expansion of i n c o e f f i d e n t o W (m + 1 ) 1 ( 1 . 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 .p ) r + 2 [(lp)(lg)]"l+1(lpg).1 ( n + m)!"[ _ (n + m + 1)! (niy J n! (ra + m)! _ (n + m)! (n .p n _ 1 ) in (n — 1)! J p"(n + m + l)! n! p n . _ „ .p ) ( l .232 CHAPTER 11. .l ) ! ( m + l)! in the expansion of p " ( n + m + 1)! p " " 1 ^ + m)! 1 _t .p ) ( l .3: /"OO I := Jo et[L^{t)]2tm+1dt.3 by one power of t. L™(t) = coefficient of qn in G(t) e tq/(lq) (1 . c positive integers.1 ) ! (n .pK — — —L . the integral / is the coefficient of pnqn in the expression _ / .9) with the help of the generating function G(t) of associated Laguerre polynomials obtained in Example 11. s = at.^~T^——^ ) = coefficient of (pn .x ( n + m)! ( n . (n + r .
c + b even) and/or j / = 0.r ) ! a ' . we have 2r = 6 + c = even. = c (q + fe)! W ° (^)!(^)!(^)! ~" (2) In the case of one power of a.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 .e.w(r) = and / a ) = r\(P ~ r .6 The Discrete Spectrum of the Coulomb Potential Solution: One considers the following integral and obtains /£. so that = (q+fc±£) .(*) from the coefficient of a1: />00 233 1(a) = / Jo dte.e. c + 6 — 2r = 0 (i.™( r > (1 + a)P r £^0y^w Jc+6_2r = .e.p) _lpq It follows that (using Cauchy's residue theorem in integrating around q = 0) 1{a) = (2^ / / ^ ^ [lpq c [ 1 . In case (a) we have u \ 7 V^ ir fb + c\ 7(a) = Jo / Tli u.1 + y w)\(a)y (r — w)\w\(j3 — r — l)!(j/ — w)\ Special cases: (1) In the case a = 0. where (a+**£)! c j n Jo = . ° ^ + a ( l . Using the contour integral representation (11. i(a) = / Jo dttaet{p/(lp)+q/(lq) + l+a} = Q . (lp)(lq) + a(l q)(l .Pq + Q(1 " 9)(1 " P)1_a_1 = ti f p^ X COeffident f ap)]"01.p) .?(t)t a e. we have to have y + (c + b — 2r) = 1.f ].p ) ] . we can reexpress 1(a) in the form: /(a) = (2^)2 J J p t + l ( l .L. i.e.(i) = i ! ( . c + b odd).p ) 0 + l 9 c + l ( l _ g ) a + l ^ a ) ' where j(a) contains the integration with respect to t which may be evaluated with the help of the integral representation (11.l ) * x coefficient of a* in the expansion of 7 ( a ) .a _ 1 = [1 + a ( l .11. — (cr)!r!(6r)! Kv.^ ( l . i.q)(l .124d) of an associated Laguerre polynomial. . (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 «. .111) of the gamma or factorial function. c + b — 2r = l (i.a t [Lg(i). 7£.
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. 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.138) " 2 ^ =2 ° ' 2 m Uo 7 = ZtZ2e2 ^T (1L139) * Screened Coulomb potentials are considered in Chapter 16. is the work of J.7. the standard principles of nonrelativistic scattering theory. Zie given by h2 _ A _ ZiZ 2 e 2 V>(r) = Ei/>(r).)!' f^)(.~* 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 .^)(. Since such a rigorous treatment is beyond the scope of this text. R. . . + (!±<)«. In view of the slow falloff of the potential at infinity. cannot really — and this means in a rigorous sense — be applied in this case. b\ V cited in Example 11. however parallel to the Coulomb case. ^A very readable source to consult. Taylor [267].234 and (only w = 0 and 1 arise in the sum) CHAPTER 11. One should actually consider a screened Coulomb potential* and consider quantities like the partial wave expansion (to be defined in a later section) as distributions. like the asymptotic condition and definitions of the scattering amplitude and the phase shift. also with regard to earlier literature. These results are seen to agree with the expressions 1^.1%^ 11. The Coulomb Interaction b + c\ 2 J (b¥)\(Pb¥w)\{a) ($±£ w)\w\(J3 ^±£1)1(111. t We have the Schrodinger equation of relative motion of particles with charges Z\e. 2mo Here we set (this defines the velocity VQ) j? E ^ 1 (11.
According to Eqs.138) becomes 235 ^A + k2 .e. where ^. for r — oo) • ipr = Aeikzf(r .. . for instance.141) Inserting this expression into Eq.11.M) = l + .140) The simple Coulomb potential permits particularly simple solutions. „ az 1.87 under Kummer functions.142) This equation is of the form of a hypergeometric equation. Our question is now: How does ipr behave for r — oo. (11.b. with the asymptotic expansions (11. a(a +1) z2 ] (11. b.z) = Wi(a. W.z). that we encountered previously. (11.145) U ' ' j " T(6a)1 J ^ 71=0 r(a)r(a6+l) n! ' (7T < axg(z) < vr). z) for \z\ —> oo? We can find the appropriate formula in books on Special Functions:* $(a..+ ^  + . z = rcos<9.109) we can therefore write the solution ipr = Aeikz$(i>y. Magnus and F.140) and using (11.z) + W2(a.7 The Continuous Spectrum of the Coulomb Potential Then Eq. i.144) *(a.143) and v = ikrj.z)). (11. what is the > behaviour of 3>(a.^ W ) = 0. v = ik(r .108) and (11. (11. A = const..e.z). (11. writing Vv = Aeik^v)/2f(ri) we obtain the equation d2 x d f(v) = 0.b. In particular the equation possesses a regular solution of the form (also ip ~ exp(ikr). pp. (11. Oberhettinger [181]. . . (11. 86 . (11.136b).146) See.b.. i.
= r cos 0.e.148) .151) exp[—ryln fc(r — z)] — exp[—ijIn = kr(I — cos 6)) exp[—ryln 2kr — ry In sin 2 (0/2)].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) . > A = Aeikz[W1(i1.l. i.ik(rz)) where ^ ~ Aeikz——^—f't^ e ife(rz) + W2(n.7 ln*(rz)) k(r .^l*7 l + O (11.e.) of a particle with incident momentum hk in t h e direction of z for large distances away from t h e scattering centre. Without t h e logarithmic phase factors.(11.a) (z) na)"~ r(la) T(ba) n\ ' (11.152) Since z = r c o s # .ik(rz))} =^ + ^d.z) + la)T(n + b.i 7 ) (11. The Coulomb Interaction W2(a. if tbr were „ikr ibr ex el"z + f{6) r (11. J\") ci(kr~/\n2kr) (11.153) The asymptotic solution represents t h e stationary scattering state (E = const.z)T(l .kz+l^ktrz)) C .147) (—7r < arg(z) < 7r) It follows t h a t for r — oo. (11. A(j)11 A where \ " r(i + \i) 1 j(kz+"flnk('rz)) 7 f ( l + Z7) _ e i(fcr. a comparison of Eq. with 2.l. We now define a quantity f(9) by t h e asymptotic relation lb O < ^(.149) (11. (11.236 n\Zzab CHAPTER ^T(n '£ fo 11. (11.150) ^(rz)]1"^ 1 + 0 T(«7) i.154) .151) with Eq.b.
2im. also called phase anomaly. 11. (11. the effective range of the Coulomb potential is infinite.n .11. 11. The problem of the logarithmic phase. The quantity /(#) 2 would then be a measure for the scattering in the direction 9. z — — oo.155) Since the gradient is proportional to the configuration space representation of the momentum. the current density is * h . In other words. This cross section is defined by the ratio of the outgoing particle current (in direction 9) to the current of the ingoing particles.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 Coulomb potential has an influence on the incoming wave even as far as the asymptotic domain.1 The Rutherford formula The above considerations permit us to calculate the differential cross section which is a measurable quantity.7.156) .4 Variation of the observed differential cross section with 0. that we encounter here is a characteristic of the Coulomb potential. For the incoming wave ipi = exp(ikz).Q hk mo v0. We define as current density the quantity h J := 2imo [^(v^)v(v^n. As a consequence of the masslessness of the photon. we see that this expression is the exact analogue of a current density in classical considerations.. do dQ K 12 K Fig. (11.
159) as the Rutherford formula which can also be derived purely classically. 'screened ZlZ<2e2 cr/r0 ° .e. That this formula retains its validity here in quantum mechanics is something of a coincidence. The Coulomb Interaction In the case of the Coulomb potential we now ignore the logarithmic part of the phase (see discussion later).154) we chose the prefactor of the incoming wave in (11. Correspondingly the other part of (11. We see that a depends only on the modulus of the potential (i. We make a few more important observations. We recognize the result (11. is called scattering amplitude. 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.238 CHAPTER 11.e. In reality this divergence does not occur. thereby screening itself off from the surroundings.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. which yields the scattering cross section.153): 4fc2 s i r  The expression /(#). since in actual fact the Coulomb potential occurs only in a screened form.152) yields as density of the outgoing current Jr = ^\f(0)\2vo(11.152) as 1. In Eq. i. (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. This implies effectively that the Coulomb potential becomes (virtually) a screened potential or Yukawatype potential of the form V . in the current this would in any case contribute only something of order 1/r. (11. the square 7 2 ). This implies that scattering takes place for the attractive as well as the repulsive potential. for 0 —> 0.
of the equation r fi. m0 m0 and V.11.i i f (''°>/Vk(x.yJ ^ j A k o W 727)3/2 e 0. which in the case of free motion (zero potential) are simply plane waves. t 0 ).e.165) ^ ( x ) = £ fe Vk(x). i. (k). 11.161) where Ak0(k) ~ A(k — ko) and ^ k (x.2 fr2 . v0 = . 11. t 0 ) .8 Scattering of a Wave Packet We saw earlier that in quantum mechanics a particle is described by a wave packet. t) = U(t. of continuum wave functions. (11163) (11.k(x) is a solution of the continuum with energy h2k2 E = Ek = — >0. be given by ^ f dk AL.^ ^ % ( x ) .8 Scattering of a Wave Packet 239 This screening of the Coulomb potential insures that the scattering cross section in forward direction is actually finite as observed experimentally and indicated in Fig.t) = e . t0) = e. t 0 )^k(x.t). The wave packet is a superposition of waves ip(x. (11.166) Let the incoming free wave packet at time t > t$. and which provides the uncertainties arising in quantum mechanics.e. 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. The particle velocity v is v = — . _ x o _ v o ( t _ toh io)_ em0^VoKtt0)^o)ix (1L167) 0 l/> () ) fxXn t)~ gik(xxo) . the wave packet in the absence of a scattering potential. whose maximum moves with the particle velocity.e. i. 2mo V + F(x) k (11. (11. i. and if H is timeindependent one has the stationary wave function ^(x.*'^ e1 piEk{tto)/h (x x i)= (2^3^ ( k ) roT ^ k o l * . / dk 7^340(k)4(xxo. 2 (11162) where Vk(x) = ^ ( x .x° ' r j .4.t).164) Ek~Eko + {kk0)v0h.
can be written A (X) = ( 2 ^ + W J ^ ^ ^ ( ^ ( x ' )  eikx 2mn f P»fexx' (11168) We saw earlier (cf.170) A ko (k) e .163) we obtain </>ko(x .154)) that for x — oo. the stationary scattering wave.171) The phase of fk(0. ip) could be carried along but we ignore it.10 we show that the solution V'k(x) of Eq. ^k(xxo) ^k(x) ~ = e. (11.xo.ip) is the scattering amplitude with respect to the incoming plane wave.t) = J ^ A k o ( k ) e .iB *(**°)/ R °'tj + 1 pikr e * .173) .169) The wave packet in the presence of the potential is obtained from the free form (11.ko) k0 (11.x + _ / k ( e j ¥ ) ) ikxo (2TT)3/2 zkxo = <>(* r y (2vr)3(27r)3/2e M ^ ) e *o)/^] e ikx 0 (11.161) and (11.i) dk (2TT (11.167) by replacing the plane wave exp(zkx) by this scattered wave. The Coulomb Interaction In Example 11.167) we have ^k0(xx0. Eq.k0) ~ k0 + and the expansion of E^ in Eq.x 0 ).ik . this solution > has the following asymptotic behaviour in which f(6.k 0 ) 2 ~ Jk20 + 2k0 • (k . From Eqs.165) ko • (k . (11. k = kez. (11.e.e.t o ) / V k ( x . Then with k = ^(ko + k .<p) + Ol Akr l (11.166). (2TT)3/2 eik* + —fv(P.169) and (11. (11. i. (11.172) h h k0 {rv0(tt0)} (11. i.x °Vk(x).240 CHAPTER 11.^ ( * . With Eqs.
however. *.167). T h e scattering amplitude is t h e same as t h a t given by Eq.y)eiko'Xoe'[fcor£. t) . .^y /k0^. t) f.174) r /k0(^.10: Schrodinger equation as integral equation Demonstrate the conversion of the Schrodinger equation into an integral equation. 11.11.^e (2^3(27r)3/2e y (27r)3(27r)3/2e (11. r 241 (0 LO) f ^ Afc o(k)rikxnri[fcor£fcoato)/ftl i(kko)x' = /k0^.5 Scattering of a wave packet with scattering centre O.r')V(rW).z Fig.*0ko (x . A particular solution of the Schrodinger equation is — J G(r.x 0 .e.171).154). the solution can be written iMr) : (2TT)3/2 ^dr'G(r. E x a m p l e 11.8 Scattering of a Wave Packet so that with Eqs. r ' ) = 47r<5(r .to) has t h e form of t h e wave packet in radial direction as indicated in Fig.(r) 2rrao V(r)V(r) 7/>(r) (2TT)3/2 ^dr'G(r. i. r ' ) by the equation (A + fc2)G(r. U(T) = ^ V ( r ) . Solution: We define the Green's function G ( r .fco(tto)/ft]^0o)(x/xo.r').(«)/ ipko(x . '. (11. (11.r')l/(r>fc(r').to). Here ^ (x' — xo. (A + fc2)t/.r')U(r')ip{r')dr' /4T7 (as we saw earlier) .r ' ) . 11. and determine the Green's function G(r. (11. Adding a suitably normalized solution of the free equation.x 0 . with the energy of the maximum momentum of t h e wave packet.5.
fc ) oo dfc' 1 d nik r 7rr dr / . G+(r) 1 d dk' lim e>0 TTT dr Jc+ fc'2 — (fc2 ~ik'r 1 d 2ni nr dr \k' + k+ ie/2kj ik r J__d_ 7rr dr 2ni N ^ residues k. These are the socalled outgoing and ingoing Green's functions.O O ™ fc2 (e*fc'r _ e ifc / r 1 fc ) =  / fc'dfc' i(fc' 2 fc 2 ) Since this integral does not exist.6. displaced slightly away from the real axis as indicated in Fig./2 dk'd(—cos 6)dip iki.Rek' kfc/2K Fig. dk' ik'*r 2TT2 fc'2 . Imk .242 CHAPTER 11. e > 0 small.=k+ie/2k 1 d f.fc2 ' 2 ^ i fc72^"fc2 Integrating out the angles we obtain 1 G roo />7r 2n fc' rZn j. Thus we consider the contour integral taken around the contour C+. 9{k') = 1 1 S(T) = j G(r) 1 ^ A j dk'e i k ' r . which is such that the contribution along the infinite semicircle vanishes. we relate it to a properly defined contour integral with the two simple poles at fc' = ±Vfc 2 + ie = ±(fc + ie/2k). « = 52 / J _ Z100 7rr 7o / A fc'dfc' 2 / 2 • fc2 (fc' *(fc' . one obtains another Green's function given by G_(r)= r eikr. r ' ) = G ( r — r ' ) with the Fourier transforms G(r) = J dk' 3 (k')e ik ' r . Hence we write G(r)G+(r). . The Coulomb Interaction We find a set of Green's functions G(r.6 The contour C+ in the upper half of the k'p\ane.e \ r dr\ k ikr 1 ikr 0 With the choice of an analogous contour C— in the lower half plane. 11. 11.
178) (11.145)) F(a.179) = = elkr{kr)l+1F(l e (kr) ikr l+1 + l + ij. b. z) = W^a. under the stated conditions 2/07 = Z\Zie2 = —Mo. (11.140) (previously we solved this only for the bound state problem) § . e.£).21 + 2. (11.9 Scattering Phase and Partial Waves 243 11.g. (11.z2 27fc 1(1 + 1)' yi = 0.0 + W2(l + 1 + iy.11. Thus we proceed as above. we define first the scattering phase for the case of the Coulomb problem proceeding from our previous considerations. where Mo is a parameter in Chapter 16. b. 2ikr)] [Wi{l + 1+11.100) — 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. Eq. 21 + 2. whose continuation to infinity is again given by (cf. the logarithmic phase as a consequence of the slow decrease at infinity and a special symmetry which permits separation in other coordinate systems.176) " 0—A ~ (I + 1 + iy)<t>i = 0.e. If we proceed from the radial Schrodinger equation. i. . z) + W2(a. (r) (11177) Similar to above we obtain a regular solution y) ' with 0j = F(Z + l + i7. Then £^<fc + (2l + 2 £ = ~2ikr.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. i. z). (11. Then yW (11.2l + 2. b. (11.175) yl' + k Here we set (as in the case of the hydrogen atom) yi = eikr{kr)l+lUi).2Z + 2. Before we introduce the general form of the socalled partial wave expansion. Eq.e. 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.
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 .244 CHAPTER 11.180) ^ ' (11. T(2l + 2) (2fcr)*T« (Alliy ikr .n) • e „—ifer+i7ln2fcr _eMr(i7Jl)/2 r(z + I +17) T(2Z + 2) /2 • ikriy In 2fcri7r(J+l)/2 r(z + 1 .ry) = e"^r(Z + 1 + ry).j syyll iy il( _ . (2kryr{i" V 7 .' . +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) (11.183) r(Z + 1 + 17) := Reie.ii)' (11. (11.l e*"" + T(l + 1 + H ) T(l + 1 . The Coulomb Interaction Using Eqs.i 7 ) I T 2l l—X—i^j ( + 2) 2ikr( ?.146) and (11.181) so that eldlT{l + 1 .147) this becomes asymptotically (r) Vi lkr eikr(u„\l+l (kr T(2Z + 2) '—(2ikr) \T{1 + 1 .«7) „ .i 7 ) „ikr—ij\n2kr T(2l + 2) e e* jTv(lliy)/2 2l+i —ikr r(z +1 . Then e2i5< = Reie Re~ J2i9 9 = Si. Then (r) r—»oo Vi 7T7/2 T(2Z + 2)e*7'*e i8i 2'+1I\Z + l + i 7 ) i5. .
a positive integer.9 Scattering Phase and Partial Waves Hence 5j = argr(Z + l + i7). (11._W2)_ ( 1 L l g 6 ) The factor e2lSl between the Jost solutions (in the regular solution continued to infinity). (11.114a) reveals). .188) in the form y/E. i. i..e. i. One defines as regular Coulomb wave Fi(j.e. 245 (11. g = e«* = rSi + 1+ *V*ri (11.e. . For instance the poles of S are given by Z + l + z7 = . i..184) The phase 6i = 8i — lir/2 is called Coulomb scattering phase.e. Analogously one defines as nonregular spherical Coulomb function or Jost solution the outgoing or incoming wave u\ . In the present quantum mechanical considerations I is. However.1. considering I —• an(E).. with Eq.11.2. where the continuum of the spectrum starts.u\ respectively with the following asymptotic behaviour: u(±) r^o e±i(fcr_7ln2fcr.n ' . (11.139) n' = 0.187) contains almost the entire physical information with regard to the Coulomb potential. kr) the regular solution with the following asymptotic behaviour: r 0 (11. Squaring this expression we obtain the wellknown formula for the binding energy of bound states (as a comparison with Eq. as we saw.185) J F} ^° sm(AT7ln2£T + ^ ) (also called regular spherical Coulomb function). indicates that — looked at analytically — the scattering amplitude f(6) possesses at E — 0 a branch point of the square root type. That the energy E appears in Eq. considering I as a function of E extended into the complex plane. The expression e 2i5j _ 2i5i—iirl is called Smatrix element or scattering matrix element.
We shall return to these in Chapter 16 in the consideration of screened Coulomb potentials or Yukawa potentials and indicate other important aspects of these.190) and the scattering solution „ikr Vwt = / ( 0 ) — + 0 V. We mentioned above that the logarithmic phase does not arise in the case of shortrange potentials. 11." These Regge trajectories can be shown to determine the asymptotic high energy behaviour of scattering amplitudes and hence cross sections.7. we obtain the partial wave expansion of the scattering amplitude as follows. 11. we obtain trajectories which are known as Regge trajectories.191) .7 A Regge trajectory of the Coulomb potential. and thus in general. The Coulomb Interaction and plotting these for different values of n. lman(E) jump from E<0 toE>0 o A LU infinity  E<0 Rea n (E) Fig. (11. For the Coulomb potential a typical such trajectory is indicated in Fig.189) where as before the incoming wave in the direction of z is ikz ipin = e (11. Singh [252]. The large r asymptotic behaviour of the solution which is regular at the origin in the case of the scattering problem can be written ^ r e g = V'in + V'scatt. (11.246 CHAPTER 11. In these cases.
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. There the expansion is given in terms of Bessel functions Jv which have to be reexpressed in terms of the asymptotic behaviour (1 2) of Hankel functions H„ ' . When we solve the Schrodinger equation and calculate the asymptotic behaviour of the regular solution we obtain correspondingly V'reg . W.193) where r\x e2l5t determines the change in amplitude and phase of the outgoing spherical wave.195) with the scattering matrix element S(l. (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 ' . r e.g. 21 .22.9 Scattering Phase and Partial Waves 247 For a spherically symmetric potential the amplitude / depends only on 9.^(cosfl) (1L194) This therefore yields the following partial wave expansion of the scattering amplitude = ^lY.r ^ V ^ Wostf).~ J > Z + l)Lil*e~ikr . The potential disturbs this phase relationship (by delay or absorption or redirection).k)=me2iS^k\ in which r\i^\ indicates absorption.196) "See e.^2l + l^S^k) lik (11. Oberhettinger [181]. 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.i f c r . (11. The mathematical expression can be found in Tables of Special Functions (there cf.eikr\P^cos9).192) Thus the incoming and outgoing spherical waves have a definite relative phase. . (11. pp.i * r 1 v^ f ne2ibl — 1 \ = k^2l W +1 \^r^)Pl{cosd)—.11. Magnus and F.
This page is intentionally left blank .
e. It is not surprising. Razavy [236].Chapter 12 Q u a n t u m Mechanical Tunneling 12. These newer methods will be presented in detail in later chapters. that the explicit derivation of such quantities in nontrivial contexts. Such a vital quantum mechanical phenomenon is socalled tunneling. M. which a particle in classical mechanics would never be able to intrude. can be found in any traditional text on quantum mechanics. In fact.* *See e. 249 .g. It is very instructive. Thus in this chapter we consider traditional "square well" potentials in onedimensional contexts. namely the instanton (or more generally pseudoparticle) method. i. in general. the computation of finite lifetimes of a quantum mechanical state. which in one form or another.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. Similarly difficult is. one of the main and now wellknown methods to perform such calculations. to explore first much simpler models in order to acquire an impression of the type of results to be expected. states of a finite lifetime. i. in texts on quantum mechanics is rare. however. that quantum mechanically a particle may have a small probability of being in a spatial domain. as — for instance — for trigonometric or double well potentials.e. was developed only in the last two to three decades. For various general aspects and applications of tunneling we refer to other literature. therefore.
2) we have = o + ~v[(vv>*)V'V>*(W)] = vj.4) (12. M^ Then dt + Vj 1=0..t) (12. i.t) that the one particle concerned is somewhere in space at time t.5) L j • dF IFo is the probability that per unit time a particle passes through the surface FQ. r (12. Correspondingly one interprets as probability current density the expression j x ( < ' >=iH v ^7 v <4 = [W{x.t)\* (i2 2)  With the help of the Schrodinger equation ihiP(x.2 Continuity Equation and Conditions The statistical interpretation of the Schrodinger wave function ^(x. (12.3) we can obtain the continuity equation which describes the conservation of probability in analogy to the case of the conservation of charge in electrodynamics.1) as the probability described by ijj(x.t).e. ihr(x.250 CHAPTER 12. . so that p = T/>2 is the probability density for a spatial measurement at time t. Quantum Mechanical Tunneling 12.t) = HiP(x. According to Eq.t)\2 =1 (12. t) was developed by Born in 1926. We assumed this already in the preceding since we interpreted the normalization integral / Jail space dx\ip(x.
Tarrach [119]. From this follows that both p and j have to be finite and continuous everywhere.3 The ShortRange Delta Potential 251 The timeindependent Schrodinger equation is a second order differential equation in x. Gosdzinsky and R. 12.33. 1 0 ~x o a= f See L. higher dimensional delta potentials do not have properties like that in one dimension. t For the singular delta function potential a modification of these demands is necessary. the problem is analogous to that of the scattering process from a potential in a onedimensional case. Schiff [243]. although in general small. the equation can be integrated at every point.12. 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. Thus it is possible that the probability for the particle to be in such a domain is nonzero. pp. as we mentioned above. If there is no absorption. If ^>(x) and V^>(x) are known at every point x. 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. where „ mpVo V(x) =+V0S(x). 2m E 2 V0 > 0. However.2a6(x)]il> = 0. It is therefore sensible to demand that V( x ) a r i d V'0(x) be continuous. even for discontinuous potentials. In much the same way that we define there reflection and transmission coefficients. since this potential implies a discontinuity of the derivative at its singularity (as will be seen below). The Schrodinger equation for this case is* ip" + [A. we can do this also here. It is therefore quantum mechanically possible for a "particle" to pass through a classically forbidden region. with regularization their study is very instructive for illustrating basic concepts of modern quantum field theory. as shown by P. 'Maybe the reader dislikes being confronted again with the onedimensional case. Such a region is for instance the domain where V is larger than the total energy E. Nonetheless. 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.2 . This effect is known as "tunnelincp. which is the case we consider here.6) . This case plays a special role. ~hr> k = n^ (12 7)  . (12. Delta potentials in more than one dimension do not permit bound states and scattering. 29 .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.
9) for ij. (12.14) One defines respectively as reflection and transmission coefficients R and T the squares of the respective amplitudes.252 CHAPTER 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.e. _ fQr x < Q^ eikx _ geikx ^=\ ~ Blkx *x . i. Consider the case of a purely outgoing wave in the region x > 0 (no reflection back from infinity). .11b) 2aC.e.' : It follows that ik .e.8) (12.11a) ikC . Quantum Mechanical Tunneling Integrating the equation over a small interval of x around 0 we see. [<//]0+ .1) . except in domains of the potential V.ikC = i.13) ~A R= for x > °(12. The continuity conditions at x = 0 are therefore for V : and according to Eq. i.12a) (12. (12. that the derivative of i/> is discontinuous there: W\U + k2 J i>dx . (12. a 2 — a 2 a — ik a2 + k2 ' . e Q (12.ikB) = 2aC or 2a(l + B).9) States with k > 0 correspond to those of free particles that move with constant velocity v = hk/mo.(ik . a — ik £Qr < { eikx a eikx •t" .= 2a^(0).e 1 + B = C. 2 (12.2 a J 5{x)4>(x)dx = 0.ik(C .: (i2iQ) Ce for x > 0.[^']o.12b) a=—^— C = a + ik and hence B = C1 Thus for i/> we obtain = ?—. (12. i.
the quantity T.i 8 ) .2 = . as coefficient of the outgoing wave.2 a C . at k2 = —a2. Proissart [223] or other books on scattering theory like R. 2m 0 (12. 'Note that the physical pole has negative imaginary part.17) da#(s) 2 = 1.e.(K)C = . Newton [219].a l s l. Thus ^ = Ce. K = a. Omnes and M. f Ce~KX c. i. See e. possesses simple poles at a ± ik = 0. z<0. and C follows from the normalization: OO A. / oo In the bound state problem the Schrodinger equation is ip" + [k2 + 2a6(x)]ip = 0. oo (12.e «x = J \ CeKX for for x > 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. poo nc s~t2 / oo dxe dxC2e~2aW = 2a\x\ _ u 2C2 / JO a Hence C = y/a and therefore ^ = VaeQK (12. . = .K C . ( 1 2 . or h2 E = —a2.K 2 < 0.3 The ShortRange Delta Potential and T= ^rr a — ik 2 253 =1R. We write the solution ^ so that ^(0+) = ^ ( 0 .g. i.15) Furthermore we see that similar to the Smatrix which we encountered in the case of the Coulomb potential.e.ip'(0) = .2 o ^ ( 0 ) . R.19) Thus the delta function potential supports exactly one bound state.) and V>'(0+) . . when normalized to 1.12. (12.
Quantum Mechanical Tunneling 12. (12. V(x) V=E a v=y Fig. . We set 2m0(E + VQ) h2 so that ip" +fc2V>= 0 ip" + K il) = 0 2 (12.22) + De Feikx + Ge~ikx *We follow here largely F. 55.254 CHAPTER 12. x G [a. 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. 12. We want to consider the case of stationary states for E > 0. The latter contains also specific applications in nuclear physics. \x\ < a. —a]. 58. Schwabl [246]. (12. p. or in the lectures of E. The Schrodinger equation is iP"(x) + 2 ^[EV(x)}iP(x)=0.4 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.1. oo].20) for for \x\ > a. p. Fermi [92]. x€[a.1 The square well potential. Similar treatments of square well potentials or barriers can be found in most books on quantum mechanics.a].
) A l ( pika p—ika pina K ina kC p—ina \ _p—ika J \ Kp—ina J ( c\ U/ (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. fc Af(a) = .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.23) Ce~iKa + DeiKa iKCe~iKa + DiKeiKa = = Fe ifca + Ge"*fca. 2 l ^i i K\ iKaika O \ (1 + %)e ina—ika (1 .ikBeika = inCeiKa + DiKe~iKa. (12.ikGe~ika. . I (12.27) v ' Similarly Eq.24) yields F ^ G.25) ^\=M{a)lCD where 1 / f l — « \ irea+ifca /i . (12.e..f )e" t\ K\0—tKa—ika . ( V C n vD ). M(a) . (12. (12.iKa ? ikAe~ika and . (12. ).28) Prom these matrix equations we obtain ( B ) = MWM(fl)_1 ( G ) ' where M(a)1 = v ^12'29) i / [i + n i K\eiKa+ika *)e (i t\ K\„iKa—ika fcje \ ] J fl2 30) • J ' 2 \ ("I — K'Jgita+Jfca Q _i_ K\p~iKaika \ .26) K^gifoa—i/ca fc. Fikeika .12.
= < Ce~lKX + DeiKX .e sin 2na IJlika e B = r)F sm2na.31) we have A AT(a) = )MFo B ^r)sin2na . (12. (12.34) into Eq.e. Quantum Mechanical Tunneling Hence with the complex quantities IK (12. (12.35) with with with x G [oo.22) we obtain ' Aeikx + AT(E) I sin 2Kae~ikx T/>(X) (12. i. the quantity T{E) := £ = A lika cos 2/ta — je sin 2/«x (12.32) [cos2«. .i^ 2 sin2 2KO 1 + ry?7* sin2 2/ta Inserting Eq. (12.e.oo].34) and as transmission coefficient (note e is complex!) \T(E)\2 = 1 .^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.33) One defines as transmission amplitude the ratio of outgoing to incoming amplitudes.a .a]. we obtain A = F( cos 2na — .—2ika [cos 2/ta + \e sin 2«a]e~ (12.36) One defines as reflection coefficient \R\2 the modulus squared of the reflected amplitude divided by the amplitude of the incoming wave.(1 + e 2 ) sin2 2na 1 .256 CHAPTER 12.a ] . x € [a. xe[a.4T(£)e ifci. \R\2 = T(E)^sm2Ka (12. i.37) .
43) Thus we set k — —i ' H . i.31): 1\T(E)\2 257 1 . cot(fca) = i— or tan(fta) = K K . (12. . However. Dwight [81]. where cos 2/ca With the relation§ cot(2«a) = [cot(«a) — tan(/ta)]. for VQ < E < 0. nor when both parameters are purely imaginary. which is satisfied if either / = g or / = — g'1.e.39) 2 i. (12. (12. formula 406. B. We see in particular from Eq. 2\k K.12. (12.\ sin2 2KQ •n 4 sin2KaT(E) — \K\ >  IRI2 (12. (12. 83.e.e.44) .e.38) as expected.40) we obtain cot(na) — tan(/«j) — i K IK . in the domain of A imaginary and K real. (12. an equation of the type f~1f = g19.34) that the amplitude F = AT{E) of the outgoing wave has a simple pole where cos 2KCL e sin 2KCL = 0. 2m0\E\ h? ' (12. (12.35) and using Eq. i. they have solutions.12. and k real.41) i. (12. Li ( — H— I sin 2na = 0. p.4 Scattering from a Potential Well On the other hand according to Eq.42) These equations have no solution for K.
(12. F. A = 0.fl = ( .34). t2 2 (12. (12. 2m0\E\ and Kcot(Ka) = y—^—L.49) Taylor expansion of the denominator in powers of (E — ER) yields .20). . "Note that dtan(2K<z)/'d(2na) evaluated at ER is 1. (12. The amplitude of the transmitted wave assumes its maximum value where (cf. = 2/rn (EER). i.45) These are the equations that one obtains from the connection relations for the case of Eq. The infinity of the amplitude of the outgoing wave corresponds to a zero of the amplitude of the incoming wave. (12.47) These states in the continuum are called resonances. . 2m0\E\ Ktan(«a) = — y——^— . Then according to Eq. t&n(2Ka)ER = 0. 2Ka = nn. Eq. The vanishing of A implies that because k is imaginary the factor exp(ikx) does not diverge exponentially for x — . 64 . (12. k\ d(2Ka) K) dE \ E R 1v/2^a 2 h 2ER + V0 ^ER~(ER + V0) r 1 (12.65. (12.35)) for E > 0 sin2Ka = 0. (12. and is itself infinite there. . T(E)e2ika = — r .46).' We observe that T(E) has poles at those values of E which supply binding energies.46) This condition implies the eigenvalues E^ER = n2^^ V0>0. .258 and obtain the equations CHAPTER 12.e.50) See e. At their positions cos(2/«z)£. „ .l ) n . (12.e. The remaining wave part becomes exponentially > decreasing. ± l .+ \tan(2Ka) where T is given by" 2 1 (K 2\k 1 (K k\ /n .c o . i. pp. Also. We now return to the scattering states. ± 2 . n = 0 . Schwabl [246].l .e. (12. Quantum Mechanical Tunneling . i.g.42) which then leads to bound states with associated even and odd wave functions. around the values given by Eq. from Eq. This can be understood.48) We expand T{E) around these energy values.
The Schrodinger equation to be solved is ^"{x) + ^(EV(x))i. . 12. V(x) x=2a x= =2a . We restrict ourselves to the period — la < x < 2a. = 0. A potential of this type is known as a PenneyKronig potential. These are precisely the resonance states. This result is physically very plausible. .12. V(x) = 9(x + a)VQ6(ax). Since we have reflection as well as transmission there are states with repeated reflection and finally transmission.' ' 5a \V / Fig.5 Degenerate Potentials and Tunneling We obtain therefore T(E)e2ika = = (^ 259 iT/2 EER + iT/2' (12. 12. a 3 a. We consider a potential consisting of a chain of rectangular barriers as depicted in Fig.' ''' 5a 3a \ a ~~~. 12.52) Here ER is the energy of the resonance and 2/F the lifetime of the state.51) i. s \ V0 /• ~\ 'X . T(E) has poles at E= ERi: (12.2 The PenneyKronig potential.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. 2a<x<2a.e.2.^ l / .
and we choose (12. This boundary condition implies the quantization of the energy with ^ ( 2 a ) = n7r. VHV1 Hip(x) = Eil){x): VHV~lV^){x) = EVtl>{x). with V as parity operator. i..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. i. (12.. . x + 3a <> ™ since the particle cannot penetrate into the infinitely high wall.260 CHAPTER 12. In view of the periodicity we have ^(x) = ^(x + 4a). E^En = ^ ^ t n = 0.2 < a . Hip(x) so that with Hi/)(x) — Eip(x): H[^{x) ± t/>(x)] = E[^{x) ± xl>(x)\.e. (12.2.e. Quantum Mechanical Tunneling (a) We consider first the case of VQ = oo..2 = W h 2 _„2 = 2mo(E .a ) = 0 = ^(a). We set . . and hence for = Eip(x).1. 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. the differential operator is even in x. = EVip(x).Vo) n.e There is the additional boundary condition (for Vo = oo) ^ ( .53) *w«w^( « j)'*^{i i ^ .55) The Hamilton operator of the problem... HV^(x) = H.
are ^_(0) = 0. ^kAcos{k2a).62) = nBe~Ka T nBeKa = ^n[BeKa q= B e " M ] . The central region of the solution is the classically forbidden domain. however.58) D sin k(3a — x) with xE[a.61) = Be~Ka ± BeKa = ±[BeKa ± Be'Ka].. This wave function is no longer restricted to the interior of a box.2). V'(O) = const. We > obtain even and odd functions ip± by setting ^ = ±1.a]. VV(°) = °. We can write the solution as consisting of the following pieces: { Asink(x + 3a) BeKX + Ce~KX with with i £ [ ..2a]. so that (W meaning Wronskian) W[tl>+. This means that i>±(x) = <  = ±1. BeKX±Be~KX with xe[a.il>]?0.2a. At x = ± a we now have to connect ip and ip' of neighbouring domains.Be Ka (12. (12. We therefore have different boundary conditions. 12. At x ~ — a the connecting relations are Asin(k2a) kAcos{k2a) and at x = a: BeKa±Be~Ka K. of course. — a].12. (12. V+(0) = const. (12. xe[a. we shall not exploit here.59) Asink(x + 3a) with x & [~2a. that these are such that for VQ — oo they approach those of the previous case.63) = KBe~ F = .5 Degenerate Potentials and Tunneling 261 We restrict ourselves again to the period — 2a < x < 2a.a]. 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. which. = Ka ±Asin(k2a). (12. ±Asink(3a — x) with x€[a. — a].2a].60) The boundary conditions that ip± have to satisfy. (12. We expect.
nir)2}} = 1 ± 2e~2Ka.. = AA. 12.65) and we expand tan(2afc) about nir. is the one belonging to the even solution.e. say.nir + 0[{2ak .nir) + OU2ak . where Akoce'2™. (12.66) Inserting these expansions into Eq. + 0 ( e .2 a . i.68) .Thus we observe a splitting of the asymptotically degenerate eigenvalues: kesympt. 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.64) we obtain ~[(2ak k .e.e. (12.even. (12.67) The value of k belonging to the upper sign.2 e . i. i. Quantum Mechanical Tunneling Fig. » we expect the eigenvalues to approach asymptotically those of the case Vo = oo. the odd solution with. Taking the ratio of the equations in both cases. i. in the present case it is easier to continue with the above equations. and is — as we see — less than that belonging to the lower sign case. (with reinsertion of nir/2a for k) 2ak ~ nir . (12.— (1 ± 2e~2A + O (—). say A. k = fc0dd.3 Level degeneracy removed by tunneling.mr)2}. However.4 x ) and cothx = 1 + 2e~2x + 0 ( e " 4 x ) . In general one now rewrites the equations in matrix form.262 CHAPTER 12.e. the limiting value for K = oo: tan(2a/c) = = tan(n7r) \ (2ak — nir) 1 — ^—.+ • • • 1! cos^(n7r) 2ak . F ^odd (12. for VQ — oo. For large values of Vb. we set tanhx = l . It is plausible therefore to use appropriate expansions.
The following example is an attempt to transcribe the quantum mechanical effect into a macroscopic situation. Fermi [92].V{x)\i.12. '. Classically the car is unable to overcome the bump. a sinusoidal bump in the road with a height of one meter at its peak as indicated in Fig. V ^ 0 are given by the equations ^o + " J . w(x0) •• = = mogy(x).4. 1pV 2ro 0 4>o '. and is a reformulated version of a problem set in lectures of Fermi. 12.69) where the potential V(x) is (with g the acceleration due to gravity) V(x) y'(xo) Setting w(x) = 7r**E. 2m0E „ 2mnE . exp r rb rb / •J a 2mo (V(x) . . ^v + ^[E.e. . as indicated in Fig.. 12.1: A car's quantum mechanical probability to pass a bump A very slowly moving car of mass mo (kinetic energy almost zero) encounters on the road between x = a and x = b. 2mn. 0. (12.v = 0.70) xo = (a + b) 2' . 12. p.3. . i.^ 0 = °' Hence in dominant order •00 — exp . Solution: The wave functions i/>o>Vv in regions V = 0.E)da amplitude divided Thus with E ~ 0 the probability P is given by the square of the transmission by the square of the incident amplitude. y(x) = sin and y(xo) = 1.2 dxy/V(x) (12. . 57. but quantum mechanically there is a finite probability for this to succeed.4 A car meeting a bump in the road.** " W ~^7~ Fig. b > a.5 Degenerate Potentials and Tunneling 263 Thus the effect of tunneling — here an exponentially small contribution — removes the degeneracy of the asymptotically degenerate states.. Compute this probability. exp . Example 12.
1000 kg) and using the following values of the natural constants: g = 9. (15.1.621.1.17)). 14.9191 ~ 1. Quantum Mechanical Tunneling = y/mog 2(1) — a) /""'' / dw sin1/2 n Jo w. n I . which give /•TT/2 dx sin'M . p. We have 8m 0 : exp ^(ba)/4 x 2_1 ^^W(s x °' 9191 (12. Jahnke and F.2 [fr/2 ~ I)'] (Ml)! 2 ' where B(x. Emde [143]. p. h = 1.2. We find™ (with ( .l . one evaluates an approximate probability of exp[6. S.y) is the beta function (see Eq. Gradshteyn and I.384.71) The integral can be looked up in Tables of Integrals.1 / 2 ) ! = v^r) TT/2 /.055 X 1 0 _ 3 4 J s .264 we have ro / dxyfV(x) Ja CHAPTER 12. 950. ^ ) = 2 ^ .807 m s .4 x 10 3 9 ]. . (12. 369 and 8.72) Assuming a mass of 1 ton of the car (i.e. **For instance E.**. dx sin ' x •• Looking up the value of the factorial in Tables. p. and assuming a length of b — a = 100 m of the base of the bump. Ryzhik [122]. we obtain ( .2 . formulas 3. : 2 ^ B ( ^ . M. We can now evaluate the probability.1 / 4 ) ! = (4/3)(3/4)! = (4/3) x 0.
Example 14. we consider the linear potential in three space dimensions that forbids the particle to escape. 144. whereas the first case permits only a continuous spectrum. in Chapter 15. 265 . 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. Then considering the probability distribution determined by the squared modulus of the wave function. 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.Chapter 13 Linear Potentials 13. Finally. This is the problem of Galilei in quantum mechanics. the last case allows only a discrete spectrum.1 Introductory Remarks We distinguish here between three different types of linear potentials.3. We shall see that in this particular case the wave function can be written as a superposition of de Broglie waves.2 The Freely Falling Particle: Quantization Under quantization of the freely falling particle we understand here the solution of the Schrodinger equation for the linear potential. for instance. * A highly abridged treatment of this problem is given. Siissmann [266]. each of which propagates with the classical momentum and energy of a Galileian particle. Thus.* In Chapter 14. in the book of G. we consider the corresponding case of a particle above the flat surface of the Earth. This is the quantum mechanical version of the problem of Galilei in one space dimension. 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. 13. p.
1).4a) f°° ${k. V(x) = —mogx.= l V 27T Joo dxelkxip(x.2) which is the case with a discrete spectrum. (13.t) = mogx + V0. 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. (13. 13. We see " that the case g < 0 can be related to the reflection x — —x. The present case is an exception and a good example of its applicability. A different > "linear potential' is the socalled "confinement potentiaF V(x. and yields rriQX = mog. from d_ m x2 0 dx + V(x) 0.266 CHAPTER 13.1 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.e.t) = m0g with potential V(x.t) = .4b) . (13.1) where g is the acceleration due to gravity and we can put Vo = 0.3) In general the momentum representation is of little importance.t). (13.2. The equation of motion follows for instance from the derivative of the constant energy.= l V 27T J_oo 1 dkeikxijj{k.t) = c\x\ + V0. c> 0. g > 0.t) = . (13. Therefore we use the Fourier transforms il){x. We consider here the case of Eq. Recall the classical treatment of the freely falling particle.t). x > 0.t). (13. i. In the present case of quantum mechanics we have h2 d2 2mo dx2 m0gx ip(x.
(13.t) the expression (13. multiplication by a test function and subsequent integration).6) (and to avoid confusion we do not introduce new functions and variables): k > k + mogt (13.6) the argument k is then replaced by this.8) With this result and the substitution (13.e.7) applied to every quantity in Eq.6).e.2 The Freely Falling Particle: Quantization 267 and consider the following partial integration in order to reexpress the last term in Eq. d \ ~f1 dkr[ k + m0gt ^r't (13. (13.13.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. To this end we consider the following substitution in Eq.5) m+tm9Wcr{ktt)=2^^ktt)t (13.t) dketkxiP(k.3) as a partial derivative d/dk: d lKx dkeAkx{imQg)—4){k. Applying the total derivative d/dt to this function. i.< (13.3) yields us now by inserting for ip(x. (13. 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.t)im0g^k.6) and wish to solve it. Equation (13. we obtain the differential operator on the left of Eq. (13.6) the latter equation becomes d ~( m0gt ^(^M<=+=?.t) h2k2 d ~J(k.t). We rewrite this equation in the following form th (13. (13.t).d m +imo9 . • fd ~(' m0gt %h .7) In the wave function on the left hand side of Eq.4a) d ~ f I dk ikx dk 2vr (im0g) iP(k.9) .t) m0gx /2TT m0gxip(x.
t In Example 13.1: Verification of momentum representation solution Verify by direct differentiation that the solution (13.6) to the solution (13.t) ok lm og^— exp dk •ghf dt'fc+^(t't)W(fc. On the other hand the second operator expression implies dtj>o imog^—ip(k.i) . We observe t h a t t h e solution is actually a function of kt — k — mogt/h. Our next step is to reexpress the solution (13.11).12) .6).t ) ' U ( f c .11) in a neater form. E x a m p l e 13.6).1 we verify t h a t this solution indeed solves Eq.= k rmt.t). Thus addition of both indeed leaves the expression on the right hand side of Eq. (13. the following result h 2moi tp[k\ ^—.e. we have to reverse the substitutions and replace in (13. We have thus obtained the solution of the transformed equation.gh [ d t ' j f c + ^ ( t ' . (13.6) is then $(k. Solution: We apply the two operator expressions on the left of Eq. We define the wave number kt as kt. the third term from the integrand) > / mog \ o exp dk h J h dt y ' 2mo k2i>(k.6).t ) = ipQ(k)exp Jo fdt' k+ m0gt' H (13. To obtain the solution of the original equation (13. t ) .10) Note the amplitude function ipo(k) in front.11) *[k?f. (13. i. (13.6). Linear Potentials This is now an ordinary differential equation of the first order and can immediately be integrated to give an exponential.268 CHAPTER 13.11) solves the partial differential equation (13.t) = i>0[k 1— ) exp 2m0i J0 k+ ^(t't) (13. (13.10) k by m0gt back : k —> k h The solution of Eq. Thus we have first (the second term arising from the upper integration limit.
} fA ' _ _tkt 2 k+ t) 2 k .o o 2 —  exp iktx h mZgH' .17) = J_/ C X ) h2 [It(k) . we have ip(x.19a) .2 The Freely Falling Particle: Quantization 269 Then the integral in the exponential of the solution (13. 27T ^ .15) we obtain / = .14) (13. we obtain rp(x.16) Next we evaluate the Fourier transform of ip(k.gfci + 2 g2m0 2/i 2 ' (13.t) = . (13.13) rn0_t+2 + mlg2t3 . Inserting ip(k. we can write the integral / = l h l h (hr + 7 7 (h 3m0g 3 m0g \ 1 H 3m0g With the definition « * ) • ! ( * = * .— [It(k) .4a). m0g ~7T' k Z k _ r ^ h t Y + i n Smog (13. .= 1 / f°° dkelktx^{kut) dkipol k V —t 1 exp iktx K ) (13. m0g (13.2„2+2 mfaH 3^3+3 .t) into this integral and recalling the property (13.11) can be written j'A ^.t) defined by Eq.t) = 1 / dkipo ( k mogt 2 h ( 1m\ig \ With . .+^f (13. (13.. _ k t  h2 3 Inserting here the explicit expression for fct.18) h =k mogt p h S' h .I0(k)]./„(*)] 2m^ig Inserting the explicit expression for the argument of the exponential.11).o ut = 2mo —kf 2mo A.13.
7 u>t'dt' (13.t) as a decomposition into de Broglie waves. the relation (13.19a) exactly as for a falling particle. (13.o o Jo This result can be interpreted as follows.t) =y= dkip0(kt)) exp i< fctx V27T J .gk— + 2h 3 2?7lo 2 2 h mQgk ^ _ mjfo2 2 i + ft2 fcr 2mf)g 03mg £3 .21) In other words.270 and CHAPTER 13.22) 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'.18) into a • different form in order to be able to perform the integration with respect to k.v = kh/mo.t) as 1 I00 ip(x.t)\2 for t — oo (distant future or past). 13.ut> = Jo t2 g2m0 i 3 k t .23) .2. 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.19b) h3 3 we can rewrite ip(x.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) expresses the wave function tp(x. fm0gt\ \ 2h ) (13.20) (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. 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. These wave vectors vary in accordance with Eq. Such a decomposition is possible only for homogeneous force fields. (13.
QX ht V ht 2ft For i — oo the definition of £ defines a distance xc\ given by > k i.e.28) (for a verification replace a by a + ifi. 2h h 24ift ' mogt^ r m.29) ./m 0 a.t) exp i\ mogxt mox2 — h 2h 2ht mo<?2£3\] r (vn§x 24ft mogt\ (13. (13.13. m0xcl ht m0gt 2h gt O v: °° 2hti (13.26) with the formula 1/2 (13. Then ip(x.24) We now set fit / 2m 0 V" 0 V m0a. integrate and then take the limit fi —> 0). We evaluate the Fresnel integral in Eq.2 The Freely Falling Particle: Quantization The integral (13. rnogt m0g2t3 tX^r~.25) The amplitude ipoi^t) then becomes for i — oo: > 4>o(h) \t\—»oo 4>o[ k \ . (13. TOO#A ft and.27) xd = vt Thus xc\ is the distance travelled by the particle as determined in classical mechanics. may be put in front of the integral. fi > 0.26) d£e ie £ kt^ m0' 0. It follows that ip(x./oo dkip0(kt) exp l x ixI exp oo . ftf fti m0gt\ 2h J' " J' m0x ht mo5* 2ft + 2mo ~ftT f.QX 2 ^ 0 7 (m. being now independent of £. /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.t) = 1 1 oo 271 27T /.t) 1 exp r . (13.18) thus becomes ip(x.
kt and ut 13. V(x)=m 0 gx V=E Fig. 13.t) ~ W —exp i V irlt m ktxd J cot. xc\).32) For x — oo the function cf){x) behaves like > 4>{x) ex exp exp dx { 2mo (E + mQgx) \ H2 . (13.272 CHAPTER 13.1) has only a continuous spectrum.t) = eiEtlhct>{x) (13.31) and obtain from Eq. 1/2  ±^(2mlg/h^x^ / 2i (13. this expression can be brought into the following form: if *\ 0 ip{x. For stationary states we set xl){x.33) oo for E <C \mogx\. These are incoming and outgoing waves.dA 4>o(h (13.3 Stationary States In this section we show that the Schrodinger equation with the potential (13. For x exponentially increasing and decreasing solutions are possible.1 A turning point at x = —XQ. (13.3) the timeindependent Schrodinger equation i^ + ^ [ E + mogx]<l>(x)=0. Linear Potentials Replacing here x by the large time value x c i.e. .30) We see therefore: The large time asymptotic behaviour of the wave function is determined entirely by the classical values of x (i.
Thus the spectrum does not contain states which are normalizable over the entire domain x e [—00. cf. H2k3 4>{k) — cexp Ek (13.35) It follows that /W tm0g \ 2m0 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.e.13. at which E — V = 0. Fig.34b) It therefore follows from Eq.t). and we can establish for the resulting continuum solutions the orthonormality and completeness relations.1. J \ „ (13. so that ih dtp{k. exponentially decreasing. i.6) that also for 4>{k.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. However quantum mechanically this is possible with a small.e.t) = dt (13.4b). The point — XQ.36) m0g \ 6m0 with c = const. at which the classical particle bounces back. t). (13. i. 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. which is not possible for real particles.e. we . i.e.0 0 dkelkxE^(k. t) = — = / V27T J . is reflected. 13.00]. its momentum p would be imaginary there. Akx (13.e.4a) and (13. The Fourier transform is given by the relations (13. probability. and hence is continuous and extends from —00 to +00. E)dk.34a) Then d &k 1 f°° ih—il>(x.t). 1 f01 dketlcx^{k. We determine the constant from the condition that the continuum solutions are to be orthonormalized to a delta function. i. t) = Eil)[x. i.t) — (p(k) > E +^og~Mk) = 1 (h2k2 ^m. but also d 00 dt Eijj(k.e.
L dxeikx(f>E(x).a : ' ) .274 have / Using d(x) = — one verifies t h a t with CHAPTER 13. i.. apart from a constant phase which we choose to be zero. / dE(j)E(x)(t>E(x') = ^ s . (1340) i.43) y=2^x+2^E.36) (with c replaced by (13. so t h a t oo / <fiE(x) J dke ikx s/2nmog exp m0g \ 6m0 Ai(s). h? h2 . the relations oo /*oo dE4>E(k)4>E(k') = 6(kk').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 .39) dk4>*E{k)4>E. (13.e. Linear Potentials OO dx<ffE(x)<i>E>(x) = oo 8(EEf). (13. (13.(k) = / J ~oo dx<t>E{x)<\>EI(x) = S(E  E') Inserting (13.£') = <*(#£')• 1 27rm0g' = /o . In a similar way we can verify the validity of the completeness relation.e.36) into (13. 13. </>E(x) = . V 2lT J Mk) = .M *K Joo Setting dt cos ( ts n Jo st)..38).41) J—oo Next we insert (13. / <>~ikx VZ7T J (13.L [ dkeikxMk).39) we obtain dkcc*exp m0g It follows t h a t cc = &(# .37) (eikxdk.38) we have for <j)E{k): OO /"OO / oo (13.40)) into (13.
For instance dxAi(y / • + x)Ai*(x + z) = 5(y — z).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.^x + ^ E (13. whereas the other is of exponential type. and in Chapter 15 in the computation of energy eigenvalues for the threedimensional linear potential. For a further solution Bi of Airy functions we refer in Chapter 14 to Tables. (13. and we shall see t h a t one branch has periodic behaviour.J * .46) This follows since with Eq. In the next section we derive the important asymptotic expansions of Ai(s) for both positive and negative values of s.^ .13./ 2m 0 E (13.45) One can now derive normalization and completeness integrals for the Airy functions.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= ^ .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. (13. / s M . . the function 4>E{X) i and setting s y = — = 1 ..
4 The Saddle Point or Stationary Phase Method The integrand of Eq. . or n <6 <TT o for r > oo.TT < 30 < 0. (13. Thus we must have sin30<O. Linear Potentials 13. i.43) contains the phase (with x — z): > *(*) := sz .f dzJ*W 2TT JC with z = reie. (13. 13. (13.50) Im z domain (b) / ' ^ ^ ~C domain (a) 0 ^ ^" ~ ^ \ x Zs=i^S Fig.47) Consider the following integral along some as yet unspecified contour C in the plane of complex z: / := i .2 The contour C for s < 0 with ends in the angular domains (a) and (b).51b) . (13..TT < 6 < 0 for r > oo.\zz.49) (13.48) We have $(z) = sre1 1 A S„3i6 r e sr cos 6 . . o (13. (13. We want the integral to exist and so desire that lim e**^ = 0. i.r 3 sin 30 ).276 CHAPTER 13.e.e..r 3 cos 30 J + i ( sr sin 9 .51a) and since with sin 30 < 0 also sin(30 + 2ir) < 0: TT < 36 + 27r < 0.
Such a piece can be found in the neighbourhood of the socalled saddle point zs. oo] as indicated in Fig.54) where a is the new variable with —oo < a < oo. Hence _ 2 / s)3/2 /•oo 2TT( s)!/4 doe J—oo CT2 i*3 3(. (13. We now choose the path of integration C at fixed O z = —iy/^s with a e f [—oo.55) s < 0. —i ${zs) = szs .z3s and V^s + . Then 6S = ±7r/2.58) .4 The Method of Stationary Phase 277 The integral (13. we choose zs = iy/^s. Then Ze TeC : V^se±i7r/2 = ±i\fzrs. (13.51a) and (13.45) exists for the contour C. It is reasonable to choose the contour in such a way that only a short piece of it contributes substantially to the integral.( \ / ^ ) 3 2z. and the main contribution to the integral comes from the a region around the saddle point. Inserting (13.e.54) i ¥~s)3/2' (13.56) 2n(s)^Jc But I daefr"?*"^.52) = exp(±i7r) Let s be real and s < 0. i.57) $"(z s ) = 2% s. so that exp(±2i9s) — 1 and — s — rl. 0 = &(z8) = s~zt s = za „2„2i6>. where the derivative of the phase function vanishes. provided its ends at infinity satisfy these conditions. Then the ends of the contour lie in the domains specified by Eqs.13. (13.48).55) into (13.53) Since we shall have the contour C in the lower half plane. at which the phase becomes stationary.2.51b). we obtain da 2KJC (*)V4 c 1 e **(Zs) exp i{$>{Zs)+l(zZsf<$>"(Zs) + (13.e. (13.S )V2 (13. i. and we set with a real: z := zs + a (S)V4' (13. 13. (13.
Ryzhik [122]. 2 s 3/2 . . S. (13.2 we show that for s — oo: > Ai(s) 1 .60) The reason for describing the point zs as a saddle point is that the integral in Eq.44) and (13. (13.63) The Airy function can be reexpressed in terms of cylinder or Bessel functions Zv(z).45) 4>E(X) ~ Ai(z). where — with e = (2/h2mog2)1'3 — we have /2^m3g3y/3 x\ = e[E + mogx}. 8. (13._ 1 .64) tSee I.s ) V 4 exp \(s) 3/2 (13. V 2h2 (13. cos . (13. According to Eqs. > so that Ai(s) 2 v ^ ( .491.. Sec.32) with the solution (13.42) into an equation with a more appealing form.2/32u = 0 (13. but would increase exponentially for imaginary values of a — thus describing a surface around that point very analogous to that of a saddle.59) In Example 13. (13.278 CHAPTER 13. Gradshteyn and I. Linear Potentials where the contribution oc er3 is obtained from &"'(zs) whose evaluation we do not reproduce here.32) therefore becomes dz2 + zd) = 0. M. as one can see from the following equation^ ri' + Ffz •2^2. We observe that the Airy function has a periodic behaviour in one direction but an exponential behaviour in the opposite direction. With the help of the substitutions (13.61) 2m0E = f 2 \1/3r 2 2 2 fi h \h mog2 Equation (13. This is an important aspect in the WKB method to be considered in Chapter 14.62) This is the Airy differential equation with one solution 4>{z) = Ai(z). V^s / 4 V3 (13. For (—s) — oo the integral converges towards T/TC.56) decreases exponentially around a — 0 for real values of a.44) we can convert Eq.
60) we have thus obtained (with the help of the saddle point method) the asymptotic expansions of the Airy function Ai(z) which play an important role in Chapter 14. P Introducing a new variable a by setting p = ai\a\.51b) are satisfied. 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 . cos  . the conditions (13._ .^ e ^ / V2 4 [°° Jo dae'2^. Eq. we obtain f° Joo dp=—[a+ i\a\]da.13. Then.43)) Ai(s) = — [ dzeiit(z\ *(z): 1 Ai(s). kl dpe*' = ~e^'4 V2 f° Joo daeW2 and f°° dpe'^ JO = .s 3 / 2  . changing to the variable p. 27T Ic Jc 3 This time s is real and positive. .51a) and (13. Choosing the contour C to lie along the real axis. * " ( z ± ) = =F2Vi. (13.59) and (13..4 The Method of Stationary Phase with solution u = z1/2Zi/2/j(7A 279 (1365) With expansions (13.2: 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. Hence the saddle points are given by 0 = * ' ( z s ) = s — z2. It follows that for s —> oo: Ai(s) ~ . Example 13. with *"(z) = 2z.z ± ) 2 * " ( z ± ) . Thus there are two symmetrically placed real saddle points on the real axis.
111). Thus here / ( z ) = — z + n l n z . n!= / e~zzndz= / e~z+nlnzdz.Be 2 Fig. since f'(zo) = 0 and (z — ZQ)2 = —x2. ~ 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.*. and hence integrating as described above through the saddle point indicated in Fig. ^/2KJW1. Solution: Briefly.e.280 CHAPTER 13. Jo Jo \ .dz = idx (observe that this implies an integration parallel to the axis of imaginary z through the saddle point). i. (11.3. . / " ( z o ) = —n/z2 = —1/n. one obtains n! ~ where we used f^° dxexp(—w2x2/2) = V2^nn+1/2en. Linear Potentials E x a m p l e 13. 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. 13. 13.3 The saddle point at z = n + ix. the method of steepest descent evaluates the integral J^dz by expanding f(z) about its extremum at ZQ with z — ZQ = ix. z o = n .
Chapter 14 Classical Limit and W K B Method 14. H. The central issue of WKB solutions is their continuation in the sense of matched asymptotic expansions across a turning point (i.* Descriptions of the method can be found in most books on quantum mechanics and in some monographs. Froman [99].§ 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. Kramers and Brillouin. 292 . will there be that of a coupling parameter.e. § See e. i. 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. and the role played by h in our considerations here. Froman and P. O. 281 . for misunderstandings pp.1 Introductory Remarks One of the most successful methods of solving the Schrodinger equation is the WKB method. however. Kramers [153] and L.g. *R. the linkage of solutions above a turning point to those below.* The method had. Wentzel [282]. named after its first promulgators in quantum mechanics: Wentzel.e. For the history see pp. of misunderstandings and other aspects may be found in the monograph of Dingle. where *G. A critical overview of the history of the method. The WKB result is therefore frequently described as the semiclassical approximation.317. Jeffreys [144]. It goes without saying that such comparisons are very instructive.295. A. ^ The central problem of the method is the topic of "connection formulas". ' T h e most prominent earlier expounder is H. been developed previously by others in different contexts. B. 316 . Brillouin [37]. N. Dingle [70].
14.3) .282 CHAPTER 14. (14. for which the observation of position and momentum is independent of the time at which the observation takes place.* We illustrate the use of this condition by application to several examples.2) dip(r = I VA{v) + ^ ( r ) V 5 J eiS^lh. We start from the Schrodinger equation ih <9* =HV at with H=£— p2 2mo + V(T). It is for these states that one obtains the timeindependent Schrodinger equation A^(r) + ^ [ E . • {VA + %AVS [ h B iS(r)/ft +^{vSVA+^A(VS)2 *See e. We shall see that in this limit the > Schrodinger equation describes a steady stream of noninteracting Newtonian particles.t) = eiEt/h7P(r) and \V(r. A(r) and S(r) being real functions of r.g. In this equation we set.1) (14. the energy has a definite value E with V(r. the derivation in Example 18. Classical Limit and WKB Method E — V changes sign).V(r)]^(r) = 0. It is then possible to derive in a general form the relation known as BohrSommerfeldWilson quantization condition. the function S being called phase function. V>(r) = A(r)eiSWn. For stationary states ^/.7.t)\2 = \iP(r)\2. so that _ Aip AA+^V• h AA+CVAn (AVS) + IVS h VS + AAS) iS(r)/h (14.2 Classical Limit and Hydrodynamics Analogy In this section we consider the limit h —• 0.
h h Taking real and imaginary parts of these equations we obtain the equations: + dt and 2rriQ 2m. (14. h2 ih— = H$ A + V(r) * .V(Y) + and 2VA • VS + AAS = 0. (14.14.t) = and obtain dA i_dS_A + dt h~dt 2 h r A(VS)2 AA h2 2mo ih A(r. We can identify Eq. These equations are equivalent to the Schrodinger equation.1) and separating real and imaginary parts. We set in this equation.Q A 2i 8A 1 (14. h h (14.* VAVS+AAS + V(r)A. V(r.2 Classical Limit and Analogy with Hydrodynamics 283 Inserting this into Eq. (14. (14. (14. (14.12) VA A (14.+ VAVS + AAS = 0.4) (14. we obtain the following two equations: E .7) h2 AA 2mo A 1 (VS)2 2mo = 0 (14.5) Apart from these we also wish to consider the equations that follow in a similar way from the timedependent Schrodinger equation.10) with a continuity equation by defining as probability density p:=*** = [A(r.11) We define as probability current density the vector quantity J:=K where V* = ijf* V* rnio i_ „ *.13) .6) 2mo dt now with A also a function of t.t)}2.10) m 0 ^ .t)eiS{r't)/h.
16) BA 1 moA—. TTIQ (14.14) A2VS. ot 2 Multiplying this equation by 2 and recalling that the function A was introduced as being real. we have mo—(A2) + V(A2)VS + A2AS = 0. (14. ^ + V • J = 0. (14.9) as 9S 1 mo^ 2 + T. m0 But now (14. (14. (14. the implications of this equation are also those of the equation of continuity.9) for the phase function S. Next we investigate the significance of Eq. (14. + h2 AA V = — ^ .17) With Eqs.21) .284 CHAPTER 14.20) With this we can rewrite Eq.10) we obtain = — [V(A2) • VS + A2AS}. %(1"1' I<2) and V J = — V • (A2VS) mo Prom Eq.15) and (14. (14.n.19) m0 m0 it is suggestive to define the following vector quantities ± = at v := I p = JVS. ie. Since J = — A2VS = PVS. .. (14.01. mo <™> (14. (14.10). (14.+ AVA • VS + A2AS = 0.J = 0.16) this becomes m 0 1 rr + m o V . Classical Limit and WKB Method This means # * — V * = — AVA + — and so J = — A2VS. Since we obtained this equation from Eq.18) ot ot This is the equation of continuity that we encountered already earlier.
i. The equation of continuity then describes the stationary flow of a fluid. and hence''' ^(m 0 v) + m 0 (v. dt 2 Constructing the gradient of this equation.23) (14.25) dt We recognize this equation as the classical equation of motion of the particle of mass rriQ. we obtain ^ V 5 +  v ( v v ) + V V = 0. or that of equilibrium in electrodynamics) and dS/dt = —E (from ^ oc exp(—iEt/K)). However. which for very small values of h yield the dominant contributions. Such series are typical for expansions in the neighbourhood of an essential singularity. they are asymptotic expansions. This result is very general. with the help of Eq. The particles have no interaction with each other. An expansion in ascending powers of h therefore starts with contributions of the order of 1/h2.14. (14. where.2) for the probability amplitude. This is the analogy of the Schrodinger equation for h2 — 0 with hydrodynamics.22) (14.20). > In the case of stationary states we have dp/dt = 0 (which is analogous to the condition of stationary currents.24) d 1 dr d •— = d Lv _ . In the first • place we require the ansatz (14. that the quantum theory implies an essential singularity at H = 0 (i. We deduced this equation solely from the phase S in the limit h — 0. . a singularity different from that of a pole). This means. v dt dt so that the equation can bedt written dr dt ~(m0v) + W = 0. d — = (14. 'Recall the formula V ( u • v) = (v • V ) u + (u • V ) v + v X rotu + u X rotv. One should note that in no way do we simply obtain the classical equation of motion from the Schrodinger equation in the limit h2 — 0.2 Classical Limit and Analogy with In the limit h2 — 0 this becomes > Hydrodynamics 285 ^ + W 2 + F = 0. V • J = 0.e. > The motion of the particle or rather its probability is altogether given by the above equation of continuity.V ) v + V 7 = 0.e. This expression with h in the denominator shows. r o t v = ( l / m o ) r o t g r a d S = 0. so that V ( v • v) = 2(v • V ) v + 2v x rotv. (14. We see therefore that in the limit h — 0 the particle behaves > like one moving according to Newton's law in the force field of the potential V. the wave function \I/ describes effectively a fluid of particles of mass mo.1/h.
30) ax *For details see R. such as "LiouvilleGreen method' and "phase integral method'. similar to our earlier procedure with A and S real functions.28) The calculations we performed at the beginning imply the following equations (compare with Eqs.3 14. i = eiW(x^h. There are also other more specific descriptions. p. is an asymptotic expansion. As remarked at the beginning. Dingle [70]. . (14.e. (14. (14. The expansion in powers of h2.26) We observe already here that an expansion in rising powers of k. z (14. or h2/2mo(E — V). i.In A(x).286 CHAPTER 14. Sec. Messiah [195]. Classical Limit and WKB Method 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. Kramers and Brillouin.3.1 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. 318.2. We set. §We follow here partially A. This means that the WKB method can make sense not only where one is interested in the classical limit but also for E — V(x) and hence E large.4) and (14.[EV(x)}y = 0. the method is widely familar under the abbreviations of the names of Wentzel. 6. B.§ y = A(x)eiS^'h where W(x) = S(x) + . We consider the onedimensional timeindependent Schrodinger equation y" + ^.27) (14. h2 l 2m0 {EVy corresponds to an expansion in decreasing powers of (E — V).5)) and 2 ^ f dx ax + All=0.
V) = h2 In the WKB method one now puts S := S' = S0(x) + h2S1(x) + (h2)2S2(x) S'0 + h2S[ + ••• .V) and (14. (14.36) Equation (14. (14. + (14.32) is therefore (S'0)2 + 2H2S'QS[ . Since the expression (14. (14.32) s" = sz + h2s'( + •••. 3 / 5 . r//\ 4V^ 2 Up to and including contributions of the order of h2 Eq. (14.34) Equating coefficients of the same power of h we obtain (S'0)2 = 2m0{E .31) \a.3 The WKB Method 287 Equation (14.\ In S'. Eq.e. Integrating this second equation we obtain ll[dX so that = \l^dX' A = c(S')" 1 / 2 .2rn0(E V) = h2 lS'A 2S'0 (14. .33) 3 (S")2 .35) can be integrated.35) .30) corresponds to the equation of continuity (i. (14.10)).28).4 (S>)2 1 S'" 2 S> (14. Li This expression can be substituted into Eq. 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.2mQ(E .14.A = const.29) becomes (note that this is still exact!) (S')2 .
i. and instead is asymptotic which we shall not establish in detail here. Dingle [70].39) and (14.41)? We note first that the expansion in powers of h2 does not converge. y/2mo(V(x) . (14. we expect that in any case So » ft2Si.27) we obtain now with Eq. (1. Eq. .33) to make sense as an asymptotic series.I In S' + In c h 2 2 2 exp ^{S0 + h Sx) . (14.± ln(S'0 + h S[) + In c (s'0y/i and hence exp :5 0 + 0(h2). The solution is seen to be periodic provided k(x) is real.31): y = exp = ~ ~iW(x) h exp L nH™ exp ^S . it is reasonable to set (cf.' For the expansion (14. y = c'{k)1/2 exp or y(x) = a \ A ( z ) cos I / dx W) dx + 13). (14.37) For Eq. (14.39) where a and /3 are constants. B.and So = ±h / — — A J k(x (14. (14.19).40) rx dx v "7 W)\. E > V(x). 1 S e e R. Classical Limit and WKB Method has the form of a momentum.E)' Then y(x) = V^(a 7 exp dx + 6 exp T(x) (14.288 CHAPTER 14.e. For V(x) > E (this is the classically forbidden domain) we set l(x) = ^==^====.41) We now ask: What are the conditions of validity of Eqs.38) h y/2mo(EV)' (14. Chapter XIII. p = h/X) k{x) :Then S'n +.
1 / 2 ]" [§0SS). Since k(xy 1/52\S'0 (14.14.36). E > V(x).V{x)) ^ This condition is satisfied provided E or E — V{x) is sufficiently large or (14.45a) . Setting R := A' = \moKV'{x)\ \2m0{EV{x)\W .42) [(^).3 The WKB Method We have 289 S0 = ±h and Si follows from (14.36)) 2s'0s[ = [(s&r 1/2 ra) 1/2 With S£ = ±ft/A.3/2 s&T + 3 {So Il« we have (cf. Eq. (14. 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.
How do the solutions of Eq. (14. around the classical turning point. and hence (with UIQ = 1/2.290 CHAPTER 14. In terms of X the original Schrodinger equation is y" + = y = 0. (14. the WKB approximation is in general invalid (see below). Roughly speaking one therefore requires E to be large in both cases. In the neighbourhood of E = V(x).46) The desired equation has the solutions y(x) = aXll2 exp ±i dx / (14. 14. i. V(x)) (14. at which E — V{x) changes sign? We note first that in this case near x = a: E . *2 A(z) = V / 2m 0 (£ .e.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.a)V'{a).V{x) ~ Or .48) The (dominant) WKB approximations are exact solutions of this equation.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) = V(a) . h = 1) . Classical Limit and WKB Method we can rewrite the inequality (14.47) Differentiating this twice — we omit the details — we obtain the equation y" + l A2 (A 1 / 2 )" *i/2 y = o.3.44) as Idxsj2mQ(EV{x))l 1 + ^R2j » ^Rh.48) behave in the neighbourhood of such a point called "turning point'. In a similar way we have for E < V(x): l'(x) <C 1.
We define solutions ?/i and ?/2 with branches to the left and to the right as follows (for . As indicated in Fig. 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. 14. i. In the immediate neighbourhood of such a point the Schrodinger equation can therefore not be replaced by Eq.1. and on the other side.e. 14. We shall not derive these conditions here in detail but rather make them plausible in Sec. in the domain E < V the WKB solutions are exponentially increasing or decreasing.a ) 9 / 4 ' 291 (v^r 1 (x — a>2 These relations imply that Eq. We assume we have a turning point as indicated in Fig.3. (14.1 The regions around the turning point at x = a.48) possesses a singularity of the form of l/(x — a)2 at every turning point x — a. since this requires different approximations there. 14. (14.14. V=E Fig. which connect one domain with the other. 14. The transition is provided by matching relations. 14. in the domain E > V the WKB solutions are oscillatory.1.3 The WKB Method with the derivative relations 1 1 a)5/4' 4(x 1 5 16 (x. On one side.3 after stating them first here.48).1 there are WKB solutions on either side of the turning point and some distance away from it.
however with the limits of integration x. 14.50) x S> 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 / —.1) is then: VI exp {~[T) Vxcos fx dx TV Ja ^ _ 4 (14.H— J =^ v/exp + m (14. This means the asymptotic part suffices only in t h e exponentially decreasing case.292 CHAPTER 14. The first W K B matching condition in the direction indicated by the arrow (and for the potential decreasing from left to right. (i x dx 7T ~~k~ 4 / One should note t h a t the solution ^4?/i + By2 for A ^= 0.49) x 3> a Vxsin and 2 exp V2 VACOS (L X dx A ra dx X for for i « a . in order to make the formulas independent of whether the potential is rising or falling. P u t differently: We could add a lot of contributions to Ay± without affecting the asymptotic form of Ayi. Classical Limit and WKB Method definiteness note the + signs): Vtexp yi ~ S + Jx I X for for i « a . 14.52) For a potential rising from left to right. (14. Thus. and E ^ V for a ^ x the same formulas apply with the same direction of the arrow. one could simply replace everywhere /•a / pa • • • b y / Jx J x . Note also the extra factor 2 in yi. (14. has the same asymptotic behaviour in the domain i « a a s Ay\. as in Fig. a interchanged.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.
3 The WKB Method 293 In that case Eq.Airy overlap* Fig. (14.51) — always with the decreasing exponential on the left — is valid in both directions.3 Linear approximation and matching How do we arrive at the matching relations given above? We observed above that Eq. Therefore we consider now this original equation in the domain around x = a.3. 1 ~ (x .e. 14. the matching relations have to be altered accordingly. y" + Toy = o. i. (14.46).54) . is singular at a turning point x = a.a).53) Since at x ~ a. whose exact solutions are the dominant WKB approximations.2 The overlap regions around the turning point at x = o. the equation V approximately linear around turning point V=E Airy solution WKB . the equation becomes approximately y" = (xa)xiy. (14. One should remember that the matching relations above are those for the dominant terms in the WKB expansion. as we saw.48). Thus around that point we can not use the WKB solutions and hence have to find others of the original equation (14.14. 2 1 X (14. If higher order contributions are to be taken into account. 14.
or M. — a ^> 0. With the substitution z = (x .57a) COS 2^2/3 _ SZ 1 1 ?(z) /2 3 es^ Bi(z)~< 1 1 for x — a C 0.a)X\/3.57b) UV2 3 11 by J dx . > for a. i I 772 \3 + > for ia>0.Bi(—z) for \z\ — oo. A.60). (13.56) We have discussed some aspects of solutions of this equation in Chapter 13.59). (13. p.1 / 4 by \^. A second solution is written Bi. J (X (i4 58)  Replacing in Eqs. we have r x * f x ' /2(x a)i/2d*=^(x .^ 4 I 7T x — a <C 0. Dingle [70].5960. Abramowitz and I. Classical Limit and WKB Method where xi is a proportionality factor which we choose to be constant. formulas 10. as > may be seen by referring back to Eqs. 448.294 CHAPTER 14.4.57a) and (14. (14. p. In particular we encountered a solution written Ai.a)3/2 ^ /2 =^ 3/2 o Q. Stegun [1].291. B. R. (14. (14. This is given by the following expressions.55) the equation becomes the Airy differential equation d2y dz2 zy. (14. In the following we require the asymptotic behaviour of the solutions Ai(—z). Cf. or to the literature:H 3/2 2^{zy/* Ai(z) ~ < ~~FL~T77 y/TTZ1/* 1 1 s e K*) n for ~. . (14.57b) To the same degree of approximation as the equation y" = (x — a)xiy.
These solutions. and hence must be proportional (in view of the uniqueness of the solution there). Example 14.14. At and around the turning point the solution of the Schrodinger equation is therefore given by Airy functions.59a) .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 / — = . As an illustration of the use of the WKB solutions we apply these in Example 14.3 The WKB Method and 295 •. e.1 the Airy solution becomes proportional to the exponential WKB solution and to the right to the trigonometric WKB solution.(zf2 by J we obtain the matching relationsX (14. 14.l v/2m0(EV(x))dx 7r = (2n + l). (14. The latter domain is a small region where the solutions from either direction overlap.1 to the quartic oscillator. In this way different asymptotic branches of one and the same solution are matched across a turning point. In the direction to the left as indicated in Fig. 14.l. F T' dx (Z)1^ by ^ harmonic oscillator V=x 2 +x 4 V=E linear approximation Fig..52). 14. As we approach a limit of this interval in going away from the turning point. In this sense we have now verified these relations.3 The anharmonic potential well. however. we enter the domain of validity of a WKB solution. n = 0. by Ai(z).51) and (14.g. We summarize what we have achieved.2. In the neighbourhood of a turning point the Schrodinger equation becomes an Airy equation (in the leading approximation). are only valid in a small interval around the turning point at x = a as indicated in Fig.
b are the two turning points. (14. . as in the case of the harmonic oscillator itself. whereas the approximation of the eigenvalues by comparison with the harmonic oscillator in the domain of the minimum.l.. /n = s u ^ TJcos^ya TJ+cos^a T J . yz{x) = c'vlexp ( [x dx\ I — / — ) for x 3> b. 14.51). 14.3. The potential now has to be inserted into this condition and the discrete eigenvalues En of the problem are obtained.e. 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. (14. ^ A = (2n + l ) . In these domains the dominant WKB approximations are 1 /j/i(x) = cVlexp / fa dx\ I —/ — \ r for i « a. Consequently only the lowest eigenvalues would be reasonably well approximated by those of the harmonic oscillator. n = 0 .59b) Solution: We consider an anharmonic potential well as depicted in Fig. j/i.y o dx TT T provided 6 _ 4 /. .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 . Let E = V(x) at x = a and x = b. . regions 1 and 3 in Fig. We argued earlier that the WKB method is suitable in the case of large values of E. Here we are interested in the discrete states (the only ones here). extends only over a length of very few wavelengths and is therefore valid for n small. . 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. (14. (14.^ . Sometimes this distinction becomes imprecise. I " ' (K rx dx iz\ _ ( (b dx_ 7T _ fx dx _ • f / fb dx\ f* dx\ ( fb dx\ . for which the WKB approximation surprisingly yields already the exact energy eigenvalue as one can verify by evaluating in its case the BohrSommerfeldWilson rule. Classical Limit and WKB Method where x = a. 2 n = 0. (14. According to Eq. i.44) this implies large values of n. In the neighbourhood of the minimum at x = 0 we could approximate the potential by the harmonic oscillator. The corresponding eigenfunction would be approximations of the proper eigenfunctions around the origin. . Hence we search for solutions which are exponentially decreasing in the far regions.3. Then around these points the potential would be well approximated by a linear potential (as we saw above). 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 ( .(x) = c ' v Acos I / — I for a < x < b. 2 .296 CHAPTER 14.. the continuations of these functions into the central region 2 are: ya(x) = c v A c o s I / J.2. 1 .61) Evidently these functions have to continue themselves into each other (since the wave function has to be unique).b).60) Using the matching relation (14. The condition for this is ya{x)=yb{x) for xe(a.
e. Dingle. .14. A. In this socalled "old quantum theortf (i. pdq = n + 1 — . ri2 I Jqi :pdq n + 1 — . (c) V Fig.x number of turning points TTH. and knows of no published form or script.63a) or. (14.x number of turning points h. Sommerfeld and W. . The corrected form* is. . equivalently. (14.63b) *The author learned this in lectures (1956) of R. 2.1. Consider the Schrodinger equation in the abbreviated form d2iP(q) dq2 f(q)t/>{q) = 0.64) (14. . with n — 0. n = 0. 2 . 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.1.4 Three cases with different pairs of zeros. B. . 14.4 Bohr—Sommerfeld—Wilson Quantization A quantization condition which is very useful in practice and in a wide spectrum of applications is — and remains in spite of its old fashioned reputation — the quantization condition established by N. Bohr. This relation is sometimes wrong.4 BohrSommerfeldWilson Condition 297 14. Wilson by supplementing classical mechanics by Planck's discretization.
where n —.4(a). Since sine and cosine have zeros spaced at intervals of TT.65) We use the knowledge of these solutions to find the eigenvalues of the equation. this is an exponential zero. —/ = K J h It follows that / pdq = (n + l)Trh= (n + l)h. where /(g) is positive. the wave function is to vanish at points q = q\ and qi with the function /(g) remaining negative in between as illustrated in Fig. this is a trigonometrical zero. We have the case of trigonometric solutions. 14.e. In quantum mechanics we have dq2 + Ki/j(q) = 0. (14. as illustrated in Fig. Before we continue we recall from Eq. together with q — — oo.67) Case (b): Next we consider. i. There are three cases.66) J<1\ Thus in this case the socalled old quantum theory is correct. Jqi ^ fpdq J =2 pdq = (n + l)h. (14. if /(g) is negative. (14. this means dqyf—f = (n + l)7r.51) the linkage between the trigonometrical and the exponential solutions across the position go in space at which /(g) = 0.298 CHAPTER 14. 0(g) ex . where /(g) is negative. / • n an integer > 0.e. In this > case the wave function for g > go is . i.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. i. Case (a): The case of two trigonometrical zeros. 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.e. 14. the case of tp(q) — 0 at q = gi. i.4(b).
63a) or (14.63b) to obtain the eigenenergies of the quantized harmonic oscillator. 14. i. Solution: As in the case of the quartic potential.e.63b). (14.4 BohrSommerfeldWilson This will vanish at q = q\ if rii / V^f{q)dq+j^ Jan It follows t h a t i 4 Condition 299 = {n+l)TT. ri\ / J do y/f(q)dq (n + 1) 1" (14. the wave function for q < q'0 must be proportional to ri'o (Z) / 1 4 sin J V W)dq+^ Z i • oc (7) 1 / 4 sin / y/~f{q)dq . (14.77T (where we reversed the order of integration to obtain a positive integral and multiplied through by minus 1).67): • the wave function for q > qo must be proportional t o (Z)1/4 sin [ ^W)dq+>x Jqa . Case (c): Finally we consider the case of i/j(q) = 0 at q = ± o o exponentially as illustrated in Fig. (14. Therefore their arguments differ only by (n + l)ir.14. the sines must be proportional. Thus in this case the socalled old quantum theory is wrong. this is the case of two turning points. or Pdq (n + l ) " 2 h. i. and • • for the wave function to be exponentially decreasing to the right of q'0.3. As illustrations we consider Examples 14. ^~f(q)dq Jqo LJqo + 7T J'% On fq / i 1 V~f(<i)dqjK (n + l)7r.e. Hence we have pdq= i n +  .4(c). Hence we obtain the condition I'o Vf(l)dq 90 (n + 1) IT. Since both cases refer to the same region qo < q < q'0.69) pdq = (n + l ) " 4 h. E x a m p l e 14.68) (14.2: The harmonic oscillator Use the relation (14. For the wave function to be exponentially decreasing to the left of qo and using Eq.2 and 14.70) We can therefore summarize the results in the form of Eq.
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.2 3 J x (n + 3 / 4 ) 2 / 3 . one obtains E ~ 2. 14. Thus there is one turning point at E = mogq.28 x 1 0 . and h = 6. whereas a turning point does. as illustrated in Fig. Example 14.V(q))dq The result is E/u. :: V=m0gq Fig. permit this although with rapidly diminishing probability. 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. where g is the acceleration due to gravity. in principle.63 x 10~ 3 4 J s . Calculate its quantized energy.5 A particle above a flat Earth.2 . The WKB Method this requires evaluation of the integral with qi = ^' E/2K2mQu2 / 2 m 0 ( B . .300 With potential V(q) = 2ix2mov2q2. 14.71) For m 0 = l x 1 0 _ 3 k g . so that E = hu{n + 1/2).5. Solution: This is the case of one trigonometrical zero of the wave function ip at q = 0 and an exponential zero at q —> oo. 4 / x CHAPTER 14.807 m s . g = 9. ^Thus a trigonometrical zero is the condition of absolutely no penetrability beyond it.
d dE V)={n+)n. 9 . and u> is the oscillator frequency in either well. 2TT W 2K J p dE V where p = ^/2mo{E Solution: We have — V).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 ( — = .2. 1 .) 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. Chapter 16) one obtains the quantization relation (for Mo = const.72) \\2dxp=Qdx^{Eh Hence.5 Further Examples E x a m p l e 14. E x a m p l e 14.7. Solution: The proof is contained in Example 18. ra 2/ = 0. (14.14. h M0 _h 1^3/a ~ 4 0 / 2 J V \ 3 _ 2N3 \M0) ~ ~Ml ' .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 £ = . with p = modx/dt.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 .l.75) which verifies the formula immediately.= — <*— = h(n+\. In the case of screened Coulomb potentials (cf.73) Jx! rp2a dx I^<E 2 v) = ^ n — = .5 Further Examples 301 14.f V ft h JXl h 2TT (14. (14.AE(q0 2 = 2n + 1) = — exp 7r dzy/E zn + V(z)/h where q=zo are the left and right barrier turning points.
e.77) Determine (a) the turning points. Example 14. (14. which determine the imaginary part of the energy.6.133)). Fig.76) in agreement with the result obtained from the classical Kepler period for the Coulomb potential (cf.6 The cubic potential. It follows that the energy between the minimum and this maximum lies in the range 0 < E < £ m a x = ^aSb2. 14.<?2i<?3 for E > 0 given by E — V = 0. and (b) evaluate the WKB exponential exp[—2/b arr j er ] and the WKB prefactor 2/ w e u.78) (a) We determine first the three turning points at qi = <?i. (11. q = y+ a . The WKB Method ^ 2 JV 3 2TT ~ ~Mjf (14.302 and T CHAPTER 14. The q term in this equation can be removed by transforming the equation to a cubic in y.a .y3) = 0. (14.79) . i. 14.~ \ (14. Eq.y2)(y . 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). Solution: The potential has a finite minimum at q = 0 and a finite maximum at q = 2a2/3 as indicated in Fig. where 1 2 • 1 2 y = Q. where .
W e o b t a i n ^ fbarrier : = 6(92 ~ 9 3 ) 2 ( 9 2 .79) a n d t h e n s e t t i n g m o = 1. i.a 2 [ l . W i t h t h e o t h e r i n t e g r a l from T a b l e s we o b t a i n (K(k) b e i n g t h e c o m p l e t e *L.g i ) ( g 2 . V3 + t a n 0 2 .cos 0 + cos(6> + 120°)] = a2 92 . p . M i l n e . . we h a v e qx = . we set 0 = 60° — e. 92.04. 92 = a2[l .14.< 1.9 3 — = = 2 " 3 " [ .0^6O°. 93.9 2 [ .3/2. (14.c o s ( 0 + 12O°) + c o s ( 0 . p .{ V 3 e } > 0. (14.q)(q .08.gi = 1.84) 92 . 14.120°)] = a 20? 2a2 3 2a 2 • cos 0  91 . 78: sin 120° = v ^ A c o s 120° = . gi ~ y {V3e} < 0. (l_fc' 2 a + fe'4)l/4. _ \ / 3 — tan(9 2 V92 . H e r e cos(<9 ± 120°) = . a n d t h e a n g l e 0 = 0° t o E = £ m a x .93).2 c o s 0 ] . p . D . g 2 ~y{3}>0.85) T h e angle 9 = 60° c o r r e s p o n d s t o E = 0. V cose' + sin 8 (14.120°)] = .1 2 0 ° ) w i t h 0 < 6 < 60° for 0 < E < £ . D w i g h t [81].93 2a2 [ . 80 a n d 361. q3 = a2[\ .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 .2/3 i n t o q = y + a 2 / 3 . F r i e d m a n [40].e.9 l ) f f / du s n 2 u en2 d n 2 u .\tan0^ v/3. M .1 / 2 . 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 .T h o m s o n [196]. B y r d a n d M .78)) c o s 30 •• 4P3 ( ' \ aeb2} . B .80) I n s e r t i n g t h e r o o t s yi.§ ( c o s 0 ± \ / 3 s i n 6 0 a n d § sin 6 sin# (14. p . 6 = 0 i m p l i e s E m a x of E q . (14. 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. 91 < 93 < 9 < 92(14.81) m a 91 . ui=sn 1 \l— — = l. f o r m u l a s 236. 0<6O°. 37. tan60°=%/3. h = 1) ^barrier := / d( }J^2~(E~v) = ^~= dq^/(q .ui=K(k).2 c o s ( 0 + 1 2 0 ° ) ] . 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 .c o s O + cos(6» .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.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 .k 2 2tan# \/3 + tan! .= s i n t 3 v 3 (14. 212. (14. q3 ~  . e > 0. (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 — 0.6.2 c o s ( 0 .g i < 1. 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 = 1.5 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. F . § T o h e l p we cite H. ' P .' : ± 6 0 ° w h e n E = 0.
qz) (14. 2 2 . . Friedman [40].91 F(TT/2. E(u = K(k)) = E(k). .?)(?3 . as in (14.2)*w+2(fe4+k'2)E{k)] v ' G(k) For one complete round from one turning point back to it. D.92) ''P. (14.2(fc2 .S . (24. 72. Eq.00. p. F.91) a2[cos0^] y/3tanfl ? L .2)K(k) + 2(fc4 + k'2)E(k)) k ^i — .86) 15 tan. E(u) the incomplete elliptic integral of the second kind. Next we evaluate the required integral across the well from 91 to 53 at energy E. the integrand of the integral across the well is effectively the inverse of this momentum.k). We can evaluate this integral again with the use of Tables of Integrals.) . V 93 .85). (14. G(fc) =86aS (ITS£ li[fe. with K(k = 1) = oo. Byrd and M. Thus Jwell : = [13 1 f13 rf 1 / dq J^^EV) 1 y/2 1 V 2 f[13 = — —.~ y ^ ) a< [ COS 5 + . E(u = 0) = 0. we obtain in the limit of E = 0: j d ( ^ ( E „ y ) = 2/barrier = ^ ! .fc'2 + fe'4)"1/4. V?2 .^ V3 / V V3 y a / c o s e J.32)). 1+72 = and J o s 0 + ^ = (1 .sm t %/3 £\. (14.^ + ^ ) (l + fc'2)2 .304 CHAPTER 14.91 . A:') = # ( * ' ) • (14.= —^ = k a2[cos0+^] v ^ + tanfl V=". sinV = q3 qi ~ = 1. E(k = 1) = 1) (•u\ = K{k) i Q G{k) = / Jo Next we set s dusn2ucn2dn2u = j[k'2(k2 15fc4 . 91 < q < q3 < 92 bJc.84) we obtain Carrier = 2a 2 sin6> 2 b( . 0 (14.89) This expression will again be obtained later with the help of configurations called "bounces" (cf. Thus Iwen=lgF(ip./ dq~ ^2 b JQl 1 \/(qqi){q2q)(q. Whereas the integrand of the above barrier integral is effectively a momentum.87) Inserting G(fc) together with the expressions for the prefactors into (14. g= 2 .90) hi JlnynEv) V(qqi)(q2 Vz b Jq q) I" bJqi where P fe = 9391 9291 = . V3 cose + ^ = V3 2 1 + 72 v/l + 3 7 4 C (cose) 2 = 1 + 3 7 4 = 4 d . The WKB Method elliptic integral of the first kind. formula 233.
114.14.** However. 905. formulas 8. (14. we cannot expect the ratio exp(—2Jt>arrier)/2J'well oc w exp(—27barrier) evaluated at k = 1.94) C °S3 2 (lfc2 + fe4)3/2  l+ g f c ( l + fc) + Comparing this equation with Eq. a°b n = 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.3.80). . S.88) in rising powers of fc' . pp.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. and hence at E = 0.99) **For the potential approximated around the origin as V ~ a2b2q2/2 the eigenvalues are E = fat)(n + 1/2) with it) = ab.85) (14.89) can be expanded to exhibit a simple pole at z = 0 allowing evaluation with Cauchy's residue theorem: 2 dz 2Ez2 a2b2 %2dz Ez2 2nib\ o?E o?b2 2KE ab tt I.4 2n + 1 fc' = 4 — ^ — . to represent a physical decay rate.! With these expansions one obtains ^barrier = 15 ahb 2 + 8A . Gradshteyn and I. (14. we obtain > .2 IE 4 fc' ln 8 (" «V iU^fc' 4 . (14.. The same expression for E is obtained by applying the "method of poles at infinity" to the BohrSommerfeldWilson condition (14. 2a 5 b K(k) + 2 E(k). . (14.3 and 8. we obtain E a6b2k'4(l + k'2). 32 (14.96) Expanding the coefficients of the elliptic integrals in Eq. Ryzhik [122]. (14.2. To obtain the latter we have to use the quantum mechanical expression approximated by E = hw(n + 1/2). M.l.95) Comparing this with the harmonic oscillator approximation E — En = ab{n + i ) . Thus one obtains from (14. 906. in fact through logarithmic contributions contained in the argument of the exponential for k ^ 1. we obtain 15 .63b).93) where w is the harmonic oscillator frequency in the well.5 Further Examples Hence (using K(0) = rr/2) 305 ab ab (14. With q = 1/z the integral (14. so that even in the case of the ground state the zero point energy will contribute to the prefactor.113.
_ exp[2/barrier] >...101) (14. . (18. also Eq. 15 a .2 7 b a r r i e r ] Z Z7T where (cf.7 .96) for the imaginary part of the energy as in the Breit. . (0) i hw En = E\> . _ T T • (14.> tf = = 2/wfiii = ±tv8a°b——P.11)) dq 2TT y/2§?(Ev) It would be interesting to derive the same quantity with the perturbation method and to compare the results. oa ba Thus with the W K B method we obtain exp[2/barrier] — barnerj = 8 s.j . i. 7 ~ —.100) 'barrier = ~0°b In [ ^ — j . /25a5b\™+5 e exp[2/barrier] ^ ( r ) Correspondingly we obtain for the full period 2/ w e ii: 2/well = ? K ( f c ' ) ( l ..102) . 2n + 1 — / 26a5b \ . +3 the result becomes exP[2Jbarrier] = ± i _&g_ (25a5br+J e . i. (14. the result agrees with the ground state path integral result (24. V (14.73).e. The WKB Method For k' —• 0 the first and the third terms dominate.Wigner formula (10.105) 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.— (14. It follows that (for one complete orbit back to the original turning point) .e.ba / 25a5b\n+2 ±l_l eT5 a f >.fc'2 + fc'4)1/4 * ± t ^ . n ~ V27T . (20.( n + l ) e _ JL a * 6 and we obtain for the ground state (ro = 0) .„. so that we obtain 4 .103) 27 w e l l 27rVn+y ' fn With Stirling's formula in the form of what we later (with Eq. We observe that with \j2je as 1. this has not yet been done./^~^l^ab ^aub ..e x p [ .178)) call the Furry factor set equal to one. 8 Bfc .306 CHAPTER 14.J + 0(VoJb).
L. Lane and T. which permits only bound states. in fact. along with inclusion of angular momentum and spin effects and their interactions. The study of this potential. led to a spectrum which contains only scattering states. T. Gottfried. [232]. that it may not even be possible to extract individual quarks experimentally with any finite amount of energy. the classification of the various states of nucleons and mesons and other particles. K. L.e. indications from the nonabelian generalization of quantized Maxwell theory. after the realization that quarks as their constituents might not exist as free particles and. See particularly E. Eichten. that this is indeed the case. now known as quantum chromodynamics. describing free fall under gravity.1 Introductory Remarks The particular linear potential we considered in Chapter 13. the Coulomb potential.Chapter 15 Power Potentials 15. 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. D. This would mean that the force binding the quarks together would not decrease with increasing separation as in the case of e. This potential became widely popular in the spectroscopy of elementary particles. Rosner [269]. There are numerous. Quigg and J. In the present chapter we consider briefly the threedimensional potential V(r) oc r = r.g. [231]. i. Yan [82]. C. [233] and H. C. Quigg and J. at least indirect. B. Thacker.* 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.M. Rosner [230]. Kinoshita. K. 307 .
(15.r acting between two particles is constant and directed towards the origin (of the relative coordinate).308 CHAPTER 15. This force maintains this value irrespective of how far apart the particles are separated.4) Instead of considering the linear potential. and the normalization [ dr\y(r)\2 = 1. Power Potentials general power potentials. i. and so u'(r) . R(r) = ^ . These particles therefore cannot be separated by any finite amount of energy.6) = (I + l)u(r)/r = R ~ rl.u ~ r i + 1 (/ + l)R(r). </>)R(r).m2.e. h2 2n V 2 * ( r ) + [V(r) . and u'(0) > [u(r)/r]0. we return to the Schrodinger equation in three dimensions. (15. and obtain 2/x 1(1 + l)h2 u"{r) + EV(r) u{r) = 0 2 2/ir (15. m\m<i M= : . For a central potential V(r) we write (15.1. In order to have a clear starting point.3) with the boundary conditions''' u(0) = 0. v > 1.2) tt(r) = Ylm{9. 15.5) u'(0) u[r) r=0 = R(0). A > 0.4>)ru{r) = Ylm(9. where \x is the reduced mass of the two particles of masses mi. we consider immediately the more general power potential V(r) = \r\ 'We have u(r) = rR(r). ~ (I + l)rl (15.2 The force The Power Potential F = .e.V 7 = const. mi + m2 and r is the relative coordinate. f dr{u(r)}'2 1. We leave consideration of the logarithmic potential to some remarks and Example 15.1) i. (15.£ ] * ( r ) = 0.
N = 0.8a) ^ JO = (2Ar + l ) £ . is to be interpreted as the principal quantum number in three dimensions.2 The Power Potential 309 In Chapter 14 we obtained the BohrSommerfeldWilson quantization condition (14. or f l H " ^ ) ! .1. Then k(x) = k(x). . We therefore proceed as follows.V(r)) = (2nl) +1 N.. and the turning points are at a — — b := — xc. We now want to find a quantization condition for the threedimensional problem. . This means that we transcribe Eq.8b) This condition implies that we have only one pair of turning points. (15.2.8a) we can write the quantization condition (14. (15. (15. (15.9) 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!).15.2.9) into the threedimensional case in the following form: / Jo C dry/2n(E . . (15.63b): 2 and V(0) = 0 (15. 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. so that with Eq.10) Here rc is the turning point and n — 1."MA. x > 0.. which has been reduced to an effective onedimensional problem in the polar coordinate r. (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. odd integer ^=(nll)vh. In order to to obtain the . We also assume that V'(x) > 0. (14..4) in the threedimensional case. d5.63a).63b) for the onedimensional Schrodinger equation.7) k(x) = h/y/2m0(EV(x)). 3 .
13) we obtain _ X_ „ _ so that A£_ (15. c (nj\irh. (15. Power Potentials eigenvalues E —• En for the radial potential (15. (15.12) i/z Jo We set dr 1 l A v ~ E \ (Et\l/v so that (15. _ —vr" *dr.y) defined by (xl)\(yl)\ y ^ (x + y .15) I ' = ^ 1 (( 1) B{x. (15.6) we have to evaluate the > integral / : = [ ° dr^2n{E\r»)= Here the turning point rc is given by E . E A fv.11) rc = (15.14) Prom Eqs.13) dt .y) T(x)T(y) _ T(x + y) ("!)/" /•! / d"(1"")/.310 CHAPTER 15. dr = .'(1*)1/2= J1F\lt)y1dt. 1W or Xv \E I It follows that (I) / ^""^(l*)1/2. . Now.17) The integral on the right is the integral representation of the beta function B(x.\rvc = 0 . s.12) and (15. (1516) (15.
$T(z) = {z. One can convince oneself now. .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 ) .2 The Power Potential 311 so t h a t by comparison y = 3/2 and x = 1 + (1 — u)/i/. Weinstein [281]. i r ( i ) = r ( l + i ) = ( i ) ! .55). for instance simply graphically as indicated in Fig.18) \v \E or 3 E2 r(§)r(i)V2M (n ' (15. T(z + 1) = z\.1)!.19) or En = _ 1)^(3 + 1)^/^2^+2) r()ir(i)^7^ • v 0 1 Fig.1.1 Approach to square well with v —> oo. and n — — ]7r/i. The case v = 4 is that of the pure anharmonic oscillator also discussed by M. t h a t in the limit v —• oo the potential approaches the shape of an infinitely high square well.* In this limit the eigenvalues become:^ E„ (nh^ni) Lr(f)r(i)^7^ rv"ir •n2 2/i 2^2 (15. (12. (15. 15. ^ '"Compare with Eq. 15.
312 CHAPTER 15. 15. 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. In the onedimensional consideration these are precisely the odd wave functions as illustrated by an example in Fig. Power Potentials In the case of v = 2.25) "Comparison with Eq. = (n>r(§)A 2/3 3TT \h n 2/3 L r(§)v^7^ J (15.oj = ^/2X/JX. Physically it does not make sense for the probability amplitude to have a discontinuity there.2.23) The eigenvalues for the linear potential can also be obtained from the zeros of the Airy function. (6. so that the appropriate Schrodinger equation becomes d2(j) + (Edx2 x)<t>(x) = 0 . (15. 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). eigenvalues (JV + h)tkj — (2n .21) (4nl).24) is discontinuous at x = 0 (this means the derivative there jumps from positive to negative). (6. the harmonic oscillator.oj2/2. since these are the ones we are interested in (cf.61) h2 = I. discussion at the beginning of this section).22) 0 which selects from the usual and with the boundary condition 0(0) eigenfunctions the odd ones.2) implies A s fj. (13. For the following we set in Eq.mo = \.41) for the . we obtain E„.1 + \)hx> = {An • l)£fiw. > and with Eq. the case of the linear potential. In the case v = 1.//L This result agrees with that for the onedimensional harmonic oscillator r2 in the form" cf>"(r) + ^(E\r2)cf>(r) =0 (15. The potential V(x) = X\x\ (15. x > 0.g — —2.> r ( 2 ) A V 2 in \)*\W \it^T& (15. we obtain En = (n .
(15. E > V or s > 0.e. and for s —> oo the trigonometric behaviour of Eq.60) applies. i. this wave function has the required exponentially decreasing behaviour at infinity.57a) one solution of this equation is the Airy function <l>(z) on Ai(z) = Ai(E . \(s?» (15. i. Ai(s) 1 2v^(s)V4 exp x —> +oo. i.2 Behaviour of an odd wave function at the origin.e.15. 15. i. (14.e.x)\x=0 = 0.59) applies.x). W i t h z :— E — x this is d24>{z) + z<f>{z) = 0. (13. we must have as a result of the boundary condition <p(x) = 0: 4>(x)\x=0 oc Ai(E . (15. In the domain —a < x < + a .e.56) and (14.28) In particular at x — 0. (13.e. Ai(s) "v^ G ' cos s 3/2 _ t (15.2 The Power Potential 313 V(x) V(x)=x turning points exponential fallofl Fig. s = E.29) .26) For z = s := E — x —> —oo.27) i. Eq. dz2 According to Eqs.
Quigg and J. It is possible.= ) = «. Table 2. (15.31) (an odd function has an odd number of zeros!).52056 6.30) ! ^ / 2 .1: SWave Energy Eigenvalues for V(r) — r (with h = 2/x = 1) from f 3 7 r ^n 2 ( i>j2/3 . to obtain the zeros of the Airy function numerically.2. i.e.3. [231]. It is interesting to note that in the dominant approximation both methods agree. so that n — 1/4 ~ n.e.08795 5. = im — (7r/4). *This Table is reprinted from C. i. but for large > values of E Eq. p.29) implies that cos (  ^ .03914 11. Co.7 T Tl * 2 V 4 Actually this expression is only valid for E and hence n large. Power Potentials Table 15. Rosner.04017 11.^ = !(2nl). L. Quantum Mechanics with Applications to Quarkonium..94249 9.78445 7.314 CHAPTER 15.82878 1 2 3 4 5 6 7 8 9 10 2..08181 5. . of course.28) is really only valid for s large and s — oo. with permission from Elsevier.78671 7.93602 12. (15.02137 10.02265 10.00852 11.00767 11.94413 9. (15.33811 4. (2/3)E3/2 or 3 / 1 2/3 E — En — .93528 12.. the eigenvalues En are determined by the zeros of the Airy function.e.En from Ai(£„) = 0 2.82814 i.51716 6. The expression (15.32025 4. n = l.1 demonstrates the quality of the WKB approximation by comparison with the exact values. 201.. Table 15. copyright of NorthHolland Publ.
We indicate briefly here — without entering into further details — how these quantities can be calculated.* We can determine the constant N in the leading approximation by demanding that = 2 / Jo i. in Chapters 24 and 26. had to be supplemented by inclusion of relativistic corrections as well as other contributions arising for instance from spin and angular momentum interactions.15. F. We saw previously (see for instance Eq. i. dr[u{r)f " f** «" (i £>i)4 Here we replace the oscillatory part cos (. van Royen and V.e. . We shall encounter WKB normalization constants at numerous points in later chapters.3 The ThreeDimensional Wave Function The linear potential has in particular been used in the investigation of the spectrum of heavy quarkantiquark pairs. *Note that this makes N a W K B normalization constant. (14. e. i. The decay widths T.• •) by a mean value. P.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. ee.e. can be shown to be proportional to the modulussquared of the particle wave function at the origin. See also J. which are inversely proportional to the lifetimes. r <53) 1 3  dr cos 2 () (15.e. Weisskopf [239].3 The ThreeDimensional Wave Function 315 15. have only a finite lifetime and decay into other particles such as electronpositron pairs.** r ( # ^ e e ) o c *(0) 2 . cos2()^— / 1 fr° 2 > w . of course. The resulting masses agree in most cases very well with those extracted from experimental observations. In general these quarkantiquark pairs.34a) **The most frequently quoted reference for this result is R. e. In order to be comparable with experimental data these investigations.g. D.g. the charmanticharm meson ^ . Jackson [140].
316 In accordance with t h e relation CHAPTER 15. (15.1 )n .32) we obtain to leading order u'{r) and therefore «'(0) = N N (1JU0) N (15. From Eq. (15.Jo dr' n M?) ~ 4 in sin sin JTCldrW2v{EV)j l \ 7T N 7W)sin = ( .l N n .38) N : sin V ^ Jr dr' 7T (15.40) VWY .35) u T (•T=2TT/OJ dt cos 2 cot 2' (15.39) in / .2) and (15. (15. (15..4) we have for I = m = 0: ^(O) 2 = n'(O) 2 y O o(^0) 2 = ^ K ( O )  2 .10) with respect to n implies (which is permissible for small separations of neighbouring levels) 2/i ldEn rrc dr ^/2ii{En V) 2 dn J0 = irk. i.rcu w X{r)dr = Kh . dE a—— / dn J0 so t h a t with Eq.36) \i dEn irh2 dn According to Eqs.37) (15.2. & (15..35) N2 . A = (15. Power Potentials this averaging implies a numerical factor like 1/2.\ \ * n = l. so t h a t approximately l N2 / Jo k(r)dr . (15..e.34b) VMEvy Variation or differentiation of Eq.
W.* Using the result (15. Hite and H. J.41) I*(0)2=_L^)^. (b) for the logarithmic potential^ l*(0) 2 ex ^ v (15. 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.23) for the energy levels of the linear potential. MiillerKirsten [147]. E. MiillerKirsten [29].15. and hence with Eq.45) 'See the references of Quigg and Rosner cited at the beginning of this chapter. The case of arbitrary positive power potentials as above has been considered by R.g.38) implies 317 (15. S. Hence Eq. Kaushal and H. J. J. 'For one such investigation which includes the Coulomb potential in addition to a confining potential see e. MiillerKirsten [28].l ) n .2. . (IM2) l*(o)r = M dEn dn irh 2 ~^WEn ~*T (15 43) ' Inserting here. we have ^'(0) = ( . W.44) (15. § S. 3 . Bose and H. K.43) or the results of corresponding calculations one finds the following dependence of (^(O)!2 on the quantum numbers n = 1. Itt^pocn2*"1^"^. 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). . .37) we obtain: 1 (2M^)1/2 47r ft.1 ^ ( 2 ^ „ ) 1 / 4 . (15. K. W. . G. Bose. (15. for instance. the expression (15.3 The ThreeDimensional Wave Function Since we are considering potentials with V(0) = 0. for various potentials: (a) For a confinement potential of power v. S. 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.
. Solution: Details can be found in the literature.318 (c) and for the Coulomb potential^ CHAPTER 15. MiillerKirsten [29]. = (I + 1/2) 2 . Example 15.133). S. Eq = i q .e. (15. 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 .L2 h4 = 4/3exp[(2a — /3)/f3]. (15. n = 0. q = 2n + l.— oo < z < oo. W..^ +•••. .j . Bose and H. and there Eq." 1 11 L . . rewrite the equation as ^ z + [ ..2. 7 With the substitutions r = exp(z — c).c = —a//3. (14. (11. = 1(1 + 1)..1: Regge trajectories of the logarithmic potential Consider the radial Schrodinger equation for the potential V(r) = g\n(r/ro). U(z) =  dz zf3e2a/?e2*. D. {S = ^ > 0.^ ( 3 ^ + l)55p^«(3^l) + .1.75).L 2 + U{z)]<f>.5. K.g ^£+(<*Pl»riy = 0. E' = E + gln(r0). Power Potentials *(0) 2 oc . J.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. We observed there in the case of the Coulomb potential that the period is proportional to n 3 .I l n f ^ p . ip = exp[(z — c)/2]. M. Landau and E. Lifshitz [157]. i. *=^f.. in agreement with Eq.46) and the classical Kepler period of Eq.
Historically this potential arose with the realization 319 pVr .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. it is. This implies. in microscopic and hence quantum physics with the possibility of real and virtual creation of particles and hence a vacuum state. r in which /i has the dimension of an inverse length or equivalently that of a mass in natural units. a charge as source of a Coulomb force leads to an effective polarization of the vacuum like that of a dielectric. In reality. that no matter how far the particle is scattered away from the source of the Coulomb potential. r where r is the distance from the source charge and the charges and other constants are collected in the coupling g2. A very similar potential is the Yukawa potential V{r) = g2 — . It is clear that for ro — oo the > potential becomes of Coulomb type. and in fact. so that like charges are repelled and unlike charges are attracted by each other. This effect sounds unphysical. The result is a screening of the Coulomb potential which thus attributes it a finite range ro.Phenomenologically the screened attractive Coulomb potential may be written V(r) = g2pr/ro . 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.Chapter 16 Screened Coulomb Potentials 16. it will always notice the latter's presence.
Since. all coefficients Mj are real and independent of the energy E = k2.3) where for the real potentials we consider here.1) and is written in spacetimedimensional Minkowskian notation l/{jpvpu + /i 2 ). (16. one frequently resorts in calculations to the expansion of the potential in rising powers of r. 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. and that. . who later — as is wellknown — deviated from this idea and invested his efforts into the study of nonlinear spinor field theory. In the literature reference is sometimes made to the socalled static limit of a relativistic propagator. from the relation Pi + P 2 + M2c2 = 0. in fact the parameter [i represents the mass of such a spinless meson.2 ) or (p = hk) 1 /" . the mathematical expression with an exponential is not so easy to handle analytically. i. What is meant is that the fourdimensional Minkowskian Fourier transform of the propagator is given by the relation dk (2TT) J 4 e x ^r—2 = 7Z^u o)^n k + fi2 4TT v y x 2 ikx ^ e^W ( 16 . Yukawa potentials and superpositions of such potentials play an important role in nuclear physics.1) We encountered Green's functions earlier. however.320 CHAPTER 16. The realization that mesons and baryons are made up of quarks does not really change that picture at lower energies.ii k . Regge trajectories — which we encounter *The Smatrix theory of strong interactions was actually initiated by W.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.* 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. Heisenberg [133]. A Green's function is effectively the inverse of a quantity called propagator which is intimately related to the expression in Eq.e. Screened Coulomb Potentials that the strong nuclear force is mediated by the exchange of mesons. and in view of their relation to the exchange of virtual elementary particles have therefore been objects of intense study in elementary particle theory.
1 in the above. Naturally it is easiest to familiarize oneself with these by studying solvable potential models. Froissart [223].3) have been considered by various authors.e. Frautschi [97] and E. (16. if>oo. (16. Omnes and M. Potentials expandable as in Eq. A.§ 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. 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. f H. i.r). Wu [47]. (16. Thus in the present case it turns out that it is easiest to calculate first the expansion for the expression l + n+1.{l. Regge trajectories were realized to play an important role in the high energy behaviour of hadronic scattering amplitudes. bound state eigenenergies and the 5matrix. only for the cases of n = 0. physical) values of angular momentum as functions of energy E. where n is an integer and we referred to these already in Chapter 11 in the simple case of the Coulomb potential. Lovelace and D. "These were first considered by C.4) where E = k2 and h = c = 1 = 2m.e.1 Introductory Remarks 321 in this context — are functions which interpolate integral (i.e.mo being the reduced mass of the system.3) in the radial Schrodinger equation for the partial wave ip(l.k. As discussed in detail by Bethe and Kinoshita^ one can start by arguing that a countably infinite number of Regge poles may be defined in the region of large negative energies E = k2 of the radial Schrodinger equation by requiring I + n + 1 for n = 0.k. l+n+l = ^P.16. . Kinoshita [21].Q. . i. . £+**PVV t/. ''The approximate behaviour of Regge trajectories for the Yukawa potential has also been calculated by H. .1. Masson [180]. C. See also the other references below. 2 . Bethe and T. J. They are usually written I = an(E). $ In the following we consider screened Coulomb or Yukawa potentials of the type of Eq.e. T.r)=0. Cheng and T. Squires [258]. (16. however. to be of the order of 1/k. for the Regge trajectories or Regge or /plane poles of the 5matrixll / = ln(K) 'Standard references are the monographs of R. 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. Mandelstam [185]. S S. S.5) where K — ik and A n is an expansion in descending powers of K.^ Regge trajectories arise as poles of the Smatrix in the plane of complex angular momentum. i.
k. for all n. W. 16. Longoni and T. Regge [31] and E.k.15) and (16. has only simple poles) in the plane of complex angular momentum and in the complex &plane (E = k2) cut along the imaginary axis. < const.z) Then x is a solution of the equation VaX= (16. J.)/j. tf H . 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. J. . In the following we follow mainly the last two of these references. Section 11. < oo for all 0 < vr/2. (16.e.4) to z = —IKv and set TP(l. Muller [201] and [202]. Screened Coulomb Potentials as a function of the energy. Bottino.** The energy is later obtained by reversion of the resulting series. one can assume an expansion of the potential V(r) in ascending powers of r. The conditions for this have been investigated in the literature* and may be summarized as follows: /•oo V(r) = / d/xa(//)e^7r. 'In the case of the Coulomb potential (cf. M. Muller [204]. J. in the present case at fc2 = 0 or k2 — Mi = 0. A. o rV(r) I dpp\V{peie)\ regular at r = 0. 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. (16.6) 2K{M°~ An(K))x+ 2K^(^K) MiX ' (16J) **H. Proceeding now as in the case of the Coulomb potential we change the variable of Eq. Under these conditions the S'matrix is meromorphic (i. as may be seen from Eqs.ndp.z) = ez2/2zl+1X(l. Schilcher [203]. Miiller and K. J.9) the cut starts at E = k2 = 0.t Considering such superpositions of Yukawa potentials with /•oo / a(/j.2 Regge Trajectories Since our treatment here aims at obtaining an S'matrix as a generalization of that of the Coulomb potential. starting with the power r _ 1 of the Coulomb potential. H. W. Squires [258]. W. *See in particular A.322 CHAPTER 16.19).
10) zm$(a)= ^ j=m Sm(a.a + j)Smi(a.j) = (a + j l. (16. a) —b — 2a.j +(a + j + l. (a. as in the case of the Coulomb problem.0) = 1.z) = Ha). all other So(a.j l) + (a + j.2 Regge Trajectories where d2 Va = z^ and (cf.a + + l). The function <&(a.a)${a) + {a.j) for \j\ > m are zero.i 0) = 0. j)Smi(a. a + 1) = a — b+ 1.7) is seen to be of order 1/K.j)$(a + j).a+ l)$(a + 1) + (a. We also observe that the ansatz (16. and all Sm(a. J.10): Sm(a. (16. .* Recurrence relations for coefficients of perturbation expansions for Yukawa. (a. K. (16. Eq.z) is known to satisfy a recurrence relation which we write here for convenience in the form z$(a) = {a. {0) X = Ha. where $(o. b\ z) is seen to be a confluent hypergeometric function which for reasons of convenience we abreviate in the following as 3>(a). By a repeated application of the recurrence relation (16. W.11b) ^ The associated boundary conditions are: So(a. MiillerKirsten [249].a — l) — a — l.b.j) (16.b.1). which means that the hypergeometric series breaks off after a finite number of terms. so that to leading order we have VaX{0) = 0.9) we obtain m l)$(a .16. Sharma and H. (16.awhere (a.11a) The coefficients Sm(a.L ^d {bz)a l+ l + ^ ^ l 2K =  n &ndb = 2l + 2 = 2n^4^ K (168) The right hand side of Eq.a + j)Smi(a.5)) a = + 323 . anharmonic and cosine potentials have been derived in L.5) is equivalent to a = —n.j) may be computed from a recurrence relation which follows from the coefficients (16.9) (16.
[a. (16. (16.a + l ] 2 [ . This follows from the fact that D a $(a) = 0.aj 2 (2K)A _ [a.o]i = M 0 . (16.l. [ 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).An{K).[a>a + J]i+i*(a + 3).j). so that 1 1 2K— ' (2W[a'a]2+(2^F[a'a]3 1 r i .a\4.1] 2 .[ a + l.a]l^a) + ^2(2K)i+1 Y.e. .7). Evaluating the first few terms of the expansion (16.13) + ••• • One can now construct coefficients with their recurrence relations for the individual terms of this expansion. (16. a . #(a + ra) fjL$(a + n) V. 1 .12a) where [a. This equation is seen to be 0 = 7n7L a ' a Jl + P a + n $ ( a + n) = 0. (16.a + j]i+l = MiSi(a. 0 < \j\ < i.<&(a + n) on the right hand side of Eq. and hence Va$(a + n) = ra$(a + n). when n = 0.324 CHAPTER 16. for the coefficients M^> of the expansion Details can be found in the references cited above.13) one obtains the quantity An(K) and hence with Eq. [a. i R »] = 2K[a.aj 2 \ [a. Screened Coulomb Potentials Substituting the first approximation x^0* = 3>(a) into the right hand side of Eq.5) the Regge trajectories I = ln(K). Va+n = Va . oo _ . + (16. the latter can be written . Any term /j. i.7) and hence in Ra may be cancelled out by adding to x a contribution /x<I>(a + n)/n except.n.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. We first cite the final result . of course.' — .
(16.^ 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. . Inserting this expansion into Eq. 16.l)n(n + l)(n + 2) + 2M 3 M 0 (3n 2 + 3n . Thus the expected appearance is that shown in Fig.1) + 3M3M02 + Mln{n + 1) +4M 2 MiM 0 ] + 0(K~8). . is an asymptotic expansion for large values of the energy — is not very useful in the very interesting domain around energy zero.1 by evaluating the first two terms. Tate [6]. P.15) The same expansion may be derived by the WKB method from the BohrSommerfeldWilson integral for three space dimensions as explained in Example 16.1. The result is AnW = M0'[n(n + l)M2 + MlM0] <2» + ^ ^ 1 + — ^ [ 3 M 4 ( n .l)n(n + l)(n + 2) + 2M 3 M 0 (3n z + 3n .1) + 3M3M02 + M22n(n + 1) (16. it is clear that one expects the trajectories to be finitely closed curves with asymptotes given by those of the Coulomb potential.5).( 2 1 ^ 6 1 ) [ 3 M 4 M 0 ( n 2 + n . With numerical methods one can achieve more. Burke and C. we obtain the expansion for the Regge poles.1) +6M 2 Mm(n + 1) + 2M2M02 + SM^Mo] .2 Regge Trajectories 325 and then demonstrate the calculation in Example 16. including an exploration of the domain of small energies E. G. 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.14) +4M 2 MiM 0 ] + 0 ( K ^ 7 ) .§ This is an important observation which indicates the equivalence of the methods.16.^ p [ 3 M 4 ( n .1) +6M 2 Min(n + 1) + 2M2M02 + SM^Mo] + g^s [3M 4 M 0 (n 2 + n . Obviously the expansion — which. However. i. Ahmadzadeh. Analytical Observe there the quadratic form of the centrifugal potential! "See in particular A.e. (16. of course.2.
a] 3 = M 2 S 2 (a.1). one can expect a resonance with the lifetime determined by this imaginary part.^ . Example 16. 114.12b) we obtain [a.—" + In = — .2 a = In See S. and the lifetime At of the resonance satisfies the relation TAt ~ h.11b) we obtain Si(a. through which the particle orbits during the course of the resonance. ±1) = 0. just as a decay width T represents the width of the resonance in energy.13). However. C. The conjugate variable to energy is time. where the imaginary part of I is very small. p. .1: Evaluation of perturbation terms Use the above formulae to verify the first two terms in the expansion of A n for the Yukawa Regge trajectories." Re I 1 0 1 2 Fig.O) = Other terms vanish since So (a.326 CHAPTER 16. In fact. . 16. a] 2 = M i Si (a. Solution: We evaluate the first three terms of expansion (16.a)So(a.1 Typical Regge trajectory for a strongly attractive Yukawa potential. this is a particularly interesting domain since at the position soon after this point at integral I (as at I — 2 in Fig. Hence Si (o. For a resonance with a long lifetime. (16. For a bound state a j = 0 and the orbit becomes permanent. so the imaginary part ai of I — an represents its width in angular momentum. ai is small and A6 is large.a) = b . (a. 0). Similarly the conjugate variable to angular momentum is angle.0) = (a. and the angle A#. 16. satisfies the relation ajAO ~ h.0). (16. Frautschi [97]. From Eq. Screened Coulomb Potentials expressions of the behaviour of a Regge trajectory in the immediate neighbourhood of E = 0 are practically unknown. [a. 0) and Prom Eq.
(a. . (16. 14. (a. 1).2 V~Az and z \ zz dr\ A 2B C dz V'A + 2Bz + Cz2 (6.1.65) 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 . 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 . A" An An A" = = = (a. 0) + a S i ( a . 0) + (a + 1.0)+ 0 + 0 = (a1).1 ) + (6 .1) + (a .0) + 0 + 0 = ( 6 .0) = (a . Sec.1) Si(a. a)Si (a.2 Regge Trajectories 327 Analogously we evaluate S2(a.1 ) + (a. We obtain the quantities S i ( a .A „ ) H Mi 2An 2K if n + (2i^): r M 2 2K Hence 1 .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". we have to evaluate (cf. Solution: Ignoring higher order terms in r. One sets z = 1/r so that. (n+l)U+^pJ . 1) Hence 52(a.0) Si(a. Thus (with terms which are 0): Si(a.a + l)S0(a. a)Si (a. a)Si (a.16.13) we obtain 0 (Mo .6 + 1 ) . using the Cauchy formula (6. .(ra + 1) [n + (2K)3 A n+1+• n K K (M0 .00 B + Cz 2ni VA + 2Bz + Cz2 B 2niI —= + vC with ambiguous signs of square roots I.11b).An) • [ M i M 0 + n(7i + l)M 2 ] 1 A n = M 0 . 1) (a .2 a ) .0) + 0 + 0 = ( a .2o)Si(a.a)5o(a.n ( n + l Inserting these expressions into expansion (16.0) (a . .6 + l)a + A.6)Si(a.b)(a .al)S0(a.65) 2m dz V'A + 2Bz + Cz2 • Z— 0.^ 2 M " + ! ) M 2 + M i Mo] + • E x a m p l e 1 6 . i ) again from Eq.
i. however. . I.14) for An(K) in order to reexpress the latter in terms of I so that i+i+ A.t First.15). Boukema [34] and [35].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. Singh [252]. This shows that 1(1 + 1) has to appear in the BohrSommerfeld—Wilson integral as (/ + 1/2) 2 . The turning points in this case are at r = 0 and r = oo. *J. (16. We know that corresponding to every Coulomb Regge trajectory we have a corresponding one in the Yukawa case.16) = n n=o 1 2 With this inversion we obtain = M0^^[l(l + ^ [ 3 ( Z . (16. t?{l + \f M0 ±2K 1/2 2TT 2J 2V^2 (n + l + l)h = 2TT Mo h. Example 16. tSee Chapter 11 and V. 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 . Here we do not perform this procedure.328 Thus here dr K' CHAPTER 16.e. we can use Eq.* 16.l. Solution: For details of the solution we refer to papers of Boukema. ±2HK in agreement with our expressions above.(iO mp~ ~ ' ' ' '+ l)M2 + M0Ml] (16. n = 0. Screened Coulomb Potentials Mo .1)1(1 + 1)(Z + 2)M 4 + 2M3M0(3Z2 + 3/ . Thus we can write down the S'matrix for the present case by exploiting the limiting case of the Coulomb potential.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..1) +61(1 + l)M 2 Mi + 3M2M02 + 3Mx2Mo] + 0(K~6).5) together with the expansion (16.2. by expanding the right hand side in rising powers of v and evaluating the individual integrals. In the second paper the second order WKB approximation is used and shown to yield complete agreement with the terms given in Eq.
.. such as. . and could then have carried out the perturbation procedure not in inverse powers of K but in inverse powers of y/K2 + M\. W. the behaviour of the scattering phase in the domain of high energies. i. Miiller [204].16.* 16.. J. This has been done§ and one obtains .19) +2M 3 M 0 (3n 2 + 3 n .2. Miiller [211].e. Paralleling the case of the Coulomb potential in Chapter 11.17).4 The Energy Expansion We observe that Ai(K) has the property MK) = MK) 329 (16. W.14).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\.l.17) and is therefore real for real values of the potential coefficients Mj. the following *See e. Now reversing the expansion (16. .19) we obtain the energy. n = 0. § See H. J.l)n(n + l)(n + 2) + 2M2M02 .18) of the Smatrix now to explore further aspects. Expanding the square root VfC2 + Mi for \Mi/K2\ < 1 one regains ln(K) with the expansion (16. H.0[(KZ + Mx)'6} (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.g. 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 . which are precisely the expressions yielding the Regge trajectories of above. for instance.l (two additional terms are given in the literature). We observe also that the S'matrix is unitary as a consequence of the result (16. One can use the explicit expression (16.
'According to S.l)n(n + l)(n + 2) . Watson [280] in 1918 and later resurrected by A. Iafrate and L.^ 16.1) 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 . E. p. J. k) as coefficient of the outgoing spherical wave in the asymptotic behaviour of the wave function tp(r). Analogous expansions for specific energydependent Yukawa potentials have been investigated by A. the transformation was introduced in the form given below by G. p.1) + Af3Mff + 1) ^ lOM 6 M 0 2 (n .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.* We recall from Chapter 11 the definition of the scattering amplitude F(9. in the following expression with an "See in particular G. Sommerfeld [256]. MiiilerKirsten and N. Zauderer [286]. C.330 CHAPTER 16.l)n(n + l)(n + 2)(n + 3) +2M 5 M 0 3 {5n(n + l)(3n 2 + 3n .1) + 2M2M. Large coupling expansions derived in H.3 M 2 V ( n + l) 2 M(o I +2M 3 M 0 2 (3n 2 + 3n .f 24(2n + l)(Z + n + l ) 5 M4M0(nz M06 M$n(n ^ M9 + n .20) Extensive investigations of the energy eigenvalues for the Yukawa potential can be found in the literature.e. Screened Coulomb Potentials expansion. J. VahediFaridi [212] contain as special case the expansions of the first pair of authors.10) + 12} + 4M3M05 +2M 4 M 0 4 (6n 2 + 6n . W.< 3 M 4 M 0 ( n . 107. Warburton [279]. A. 282. . Frautschi [97]. i.11) + 2M22M03(9n2 + 9n .2)(n . 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) ^ — . B. N.l)n 2 (n + l) 2 (n + 2) 1 + • • • (16. Mendelsohn [135] and E.
k) = 2zfc L " v "'' v . \x\ <C l . z=o in which (cf.iJ 11 —[S(l.23) f(l. 16. Eq.184) to (11. (11.kmcose). c = 1.21) The scattering amplitude determines the experimentally measurable cross section o given by da (16.24) Si(k) being the phase shift. dn Iml 1 I c ( \ x o x 1 A 2 A 3 x4 X 5 Rel Fig.k) = ^T(2l + !)/(*.k)\2..5 The SommerfeldWatson Transform 331 ingoing plane wave in the direction of z (here with h = 1. Actually 5j lrr/2 as in Eqs.k) 0ikr (16..k)~l] = ~ k" PiSiW sin 5i(k) and S(l.187)..23) as the contribution of the residue of a pole in the plane of complex I of some suitably constructed contour integral.''' The idea is now to consider each term of the expansion (16.2 in the complex Zplane: *<«•*> = s / c c K <u(21 + 1) vf{i.fe)P . (cos 9). For this purpose we consider the following integral taken along the contour C shown in Fig.26. The scattering amplitude possesses the partial wave expansion F(0. ~:' sin nl (16. .k) — e 2i5[(k) ~" p—iirl (16.187)) (16. so that sin KI = sin7r(ra + x) ~ (—l)nKX = (—l)lir(l — n). n = 0.22) = \F(9. (11.2..TOO= 1/2): ip(r] r ~*°° Jkz elkz + F{0.1.2 The integration contour C. I = n + x.25) Setting Note the minus sign in the argument of the Legendre function. 16.
3 Reaction channels and their respective Mandelstam variables.3.—. be elastic scattering of a off b. we have to digress a little and introduce a few simple ideas which played an important role in the development of particle physics.26) P((+COS0) = 2 ( 2 n + l)/(n.—. for instance. t = (qq')2. 16. as indicated in Fig. For the momenta indicated we set (with metric +. In order to see the relevance of Regge trajectories in a reaction of particles. we obtain m® = ± I dl^±^f(l.fc)P„(cos0). Screened Coulomb Potentials Then integrating with the help of Cauchy's residue theorem.k) Zi Jc TT{L . a + c —> b + d.—) s = (p + q)2. 16.28) . n=0 which is the usual partial wave expansion.3.27) a(p) © b(q) Fig. a + d — • c + b. Such a direct reaction and its "crossed channel reactions" are shown schematically in Fig. in which case c = a and d = b. u={pq')\ (16.332 CHAPTER 16. 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'). 16. The reaction therefore describes the following processes which are also described as (reaction) "channels" and in which an over line symbol stands for the appropriate antiparticle (like IT meson with positive charge and that with negative charge): s: t : u: c(p') a + b —> c + d.n) oo {l)1 Pt{cosO) v ' (16. Thus the direct reaction may. d(q') (16.
t. e. F.g. Chew [48].^ (b) Chew's hypothesis: All (composite) particles lie on Regge trajectories. Thus. B. u.30) with partial wave expansions (observe the noninvariant factor k has been removed) Fs = ^ ( 2 2 + l)F z (g s )PKcos 98).u are not independent. (a) Mandelstam's hypothesis: All three processes (described by the s. Treiman [25].16. the quantities carrying these quantum numbers are the Regge trajectories. one then has A(s. if we had been considering the s reaction originally.t) = y^Fs(s. the variables t and u would describe momentum transfers in the crossed channels. s = 2qf(lcos0t)(16. (16. T See e.t). Mandelstam [184] and R.u) (if any of the external particles has nonzero spin. Khuri and S.(<fe)^(cos 6t).9s)\2. 16.§ Since the three variables s. § See G. A(s. One now makes two hypotheses. except for poles and cuts which characterize the reactions in the three channels..31) i l Taking all particles to have the same mass mo. Symbolically we then have the situation shown in Fig. the momentum k there would correspond to qs) s = ~4(ql + ml).c o s 0 s ) . Ft = £ ( 2 J + l)F.u are known as Mandelstam variables. there will be several such amplitudes which together describe all three processes).t.t) = VtFt(t. N. etc.29) and with selfexplanatory meaning for the cross section of the reactions ~ = \Fs(s. (16. . In the Yukawa picture of a reaction the dynamics is described in terms of the exchange of mesons IT (called pions).The variables s.5 The SommerfeldWatson Transform 333 and for simplicity we assume here that all particles are spinless and have the same mass mo. N. the only kinematics we require in the following is given by the relations (e. if s describes the square of the total energy in the s channel.t.u channels) are determined by one and the same relativistically invariant scattering amplitude A(s. t = 2q£(i . one can write the amplitude as depending only on two. For a Lorentzinvariant normalization of the initial and final states.et) (16. Goldberger.4. the dynamics is described by the exchange of quantum numbers.0s). t. M. A(s.g. which is an analytic function of the variables s. u = .g. L. S. In a Regge theory on the other hand.2 ^ ( 1 + cos6 s ). Blankenbecler.32) t = 4(q? + ml).t.
But for integral values of I. Screened Coulomb Potentials Regge oc(t) Fig.)e~ •fj. 16. It is known from the Mandelstam representation that the amplitude possesses a branch point at cos 9 = 1 + TOQ/2(^. For s — oo : cos 9t = 1 + s/2q^ — oo. Lehmann [162].e. .334 Yukawa CHAPTER 16. 16. However.5 for Yukawa potentials V{r) u: lm cose dr a(/j. Thus one is interested in establishing a representation of the amplitude in terms of ichannel Regge trajectories.((fc)/Kcos0t) schannel ichannel schannel variables the partial converges only within the socalled Lehmann ellipse^ which is shown in Fig.4 Yukawa versus Regge theory. So what can one do? 1 H. in the domain of schannel physical values of the kinematical of the ichannel).5 The Lehmann ellipse.r 1+m 2 /2q 2 o ^t "" Re cost: Fig. 16. the partial wave expansion is not valid in the physical region of the (i. the Legendre polynomial Pj(cos 9) does not possess a cut. But > > wave expansion X)(2Z + l)F.
16.6 The integration contour C". 16.5 The SommerfeldWatson Transform 335 T h e answer is.33) . k) have to the right potentials one can show t h a t the integral along the curved porA' of the circle at infinity tends to zero. contour C / / D ^ ^ . and there is only a finite number N + 1 in the domain of 9W = .^ B 1 A Fig.e. B Iml 1 \ i All \ r\ 0 / closed . Thus consider now the same integral but taken along the closed contour C shown in Fig. „ i.6. the contour representation (16. Then / JDD' ••• + j JBB' • • • = 2TTI V ^ residues f5n. Moreover. to use the SommerfeldWatson transform.26) but for a different choice of the contour. N F(6.16.k) »E n=0 [2an(fc) + l]/3n(fc) p sin 7ran(k) an(k){ cos 6) 2i y _ i _ i o o = F(s.t). all poles of Ql > 0. sin nl [16.1 / 2 . Again we have / Jc> • • • = 1ni y^ residues j3n „ For Yukawa tions A and f(l. i.e.
(16. Bose and H. at high schannel energies. 42.35) This result is valid for s — oo and t negative (cf.e. this behaviour has been confirmed in high energy hadronic reactions. we have (16. ^M)^E i 2 a f: a „t ( t ) ^ m <^'F(s. Regge [229]. Predazzi and T. Regge trajectories have. Apart from more refined details. 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.e. Eq. if ao(t) is the Regge pole with largest real part. 16. K.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 particular.e. (IM4) This is the asymptotic Regge expansion of the invariant amplitude for s —>• oo. i. The logarithmic potential has been investigated in a way similar to the Yukawa potential above by S.336 CHAPTER 16. Potentials with a shortrange repulsion more singular than 1/r 2 have been considered by N. we have P\/2+IR{Z) ~ 0(l/y/z) — 0 for s — oo (i. for high energies in • > the schannel). i.t)&B(t)sa°V. In subnuclear physics they played an important role before the advent of the quark idea and thus of quantum chromodynamics. as we discussed briefly in the above. The slope of the Regge function an(k2) is an " Simple cases are the harmonic potential and the squarewell potential. Screened Coulomb Potentials Since for s — oo : cos Qt = 1 H n ~* °° > 2?* and (from Tables of Special Functions) farMoo: Pa(z)* ^+l]\(2zY. Foissart [223]. 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. p. W. It follows that under these conditions. . for plots and discussion see R. Limic [177] and E.32)) and finite. J. of course. the amplitude can be represented as a sum over ^channel Regge poles. MiillerKirsten [29].
The way to do this is similar to calculations in Chapters 18 and 20. and is composed of quarks.16. and also as socalled "fixed poles" . H. W. like a meson or a baryon.** Whereas the former are related to absorptive properties of a reaction. The small imaginary parts of eigenvalues or lifetimes of resonances have not yet been calculated. J. a composite particle being one with structure. MiillerKirsten [209]. Regge poles are not the only possible singularities of a scattering amplitude in the plane of complex angular momentum. 20.1. Singularities appear also in the form of cuts in the plane of complex angular momentum.g. Related aspects are discussed in Sec. For an overview see e. The above treatment of screened Coulomb potentials is incomplete. called "Regge cuts". the latter are related to the distinction between "' elementary and "composite particles". .6 Concluding Remarks 337 important parameter in string theory.
This page is intentionally left blank .
Chapter 17 Periodic Potentials 17.1 Introductory Remarks Prom the beginning of applications of quantum mechanics periodic potentials of various types were immediately considered. Apart from the discontinuous KronigPenney potential consisting of a periodic repetition of rectangular barriers. 25.* These elliptic equations involve Jacobian elliptic functions like sn(a.lP = 0 in ellipsoidal coordinates leads to the ellipsoidal wave equation with sn 2 z and sn 4 z terms. M. In essence these periodic Jacobian elliptic functions. called elliptic modulus. 2K or AK. p. Arscott [11]. 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. Thus Jacobian elliptic functions also pro*To be precise: Separation of the wave equation Aip\x2. however.g.) which are functions that interpolate between the n. Taking in the ellipsoidal wave equation the limit k —> 1 (fc elliptic modulus) and putting tanhz = sin 6. depending on the parameter k. The reason for these difficulties is. the most immediate candidate is a trigonometric form like that of the the cosine function.0 < k2 < 1. motivated mainly by the regularities of the crystal structure of matter. separation of Laplace's equation Ai/> = 0 in ellipsoidal coordinates leads to the Lame equation (no sn 4 z term). are not much harder to handle than trigonometric functions.or 27rperiodic trigonometric functions on the one hand and the nonperiodic hyperbolic functions on the other. See F. 19 and p. The Schrodinger equation with this potential. and are themselves periodic with period e. since many analogous relations hold. example 14. that the Mathieu equation lies outside the scope of equations which can be reduced to hyper geometric type. 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. 339 . K being the complete elliptic integral of the first kind.
cosine and tangent. This is a significant difference compared with the Mathieu equation. these periodic potentials may be approximated by series of degenerate harmonic oscillator potentials. The Mathieu equation. which is conveniently written ^ + [\2K2sn2z]y = 0. or Sw&ve Schrodinger equation with elliptic potential.. need not be afraid of them. k2 — 0 which means that in > > this limit the Lame polynomials degenerate into the periodic Mathieu functions.2) where K2 = n(n + l)k2 and n real and > —1/2 and 0 < z < 2K. W. c n z and d n z are handled with analogous formulas.e. the three Jacobian elliptic functions s n z . and to relate these to the weak coupling bands or regions of stability.1. B. In fact. Like the trigonometric functions sine. Dingle and H. each with its own characteristic eigenvalue.1.7. Lame equation: H. or 5wave Schrodinger equation with the cosine potential is conveniently written d2v —  + [A . J. the Lame equation possesses 2 n + l polynomial solutions. [205]. 17. 8. 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. (17. such as sn 2 z + cn 2 z = 1 and double and half argument formulas etc..340 CHAPTER 17. those of integral order. which does not possess any polynomial solutions. J. Here the Jacobian elliptic function sn z is a periodic function analogous to a sine function and has real period 2K} For n = 0. 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. 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. See also the discussion in Sec. 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.4. The first of these conditions is already tMathieu equation: R. ''The reader who encounters these Jacobian elliptic functions here for the first time.2 . Miiller. . i. Periodic Potentials vide periodic potentials like trigonometric functions.1) where h2 is a parameter and — IT < z < IT. Muller [73]. For large coupling. W.2h2 cos 2z]y = 0. (17. Both of these domains are important for a host of other considerations. which one looks up in books like that of MilneThomson [196] or Tables when required. We might mention already here the much less familiar but more general Lame equation. The most important formulas for our purposes here are collected in Appendix A.
17.^ 17.1 The Floquet exponent We consider the given Mathieu equation with argument z and compare this with the same equation but with z replaced by z + ir. Although the mathematics literature on the subject uses the physical concepts of stability and instability. we see that in a plot of h2 versus A. these concepts are rarely explained as such there. .17. i. Our considerations below are deliberately made simple and detailed because their treatment in purely mathematical texts requires more time to become accustomed to. y" + \y = 0(h2). Vachaspati [242]. i. since the solution there is of the form cos vAz. The orbits of planets are stable in the sense that deviations from the recurring elliptic orbits are small. Here an important aspect is the determination of the domains of stability of the solutions and the boundaries of these domains.2 Cosine Potential: Weak Coupling Solutions 341 seen to rule out polynomials. In classical mechanics stability is treated for instance in connection with planetary motion.e.e. 17. i. and one can imagine that a nontrivial parity factor exp(in7r) then turns into a complicated phase factor exp(iz/7r). we see that for sufficiently large values of h2 the periodicity or boundedness of the solution is destroyed and hence becomes one of instability.e. the semiaxis A > 0. The instability of an orbit would be evident either from an unbounded spiralling away to infinity or from a collapse into the centre. Salem and T.1. a bounded function. as in Fig. 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. Thus a bounded trigonometric solution is indicative of stability and an unbounded exponential solution of instability. h2 = 0 belongs to the domain of stability.e. Looking at the Mathieu equation for h? —> 0. i. Correspondingly quantum mechanics — which requires the electrons of an atom to move around the nucleus — explains the stability of atoms. Thus the solution of the appropriate Newton equation is essentially a periodic function like cos 8 (in the case of planetary motion this yields the polar equation of an ellipse).2 Cosine Potential: Weak Coupling Solutions We consider first the weak coupling case. we compare the For interesting related discussions see M.2. Adding the negative term — 2h2 cos 2z to A. the case of h2 small.
— • . we obtain y(vr) y(2?r) a = — — a = —^^r. 2/(0)' yfr) . TT. i. Setting z = 0. (with Tn as translation operator) Tlxy{z) = y(z + TT) = cry{z). i. Periodic Potentials equations ^ p + [A . y+{z) . y~(z) = 6sin vXz.B and a.2/i2 cos 2(2 + vr)]y(z + vr) = 0.2h2 cos 2z]y(z) = 0. (17. 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). sin V Az or linear combinations. (17. b: y(z) = A cos vAz + B sin \f\z and y+(z) = a cos yXz. a = const.4) with (for convenience in connection with later equations) y'+{z) y'_(z) = avAsin\/Az = = bv\cos^f\z = —y(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 set (here and in the following frequently apart from contributions of 0{h2)) with constants A.342 CHAPTER 17.3) 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.e. ° ^ [1 + 0(h2)}. d(z + TT) cos 2z We see that there are solutions which are proportional. d2y(z + 7r + [A .e. y(2vr) 2 a = —7—^ = 1.
we choose these with the following set of boundary conditions: y+(0) = a. (17.y_] = y+(0)y'_(0) — y_(0)y+(0) = ab\f\.a2b2\ a± ~ 2a6\/A . y'_(Q) = bV\ with Wronskian W[y+. ay'~(ft) — bVXy+(n) and the Wronskian w + {y+(7r)y/_(7r) .2 Cosine Potential: Weak Coupling Solutions 343 The solution y(z) is an arbitrary solution. \y+>y\ = y+{ir)y'{ir) .oWX] . 7 y\z) = ay'+(z) + (3y'_{z). Setting in these last equations z = 0 and using the previous pair of equations and the boundary conditions. [y+(v) . i.y+(7r)y_(7r) = 0 or abVXa2 . y'(z + n) = a[ay'+(z) + py'_(z)}. (17. Using Eq.W or fy+(n)aa J/_(TT) = W ^ ^ J a ^\ = n For linear independence of a and /3. This expression is equal to the coefficient of a2 in the quadratic equation for a. the determinant of the matrix must vanish.17. This means. and by'+(?r) = —ay/Xy{n).3) we obtain y(z + ir) = a[ay+(z) + @y(z)].4).e.aa][y'_{n) . 2/_(0) = 0.y{ir)y'+{K) = abVX in agreement with its value at z = 0 above. The roots a± are therefore obtained as 2bV\y+(ir) ± J4b2\yl(Tr)4. yV(0) = 0. whereas the solutions y+ and y_ have here been chosen specifically as even and odd around z = 0 respectively. with constants a and /3. We can write therefore. y{z) = aty+(z) + 0y(z).{ay'_{ir) + bV\y+(ir)}a Now from Eq. we obtain V (TT) = ay+ (TT) + (3y _ (TT) = < [aa]. 2/'(7r) = a ^ W + ^ .y+(7r)j/_(7r)} = 0.
e.1. <T__ + <T_ = 2y + (vr) .7.2 + 2 9 ) ^ 2 + 64(1/2 _ 1 ) 5 ( z / 2 _ 4 ) ( l / 2 _ 9) + u ^ ^ ^ . (17. 8. This is the easiest application of t h a t method since only simple trigonometric expressions are involved. 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 . ** An important consequence of Eq.5) a The parameter i/ is known as Floquet exponent. (17.= — .6) We can derive various terms of this expansion perturbatively with the method of Sec. Setting cr+ = e .344 Setting / = 2by/\y+(Tr). This calculation is demonstrated in Example 17. . Periodic Potentials g= 2aby/\.+ 2 2(i/ l) 32(i/2l)3(v24) ( 9 ^ + 58. **This point is not immediately clear from mathematics literature.22)) the boundary conditions of periodic solutions.1169)/i 12 64(Al) (A4) (A9) 5 2 + 0{hw) r 16 (17. 2 _ hA 2(A1) (13A25)/i8 32(Al)3(A4) 12 ift (45A3 . i.5) is that for h2 = 0 we have v = A/A and hence more generally v2 = \ + 0{h2). (5i/ 2 + 7)/i 8 vz + — . we have <7± — CHAPTER 17. _ / ± v 7 ^ V .0 This series may be reversed t o yield the Floquet exponent.8) "Observe that for integral values of v (in lowest order of h2). (17. where the constants are usually taken as unity from the beginning.. cr+ = 1/<T_. T h e result for A is the expansion A = h4 2 . (17. this implies y+(Tr)/a = ± 1 . Prom this condition we determine later (cf. Eq. 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.e. i.455A2 + 1291A . <7__cr_ = 1. . 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.
35A + 8 .35A + 15A2)/l8 /. p. Schiifke [193].7 to obtain the eigenvalue A as a perturbation expansion for nonintegral values of v and b? small. 8. These cases therefore have to be dealt with separately.9a) 4(1A)>/A 64(A4)(A1)3A^A One observes immediately that these expansions cannot hold for integral values of v or A.„„ x n.11) or smvz or e±ivz. Solution: We write the eigenvalue equation A = v2.17. which can then be rewritten as Dvy = 2fe 2 (A + cos 2z)y.1 below one obtains the following expansion for cos7rz/ which is fairly obvious from Eqs. (17. n an integer. (17. For these integral cases of v we demonstrate in Example 17.12) S e e J. rr +h* 3 =7rsin7rvA 64(A1) (A4)AVA +Q(h12). — 2/i 2 A and insert this into the Mathieu equation y" + [A — 2h2 cos 2z]y = 0. Meixner and F. (17.1) 15A2 .1: The eigenvalue A for nonintegral v and h small Use the perturbation method of Sec. = J—^ + o(hr). (17. . W. To lowest order the solution y is j / ° ) = yv = cos vz n where d2 Dv:=—~+u2.9a) and is therefore not derived here in detail^ ri/ = VA + rCOS7Tf = COS7TVA + /l 4 h4 7rsin7rVA = 4\/A(A . With a perturbation theory ansatz for y+ around h2 = 0 as in Example 17. cos TT\/A 32A(A1) 2 .2 how the expansions for small values of h2 may be obtained perturbatively. Example 17. as will be done later.10) A = n2.jl2.3). (17.5) and (17.e. VA = ±n.9b) TT2 Next we observe that if we demand that the solutions y+ (z) and y_ (z) satisfy the condition (17. It follows that in these cases i? = n2[l + 0(h2)]. 124. we must have a2 — 1 and hence e 2iv/A7r = 1} i.2 Cosine Potential: Weak Coupling Solutions and so 345 (8 . (17.
h2). (17.e. v . The recurrence relation can also be obtained by substituting the right hand side of Eq. Next we treat terms yv+a in j / 1 ' in a similar way. Periodic Potentials The complete solutions of these cases are written respectively in selfevident notation cev(z. i.14) = = = cos(y+ 2)z + cos{y— 2)z. 17. when a = 0 or 2v + a = 0.v±2) = l. (17. v .16) directly into the original equation in the form of Eq. or A = 0(/i 4 ). . With our perturbation formalism — as demonstrated in cases considered in Sees. v)yu + {v.346 CHAPTER 17. in each case 2cos2z2/„ = 3/„ +2 + j / „ _ 2 . '. by J. h2). siniy + 2)z + sin(i/ — 2)z. me±v(z.11) and so in R}?' may be cancelled out by adding to j / ' 0 ' the new contribution a(2v + a)' = 0. The right hand side of Eq. sev(z.. se and me as* oo i y = „«>) + yW + i.11).h2). We assume for the time being that a ^ 0 and 2v + a ^ 0 (the latter case requires separate consideration. We observe that in these cases 2cos2z cosvz 2cos2z smvz 2cos2ze±il/z i. Schafke [193]. Proceeding in this manner we obtain the solutions ce.(2) + . Meixner and F. of course.e. . up to 0(h2) we have A = 0.16) *These smallh expansions are convergent.2) — 2 "^ 2(2i/ + 2) " + 2 vanishes.2)j/„_ 2 + {v.v + 2)yu+2]. for instance. (17. i.13) The first approximation y(°) = yv leaves unaccounted on the right hand side of Eq. (17. = .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. v) = 0. so similarly yW leaves uncompensated / # > = ft2 ("• v ~ 2 ) R (o) .!/) = 2A We now observe that DvVv = 0. (17. ("• ^ + 2) (o) 2(2^ .11) terms amounting to i?i 0 ) = = 2 h 2 ( A + cos2z)y„ = 2h2Ayu + h2(yl/+2 + y1y2) h2[(v. as shown. (17. e ± i ^ + 2 ' 2 + e±i("2)z.3. .3 and 19. (i/. where except.2) _ __ _ (y. i. (17.„ + £ h2i £ P2i(2j)yv+j (17. since j / 0 ' = yv leaves unaccounted R\. so that also Dv+ayv+a and hence Dv+a = Dv + a(2v + a) and Dvyv+a = a(2u + a)yv+a(17. W.2 — we can write down recurrence relations for the perturbation coefficients pa(2j) together with boundary conditions.e. h2(u. and (v.e.2) which applies when v is nonintegral.2. see Example 17.14) therefore leads to the following nextto leading order contribution (v.15) Hence a term fMy„+a on the right hand side of Eq.
. Example 17. See e.2—r——— (i/ + 2.. 2 ( 2 i / . + y~i) (17..2 Cosine Potential: Weak Coupling Solutions 347 together with the equation determining A. (17.. 0 = h2(u.7).18) Again applying Eq.. For v = 1 we then have yi = cosz = y~\. * Note. Hence we obtain 0 = /i 2 (2A + l ) f h4 + ••• . one has to deal with each integral case separately as in the following example. We consider the specific unperturbed solution yv = cosvz which is the dominant contribution j / ° ) of our solution y. Schafke [193].2 ) v(y2.e.1 with A = v2 — 2h2A but with v = 1. „ . 121. is set equal provided the coefficient of the sum of the contributions in 3/1 contained in R\ to zero. v) ' ( 2 v + 2) (17.17) Inserting 1 for the stepcoefficients of Eq. (17. thus determining the quantity A.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. i. p."2).14).u) + h4 (". . Meixner and F. Solution: We start as in Example 17. (17. These expansions are actually convergent with a definite radius of convergence as explained in the mathematical literature. J.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)+. we obtain immediately 0 = 2Ah2+ 2(^T)+' which verifies the term of order h4 in Eq. Following the first few arguments of Example 17.v)\ . fo^ + 2) . The higher order terms naturally require a little more algebra.. . however.g.R{ .1 we now have the situation that the contribution y\ to the entire solution leaves uncompensated on the right hand side of the equation the terms amounting to R(°li = = 2 h 2 ( A + cos2z)3/„=i =2h2Ayi+h2(y3 h2[(2A + l)yi+y3].17. This completes the determination of the solutions for nonintegral values of v in the domain of small values of h2. In the case of integral values of i>. W. the solutions are not yet normalized. .
The specific normalization here is taken as I /*7r rn 1 = — / dzce2(z. Schafke [193]. cos z h cos Zz 8 h2 h4 ( cos 3z cos 5z — cos iz A 1 8 4 l 8 48 h4 (cos 3z cos 5z 1 4 I 16 48 (17. The reverse situation is later of importance. the case of A an integer.* The associated solution is ym+ym. In the above we considered the case of v an integer and calculated the eigenvalue A.19b) This expansion agrees with that given in by Meixner and Schafke [193] (p. W. 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. Eq. i. 120.19a) This expression agrees with the result given in the literature (there the eigenvalue is called ai). With normalization (not to 1. see below!) one obtains the additional terms given by Meixner and Schafke [193]. J. 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 . *" Jir Thus we have and uses / dz cos mz cos nz = — <5 mn . 123) (there the solution is called cei(z.24c) below) .348 From this we obtain 2h2A = h2 CHAPTER 17. Inserting this into A = v2 — 2h2A 8 for v = 1 we obtain (cf. and v is to be found in ascending powers of h2. 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].n integers. h4 (17. (17.e. h2). We consider this case again in an example. We introduce a normalization constant c and rename the normalized solution y then yc. Periodic Potentials h4 r ••• . Meixner and F. m. . p.
W. We thus obtain the expansions cos"\/A7r sinvAvr = = cos27r + (A — 4 ) ( — s i n V / A 7 T ) A = 4 — = + ..9b) and considering the approach A — 4 gives > (A4)2 + .3 1 x 4 + 8) 274233 (. o 1• • • .1) (A .1) .9b) about A = 4..9b) v — 2 + <5. Gubser and A. H.17. . (17. S. (17.3: 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. 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 . on the left hand side of Eq.l ) 2 (A . (17. (11A2 . so that 2 2 cos nu = cos 7r(2 + S) = cos 27r cos TC8 = 1 and comparing with the above.Q .• 1 + 5n h 2 s 2932 Setting in cosm. Substituting these expressions into Eq. Liang and Yunbo Zhang [189].4) 2 7T 2 /l47T2 + • 8 h87T2 8A(A . J . fe47T2(A4) 4%/A(A .4)Av A2v A ~ 32A(A . MiillerKirsten. ^S.3 5 A + 8)TT2(A4) _ 3 / / 64(A .1)2/A TT2 +h8 (15A2 . We set A — 4 ~ 4e and expand the cosine and sine expressions appearing in Eq. Hashimoto [124].= l 2vA sin27r + (A — 4)(cos vA7r)x = 4—= + ••• = j= h •• • .4 ) + .('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 .2 Cosine Potential: Weak Coupling Solutions Example 17.3 1 A + 8) ( A . J. . Manvelyan. we obtain^ S 3 V 2 and therefore i/ = 2  iy/E ( h 3 \2J 7i .
10) applies. M. in F.21a) and (17. We verify these for the large/i 2 solutions in Example 17.350 CHAPTER 17. i.w^. 28 . for instance. We now derive these boundary conditions. "These conditions.29. may also be found. Periodic Potentials Our next immediate aim is to specify the solutions for which Eq. Arscott [11].e. we have" y+{z ± 7r) = a cos vA(z ± TT) = y+(z) cos vXir = TV{Z) F sm v<\7r. These are defined by specific boundary conditions. Taking in particular Eqs. (17.W4W=UtWSta±I(.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.6.4) into account.. bV\ y+(0) 2/40) 07. solutions for integral values of v. 17.1 Boundaries of domains of stability.2. concentrating again on the leading term in the even and odd solutions y±(z) for ease of understanding. Fig. . = and s+w 2tW±I.21b). 17. (17.2*0 y~(z ± 7t) = 6 sin VA(z ± TT) = y_(z)cos VA7r ± 6 cos vXzsin v\ir. pp. as also conditions (17.
f^\y+(j) /vr\y_(7r) and for the derivatives ^(i)^^'(D^ . 1+ y+(f)^(f)y_(f)yV(f)' 7T Using the Wronskian (which we actually had above!) W[y+. .y V+ \ 2 7"^ V 2 fla^+U a cos V A  ] 6^A + a&( sin\/Aj ^A = abV\. /7r\j/!_(7r) y L l 7TJ . /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.4(f) (17. ^\y±M /vr\yL(7r) . which after some rearrangement can be written y+(vr) _ . o y+(fy(f)ob\/A _ = 1 + 2I&A/X . Then y+ K 1 _ 2/+00 7T y V + (f)[i + 2/±Ml ^ ] y'il) 2y_(f )i/+(f. Then y+ y 351 U .2 Cosine Potential: Weak Coupling Solutions We select the equations with lower signs and put z — TT/2. .17. we obtain y+(7r) _ i  2y_(f).21a) Since (with the help of the Wronskian) *W4)MM?)^ we also have y+(v0 _ .
24b) 8 h2 z —— cos 3z + • • • . 19. (17. These may be subdivided into classes depending on whether the integral value n — 0. (17.24a) ao = hA h6 b\ = 1 — h —— + — — • • • .1 for the first few eigenvalues which are boundaries of domains of stability (as discussed in Sec. y > se2n+i=qo..T + . in this order. the case of ai with solution cei (of period 27r).24c) 8 h2 sin 4z \ . Periodic Potentials and hence** a afc\/A For integral values of v (in lowest order of h2) we know from Eq. The ordering of the eigenvalues which then results is a0 < bx < ai < b2 < • • • < bn < an < • • • for h2 > 0.. 27r.3.' fh2 V2ceo = l . Finally we cite here some expansions of the eigenvalues along with those of the associated solutions (an explicit example. y > se 2n +2s 9o +i The functions so defined have respectively period 7r. of v is even or odd.352 CHAPTER 17. A 2 y • ce2n=q0i. sei = sin 8 64 /i4 h6 a\ = 1 + h2 — . .. (17. y > ce2n+i=qo. 17.5) that the left hand sides of Eqs.23) We can now see how the domains of stability and instability arise. 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. There the second term on the right is nonzero. IT. 2K. 17.1.2): hA 7h8 29h12 + Y l28^04.1). 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 . was treated in detail in Example 17. (17. and whether the function is even or odd. (17. A *• b2n+2(h2).— — — • • • ..+ . Mr7nA (17. A > b2n+i(h2).21b) must be ± 1 . We see that there are four possibilities and hence four different types of functions.2.• a2n(h2)..2 we require this relation for nonintegral values of v. This is shown schematically in Fig. . se2 = sin 2z 2 12 13824 ' h2 z —— sin 3z + • • • . ce± = cos 8 64 h4 5h8 b2 = 4 1 . (17..21a) and (17. (17.24d) v 12 ' **To avoid confusion we emphasize: In Sec.22) * a2n+i(h ). These are defined by the following boundary conditions.
Thus in Fig. In the case of the cosine potential. The cosine potential cos 2z depicted in Fig. 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. Dingle and H.3. B. W.17.24e) 17. . the walls are not infinitely high and hence tunneling occurs from one well into another.17„.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. 353 .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.z3 ce2 = cos 2z . 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. Consequently one expects the eigenvalues in the large h2 domain to be given approximately by those of the harmonic oscillator. 17. 13824 .> (17. 17.2 or by calculation that the potential 2h2 cos 2z has (harmonic oscillatorlike) We follow here R. Miiller [73]. We can see from the inverse of Fig. 17.hz — 12 .2cos4. . .1 Cosine Potential: Strong Coupling Solutions Preliminary remarks In the case of large couplings h2 we observe from Fig. 17.1 that the boundaries of regions of stability merge to lines. J.3 Cosine Potential: Strong Coupling Solutions „ 5/i4 a2 = 4 H 12 763/i8 .2 The cosine potential. of course. 17.
Sees. 8.27) .. (17. We therefore wish to expand it about these points.3) and is.25) This is a Weber (or parabolic cylinder) equation which we encountered earlier (cf.2.3. and we set this equal to q.26) 17.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. simply the onedimensional Schrodinger equation for the harmonic oscillator. where A / 8 is a remainder. Periodic Potentials minima at z — ±ir/2. In the case of finite heights of the potential barriers.1. Taking the remaining terms of the cosine expansion into account. This equation has normalizable solutions for X + 2h2 2h q0 = 2n + 1. Ion (17. (17..2/i cos z V = 0. we therefore set 9 A A = 2h2 +2hq + —. n 0...2 The solutions We insert Eq. 4 (17.26) into the original Mathieu equation and obtain y" + 2h q + rrxr .354 CHAPTER 17. Thus in this case the quantity on the left is not an exact odd integer. but only approximately so. of course. as in the case of the cosine potential. there is tunneling from one barrier to the next.2 and 8. Thus we rewrite the Mathieu equation as y" + A + 2hz cos 2 \z ± 7T y = 0 and take c o s 2 [ z ± .
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). (17. another solution. and substituting this into Eq. the linearly independent one. (17. Choosing y(z) = A(z)e2hsinz.31) We make the important observation here t h a t if one solution of the type (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.27) we )A = 0.3 Schematic picture of Mathieu equation eigenvalues.30) D A): COS « = 4z + 1 iq sin 2. 17.28) A" ±4hcos zA' + 2h[ q =f sin z \ A. z).17.29) is known. (17. can be .29) the equation can be written in the form D^A where 1 2 6 /i (16A" + 2 A ^ ) . 2 (17.
94) = (9 + l)(9 + 3).30) amounting to Ri ] q = . 2 /i D^Ag = 0. Periodic Potentials obtained by changing the sign of z or alternatively the signs of both q and h throughout. of course. 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 (17.30) can be cancelled out by adding to A^> a new contribution — fiAq+4i/2i — except. (17.356 CHAPTER 17.4 ) ^ .4)A. ? .^ 6T [(«>« + 4 ) A ? + 4 + (Q.32) cos^'+i) [\K . 2 A A = («. ? .33) (17.36) Then ^4.35) can be cancelled out by adding to A^ the next order contribution A& = 1 (g. (17. 2[(92 + l) + A]. 1 + 4 ) ^ + 4 + (^. (9.e.4 ] .\z To lowest order the solution A is A^> = Aq. Aq+4 2Jh . (9.9) (9. (17. ^ ) A + ( ? .9) = 0. i. when i = 0 — the terms not involving Aq in (17. D{q%Aq+4i = 0.35) But since and so a term iiAq+n on the right hand side of Eq.35) not yet taken care of is set equal to zero.Q)Ag + ( ? . (17._ 4 . With some algebra one finds that 16^ + where (9. .34) The leading approximation A^ = Aq therefore leaves uncompensated terms on the right hand side of Eq.94) \ — Aq4 (17.(°) + A^ is the approximation of A to that order provided the coefficient of the term in (17.g + 4) .9 + 4) = (9. (9l)(93).
40) . h).38) Aq8 + 1 2 Clearly we can continue in this way. q..37) Here all the terms. ) ^ 4. (q. (17.g4) 4. (17. q . +4.q + 4) (g._ 8 (17. h) = A(z. h) = A(z. (17. . Thus 0 = (g. .4. h).. q . (q. We note here again that if one solution y = A(z)e2hsmz 2hsmz is known. q (17. an associated solution y — A(z)e~ with the same eigenvalue and thus the same expansion (17.q + 4)(q + 4.30) and hence y = Ae2hsmz to be a solution of the Mathieu equation.)+ & ! = « ( . A. q . Then for A = A(0)+A(D+A(2) + .q 4)Aq+4 q + 4 1 q + + + + + <^(.. can be taken care of by adding to A^ + A^ the next order contribution A^2\ where A (2) 2h u 2 (q.4 . In its turn the contribution A^1' leaves uncompensated terms on the right hand side of Eq. A(z.37) and so on — left uncompensated — must vanish. . the sum of the terms in Aq in Eqs.39) is obtained by changing the sign of z or alternatively the signs of both q and h throughout. q.4) (q .3 Cosine Potential: Strong Coupling Solutions 357 This equation determines the quantity A in the eigenvalue equation to the same order of approximation. Thus in particular A(z. except those involving Aq.4) 1 1 (q. to satisfy Eq. (17.g) 26/i L_ 2 13 /i 2 ^±A{q + ^q) + —1 ^A{q^q) '17.8)A.39) which is the equation from which the quantity A and hence the eigenvalue A is determined.q 8)Aq+8 (o)' ^±V(q 1 4.17.30) amounting to RW 9 = fag+ — 2?h 1 ^ 2™h? +< 4 ) R(o) \ .35) and (17.g + . q.4. Q . q.4) (q .g + 4)(g + 4.<J+4 Aq+& + 1 2 1 .
where Hen{w) is a Hermite polynomial when n is an integer. Before we study the eigenvalue equation in more detail.d2B w dw2 + w(l w dB_ dw d dw 1. 2 w2 + H* (17.i ) We encountered Hermite polynomials earlier in Chapter 6.42) (17. We return to Eqs. The differential equation there is Eq. (17. We observe that for the large values of h we are considering here. this means the solutions are valid in particular around z — 0 since then l / \ / 2 3> l/Vh ~ 0.e.46) 2s(«.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 .41) so that the early successive contributions decrease in magnitude.358 CHAPTER 17. 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. a solution Bq(w) of the equation Dq Bq(w) — 0 is Bq(w) oc He^^^iw).45) {B) We recognize Dq as the differential operator of the Hermite equation. it is found that D(B) z+ — ]B = 0.43) 9 2eh .67)." 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.) i ( Q . and discussed there. Note different definitions in use. .44) where D(B) 9 = dw2 $_ !)• (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. (6. i. (17.
which is an amazing and unique property of the Mathieu solutions — cf.e.q + 4 ) £ g + 4 + (q. i. (17.e.3 Cosine Potential: Strong Coupling Solutions 359 because then as shown in Example 17.. . 1 COS I .47) where the coefficients are the same as before.50) C = 0.3 ) \He* ! (9+1) H. Finally we obtain a third pair of solutions. The solution with contributions B^°\B^>. 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).4 (using the known recurrence relations of Hem{w)) the perturbation remainder can be linearized. with the equation C" + 4h cos z C + 2h ( q ..qi)Bq_4.d2Bq + w{l dw2 "» 2 >^(» 2 + ^ = {q.17.. 16hJ (17. i. q)Bq + (q. also the same coefficients in the case of functions C below. 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. y — Cexp(2hsinz) and Cexp(—2/isinz). A{wz (17. forms a rapidly decreasing expansion provided that \w (z)\ « Vh.34).51) . i.e.49) Then the dominant order solution is C^°\w) = Cq{w) with and — again choosing the multiplicative factor suitably — Cq(w) = 2i(" +1 ) ^ .48) This shows that this solution is valid in particular around z = ir/2. (17. i.sin z + by changing the independent variable to w(—z) = 4v/icos I 7r z (17.7 T H 1 2 Z 4 <1. (17. as in Eq. the expansion for the quantity A being identical with that of the previous case.
we have a remainder •R:=w2He„+w(l Using the recurrence relations wHen(w) = Hen+i(w) + nHeni{w).44).A Hen . Again the solutions have the same structure as in the previous two cases and the eigenvalue expansion remains unchanged. (17. (17. ie 1 COS I .52) and the coefficients are again the same as before. Periodic Potentials since then (using the known recurrence relations of He^) ~Mw2^ +w(l+w2)^ + (w2 + o dwz dw V 2 (q. (17. Example 17.4.h) ocB{z. show that the perturbation remainder of Eq.q. These domains of validity are indicated in Fig. .h). (17. A closer look at that expansion is our next objective. (17. Solution: Inserting the dominant Hermite function solution into the right hand sid of Eq. We have thus determined three pairs of solutions of the Mathieu equation. can be reexpressed as 72 = (n + 1 ~(q + l)Hen+2 l)He.59d) of Hermite functions Hen(w).4)C g _ 4 . 2 J where n=(ql). q + 4)C 9 + 4 + (q. (9+3) (2n 2 + 2n + 1) + . He'n(w) = wHen(w) Hen+i(w).53) This solution is valid in particular around z = —7r/2. q)Cq + (q.360 CHAPTER 17. 17.q. 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/2.4: Hermite function linearization of perturbation Using the recurrence relations (6.q.(w2 + A)Hen.n2(n l)Hen2 .w2)He'n .7 T 1 2 Z 4 <1. The solution y = C exp(2/i sin z) is a decreasing asymptotic expansion provided that w(z) ^ v^/i. 2' the remainder 72.54) where the proportionality factor is complex.44) can be linearized with the same coefficients as in the case of the trigonometric solutions. each of them associated with one and the same expansion for the eigenvalue.(ql)2(q3)HeX (95) i(« 2 + i) + iA Hei (91) . We note that C(z.
57) .3. (q.g) 1 1 and so on. q) = 2[(q2 + 1) + A]. The coefficients have thus been constructed in a way which now allows the formulation of a recurrence relation. ^ 1 Hv" '^ ^' ^ ' .3 Cosine Potential: Strong Coupling Solutions It follows t h a t with the definition of Bq (w) as Bq(w) that He K .l)[l(q _!)]. the solution y = exp(2/i sin 2)^(2).56) (94..)Bqi(w). (?) 1 1 + (9.j): y = = = e e2hsinzA{z) oo 2ftsinz r=0 V .3).' 361 in 2*f*1>[i(gl)]! = (q.q4) (17. 17. 2 + ^ M 3 + (17. ) .i ) ( w ) 2l(.55) M2 = M.g4)(q4. We see that (a) every coefficient Mr results from a sequence of r moves from q back to q. In addition (c) each move except the last has to be divided by the final displacement from q divided by 4. (b) The allowed moves are +4. r j=r. .q 4.g. l)Ag4 + P2(q.q + A){q + A. (q. .3 T h e eigenvalues We saw above that along with each solution the following expansion results: 2A = M 1 + ^ M where Mx = 2(q2 + l).q + 4)Bq+4(w) + (q.1i . q + 4) = (g + 1)(? + 3).l)(q . e.0 and —4.17. no intermediate move to or from q being allowed. ~l)Aq4} 2h 1 { P 2 ( < Z 2 + + P2(<? lAq+i ) 8 A ) ' " ' +P2(q..4) = (q . 2)Aq_8} + • (17. We can write the solutions obtained previously. q . with (. in the following compact form with coefficients Pr(q.q + 4) (q + 4. q)Bq(w) + (q. (q.j^0 e 2ft sin z A + ~ww 1 7 i + ^7ir{ p i(9' ^ ^ + 4 + Pi(q.
g + 4)(g + 4.g + 4)(g + 4. P2(q 2) Pl(g.j) = (l)rPr(q.g + 4) + 1 1 1 and so on. B.g + 12) 1 2 3 (g. Dingle and H. g + 4 j ) P r . x)_ (g.4. Miiller [73]. l ) .3) ^3(9.P 2 ( g .g + 4 ) ( g + 4.59) One can now write down formulas for M^r and M<ir+\ in terms of the coefficients Pr{q. 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. From the coefficients (17. j + 1) (17. M2 = P 2 ( g .j ^ 0) = 0.g + 4) 2 1 (g. 0) = 1 and all other P0{q. .g + 4) ^3(9. (17. as jP r 2 (g. j .362 Clearly CHAPTER 17.60) M2r+l = = *R.58) with P 0 (g. i.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.l ) . g + 4j)P r _i (g.j).g + 8)(g + 8. the factor j removing a duplication of factors in the denominator at the junction. Periodic Potentials PlM)=(M+i). (17.j)^2(9.j)Pr+1(g.g + 4)(g + 4. g + 4j)P r _!(g. J. and in general* M2r = J]j[P2(g. 2) 1 2 ' ' ^ ' ^ 1 (g. . and the contribution from a sequence of r moves from q + 4j back to g.j).j). p = (<?><? + 4 ) ( g + 4 >g + 8) . Thus we have.j).g + 4)(g + 4. j) = (g + 4j .e. W._1) = Mzil.i (g.j).34) we deduce that Pr(q.1) + (g + 4j.g + 8)(g + 8. j) +(g + 4j + 4. r ^JtPr^jOPr+lfejV^rfejO^r+l^j)] r 2^jPr(g.j)] = 2j]jPr2(g. for instance.
62a) f R .4.3. We observe that there are regions. 17.^ . J. Goldstein [115]. These domains in the interval —7r/2 < z < 7r/2 are indicated in Fig.284g + 57) + (17.* We observe that the expansion remains unchanged under the replacements q — —q. *E. It will be seen in Chapter 18 that in the case of anharmonic oscillators. a= (8/i)3(9x) [1(91)]! 1 . Dingle and H. . t Terms up to and including those of order l//i 5 had been obtained by others and by different methods. Ince [136]. h — —h.61) +288161796g2 + 130610637) These are the terms given explicitly in the reference cited above.5 the following proportionalities there: B[w(z)] = aA(z). W. S.3 Cosine Potential: Strong Coupling Solutions In this way one obtains the eigenvalue expansion A = 363 _2h2 + 2hq^(q2 1 . In fact we demonstrate in Example 17. Up to > > this point. where solutions overlap. as pointed out earlier. Its precise deviation from qo. remains to be determined next.2 * +3) + ^ 2 (9g4 . L. [137].( 3 ^ . [116].4 The level splitting Above we obtained three pairs of solutions along with their domains of validity as decreasing asymptotic expansions. the solutions and eigenvalue expansions again have the same type of symmetries.(5q 4 + l)^Jiq(q2 + 3) + 410g2 + 405) + 34q2 + 9) .17. 17. of course. This means that the in approaching such a domain the appropriate solutions must become proportional in view of their uniqueness.92q3 + 70g2 . q is only known to be approximately an odd integer qo = 2n + 1. Miiller [73]. B. as a result of tunneling between wells.^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.
1 ) ( 7T+ z 1 — COS 2 1 \l^+1> 2 Z C O S S ^ . Example 17. The first term of the expansion of A(z) is the function Aq(z) given by Eq. Solution: We begin with the solution containing the function A{z).4 Domains of solutions and their domains of overlap.5: Proportionality of solutions in domains of overlap Show that in their common domain of validity B = aA and C = aA.COS 7T 2!42 \4 2 COS /l 7T V4 Z H 1 2 Z H 1 2 .7T H Aq(z) = = C O S ^ .62b) *z domains of overlap Fig.1 4 . and determine the constants a and a. a ._ ^ l CHAPTER 17.(17. 17.32). Periodic Potentials + _(3<?+2<? + 3) + 2 15 /i 2 (9g 4 + 92g3 + 70q2 + 284g + 57) + • • • (17. We write and expand this function as follows: .Z (g+1) 1 H 1!4 4 2 1 (g + l)(g + 5) .1 ) ( 7T + .364 and C[w(z)] =a~A(z).
This is essentially the expansion of the parabolic cylinder function with the exponential factor removed — see Eq. i.63) In Example 17. s .92q 3 + 70q 2 . the coefficient of the dominant factor cos' 9 _ 1 ^ / ' 2 (7r/4 + z/2) in the result is seen to be J (8/i)3 ( g . (6. with (8h)li''1) (3g2 . For further details and explicit expressions of higher order terms we refer to Dingle and Miiller [73]. In order to link our asymptotic solutions with the Mathieu functions of integral order defined earlier in the consideration of small values of h2. (17. the coefficient of the dominant factor c o s ( .!)(.6 at the end of this section we derive for these \axgeh2 solutions the conditions similar to those given in various equations in Sec.q) 2 7 /i 1 1 4(g + 3) 1 (g + l ) 2 ( 9 + 3) + 29ft ti(9l)]l i (8/1)4 ( 9 . _ 1 " 2 ( 7 r / 4 + z/2) in the result is seen to be a = 1 2 /i 7 fa.22).g. A*1) given by Eq.2g + 3) 1 [£(91)]! 2?h 215h2 (9g 4 . with e.3 Cosine Potential: Strong Coupling Solutions 365 Inserting this into A = A(°> + A' 1 ) + • • •. 17.53) for the expansion. these are conditions imposed on even and odd solutions at the point z = it/2. we obtain Bq[w(z)} H.36). \ql)/2(w) 2i(9D[i(.i ) 1 (g. (17.2 for the case of small/i2 solutions and thus insure that in both cases the same boundary conditions are obeyed.46).2. These are the conditions of Eqs. we have as even and odd solutions y±ocA(z)eZflsmz±A(z)e 2/isin. As we saw there. (8._i)]! w*te» 2l(91 (. Hence we have to construct even and odd solutions and impose the boundary conditions at z = 7r/2.17.3) 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.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 (17.61) for the connection and Eq.e. (17. Since the solution with A is obtained from that with A by a change of sign of z.i ) [(91)]! It follows therefore that over their common range of validity B = aA with a = b/a. we have to impose the appropriate boundary conditions.284g + 57) + • The other relation is obtained similarly.g + 4) (q + 4.
80.g.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' (17.66) then follows from the recurrence relation (6. .62b).66) ^E.59d). W. Periodic Potentials We can extend the solutions (17.63) to the domain around z = ir/2 by using the proportionalities (17. 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\ ! ' . a ± C[w{z)}c_2hsinz^ a (17.64) These are now the solutions around z — ir/2 which permit us to apply the boundary conditions (17.62a) and (17.366 CHAPTER 17. i.e. The functions defined as in Eq. Thus y+ = ce and y_ = se.e.» a \dw _ C[w = 0}c_2h a w=0 = Q (17. He2n+1 (0) = 0.66). With the help of the duplication and inversion formulas of factorials. and then replacing 2n by i>. 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. He'v(0) = Hev+1{0). The second of relations (17.22) and to correlate the solutions to those for small values of h2. Magnus and F. p. (17. one obtains the first of relations (17. He'u{w) = wHev{w)He„+1{w). i. Oberhettinger [181]. Then y± oc ^ M e ^ i n .
82g2 .! ) ] ! [ .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 .1) 7T . and using the reflection formula (8.67c) 4 2 \dw)w=Q C[w = 0] [\(q .Qo) + 0[(q . the solutions would be given by q = go = 3. Hence expanding the cotangent about these points. and similarly for C and its derivative.39 (17.3 Cosine Potential: Strong Coupling Solutions 367 With these expressions we obtain by insertion into B and its derivative.67b) <f ~ 82g^ . f^(16h)i/2e4h 2 [Uq1)]\ 3(g2 + l) + 2^h2 (9g4 . .7.67a) and (17.136g + 9) + • • • .\4 q + m { 4« VJL ^[±(g3)]!sin{f(gl)} [J(ffl) "(^L^»{^': _ _q_ ¥h + g . we have cot j ^ ( g ~ 1)} = ~\<(l .106g .11.» 2 6 /i the condition reduces to the equation cot{ .67a) g 106g 87 2 14 /i 2 (17.g0)3]. (17. B[w = 0] 2TT _ _g_ ¥h+ g 24/* _ JL g4 .67c).87 4 2 ¥w (17.69) If the right hand side of this equation were exactly equal to zero.65) we have B[w = 0] _ a C[w = 0] ~ ~ ^ 4/i (17. from which one obtains "1 1) 4<*3) 2(9!)/2 2 ^ \b.68) Inserting here expansions (17.( g .39 2~^h? (17.40g3 + 18g2 .17.67d) Considering now the second of conditions (17.
136g0 + 9) 2 13 /i 2 1 4 2 19 3f(9gg .528gg + 3307Q . Periodic Potentials ' " Vn [i(gol)]' 3(g02 + 1) 26/i .65) we have (dB/dw)w=0 (dC/dz)w=Q Proceeding as above one obtains tan{(ql)\ 7T .136g0 + 9) + • Clearly the last of conditions (17.65) yields the same result except for a change in sign.136q + 9) + .5. = _a a 6 _4h (17.h qqo = * [£(*>!)]! 1 3(902 + 1) 26h +^^{9q^ .368 Hence we obtain qq0 a [2(16h)i°/2e*h CHAPTER 17.70) except that now qo = 1.9.40g3 + 18q2 .(17.73) We now return to the eigenvalue X(q) and expand this around an odd integer qo..72) This leads again to the result (17..70) + ^ 2 (9?o .65) leads to the same result except for a change in sign. . .71) = 2 13 /i 2 fc{l6h)i/2e4h 1 3(g2 + 1) 2 [4(51)]! L (9g4 . Considering now the first of conditions (17. i.120g£ + 467g ... The remaining condition (17.408g0 + 1089) 2 /i +• (17. Thus we have 2 (16/i) qo/2 T2 4.40gg + 18gg .e.40gg + 18gg . (17. we use \(q)~\(q0) + (qq0)(^J .
3.61). In the literature that we follow here* in all cases several more higher order terms have been given. Arscott [11].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. Effectively the degenerate eigenvalue expansion results from the anharmonic terms contained in the power expansion of the cosine potential. J. . M.88q0 . plus sign) give the socalled level splitting AA(go) as a consequence of tunneling indicated schematically in Fig. *In Sec. quite apart from applications. Miiller [73]. solution of the Schrodinger equation for the basic and nontrivial periodic potential is important for a thorough understanding of its quantum mechanics. Dingle and H. results from the boundary conditions. In particular we saw that the perturbation expansion yielded only the degenerate eigenvalue expansion.. The complete.87) = Azpteo).t From Eq. p. The dominant contributions were also found by S.3 Cosine Potential: Strong Coupling Solutions Differentiating \(q) of Eq.17.A_(g0) ^ . See also F. + ^ 2 (9<?o + 89o " 78g02 . We see here that *R. i. .3 we call this difference 26E„n.2.e. 120. just as any solvable problem is important for the insights it offers in a concrete and transparent form.74a) we obtain for the level splitting in dominant order* x r ^ \ / N 2(16/i)>+1 _4h A+(go) . (17. Goldstein [116]. and that for the odd Mathieu functions se qo+ i and se9o (lower.. example 3. (17. and not the separation of these as a result of tunneling.e.1. i. The tunneling effect. 20.3. (17.74b) The above results for the cosine potential are in many respects important. (17. 9o = 2n + 1. 17. W. minus sign). n ' r 1 / rrre 4/l ^odd(go) Seven (<Zo) Wfaqom (16/i)T n\V2n e~4h. made evident by the splitting of the asymptotically degenerate (harmonic) oscillator levels. n = 0.^ ( 3 g 0 2 + 8go + 3) A ( Q 0 ) T (87r)V2[i ( g o i)]l L2 w . though approximate. In the literature one finds frequently the statement that the splitting is a nonperturbative effect. this becomes 369 = (16/i)^+1e4h l . B..
. y^{z) = Aq(z)e~2hsin *. results from the imposition of boundary conditions. the exponential factor we encountered in the semiclassical method where it has the structure of the factor exp(classical action/ft). 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. From Eq. . yA{z) = Aq(z)e2hsin '. _3/+(z) + ____2/_(z).74b).§ The boundary conditions we imposed here are those of a selfadjoint problem which has real eigenvalues. 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. Periodic Potentials this nonperturbative effect. ^ ± n ) = _ ( _ 1 ) T i(<JTl) A 9 ( 2 ) (the extra minus sign in the second equation coming from Aq(z ± 27r) = — Aq(z)). the Schrodinger equation with the cosine potential is a special case of several other more general cases.2: y+(±ir) J4(±TT) y+{z±n) y(z±7r) = = ± —y+(z)± 2/+(0) . (17. The splitting can also be derived by the pathintegral method as will be shown in Chapter 26. The splitting can also be obtained with some methods of large order of perturbation theory. 17. We begin with the solution containing the function A(z). y+(0) 2/_(0) Solution: The derivation requires a cumbersome tracking of minus signs. The exponential factor represents. (17.370 CHAPTER 17. evident through the exponential factor exp(—4/i) in Eq.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. which one can then try to solve in order to obtain the behaviour of the coefficients of the late terms of the expansion. T y'_(0) y'A±*) y{z). and one can compare the methods. Naturally one expects the solutions there to be related to those of the periodic case considered here. 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. Example 17. As mentioned at the beginning.32) we obtain Aq(z ± TT) = (l)±i<9^'Aq(z). 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. M±TT) .74b) is also important for various other reasons.6: Translation of Solutions For the largeh 2 even and odd solutions (17.2. Thus we do not reproduce every step. The explicit derivation of the level splitting (17. in effect. Then we obtain first y±(z) = yA(z)±y^(z).
%/2(g .3 The Ellipsoidal Potential Replacing VA. the operator having solution Thus ^ ( 0 ) = (q/2)Aq(0) = A~'q(0) and A'q(z = ±TT) = ~qAq(z qAq(z = ±7r).( z ) From this we deduce for z = 0. A' (z = ±ir) = (.1 )  .4ft) sin j ^n(q . y'_ (0) = \/2(4h . since A q (0) = \/2 = Aq(0): y+(±n) = ±isin^(9=Fl)J3/+(0). ( .2hAq(n)] . (l^ifr^i^O + Cl^ifo^j/aCO cos  . y+(0)~2V2. which is easiest by comparison with the foregoing treatment of the Mathieu equation.4ft) ~ 8(q . Similarly we find y(z±ir) = = and by putting z = 0: V.4ft.4.q).yV+\z^ = [j/+(0)] 2 (<? .T ( ? =F 1) p + (0). 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+.1 Elliptic and Ellipsoidal Potentials Introduction Our intention here is partly to deal with other periodic potentials.4 17.7 r ( q = F l ) } j / .[A'qM + 2hAq(n)] = iV2(q . but also to enable a brief familiarization with the Lame equation.• Then y'+ (0) = 0.. etc. where Aq(0) = Aq(0) = V2. Aq(z) The derivatives can be handled with the help of Eq. Finally for the derivative of the odd solution we obtain y'_{ir) = K ( T T ) .VA DV 371 1 V±. we obtain the conditions stated at the beginning. The Lame equation has not been a widely known equation of mathematical physics so . 17.17. we obtain — with Aq{z + 7r) = (1)^ ^ 2 A .y]z=„ = [y+yL .) cos < 7r(q 1) [ = . (17. Replacing sines and cosines by the functional expressions.z ) .g A ? ( z = ±TT).4ft) cos i TT(<J + 1 ) 1 . and y+(z = ±TT) = V2(q .7 r ( g ^ l ) U + ( z ) q : i s i n { .31). = ±ir). A ?q( Z = ±TT) = .Ah) ~ 32ft.c o s  ~(<1 T 1) f y . — j / + ( 2 ± 7 r ) = ± i s i n  ^ ( g = F l ) U + ( z ) .( ± 7 r ) = c ° s  .( 2 ) .
5. as alluded to at the beginning. 19. This line of investigation. Ince [138]. 194) remarks. inverted double well and cosine potentials (J. Arscott [11].* As Arscott [11] (p. so that Floquet's theorem cannot be immediately applied . L.372 CHAPTER 17. Erdelyi [85]. H. Arscott [11] (p. Tchrakian [165]). H.. however. § F . 194) remarks to parallels with the Mathieu equation:".. J. 17. Periodic Potentials far.98. If one separates the wave equation § V 2 * + J2* = 0 in ellipsoidal coordinates (which we do not need to consider here). M. and was continued by Erdelyi. (17. for general n the solution of Lame's equation is not singlevalued owing to the singularities in the finite part of the uplane.. k) for k = 0.Q.5 < z < 5. the nonoccurrence of the Floquet exponent.fi2A..0. The most conspicuous difference compared with the Mathieu equation is. but also cubic potentials.0.0 and . In particular the Lame equation was recognized to arise as the equation of small fluctuations about instanton solutions for practically all basic potentials.. But more recently it has been observed to arise in various contexts. f E .K2sn2u .. one arrives at three equations of which one is the ellipsoidal wave equation pt + [A . MullerKirsten and D. the new stage of the development of the investigation of the Lame equation was really initiated by Ince^ around 1940. ." Fig.1.* and the work of both prepared the ground for presentday investigations. p..5 The potential sn2(z. *A.75.4sn4u]y = 0. W.0.75) ' T h i s means double well. Liang. 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 .
The range of the independent variable u is 0 < u < 2K. Jianzu Zhang and Yunbo Zhang [206]. The function sn2u is plotted in Fig. and of the derivation of the level splitting'. The second step is to insert (17.78) We follow in this brief recapitulation the description in H. i. i. (17.a > b > c being related to the lengths of the three axes of the ellipsoid in a Cartesian coordinate system.2k.\k\ < 1. The first step is to write the eigenvalue A as A(q.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. (17. . cmx and dnu. In order to distinguish the above equation from that with the Lame potential. 2 . W. is the elliptic modulus of the Jacobian elliptic functions snu. Although the results have been calculated for the ellipsoidal wave equation. (17.77) where q —> go = 2iV + 1. (17. MiillerKirsten.17. Eq. we refer to the potential consisting of the two terms with sn 2 n and sn 4 n as the ellipsoidal potential. 17. .5 for several values of fc. we consider mainly the Lame equation.76) and to write the solution y = A(«)exp  f Ksnudul = A(u)[f{u)]K. J.K) = qK+^±. where n is real and > —1/2 (and n is an integer in the case of solutions called Lame polynomials) where k. If we put fl = 0.75) reduces to Lame's equation and one writes K. K being the complete elliptic integral of the first kind.K2sx?u}y = 0. N = 0.2 Solutions and eigenvalues In the following we sketch the main points of the method of deriving asymptotic expansions of the eigenvalues and eigenfunctions. In comparisons with the Schrodinger equation. in the case n2 — oo. For barriers of finite height the parameter q is only approximately an odd integer qo in view of tunneling effects. the equation ^  + [A .e.e.77) into Eq. for very > high barriers (harmonic oscillator approximation around a minimum of the potential).3 The Ellipsoidal Potential 373 Here K2 and A are separation constants and U2 = l2(a2 — c2). .2 = n(n + l)k2. (17. in front of the second derivative has to be kept in mind (mo being the mass). 17.1.4. . the usual factor —h2/2mo.
one has to derive new sets of solutions there and match these to the former (i.80) The domain of validity of these solutions is that away from an extremum of the potential. Periodic Potentials . k' = Vl~ k\ \dmi±cnu/ 17. by transforming the equations for A. the other in terms of those of an imaginary variable.82 . more precisly for dnu = cnu F 1 » . i.K — > A(u)=A(u + 2K). C which are valid for dnu±cnu <1.374 where CHAPTER 17. one obtains again the same expansion for A. (17. /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. A. dnu =F cnu . A(u.K)=A(u.e. (17.e. Thus in the third step two more pairs of solutions B. A are derived. determine their proportionality factors) in domains of overlap (their extreme regions of validity).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.77). dnu± cnu K Thus one can construct solutions Ec(u). (17.81) Since these expansions are not valid at the extrema of the potential (where the boundary conditions are to be imposed). \/8K (dmz = cn« \ ' F . i. C replacing yl.K). . B and C. ll^ ^A(u)[f(u)]^k±A(u)[f(u)}^k.q.Es(ii). which are respectively even in u (or snu) or odd. one pair in terms of Hermite functions of a real variable. A into equations in terms of the variables z(u) = ^—{k' . .e.. ^ (17. / z J . A second solution is written y = A(u) exp < / nsnudu >. _ Solving the resulting equations iteratively as before.q. but solutions B.
2)4(33g4 + 410g2 + 405) 24/c 2 (l + k2)2(7q4 + 90q2 + 95) + 16fc4(994 + 130g2 + 173) +512f22fc4(l + k2)(q2 + 11)} . £2 = 0).^ ^ { ( 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£.86) \duj0 o h r 2K > \duJ =[sr 0 =0\duj (1787) H.17. Magnus.3 The Ellipsoidal Potential 375 In their regions of overlap one can determine the proportionality factors a. *A. . G.3 T h e level splitting In the fourth and final step one applies the appropriate boundary conditions on these solutions.e. (17.75). Erdelyi.38g2 .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. (17." 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) .85) The first three terms of this expansion were first given by Ince [138]. the ellipsoidal equation.384n2k4(q2 + 1)} q r {(l + A. Es(2f0 = Es(0) = 0. i. p. one sets at it = 0 and u = IK altogether** Ec(2A") = Ec(0) = 0. W. Miiller [205].6(5g4 .4. W. (17. 17. Tricomi [87]. 64. F. C = a~A. This expansion is found to be in the more general case of Eq. Oberhettinger and F. as well as \duJ2K 11 (17. who obtained this expansion for the eigenvalues of Lame's equation (i. a of _ _ B = aA. J.e.63)} + ••• .
136g0 + 9) n 3. Finally expanding A(g) ~ A(g0) + ( g . Periodic Potentials These conditions define respectively functions EcX°.j 26K2 (1 85) = A(g0) + (q . Muller [205]. J.:J {3(1 + k2)2 (9q* .88g0 . go being an odd integer.90) + 3.4k2(q2 + 2g0 + 5)} + • • • ].l n 2^+ i J + 0(«i) (17.Ec£ 0_1 and Es2° of periods 4K.2 /c +256fc2g0(g2 + 5)} (17. and the lower to Es*>+1 or Es*5.1 / 2 (2TT) 1 /22« l + (l_fe)^«( i .89) one obtains the eigenvalues from which the level splitting can be deduced.Es^ 0+1 .go)« qo(l + k2 22K {3(1 + k2)2(q2 + 1) .40g03 + 18g02 .376 CHAPTER 17.87) (17. 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). .g0 is obtained by expansion around zeros.91) tt H. W.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)*. 2K.78g^ . (17. One obtains with go an odd integer QQo =F2 2fl + ir\lkj 3(g02 k\K/k 8K qo/2 \lk ) 2 1 [i(g0l)]!L + l)(l + fc2) 25 K ^ 2 J J . 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.g 0 ) ( ^ . + fc )(3g + 8g0 + 3) k2)2(9q% + 8g^ .i Here the upper sign refers to Ec*> l or Ec£°. from which the difference q .
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./2i and hence A^ A/ A AU I2 .* Apart from the choice of notation.e. V. + E. however. K2 = h2 (say). h .* M G . . (17.2h2 cos 2x}y = 0.4 h ( i e M 9 0 7 2 \. that calculations with the Schrodinger equation are not only easier. A = 0.86). Replacing u by x ± IT/2. Thus in the case of the dominant contribution of Eq. (17. J.4 Reduction to Mathieu functions Under certain limiting conditions the ellipsoidal wave equation reduces to the Mathieu equation.92) reduce the periodic ellipsoidal wave functions and their eigenvalues to corresponding Mathieu functions and their eigenvalues. Rao [79]. .90) one has in this limit: /l±±\ ~K/U = e . Miiller [206].2h2 . K = 2/i.4.93) (17. *H. Dunne and K.87) agree with those of Ince. but also more easily generalizable. (17. then snit — sin u and Lame's equation (f2 = 0) becomes > y" + {A . in which they dubbed the classical configurations Lame instantons.17. i.Ah2 sin2 u}y = 0.2h2 = A.J1 O. Ince [138]. this equation becomes y" + {A . (17. 17. W. the conditions (17.3 The Ellipsoidal Potential 377 This result agrees with a result of Dunne and Rao** who calculated this expression using instanton methods.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. Hence the conditions fi = 0. A(g)>A(g0)T4/i\/e ^ [Hqo ~ !)]•' in agreement with Eq. One can see.74a). it = x ± ^ . A .f ln[(l + fc)/(lfc)] _ e 4h) ( 8" \ ^^ __ (16h)«. L. Thus if k — 0 and n —> oo in such a way that > K2 = n(n + l)k2 ~ finite.
S. N. W. Periodic Potentials 17. Thus. Liang. 1 1 See Z.Y. H.g. 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. thus revealing an unexpected significance of this not so wellknown equation of mathematical physics.5 Concluding Remarks The potentials we considered above will reappear in later chapters. for instance. A host of related elliptic equations has recently been discovered and studied. J. J. all basic potentials. H. Rana [287]. M.378 CHAPTER 17. MiillerKirsten and D.§ This is therefore a very important equation which in the limit of infinite period becomes a PoschlTeller equation. D. J . H. Cho.e. § .Q . de Veigy and S. See Chapter 25.W. Ganguly [103]. A. double well. A. In particular we shall encounter the Lame equation as the equation of small fluctuations around classical configurations associated with cosine. Quian [123]. Shiraishi [49]. Tchrakian [165]. inverted double well and cubic potentials. See e. .** The associated Mathieu equation appears also in string theory in connection with fluctuations about a L>3brane (see Chapter 19). i. Kan and K. Gu and S. Sukhatme [148]. MiillerKirsten and J. Khare and U. W.^ The equations considered above also appear in diverse new problems of physics. This is an interesting problem. A generalized associated Lame potential has been considered by A. **Y. Ouvry [276] and Jianzu Zhang. 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.
J. Wiedemann [4] and a revised version of parts of this reference by J. 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. Bender and T. In the cases treated most frequently in the literature the anharmonic *We follow here largely P. Achuthan.1 Introductory Remarks The anharmonic (quartic) oscillator* has repeatedly been the subject of detailed investigations related to perturbation theory. Weinstein [281]. Priedberg and T. Bender and T. T. M. MiillerKirsten [163]. W. [19]. Lee [98] referred to it as a "long standing difficult problem of a quartic potential with symmetric minimd'1.* 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. demonstrates that derivations of such a quantity are much less familiar than calculations of discrete bound state eigenenergies in quantum mechanical problems. Liang and H. 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. 379 . Wu [18]. MiillerKirsten and A. Wu. W. T.". This seemingly simple problem revealed extremely rich internal structure . There is no end to this: An entirely new approach to anharmonic oscillators was recently developed by M.t which related analyticity considerations to perturbation theory and hence to the large order behaviour of the eigenvalue expansion attracted widespread interest.. ' T h u s A. both of which make the calculation more difficult.. f C .Q. J. In particular the investigations of Bender and Wu. The recent work of R. D. H. Turbiner [273] remarks: "It can not be an exaggeration to say that after the seminal papers by C.
V(z) = \\h*\z* + \\<?\z*.1 The three different types of anharmonic potentials.380 CHAPTER 18. (3) Complex eigenvalues with tunneling: In this case. Anharmonic Oscillator Potentials oscillator potential is defined by the sum of an harmonic oscillator potential and a quartic contribution. described as the case of the double well potential. 18. 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. and thus lead to very different physical situations. with the potential described as an inverted double well potential. *~z (1) (2) (3) Fig. These contributions may be given different signs.1. which are nonetheless linked as a consequence of their common origin which is for all one and the same basic differential equation. 18. 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. V(z) = \\h*\z2 I 2i 4 \cr\z .
and this behaviour is that of asymptotic expansions our treatment largely terminates this muchdiscussed topic. We begin with the latter. (18. The eigenvalues of this case are given by Eq. as is sometimes hoped. The result is an expansion in descending powers of h?. The imaginary part will be rederived from path integrals in Chapter 26. Calculations of complex eigenvalues (imaginary parts of eigenenergies) are rare in texts on quantum mechanics. however with decay as a consequence of tunneling. The boundary conditions are nonselfadjoint and hence the eigenvalues are complex. 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. The eigenvalues of this case are given by Eq. Since these exponentially small contributions are related to the behaviour of the late terms of the eigenvalue expansions (as we shall see in Chapter 20). i.86) below.175) below. . In this case our aim is to obtain the aforementioned complex eigenvalue. 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. 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. The level splitting will be rederived from path integrals in Chapter 26. If the barriers are sufficiently high we expect the states in the trough to approximate those of an harmonic oscillator. Thus we are mainly concerned with the double well potential and its inverted form. This type of potential allows tunneling through the barriers and hence a passage out to infinity so that a current can be defined. since the potential decreases without limit on either side of the centre. We therefore consider in this chapter and in Chapter 20 in detail some prominent examples and in such a way. We do not dwell on Case (1) since this is effectively included in the first part of Case (3). The shift of the eigenvalues is best calculated with straightforward perturbation theory.18. Case (3) is seen to be very different from the first two cases.e. and we present a fairly complete treatment of the case of large values of h2 along lines parallel to those in our treatment of the cosine potential in Chapter 17. it will not be possible to obtain the exponentially small contributions with convergent expansions. that the general applicability of the method becomes evident. Case (2) is also seen to allow only discrete eigenvalues (the potential rising to infinity on either side).1 Introductory Remarks 381 normalisable wave functions. 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. except for a change of sign of c 2 .
2 18.2 The inverted double well potential with (hatched) oscillator potential.3) "1/2 harmonic oscillator h8/25c2 . If suffixes 1/2. The potential in this case is given by V{z) = v(z). (18. 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 . we can pass from one case to the other by making the replacements: 4 E1/2 = 2Ei. 18. h 4 2 2h1> ~1/2 2c?. Anharmonic Oscillator Potentials 18.2. (18.z Fig. v(z) = ^h4z2 + ^c2z\ (18.382 CHAPTER 18. 18.1 refer to the two cases.1) for h4 and c 2 real and positive. We take here h = 1 and the mass rriQ of the particle = 1/2.1 The Inverted Double Well Potential Defining the problem We consider the case of the inverted double well potential depicted as Case (3) in Fig. .2) We adopt the following conventions which it is essential to state in order to assist comparison with other literature. a point which has to be kept in mind in comparisons. and the Schrodinger equation to be considered is C L^ + [E + v(z)]y = 0.1 and more specifically in Fig. 18.2.
4) we can rewrite Eq.8) h8 v"{z±) = h and V(z±) = 5 2 2 c ' 2 2 Thus for c > 0 and relatively small... y{w) oc Di.2) as Vq(u >)y(w) = with T)Jin\ 1 (A + c2w )y(w 0 (18. 18. The problem here is to obtain the solutions in various domains of the variable. and a variable w defined by setting E= b* 1 + A A and w = hz. to match these in domains of overlap.. (18. The result will be that derived originally by Bender and Wu.5) becomes . the harmonic part of the > potential dominates over the quartic contribution and Eq. then to specify the necessary boundary conditions and finally to exploit the latter for the derivation of the complex eigenvalue.2 The Inverted Double Well Potential 383 Introducing a parameter q and a quantity A = A(qr. (18.2.5) (18 6) "^'"dw* ' * 2' 2 In the domain of w finite..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 = 0.18.7a) The problem then reduces to that of the pure harmonic oscillator with y(w) a parabolic cylinder function Dn(w). _i\(w) and q = qo = 2n + l.d2 . \h?\ — oo and c finite. and h large the eigenvalues are essentially perturbatively shifted eigenvalues of the harmonic oscillator as is evident from Fig. (18. (18. /i).e. (18. although our method of matched asymptotic expansions here (which parallels that used in the case of the cosine potential) is different. 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. n = 0.1/7  Vg(w)y(w) = o(J^j.2.1. i. 4 .
•A~2 hrz" 2^\ c^z V2 —— + (18.4(2.2 A (18.9) z =0 these Thus these c2z4 1/2 + • 2 (18.) exp Z ifdz[ifdz[ h4z2 4 h"z2 + + cV cV 1/2' (18. as we shall see. Anharmonic Oscillator Potentials T h r e e pairs of solutions 18. We observe — before touching the square roots in Eq. these solutions are not valid around z = 0.11) y(z) = A(z)exp ± z dz h4z2 4 + E hAz2 + cV y(z) = 0 (18. (18.2.12) Then A(z) is found to satisfy the following equation r /iV c V V/2 A"(z)±2<^. 1 d r A!{z)±iA{z)±{ 0. This construction is simplified by the consideration of symmetry properties of our solutions which arise at this point.14b) .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 ) . (18. h) is obtained from the perturbation expansion of the eigenvalue.9) we obtain 1 h4z2 c2z4 A L2 (18. Inserting (18. In order to arrive at solutions we set in Eq.13) — that one equation (of the two alternatives) follows from the other by changing the sign of z throughout. Before we return to solutions we derive a new pair which is valid in the adjoining domains. be and A = A(q. This observation allows us to define the pair of solutions yA(z) = .10) E= qh2 ^ j. as encountered and explained earlier.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. y 4 2/i Here again q is a parameter still to 2 determined from boundary conditions.2 We are concerned with the equation d2y(z dz2 where ^ 1 .384 CHAPTER 18.+ ^ .10) into (18.14a) 1/2 VA( ) = A(z)exp (18.
17) Aq(z) 1(91) Z2 (Z )V4 exp 2 We define correspondingly w dz (z2)l/2 (18.16) For large h2 we can write the Eqs. One finds that these solutions are associated with the same asymptotic expansion for A and hence E (given by Eq. 1 2 4 2 Aq(z) + 0 (18.e.zh2 =()*2"(18.1 and as a verification again in connection with the solution y# — as the other solutions. 385 (18.34) below) — to be derived in detail in Example 18.20a) These expansions are valid as decreasing asymptotic expansions in the domain \z\ >0 h.. Since these higher order contributions are of little interest for our present considerations.13) and A(z) that of the lower of these equations. (18. Clearly Aq(z).Aq(z) approximate the solutions of Eqs. We take the square root by setting z2h^1/2 +.20b) . 1 (18.15) where A{z) is the solution of the upper of Eqs.4 . i. /H 4 Z h +~C Z 21.18) M*) zf(9+l) A_ g (s) = (^2)1/4 exp ?/" dz (z ) / 2 1 2 (18.2 The Inverted Double Well Potential with A(z) = A{z) and yA(z) = yA(z). (18.18. Thus we now have the pair of solutions yA(z) = exp yA(z) = exp dz I^4 + l„2„4 1/2.19) We see that one solution follows from the other by replacing z by — z. / c z Aq{z)+0 1/2 . (18. we do not pursue their calculation. (18.13) TzA'(z) T ^A(z) + qA{z) = O We define Aq{z) as the solution of the equation zA'(z)(ql)Aq(z)=0.13) and we can develop a perturbation theory along the lines of our method as employed in the case of periodic potentials. (18.
04 = 5 .:=*£ +i(5l). 03 = .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).. . A A _ 3 . (18. „ .• Using Eqs. h2.e. away from the central minimum. As always in the In this way Rq is expressed as a linear combination of functions Aq+4i(z). one observes that with £>.1: Calculation of eigenvalues along with solutions of type A Use the solutions of type A. W i t h proper care in selecting signs of square roots we can use the solutions (18. .2 ( * ) 2 and from this or separately A"(Z) : 1 (q . A = . we write > V±(z) = . * = — .^A(z) °° /1r2z2\i + J2 { fir ) \**A'(*) 1 + jA^Wl.f .. h2. zA'(z) where the expansion coefficients are given by ao = l. 1 . h2 —> —h2). a2 = 2 1 .2 .20b).[yA(q.20b) to construct solutions y±{z) which are respectively even and odd under the parity transformation z —> — z (or equivalently q —• —q. 128 8 16 A = i.17) and (18.20a) and (18.! ) ^ . „f 1 \ S o l u t i o n : We rewrite the upper of Eqs. (18. z)).21) Example 18.386 CHAPTER 18. Vq+ii = Vq + 2i. (18.e.v ' = o (« . i.e.18) we obtain A » = H^rL^z) 2 z . to obtain in leading order the eigenvalues E. (18..20a).. c*i = 1 1 .l)A(z) 1 = tfA"(z) A . i. Anharmonic Oscillator Potentials i. (18. ft = . 2 . procedure. z) ± yA(q.13) in the following form with power expansion of the square root quantities and division by h2: 1 + (q . X » / A q + 4 i = Mg+4i .
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.(q + 3)(q + l)Aqvri(q '^ ' . In this way we obtain the next order contribution A^1' to A^°>. Rq '. +1J±iw) (observe that Eq.Aq+4i in Rq 387 can be taken care of by adding to A(°> the contribution — _ ^ 4 ' except.'2H9i) ^faDfr™) and .18.w — > > ±iw) or functions R B w q( ) . i. We return to Eq.2 The Inverted Double Well Potential Thus a term /j.e.23) .. when i = 0.. (18.22) are parabolic cylinder functions D i . 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). Hence in the present case In its turn A^ leaves uncompensated terms amounting to A „ /2c 2 \ 2 9 „ M 2ft4 2 { 4/i2^ .. of course.. c V + It follows that l) + 0 (i The same result is obtained below in connection with the solution of type B. (18.22) is invariant under the combined substitutions q — —q. those in Aq(z). ^f( g + i)(±H2^ + 1 > \ — 1. and the coefficient of terms with i = 0.—77 [i(9l)]. give an equation from which A is determined. = •>— r. _1J±w) and D_\. The solutions yq(z) of the equation Vq(w)yq(w) = 0. r ( CJw) = ^ . [1(9 + 1)]! . (18.5). Hence to the order we 0: {^)^) [{ 1){ 2 ^ ^ ^3)^ 0 +1 ^ + 2^ + 3)]+0 {h) 2ft6' A = 24 1 2 ^ + 1>2^+°G9A= . (18. w = hz.
(18.388 CHAPTER 18. (18.±8) 5 4 (g.q + 4j] = c2SA{q. (18. j=i (18. F. (18. (18. . we also have Vq+^yq+tj = 0.26) The first approximation y(w) = y(°>(w) = yq(w) = Bq(w) therefore leaves uncompensated terms amounting to 1 4°) = where [q. (q±2)(q±S).3)y g _ 4 . Actually these factors can also be extracted from W K B solutions for large values of q. Erdelyi.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 + . 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 .q + 4j}yq+4j. Vq+4j = Vq + 4j. Magnus.^ For higher even powers of w we write i w2l V<i = Y.23) have been inserted to make this recurrence relation assume this particularly symmetric and appealing form.4j)yq+Aj.0) = = = ±(q±3)(q±7).0).±4) S4(9. and for j ± 0 : [q.Aj)(18. Comparison with our notation is easier if this reference is used. G. and so Vqyq+Aj(w) = Ajyq+Aj(w).28) (A + c w )yq(w) E 2 4 1  2 ^ [q.l)2/?4.3. 115 .29) ' A . pp. since Vqyq = 0.27) Now.24) The extra factors in Eq. W. Tricomi [86]. l(q2 + l).q}=A + c2S4(q.123. S2i(q.25) where in the case i = 2: SA{q. See Example 18. Oberhettinger and F. (18. As an alternative to Bq(w) in Eq.( < ? .
31) Proceeding in this way we obtain the solution y = yW („.^ g ( 4 g + 29) + O(^ 2 1 2 Qr2 2 4 2 / 1 \ J.27) can be removed by adding to T/°) the contribution (—^/4j)y<j+4j.10) we obtain E(q.32) Evaluating this expansion and inserting the result for A into Eq. Thus the next order contribution to yq is y H = ie 2^ —z. .18.34) We observe that odd powers of q arise in combination with odd powers of 1/h2. (18.—A(q + 1) .1.35 ) . Equation (18.34) is the expansion of the eigenenergies E of Case (1) with q — qo = In + 1. A = ^(q2 + l)c 2 + o(J^. so that the entire expansion is invariant under the interchanges q . (18. . and even powers of q in combination with even powers of 1/h2.e. p i(9»9 + 4 J'WK7». and c 2  replaced by — c 2 . In Case (3) the parameter q is only approximately an odd integer in view of tunneling.—v*+*j ( 18 . This type of invariance is a property of a very large class of eigenvalue problems. We can now write the solution y(w) in the form y{w)=yq{w) + Y. h2 —> h2. .) ty^^w) to be a solution to that order we must also have to that order [q.) + j. 2 . (18.) + j/ 2 ) (W) + • • • with the corresponding equation from which A can be obtained. i. h ) = qh . n — 0.[rp ) Y.• q. ( 18 .3 °) For the sum y{w) = y(°)(«. (18.2 The Inverted Double Well Potential 389 Hence a term (iyq+4j on the right hand side of Eq. i.e.q]=0.(!) («. (18.
q + At] (18.argz=0 (18.37) For further details concerning these coefficients. we may infer that given one solution y(z). q) = = 1. w — ±iw.i 1 + 4j) in complete analogy to other applications of the method. Anharmonic Oscillator Potentials ±4 ±4 ±8 ±4 P2{q. 2 4tPi{q. ^VO(<L<?) ^i(?.—.38) [yB(z)]argz=7T = [?/s(2.q±A) = MM!i±MM+[M±8][g±8. there is another solution y(—z). Achuthan. and for j ^ 0 all other P0(q.11) is invariant under a change of sign of z. / l \ 1i i( ) + J2 ( ^e ) J2 P ^q + 4j)Bq+lj{w) iu=/iz. Our third pair of solutions is obtained from the parabolic cylinder functions of complex argument. MiillerKirsten and A. q + 4j) = 0. W.36) with the boundary conditions p o(q. We observed earlier that these are obtained by making the replacements q —> —q. their recurrence relations and the solutions of the latter we refer to the literature. ? T 4 ] [ g T 4 . g ± 4 ] T4 ±4 and so on. J. o. H. (18. Wiedemann [4] .390 where for instance v J CHAPTER 18. We thus have the following pair of solutions VB(Z) VB(Z) = B w .)]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. Again we can write down a recurrence relation for the coefficients Pi{Q. q + tt)= J2 i=2 p il(?> Q + ±3 + 4*)[q + 4j + 4t. > "p." Since our starting equation (18. 5 + 4j) = 0 for \j\ > 2% or \j\ > 2i + 1.g±4] ±4 ±4 [g.e. i.
(18.39) are with the same coefficients Pi(q.g. . we obtain: exp 8c2 3 < 2cVl3/2 /i4 h2z2 „(zA exp 12c2 (18. Mauss [192]. We add parenthetically that all our solutions here are unnormalized as is clear from the fact that the function in the dominant term (e. see e. Such contributions arise. In this bordering domain the adjoining branches of the overall solution then differ by a proportionality constant.40) "Variables like those we use here for expansion about the minimum of a potential (e.3 M a t c h i n g of solutions We saw that the solutions of types B and C are valid around the central minimum at \z\ = 0.4)).g.** First we deal with the exponential factor 1/2 exp exp dz z z h* + c z J2c z 2 2 .z2 .5) are known in some mathematical literature as "stretching variables" and are there discussed in connection with matching principles.h^ih. one has to stretch each by appropriate expansion to the limit of its domain of validity.The solutions yc.2 The Inverted Double Well Potential These solutions are therefore defined by the following substitutions: VC(Z) = [yB(z)]q^q. Integrating and expanding as follows since h? is assumed to be large. 18.2. like w of Eqs.g. Vc(Z) = [V!B(Z)}q^~q. in fact. (18. 4 .2 _4 A } 2cVl1/2 h4 8c 2 1 in the solutions of type A which are not valid around z = 0. h2) in which odd powers of q are associated with odd powers of h?. In order to be able to extract the proportionality factor between two solutions. h2 — — h2 as long as corrections » resulting from boundary conditions are ignored. Thus in the transition region some become proportional.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. Aq(z)) does not appear in any of the higher order terms. so that the eigenvalue expansion remains unaffected by the interchanges q —> —q. J. with normalization since the normalization constants are also asymptotic expansions. the solutions of type A being valid away from the minimum.h^ih 391 (18.18.q + 4j) as in ys.
41) with the solution ys(z) of Eq. 4/T 4 Prom (18.0x ! [ £ ( g . (18. ^Jw) has a similarly complicated expansion for 1 5 — — > argw > .2 « . (18. 2 ^i![i(g4il)]!(2«.44) In the solution y~g(z).44) we see that in their common domain of validity yA(z) = ~yB{z) a (18. with z in ys(z) replaced by — z.ir_I + M + 1)]! (2w*y U i\[~W • i 5 with 7r > argw.42) H W2 (ql)e$W I „ . We do not require this at present. (18. 2 ) i ' ir! i iHim 1 3 j argw < 7T.. y^(z) we see that in the direction of z = 0 (of course.45) .2)i (27T) 1 /2 e f(9l) 1 eb Uq + m (18.43) (since z — — 2 implies > argz = ±7r).l ) ] ! ( . we would have to substitute correspondingly the expression (18. Anharmonic Oscillator Potentials Considering the pair of solutions ju(.(«. not around that point) VA(Z) VA{Z) = ehVl2c 2 e\z*h* z^) + o(± zfr+i)+o(±\\. Thus from the literature [86] we obtain Di_{q_1){w) = w*{ql)e\w* but D Wi) ^=.w yB(z) ~ B.42) we obtain for the solution ys(z). > n.) ( ^2 „22)\ z £(</!) e 4re z 4 4 = hz: [i( 9 l)]!24(«D l+O h? (18. eh* /12c* e\z*h? (18.4 i .43) W + 1)]! w^+V A . . The function Di. Comparing the solution VA{Z) of Eq.7 r .z). 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.392 CHAPTER 18.
11.5. the ratio of yA(z)iVB(z) i s n ° t a constant.. hence we have to . Looking at the potential we are considering here — as depicted in Fig.49) d yc{z) Again there is no such simple relation between yA(z) 18. — For instance qo = 1 (or n = 0) implies a ground state wave function with the shape of a Gauss curve above z = 0.47) yA(z) Comparing this behaviour of the solution yc(z) with that of solution of Eq.46) However.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.7. The first of the conditions (18. i. (18. large probability for the particle to be found thereabouts.42) into y~c(z) w e obtain Vc(z) = (h*.21) as even and odd about z = 0. (18.18. Recalling the solutions y±{z) which we defined with Eq.50) and y+(0) ^ 0.41).y'_(0) ^ 0. 18. ^^+i)Ah^ze* [_i( g + l)]!24(9+D Inserting the ex l +O h2 (18.2 — 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 h2.2. At z = 0 the solutions of type A are invalid. we see that in their common domain of validity yA(z) where a = =Vc(z).yc{z)pansion (18.50) will be seen to imply qo = 2n + 1 = 1.2 The Inverted Double Well Potential with a 2 (K ^ ( 9 . We proceed similarly with the solutions yc(z).1 ) e 12c 393 Kgl)]^"1) l + O h? (18. we see that at the origin we have to demand the conditions yV(0) = 0 and y_(0) = 0 (18..48) (h2y 4(«+i) [i((? + l)]!24(^i) l +O /i2 an (18..9.e. and the second qo = 3. (18. Thus the boundary conditions to be imposed there are the same as in the case of the harmonic oscillator for alternately even and odd wave functions.
 ( 9 + 3)]! s i n { .. = = V^ >—= = — i i ^ i [1(93)]! ^_ sin{_(q + i)} (18.58) diu [  ( g . [ .394 CHAPTER 18.l)(°) Thus with the help of the reflection formula cited above: D B.. Abramowitz and I.l ) ] ! [ . A.i)H = e.SxW) = \ i»fl(0) .( g + 3)}..(0) = .i ) .(«. y'B(0). Expressions for C 9 (0).2. [C.( g + 3)}.F ? .) [i(«l)]![i(?+l)]! 2TT V5F[i(gl)J! V (18. Z (18. (18. + 1)]!' D ' j . one finds that [ 1 IT (q + 1)1! = .i ( g + 3)]l (18. M. — C. Example 18.1)('») + ^ r + l)]! ^. we obtain i V7T24( g .oi(. 4W >S [ I ( ( ? _ 3 ) ] ! s i n { f ( g + 1)} and hence =^= 1.^{(9+1)}.s i n { .! ( .57) [i( 9 _l)] ! 2 i(9i) B.55) Solution: Prom the literature. .52) # 1 = ^ y' c (0) a and 2/B(0) = a (18.53) yc(0) a' Clearly we now have to evaluate the solutions involved and their derivatives at the origin.54) dio q(w) .(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)}.5 ( 9 + !!)(*"')• i ( 1) ..56) D 4(.(0) + ^ ( 0 ) l z—>0 2./o^Tpif (q+i) ."*(«!) ^ ' ~ [ . Anharmonic Oscillator Potentials use the proportionalities just derived in order to match these to the solutions valid around the origin. y c ( 0 ) .g. C.(0) = and htoi)(°) ^—'. 1 [7(93)]! n «(°) = .2: Evaluation of yB(0).u=o follow with the help of the "circuit relation" of parabolic cylinder functions given in the literature as (q r — ) Di(. with the reflection formula (—z)\{z — 1)! = IT/ simrz. We leave the detailed calculations to Example 18.' '. Stegun [1].fc(0) Thus we obtain the equations (18.(iu)]. Then imposing the above boundary conditions we obtain 0 = y'40) = limJ[y'A(z)+y'A(z)] = i»i.(0) (18. IT 4 dio In fact.51) and 0 = V(0) = Jim i»xM . e.
49). using the above reflection formula and the duplication formula ^/TT(2Z)\ = 22zz\(z . _ l)]!2l(91)(/J2)(9+D We rewrite the left hand side as 1 sin{f(g + 3)} r„. 4 r T (18. ~ ^ 1 5(9+1) 2i(''.60) (1861) [C>)]o = iV2i [1(9 + l)]![i(9 " 3)]! (18. we obtain 2 i(»+i)[_i(. isin{f(g + 3 . + i)]!2i(«+ ) 1 [(«3)]! V5F[i(9l)]! sin{(q + l)}.1/2)!. (18.2 The Inverted Double Well Potential From this relation we obtain 395 ^^^i^w^vy ^(9i)(°) (18 59)  Inserting from (18.6 ) } ^ — i t a n < — (q + 3) >. W .(0) = *>i(„+i)(0> 2 ._l) From this we derive C.46).53) in dominant order and insert the appropriate expressions for a and a from Eqs.i . (18.+1iH]«=o = . + i)] V* H(?+l)]'[£(9l)]l Similarly we obtain [°'i(.3)}.62) .56). Starting with the derivative expression we obtain (apart from contributions of order 1/h2) lsin{f(g + 3)} i Sin{(g .+i)(°) [ _ I ( .sin{^(g+ 1)} = J .)[i(g3)]! and ^i(.sin{^( 9 . 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) _^_ .i ( « + i ) [ _ i ( . (18.3)} (h?)faV[\(q [Uq U + l)]\2*(q+V _£ e 6c^. 7T 4 V 7T 4 (B) Evaluation of the boundary conditions We now evaluate Eqs.18.
and the left hand side of (18. The analytic continuation of one case to the other is accomplished by replacing ±c 2 by =Fc2 or. . This is Case (1) of Fig.._ e =*.i . the vanishing of the wave functions at infinity).5..1. With a Taylor expansion about go the left hand side of (18. equivalently. (18.63) vanishes for q = go = 1. (18.63) Proceeding similarly with the second of relations (18. ^ . (18. (gg 0 )^±^ y e5?. For c 2 < 0 and the solution y(z) square integrable in —oo < ?Rz < oo.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. . the energy E is real.65) but now for the odd function with these values of goWe have thus obtained the conditions resulting from the boundary conditions at z = 0.64) about go — 3. 7.. . 1 1 .2) with potential (18. we obtain cos((g + 3)) = .64) In each of Eqs.11. Our next task is to extend the solution all the way to the region beyond the shoulders of the inverted double well potential and to impose the necessary boundary conditions there.396 CHAPTER 18.64) the right hand side is an exponentially small quantity.5. + 3)} = . 7 . In fact the left hand side of (18. ..( g 0 + 3)  + • • • ~ (g . z^ ein/2z. (18..63) and (18.1). 18. Thus we have to determine these conditions first.53) becomes 8 m(I(.^ . we again obtain (18. (18. this is the case of the purely discrete spectrum (the differential operator being selfadjoint for the appropriate boundary conditions..L ^ i .9.9. 18.65) Expanding similarly the left hand side of Eq. i. .. z2 > z2.e..^ >^_>m e J*.9o) ^ cos  . Anharmonic Oscillator Potentials Then the derivative relation of Eqs.53)..go) ^ (1) It follows that we obtain for the even function with q — go = 1...2.64) for q = g0 = 3. Recall > the original Schrodinger equation (18. (18. by the rotations E > einE = E.63) becomes (? .
18. we see that the solution with the exponential factor is exponentially decreasing for \z\ — 00 provided that cos 39 > 0. i. It is then necessary to insure that when one rotates to the case of the purely discrete spectrum without tunneling.cos 36.68) . i.e.e. we set z w r = \z\eie.e. + i s i n 3 6 0 i y ) ~3~^ c2\l/2\z\3 oc exp <M — J — . or — < 6 < . the case of complex E). i. Now. i. 6 6 In the case of the inverted double well potential under consideration here (i.sin 3(9 This expression vanishes for \z\ —• +00 if the angle 0 lies in the range — ir < > 36 < 0. 2 c2>0. y{z)~exv\i[) 7T dV'V. Then c 2 y/vi ^p^My) yf = exp _ r HvcV2!*!3 cos36. Thus in our case here the behaviour of the solutions at infinity has to be chosen such that this condition is satisfied. —}. we therefore demand that for $lz — +00 and • —7r/3 < arg z < 0 the wave functions have decreasing phase. (18. (18.2 The Inverted Double Well Potential 397 One can therefore retain c2 as it is and perform these rotations. 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. replacing sin 30 by " 0 + ^) 1 IT = .67) Rotating z by vr/2. if TT/3 < 6 < 0.66) z3 = \z\3em. r / c 2x 1/2^3 = exp{ ± i — —\. in the domain — TT/2 < 36 < > 7r/2. 1/2. the resulting wave functions vanish at infinity and thus are square integrable.e. Thus exp exp (<£ff} [+<£ )%} sin * 0 for $lz —s.+00 in arg z € (!•" * 0 for 5ftz — —ex) in arg z € » N> (18.e.
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. Equating to zero the coefficient of the term with sign opposite to that in the exponential of Eq.68) for c 2 > 0. (18. Looking at Fig. (B) Evaluation of the boundary conditions The following considerations (usually for real z) require some algebraic steps which could obscure the basic procedure. For c 2 < 0 we have correspondingly y(z) ~ exp I ± ( — 1 — y c2 < 0. for z > Too. 18. i.Z\. Anharmonic Oscillator Potentials V(z) Fig.68) on y±(z) (by demanding that the coefficient of the solutions with other behaviour be zero). We have to remember that we have various branches of the solutions y(z) in different domains of z.Z\.21)) were defined in terms of solutions of type A which have a wide domain of validity.3 The inverted double well potential with turning points ZQ. (18. we match the WKB solutions .e. 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. Our even and odd solutions y±(z) (cf. 18. We therefore explain our procedure first.398 CHAPTER 18. This is the boundary condition also used by Bender and Wu [18]. Our procedure now is to continue the even and odd solutions (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. Eq. This is not the asymptotic behaviour of a wave function of the simple harmonic oscillator. We do this extension with the help of WKB solutions.21) to + infinity and to demand that they satisfy the condition (18. and there impose the boundary condition (18.
6. We begin with the exponential factors occurring in Eqs.=c z 4 dz 1 L2 9 + 2 4 1/2 x exp . 1 2i4 1 2 4 / ~ 2 4 ~ 2 C * (18.71) We now come to the algebra of evaluating the integrals in the above solutions. A J24 1 ? 2 • "^ •z h \—c z ~ — q h .69) The W K B solutions have been discussed in Chapter 14. 291. (18.B.18.. (22) or A. p.e. 18. The distant turning point at z\ as indicated in Fig. i.zi)/ \ 1/4 ywKB\ ) z — g/i* . Dingle [70]. Dingle [70]. 291. equations (21). In using these expressions it has to be remembered t h a t the moduli of the integrals have to be taken.V 2 iL44 + <?z / „2 4 X COS < f / il/2 G?Z g/l 2 1 2L4 z 4 M 1 2i.^ T rz\ 1 + yh* . f R . ^2? 1 2qc2 /2? . z2h4/A > c2z4/2.70).*l) Z 1 2L4 1 qh? + zzh* 1/4 _2i4 VWKB\ ) . 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\. 4 2 T ' i.2. t To the right of the turning point at z\ these solutions match on to (r.z\) ( \ y^NKQyz) qh2 z 4 dz h H—c z 2 x sin ^ . p. Vol.^ 4 1/2 + icV 7T Jz + (18.4) for E and ignoring nondominant terms) t o . ^ 2 4 •K + —(r. h2 (18.e.3 is given by (using Eq. .70) where z < z\. (18. Sec. 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.4 h + c z 2 1/4 < . z\ ~ . I [195].2 The Inverted Double Well Potential 399 to the left of z\ to the solutions of type A.4. *R. Messiah. B.
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. (18.^ i ^ ^ c 2 .40)) E± = exp _/^_2 f ± In A 2c2 . .4 In 2c 2 J + 1/2 ^1 2c2 1/2 21 and hence (with use of Eq.^ I 2 2 1/2 zV = F 8c2 Jz [i .72) Here the first part is the exponential factor contained in 2/A(Z)) J / A W respectively (cf. Eq. (18.73) .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.^ ] V 2 and so /^2f 8c2 3 1 2cVl3/2 h* f q fZ\ ck h^y/ E± = exp T 2 jJ Z ^ [ g .211/2 1 \ /j. In the remaining factor we have (looking up Tables of Integrals) 21 2 dz h 4 2 2c.±9/2 = ^ V M z^l2 21 2c2 exp ± i dz 1 4 1 1 V2 2 (18. Thus the above factor yields '2c2\l/2 h4 so that (cf.2^3/2 2_(h^\ll2 2czz 4 N 1/2 8?3\ 2 ~hf ' =F9/2 ^ 2 ^ .. Eq. CHAPTER 18. zi 2q 1 c 2 [1 . We obtain this factor by expanding the expression in powers of zjz\ (since in the integral z < z\).\qh2 + \z2h± .e.2 \ 1/2 .400 i. Anharmonic Oscillator Potentials E± = exp = exp .40)). (18. (18.
Zl) ( \ yWKBV^J \z h* 2 1/4 and ywKB(^) \z h* 2 1/4" (18.20b).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. we see t h a t in their common domain of validity VA (z) = /3ywKB 0)> where .75) 9/2 1/2 0= or '(2c )V2 2 and (3 = — 2 2 (fr2) (2c 2 ) 1 /2_ 9/2 (18.79) +S_(±)exp . CZ2 JZ1 cz 3^2 V2 h6 12c 2 ' Ac2z2 cz" 3y/2_ (18.*l) (3y^(z) (18.78) Inserting this into the solutions (18. In these expressions we have chosen the signs of square roots of h4 so that the conversion symmetry under replacements q — —q.\z2h4 + J Z\ nl/2 \M fZ . h2 — — h2 is maintained.21) we now have V±(z) 1 2 \yA{z) ± yA(z)} = \W^(z) WI ± _(J.1 ) 9 / 2 _(2c 2 ) 1 /2_ 1 (18.18. (18.2Z 4 C S ' + ( ± ) e x p I i\ cz / cz3 V3\/2 U^2 h6 12c 2 12c2 J (18.21 1/2 .76a) P P r 2/1 2 " 9 ( .*l) (18.71) and these into (18.(J*i) VA 0 ) = PVwKB (Z C.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) . > » Returning to the even and odd solutions defined by Eqs.76b) apart from factors [ l + 0 ( l / / i 2 ) ] .74) Comparing these solutions now with the solutions (18.20a) and (18.77) Now in the domain 2 — oo we have > \qh2 .
(18..76a) for (3 and /3.^ j . Eq. (18.68).^ § ( 4 g 2 + 29) + O ( ^ ) ..82b) This is our second condition along with Eq.. this equation can be rewritten as or as the replacement + / u6 \ 90/2 (/i 2 )W 2 =^(_) 2®"1 f — J .6 The complex eigenvalues (1883) We now return to the expansion of the eigenvalues. Anharmonic Oscillator Potentials where 5+(±) 5_(±) = = Q / ? ± ^ e x p ( ^ ^ T ^ ) e x p ( ./i2) = E{qQ.402 CHAPTER 18.84) Expanding about q = qo we obtain £(g.80) Imposing the boundary condition that the even and odd solutions have the asymptotic behaviour given by Eq. (18..^(q2 + 1) .h2) + {qqo)(^pj = E(q0. 18.5. Inserting this into the latter equation we obtain (the factor "i" arising from the minus sign on the left of Eq. (18.h2) + (qqQ)h + (18.e. (18. E(q. i / 3 ± .82b)) /j6 \ 9o/2 («*) = W? mam ''*• with qo — 1.85) + . we see that we have to demand that 5 + ( ± ) = 0. .81) Inserting expressions (18. (18./ ? = 0. i..e.34). (18. i.2.3.65). (18. h2) = \qh2 .
S W K B ( 2 ) = PVA(z). ^ 2 4 y qh 2H ""{f' z h H—c z 4 2 1/2 1 ^ 4 + IC2Z4 dz qh2 4 2 .h6/6c2 E = E(q0. o 2 + l)^(4. Example 18. [19]) for ft = 1 and in their notation 90 = 2 ^ + 1.g ( .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 . The large order behaviour of the eigenvalue expansion (of Case (1)) which Bender and Wu derived from their result (18. ft6 ^ = e. !*WKB(Z) ~ 2 .36) of their Ref.ZQ) . zo} means "to the right of the turning point ZQ" .~p{q. In the above we did not require the matching of WKB solutions at the lower turning point ZQ to the solutions yA(Z).3: Matching of WKB solutions to others at ZQ Determine the proportionality constants p(q. Solution: The turning point at ZQ close to the local minimum is given by qh2 2* 4 z2hA c z ~ 0.yA{z). ZQ (2g)V 2 / h \ 2qc2 he (2q)1/2 In the domain 9ZE > V to the left of ZQ the dominant terms of the WKB solutions are (cf. above) (1. * _ qh2 1/4 *• 2 j _ 4 i 1 2 4 z h H—c z dz qh2 2 1 2L4 •{/.18.3. Since this is not without interest (e. in comparison with the work of Bender and Wu [19]) we deal with this in Example 18. h2). where the superscript {r.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 (3.4 1 2 4 1/2 ft + i c V 2 1/4 + f}' z.86) will be dealt with in Chapter 20.2 . h2) of the relations KWKB (Z) =PVA(Z).g.l ) ] ! > + = ^ 2 .h2) The final result is therefore E () i ( 2 7 T ) V 2 [ l ( g o .e. i.
404 CHAPTER 18. f x Zn 4) h2 2. cf.O." / 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. N.zo)/ \ qh2 2 ZO + z2hA 4 2 . 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. J ^ K B match on to the W K B solutions with exponential behaviour to the right of 20.g In other words.ic224 2 4 qh2 + i^fe 4 . 2 z 2 ) / 2 e \ . Anharmonic Oscillator Potentials where zo > z.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.] { hz combined with the particular constants contained in y ^ ' ^ g . Schwind [247]../ ^ 2 X 1 / 2 exp(ifi. . and we have 1/2 dz h?_ 2 J /i2 qh2 + iz2h4 _ ^ 4 J dz zn qh2 + z2^ 1/2 . 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 . 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 2.1/2 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 . 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.
23)) appear quite naturally. we shall employ basically the same procedure as above or.88) + [EV(z))y{z)=0 (18.87) The minima of V(z) on either side of the central maximum at z — 0 are located at h2 h8 z± = ± with y(. i.e.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.89) .(r." / 4 ( ? « / 4 .5 "l _ o f . Then .3 18.z ± ) = .DfJ»v we see Comparing now the expressions (18.18.J ' i J .3. in Eq. as in the case of the Mathieu equation. (18.kyil(.13) of Bender and Wu [19]. But there are significant differences.41) for yA(Z). (j . ~p requested at the beginning are given by /1y/2[j(gi)]i2^c^ h 6 /12e3 /ly^ih+iwi"" w 18. 4" ) (h?z2)^q+1) for q j£ odd integers.1)! = ( 2 7 r ) 1 / 2 e .yA{z) with expressions 27WKB i J'WKB ' that the factors p. in fact. 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. J/J Hence ^ and h so that \]q] \2"'2 ~ ( 2 7 r ) 1 / 2 e . We consider the following equation ^f& with double well potential V(z) = v{z) = jz2h4 + ^c2z4 for c2 > 0. „ / M 1 / 2 i . In this form the WKB solutions reveal their similarity with the solutions VB i VB > Vc i Vc a n c ' demonstrate that the factors which we inserted (e. (18. (18. We can reexpress these relations with the help of Stirling's formula.. h4 > 0.^ 1 .g.3 The Double Well Potential 405 It may be observed that 3/Y/KB ( z ) corresponds to solution (3._ .zo).
93) w± = h±(z .i>5. the previous equation (18.z Fig.91) (18.V(z±) l{z z±fhA + 0[(z .95a) .91) becomes T>q±(w±)y(w±) = o(^)y. 18.z±) y = 0. vW(z±) = ±Qch2. . In order to obtain a rough approximation of the eigenvalues we expand the potential about the minima at z± and obtain y dz2 We set + E .90) V(i)(^)=0. (18.4 The double well potential.406 CHAPTER 18. (18. (18. (18.92) (thus we sometimes use h4 and sometimes Ii±) and EV{z±)=lq±hl With the further substitution + ^.z±). Anharmonic Oscillator Potentials and V{2\z±) V(4)(z±) = = h\ 12c2.94) (18.
(18. i.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) = 0.93) for E into Eq. given one solution.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. Inserting the expression (18. (18. (18. n = 0. (18. d2u{z dz2 + v+ 2~r u(z) = 0. where U{z) = and near a minimum at z± 4_ 4 jp[V(z)V(z±)].. (18.3 The Double Well Potential where ^ 407 d2 •« 2 2  d8.1.87). (18.18.i 4 f W » = o. 18. Thus again.97a) U(z) = (zz±)2 + 0[(zz±)3}. Eq..95a) and (18.95b) with the equation of parabolic cylinder functions u(z) = D„(z)..95b) By comparison of Eqs.96) dz2 + 5 « 4 +  . we conclude that in the dominant approximation q± is an odd integer. qo = 2n + l.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. we obtain d*y (18. we obtain another by replacing z by —z.97b) Our basic equation.3. is again seen to be invariant under a change of sign of z.96).e.100) ..2.
c W + I ) 2C yfi 2h 5 2 A V2c* g(17g2 + 19) + 0 ( / t . ~h2.106) . We leave the calculation to Example 18. these equations can be rewritten as + (qz+z)A{z) = . z) may be obtained from the solution yA{q. 8hw (18. The result is given by A and E(q.^ A"{z) + 2c V2 *A(z) (18. h2. we observe that the solution y^iq.z) by either changing the sign of z throughout or — alternatively — the signs of both q and h2 (and/or c). (18.4.1 6 ) . (18.c W + 1) .101a).e. (18. yA(q. z) = yA{q.105) "8 + 4 ^ . z) = yA{~q. (18. i. z).101a) {z2+z2)A\z){qz+ + z)A{z) = ^ A"{z) + £A{z) (18. Vq = {z2+z2)— + (qz+z).103) Looking at Eqs.102) We observe that a change of sign of z in this equation is equivalent to a change of sign of q.102) yields the following expression M*) I z2 _ z2 11/2 *+ z + z+ q/2 z+ 1(91) \z + z+ 1(9+1)' (18.yA(z) are associated with one and the same expansion for A and hence E.e. (18. Anharmonic Oscillator Potentials and selecting z+ (z2+z2)A'(z) h z+ = —.^q(l7q2 6 4/i + 19) + (18. n± = V2h2.408 Since CHAPTER 18. but the solution is a different one.h2.h2) .101b) To a first approximation for large h2 we can neglect the right hand sides. h2 h4 2c' with g+ = q. h2.104) Both solutions yA{z). Integration of Eq. h2. The dominant approximation to A is then the function Aq given by the solution of the first order differential equation VqAq(z)=0. i.101b). Aq(z) = Aq(—z) = Aq(z).
± 2 . J = 0.111) {Aq+2  Aq2)2 (4zz+)2 {z2  2 . <4q+2 — z — z+ Aq+2iAq+2j (18.110) Aq+2+Aq2 9+2 A.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 .103): .101a) with A replaced by A . z • . (18.102). . The first such step would be to reexpress the right hand side of Eq. . such as periodic potentials. (18. i. (18.105) and (18. ± 1 .z\) + 2z(z ..(l + + 2  z y* .2 z\) • Aq$4 £Aq + Aq—4.108) We wish to rewrite this expression as a sum y coefficientiv4. however. as a linear combination of terms Aq+2i. 92 ZZJf (18.103) for Aq{z) is very similar to that of the corresponding solution in considerations of other potentials.*l) (* ~ "+> ) A (z)exp q /2 \ 3 4c J (18.4: Calculation of eigenvalues along with solutions of type A Show in conjunction with the derivation of solution y^ that the leading terms of A and hence E are given by Eqs. (18. (18.e. Similarly we obtain _z+_ Z _ Aq + 2 — Aq Aq+2 + Aq 12 ^g+2 A ± ± _2A 2 ± 6 + (18.qz+) [2z2 .{z2 .+ 2 ^ ± l + 2^±i + .qz+)2 . Thus it is natural to explore analogous steps. .112) From these we obtain.114a) .106).z2 ) N t. A Differentiation yields » = ^^w (18. . .18.109) We observe some properties of the function Aq{z) given by Eq.107) KV) ~ Aq(z) {z*z\y Aq{z) (z 2 z\f (z . = Aq+2i+2jAq. + Aq+2 + Aq Aq+2 .113) 2 .Aq 1 + 2^± 2 Aq+2 (^M^f (18. (18. Solution: The structure of the solution (18. . First.4zz+q+ z\{q2 + 1)]. we reexpress A'q in terms of functions of z multiplied by Aq. We know the first derivative of Aq from Eq.3 The Double Well Potential 409 Example 18. for instance by componendo z z et dividendo.
(18.z\Y Z2 + {~)\Aq+42Aq 1 224 Aq.4 ( g .3 ) A .4 4+* + .q2)Aq2 + (q.116) [Ag4 . (18. (q. <?) = 6(<j2 + 1) + z: 2 2A +72"' (18.410 and from this = Hence with Eq.2Aq + Aq4] 1.116) into Eq. q + 4)Aq+4 + (q. (18.4Aq+2 + Aq+A Inserting (18. (18.q)Aq + (q.e.4)A„_ 4 + (q. (18.108). where the lowest coefficients have been determined above as (9. i. . we obtain 5 K = ( ^ f ) [ ( 9 . Vq+2iAq+2i=0 and Vq+2i = Vq + 2iz+. (17. cf.112).112) CHAPTER 18.<? +{q. _ 2 + 6 ( 9 2 + l)A ( ] + (q + l)(q + 3)Aq+4.(0) V2 ' 2c K + h4+Ao .aAg+S 12+ 16 (18. leaves uncompensated on the right hand side of Eq.2 ^9+2 „^9+4 + \4z \4z+J z+ Finally we have also [A q _4 — 2Aq2 + 2Aq+2 — Aq+4 (18. ' T h e reader may observe the similarity with the corresponding coefficients in the simpler case of the cosine potential. Anharmonic Oscillator Potentials A 1 . .2\2 (z* .l ) 2 A . (18.115) and (18.34). _ 4 . q) = 0.118) 2s/2c ~^j[(?.q + 2) = 4{q + l ) 2 .z%) 1 n z+z 2 \4z+J [Aq+4 [Aq+4 .q + 2)Aq+2 (18.114b) . 1_4^±^+8^±±12^±^ +.115) (z*z%) 2 i2 (z 2 .101a) the contribution R . Since VqAq = 0. A(°~) = Aq.117) 4(q + l)2Aq+2 Here the first approximation of A.9=F4) = (g=Fl)(?=F3).l ) ( g . •••].+6 + • • • ] . QA?+4 _ 10 A + 6 .33) and (17.q + 6)A<.4]j.4A ? _2 + 6Aq . In particular the dominant approximation of A is obtained by setting (q. (q. Eqs.119) It is now clear how the calculation of higher order contributions proceeds in our standard way.2A g + A g _ 4 ] .
thus yielding the next approximation of A as given in Eq.e.{^ h2.g) 4z+ {q q)+ 2r (9.101a) the contribution Rq .g ( 1 7 g 2 + 19).g) 8z+ (g.h2.e. The first approximation A(°> = Aq leaves uncompensated on the right hand side of Eq.0) = y^(q. z)\ = 2^A^ h2 '>z)± VA(Q.0) = ^ . y'+(q.g2)(g2. where Rq is the sum of terms left uncompensated by AW. for i = 0. of course. g) 8z 4 . 9 + 2) Aq+2Az+ (9.28. g + 4)(g + 4.2 ) (0) (18. 9 + 4) ^+48z+ (18.h2.(0).h2.h2.122) The solutions yA(z). i.121) This coefficient of Aq set equal to zero yields to that order the following equation 0 £) ' {^) (g. y'_(q. g .Aqjr2i in this may therefore be eliminated by adding to A^0' the contribution A^1' given by (9. h2.g + 2)(g + 2. A' (0) = q/z\) y+(q. \ZZ±\ > 0 ( y We can define solutions which are even or odd about z = 0 as y±(z) = = £ fe^te' h2> z) ± y A.0)=0.0)=0.yA(z) derived above are valid around z — 0. which reduces to 2 \/2c 2 0 = 2(3g 2 + i) + — A + ^ g . (i) _ / V2c\ [ ( g .3 The Double Well Potential we have T>o 411 2iz+ J Aq+2i except. in the domains away from the minima. i.94)(g4. *)] (18.92) ig4 ' H ~ 23/X4 iz+ An2 H + (9.105). (18. (18.h2.120) 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.123) ^[yA(q. (18. h2.z)±yA(q.94) 8z+ (9. (g.124) . Terms jj.4 ) ( 0 ) ( g .z) Considering only the leading approximations considered explicitly above we have (since Aq{0) = l/z+ = A.g) Az. h2 2y^ Age? h4 (18. g .
(18. 18.yB(z) is obtained around a minimum of the potential. Moreover. yB(z). (18. Anharmonic Oscillator Potentials Our second pair of solutions.^ % [i(?l)]!24(«.95b) that the solution there is of parabolic cylinder type.Evaluating the exponential factor contained in VA{Z) of Eq.1 ) D (18. we have w±(—z±) = —2h±z±.3 M a t c h i n g of solutions Next we consider the proportionality of solutions yA{z) and VB(Z).126a) hjW + l)]! VM = Vc(z) = Cq[w±(z)] (18.We draw attention to two additional points. Inserting (18.126b) These are solutions again around a minimum and with yB(z). This means. (18. (18. the equation is — with differential operator T>q as defined by Eq. + 0(hl2).99) for z — z±. Bq[w±{ ^ . and another VB(Z) ± 25^ch3wl + c2wi .95a) and (18.125b) It is clear that correspondingly we have solutions yc(z). Since w±(—z) = —h±(z + z±). but w±(z±) = 0.95b) — 1 T>q(w±)y(w±) = j£ Thus we write the first solution yB(z) = Bq[w±(z)] + 0(h±2). Thus yc(z) Cq[w±(z)} = = Cq[w±(z)]+0(h±2).3.125a) = VB{~Z) = Bq[w±(z)] + 0(h±2) = Bq[w±(z)] + 0(h±2) (18. in this case we use the Schrodinger equation with the potential V(z) expanded about z± as in Eq.yc(z) with complex variables and Cq(w) given by Eq.91).412 CHAPTER 18. in view of the factor "i" in the argument of Cq the solutions VAJVC have the same exponential behaviour near a minimum.93) and setting w± = h±(z — z±). (18. (18.2A y(w±).23) with appropriate change of parameters to those of the present case.yc(z) providing a pair of decreasing asymptotic solutions there (or correspondingly y~B(z)iyc(z)). we > . 2(9+) . We see already from Eqs.
Allowing z to approach z+ in the solution IJA(Z). .129) . Here z± = ±h?/2c are the positions of the minima of the potential. (18. ein+z+e in+(z 22+(<?+i) .3 The Double Well Potential (cf. .127) Recalling that around arg w± ~ 0.97b) ~ exp . (18. (18.z±)dz .frW.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.^/w + ± 1 . We observe that for z — z± the approximation yields exp[±/i? t z 2 t /4].i*l(«^ 4'"+^+ 4"'+v ( « . 24(9D[i(gi)]! hj (18. (18.• • " .• )2 (^)itoi) ( 2 z + ) 5 ( 9 + i )e ' yA(z) z+> lft2 2 _ I .^(±>(^*4 ^\h\{zz±f exp 7>2 r 2 ^n±z± For later reference (after Eq.ZI and find that these are given by z+ + 0(l/h) for finite q (cf.147)).125a) we see that (considering only dominant contributions) in their common domain of validity e ^+z+ l + O 1 yA{z) = yB(z). (18.172)) we add here comments on this approximation.128) Similarly we obtain in approaching z+: VA(Z)  ( 2 ^) (9 " 1} e.164)) only in a nonleading contribution and hence implies the equivalence of the exponentials in the relation (18. . the dominant term in the power expansion of the parabolic cylinder function is given by Dv(w±) ~ wv±ew±>4. we have \z ~ Z+\^Q ^ l f e 2 2 2 _kh2( ~ i .« + ) 2 (9+1) e (18. (18..146). Later we calculate the coordinates of turning points ZQ. a= ^ . . 2 An+z+e 4w+ (18. Eqs.18. . and comparing with Eq. Thus for h very large these turning points are very close to the minimum at z+.97b)) exp 413 '\hlJQ'uV\z)dz = exp = exp (18.
but it is clear that none of the above solutions can be used at the top of the central barrier. since.z+)]fr+V ' (18.4 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. de Deus [66] and W. basically. The pair of solutions of type B is defined around argz = 0. Furry [102] .131) We have thus found three pairs of solutions: The two solutions of type A are valid in regions away from the minima.130) Therefore in their common domain of validity a a = dte+vM' {2z+)^. we naturally expect the wave function there to be similar to that of the harmonic oscillator. R. S. The next aspect to be considered is that of boundary conditions. D.7r and the solutions of type C around argz = ±7r/2. and this means at both minima. 18. g . H.(18. Rau [22]. The involvement of these WKB solutions leads to problems. Bhattacharya [23]. K. Thus 1 E . We have to impose boundary conditions at the minima and at the origin. 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. The solutions in terms of parabolic cylinder functions have a wide range of validity. P. Bhattacharya and A. 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.414 and CHAPTER 18. even above the turning points.3. S.+)2 [1(1 + I)]' [i(q+ l)}\[ih+(z . K. Various investigations^ therefore struggle to overcome this to a good approximation. Anharmonic Oscillator Potentials D_h{q+l)(iw+(z))2i(o+V Vc( ) z 2i(g+1)ei^(z_. they assume large quantum numbers.\h\)^+l\\{q l + l)]\ 1+ 0 J_ . Thus it is unavoidable to appeal to other methods such as the WKB method to apply the necessary boundary conditions at that point. Since it is more probable to find a particle in the region of a minimum than elsewhere. J.
18. y{z±) £ 0. (18. 18.5 Behaviour of fundamental wave functions. y(z±)=0. The odd wave function then exhibits a correspondingly opposite behaviour. as indicated in Fig. (18. as indicated in Fig.Zii yB{z±) a ± a yciz±) (18. y'_(z±) = 0. However.136) . and y'±(z±) V'B(Z±) ± =y'c(z±) (18.5. At dtz — ±oo > we require the wave functions to vanish so that they are square integrable.132).134) y'+{z±) ± 0. as indicated there. ^(*±)#0 (18. 18.3 The Double Well Potential 415 the most basic solution would be even with maxima at z±. We have y±(z±) = ^VA{z)±yA{z)]z^. and y+(z±) = 0. We have therefore the following two sets of boundary conditions at the local minima of the doublewell potential: y'+(z±) = 0. an even wave function can also pass through zero at these points.135) Hence the conditions (18.133) imply yB(z±) yc(z±) « a' and y'B(z±) y'c(z±) a a (18. 18.133) y+(z±)^0.5.132) Fig.
(1.142) (<Z9o)J(l)^°+1)(l)1/2 . _ 8)} s . (18.138) Using formulae derived in Example 18. Anharmonic Oscillator Potentials (B) Evaluation of the boundary conditions Inserting into the first of Eqs.«*{=(. . .. +1)} = 4 m. .136) as  M . (18.141) 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.„*{!<. (18.54) +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.136) the dominant approximations we obtain (cf. 7 .2 we can rewrite the second of (18.9.5..416 CHAPTER 18. (18.134) this equation can be written Now sin  ^ ( 9 + 1 ) 1 oi sinl j(q0 +I) \ + ^(qq0) ~ (_l)^(9o+i) ( g _ g o ) for cos I ^(q0+ !)[ + ••• 50 = 3 . also (18. . Thus sin {^ +1) } = ^^>^30fe^i. Eqs. 1 1 ..m Using Eq.
3 The Double Well Potential Thus altogether we obtain 417 ^ • . Hence we have to impose at z — 0 the conditions j/_(0) = 0. We emphasize: We needed the solutions of type A for the definition of even and odd solutions. We do this with the help of WKB solutions.1 " <813 1 4)  In Example 18. y+(0)^0. 18.18.3. we must also demand this behaviour here along with a nonvanishing Wronskian.» ^ B ^ ' .144) Thus we require the extension of our solutions to the region around the local maximum at z = 0.5 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.6 Turning points z$. y'+(0) = 0. Since the type A solutions are not valid at the minima. we matched them to the solutions of types B and C which are valid there and hence permit the imposition of boundary conditions at the minima. y'_(0)^0. V(z) — • z E=V Fig. 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. (18.z\. We deduce .
418 CHAPTER 18. h8 25c2 (18.e. we see that the turning points ZQ and z\ are very close to a minimum for reasonable values of q ~ go> the latter being an odd integer. As a consequence.147) Since the minima are at z± — h2 /2c with /i 2 very large. (18. i.145) \qh\ + \zW Using z+ = h2 /2c.96) that the two turning points at ZQ and z\ to the right of the origin are given by \qh\ i. 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. We also note that the height of the potential at the turning points is h8 qh V(z) 20. to the left of ZQ where V > E.^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.24 C Z O 1/2 ZQ = Ac2 + + °(r A_ 2c 1 2 2q\ 1/2 2 / cq 5 + h± + 0(h~ ) h%) (18. In the domain 0 < z < ZQ.e. in leading order the integrals to be studied below from ZQ to z = 0 are equal to those from z+ to z = 0. ZQ) means "to the left of 20" . Anharmonic Oscillator Potentials from Eq.\qh\ + \h%U{z) (18.144).148) ("f * "The superscript (I.146) and Iqh* 1/2 z\ Ac" + +o ^ (2g2)V4 2c + /i ' (18. + £\h%U(z)=Q. one finds that .
151) In identifying the WKB exponentials we recall that y^KB is exponentially increasing and ywKB is exponentially decreasing.^ j O e . )+^*+)2lr} rnD h\z'2)}z Q )\ = .y^(z) to 2/VVKB(Z) anc ^ ^ W K B ^ ) .* 4+ .3 The Double Well Potential and 1/4 419 ywKB\ ) z — \qh\ x exp I / + \h\U{z) 1/2 I. 20 dz \qh\ + \h%U{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 >'^. .^ + ) 2 + igln2(zz+).149) and consider both types of solutions in a domain approaching but not reaching the minimum of the potential at z+.149) In order to be able to extend the even and odd solutions to z = 0.(29/^)1/2 1 ^(zz+){(zz+) 1/2 2 2q_\ 1/2 hJ 2 l l q+ q  ^ ( ^ .150) Evaluating this we have ±h ~ h2 GMln 1 In 2g q In +4 1 (z .148) and (18. (18. we have to match yA(z).18. Thus we consider in the domain \z — z+\ > {2q/h2+)l/2\ h 1.^ e therefore have to consider the exponential factors occurring in (18. (18.i (2q ln (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) .
e.125a). \ yWKB\ ) z 1 fl(g3)]! V*(h%)V*\zz+\h(*+V\2 q/A h + e* \(zz+)W+t (18. t h a t the Stirling approximation is amazingly good even for small values of the argument). 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 (18.153) This expression is valid to the left of the turning point at ZQ above the minimum at z+. l/2 ~h\U{z \(zz+?h\ (18.420 CHAPTER 18.152b) [\(q 1)!] V2 K7: hi Here q/A was assumed to be large but we write [\{q — 1)]! since this is the factor appearing in the solution (18. the results necessarily require adjustment or normalization there in the q—dependence. Anharmonic Oscillator Potentials We observe here t h a t the W K B solution involves unavoidably the quantum number dependent factors exp(g/4) and (q/A)qj/4 which do not appear as such in the perturbation solutions. of course. Since correspondingly 1/4 2 l/2 \*% we obtain + 4/1W*) (*^)i/'(M)i/*' 27T 1 / 2 {h\)^{\{q1)]\\2"+ *+ 1(91) ~hi l2 q/A e \{zz+)*hl for \z — z+\ > m 2q \ (1 ' (18. since there is no way to obtain an exact leading order approximation with Stirling's formula for small values of q. Stirling's formula is the dominant term of the asymptotic expansion of a factorial or gamma function and thus assumes the argument (oc q ~ 2n + 1) to be large (it is known.126a). We see therefore. This is the aspect investigated by Furry [102]. In a corresponding manner — i. we can write the exponential as exp (27T) ("f 1/2 I dz Z+ 1 qh\ + q/A 1. using Stirling's formula (and not the inversion relation) — we have —(1.ZQ) .125a) and (18. However.154) . Thus using the Stirling formula z\ ~ e~(z+1\z + l)z+2y/2~rr.
K?+1). . it is our philosophy here that the factorials with factors occurring in the perturbation solutions are the more natural and hence correct expressions.143) T^^*)) 27T (18. (18. .159) 7 i(«i) r(9^3) ^2~ j(9l) . and is shown to reproduce correctly the result (18.153). Returning to the even and odd solutions (18.143) which was obtained with our perturbation solutions from the boundary conditions at the minimum. 27T1/2 9/4 1VA{Z).155) (2z.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. 7 of t h e matching relations y^NKB\z) i. (18.n('>2o) 3/WKB(Z) =1VA{Z)I (18.103) and (18.158) Applying the boundary conditions (18. Comparing for z —• z+ the W K B solutions (18. multiplied by (1 + 0(\/h\))) the proportionality constants 7 .157) is used in Example 18.156a) and 7 _ i\(Qm •K •/2(M )l/4 ^ 2 n« q/4 (2*+) •(9D e ^^4.157) Since the factorials [^(q — 1)]!. [\{q — 3)]! are really correct replacements of [\q}\ only for q large. However.99).144) we obtain fi(0) Tin the present case as 1 (91) 1 — 5 (18.18. \hl. linearly matched) W K B solutions.e. The relation (18. (18.e.123) we have V±(z) 1 [yA(z)±yA(z)} = ^y^l(z) ± ^j#&(*) (18.154) with the typeA solutions (18.5 for the calculation of the tunneling deviation q — go by using the usual (i. this result is somewhat imprecise. we obtain in leading order (i.e. as the results also seem to support.3 The Double Well Potential 421 where [\q}\ was written as [\(q — 3)]! for q large. (18.104).
Friedman [40]. Anharmonic Oscillator Potentials Thus we have to consider the behaviour of the integrals occurring in the solutions (18.3 evaluated at z = 0 is then given by = y/T+u[E(k) uK{k)} (18. Byrd and M. 3G2 12V2c ^See P. .163a) (18.160) Setting 6 =4?^T"z°' rb Ac2 + VQ—> (18.162) T h e integral appearing here is an elliptic integral which can be looked up in Tables. 213..163b) The integral hi?) . .k'/2. F. p. (18. G2 K [i 4.148). formulae 220.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. 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).iV2 ! i . 60 and 361. p.422 CHAPTER 18.164) / 2 (0) z ca [(1 + k2)E(k) .149) near z = 0.161) we can rewrite the integral as r I2{z) = = dz^{a2 .. K(k)} 3 ^ 2 y/T+u[E{k) uK(k)\. D.19. n + uY 2c „ ^6 8V2c2 (18.05. (18..
.166 ) ^^iffil/^(2*)1/affiii£ Correspondingly we find ^ (18168) S W I 2 = o .r1)/10//2 ( f g 1^ yS» /L ( ^ ) l^4 . n = 0. Here we are interested in the behaviour of the integral in the domain of large /i 6 /c 2 . where the result is shown to be (with q ~ q0 = 2 n + l .± ' 3G 2 ' T ( ^ (18.e T ^ 5 U W ( 2 „ ) i / 2 ( ^ i W4C • (18 170) (18 1?0)  ^The expansion of I2 used in Ref. [167]. [4] misses an n—dependent power of 2 in the result. .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 ..( 2V + l)ln V 4 / 4( 2 n + l ) l nV 2 n .. . .3 The Double Well Potential 423 where K(k) and E{k) are the complete elliptic integrals of the first and second kinds respectively. n.2 A( L ) 4 s / and z dz yWKB\ J and d^WKBW 2=0 (*?) a ( . /2n + l 2 ± — .1)]' Gfo.165) in agreement with Ref. ^ ^ .3 )l ! .( ft 7. 1. The nontrivial expansions are derived in Example 5.) = 2 \..§ Since the integral is to be positive (as required in the WKB solutions) we have to take the upper signs. ± . (G\ .1.18.1 " e " O T (18 169)  Y^V / 4 tfo. which implies small G2.
175) ~E0(q0.1.173).h2) + (qq0)^=. . 18. .. (i*™) We obtain the energy E(q..^ W ^ / 4 t o . Anharmonic Oscillator Potentials With these expressions we obtain now from Eqs. (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.3.n = 0.. . h2) the split expressions 29o+1/i2(—W2 h6 (18.i ) ] ! e ~ s A ' . . 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.157).172) A corrresponding relation holds for h+ and z+ replaced by /i_ and z. we obtain for E(q0. 2 .3. (18.. (18.424 CHAPTER 18.»W. (18. 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).106).174) . Expanding this around an odd integer qo we have ( 8E\ h? —J Inserting here the result (18. Inserting the expressions for h+ and z+ in terms of h.143) the replacement: {hlY'2{2z+fe^zl + 2 9 ( ^ y Z e"A. and obtain ( * . (18.143) with qo — 2n + l. 18. with the help of Eq. (18.h2) and hence the splitting of asymptotically degenerate energy levels. (18. we can therefore impose the boundary conditions at z = 0 by making in Eq.3 concerning the exponentials.6 Eigenvalues and level splitting We now insert the replacement (18.172) into Eq.
K.e. The result (18.n = 0.157). for finite h2.143) with (18.^ 2^ h* .. 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 / » : = \ .h2) / h 6 \ —K />6\n+l/2 q o / 2 2«>+2h2 — —^= m 1.106).q2 7. the following expression for large q ~ 2n + 1.V2^\ \n+ i .1 7 6 ) (2n+9)/4^/ft V7T«! V C / \ eiI72^.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 . (18.^ e 21/^6c2 fe6 ( 1 mass mn = (1 8 .2.. i n U n / ! Thus AE(q0. .7 =( e N n+1 .e. in the WKB restricted sense) defined by ^( n +2^KB(0) 2 Here the derivative dE/dn corresponds to the usual oscillator frequency.1. Bhattacharya [23] the "usual WKB approximation" of this expression is — with replacements h4/4 <» k.e.174) and (18.h2) = E_(q0. i.18.e.176) is described by Bhattacharya [23] as a "modified well and harried result AjfWB.2 = 2 ..fr2) is given by Eq.e. T \ *n ! 1 ' (18. 2h4' i. the difference between the eigenenergies of even and odd states with (here) qo = 2n+l. 1 u2 c 2 (3 9 2 + l) ^ « b .3 The Double Well Potential where £0(<7o.. i.h2)E+(qQ. can be given by In the work of Bhattacharya [23] the WKB level splitting is effectively (i. the level splitting. which evidently supplies some correction terms. the pure WKB result of Banerjee and Bhatnagar [14] (i. Combining Eqs. 1 7 425 2 . (18.
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. K. The Furry factor represents effectively a correction factor to the normalization constants of WKB wave functions (which are normally for mo = 1/2. Then E^h2) and AE becomes 2^+U 2 (^)W 2 h6 = Eo{q0. (18. Then the splitting A ^ W B of Eq. however. as we do here.3)). as is also explained by Bhattacharya [23]. h2)\mo=i = . that if this symmetry is taken into account from the very beginning. H." Of course. (66). Anharmonic Oscillator Potentials which is unity for n — oo with Stirling's formula. 2/z4 and 2c2 respectively (see Eq.426 CHAPTER 18.h )/2 2 2 ( — ' 3 6\/2c A h6 n+l/2 h6\ 2{2n+5)/if \ exp 2 \ c / of Eq. Eq. Bhattacharya [23]. the Furry factor corrected WKB result follows automatically.** Had we taken the mass mo of the particle in the symmetric double well potential equal to 1 (instead of 1/2). (18.2 / 7. supplies the bridge to the perturbation theory results and removes the unnatural appearance of factors e and nn.180) h8 E0(qo. 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(q0.6 \9o/2 e~h&l^\ with (m 0 = l) (18. and A there with c 2 / 2 here.h2 <> — h2.6) to yield an improvement of WKB results for small values of n. This. The factor set equal to 1 represents a somewhat "mutilated" Stirling approximation.4/4 here. ** Comparing the present work with that of S. Furry [102]. (18. 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. .2X2 V(z) = \z2tL) .. we identity the parameter k there with ft. h = 1.176) together with Eq. A>0. we would have obtained the result with E.h2) * ^ 2 q o / ^ q l _ m ^  (18179) l~o o». h4 and c2 replaced by 2E. these deriva> tions do not exploit the symmetry of the original equation under the interchanges q <> —q.181) "W. (18.^ 2 + ^h2 If in addition the potential is written in the form A / .178).
Jentschura [293].—Q. Jentschura [293]. D. (18. provided their result is multiplied by a factor of 2 resulting from a corresponding inclusion of antipseudoparticle (antiinstanton) contributions.^ *See E.182) implies AlE(l. See R. Miiller—Kirsten [167] the potential is written as V2 VW = \ for mo = 1.180) agrees similarly also with the result for arbitrary levels obtained with the use of periodic instantons* and with the results of multiinstanton methods. (18.^ Thus for the case of the ground state Eq. J. Eqs. (18. Liang and H. however. ZinnJustin and U. We considered above only the symmetric twominima potential. 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.h>) * 2V2^^^y/2e^3/3A. The asymmetric case can presumably also be dealt with in a similar way since various references point out that the asymmetric case can be transformed into a symmetric one. carried out along the lines of the corresponding calculations for the cosine potential and thus of the wellestablished Mathieu equation. Patrasciociu [110]. M.181) therefore implies the correspondence rj1 <> A/2.3 The Double Well Potential 427 a form frequently used in field theoretic applications. K.7 General Remarks In the above we have attempted a fairly complete treatment of the largeh2 case of the quartic anharmonic oscillator.3. ZinnJustin and U. Friedberg and T. In principle one could also consider the case of small values of h2 and obtain convergent instead of asymptotic expansions. first paper.* A similar correspondence is also found in the case of the cosine potential. Gildener and A. ^To help the comparison note that in J. m 2 <• p 2 / 2 . and g2 is given as g2 = r]2/m3. Equation (18. 'This will become evident when the perturbation theory result of Chapter 17 is compared with the path integral result in Chapter 26.34) and (E. MiillerKirsten [186]. H. Banerjee and S. J. formula (4.11). The comparison with Eq. J. these are presumably not of much interest in physics (for reasons explained in Chapter 20).15).W.§ 18. Lee [98].c2 = A/2. Bhatnagar [14]. P. M.88) h4 = 2/x2. §J. W.18. (2. D. Mansour and H. so that by comparison with Eq. D. .
Mahapatra. Santi and N. Very illuminating discussions of double wells and periodic potentials. [19] — that the expansions considered above are asymptotic. Perturbation theoretical aspects are "See e. In fact. The ground state splitting of the symmetric double well potential has been considered in a countless number of investigations.[131]. Tables of properties of Special Functions are filled with such expansions derived from differential equations for all the wellknown and less wellknown Special Functions. as some relevant references with their view we cite papers of Harrell and Simon [129]. Simon and Wightman [178]. as he discusses. we cite papers of the Uppsala group of Froman and Froman [100]. for instance. like those for anharmonic oscillators. though not exclusively numerical.g. has also been the subject of numerous numerical studies. [132] and Loeffel. As references in this direction. 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. P. Ashbaugh and Harrell [12] and Harrell [130]. in his recent paper on simple uniform approximation of the logarithmic derivative of the ground state wave function of anharmonic oscillator potentials. Sophisticated mathematics — like that of the extensive investigations of Turbiner [273] over several decades — seems to approach the problem from a different angle. The double well potential. 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]. 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. The wide publicity given to the work of Bender and Wu made pure mathematicians aware of the subject.428 CHAPTER 18. Martin. in both symmetric and asymmetric form. . A reasonable. Anharmonic Oscillator Potentials Every now and then literature appears which purports to overcome the allegedly illnatured "divergent perturbation series" of the anharmonic oscillator problem. 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. Pradhan [182]. N. mostly in connection with instantons. can be found in an article by Coleman [54]. 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]. There is no way to obtain the exponentially small (real or imaginary) nonperturbative contributions derived above with some convergent expansion. There is no reason to view the anharmonic oscillator differently. B. B.
183) S o l u t i o n : We have k2=1^.f + • • •) 2u\ . 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.2G^2q + 4G2q .G2q + .3 The Double Well Potential 429 employed in papers of Banerjee and Bhatnagar [14].\G\ + • and 1k2 =2Gv/2q4.5: Evaluation of WKB exponential with elliptic integrals Show that (l + u)^2[E(k)uK(k)}'.192) 'F 12 16 . Gutschick. + (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. Bender. (18.. (18. 2 a.190) Hence ln]=ln : l n ' 7 2 ^ + 2 (18. which is assumed to be small. Cooper and Strottman [221]. [122] as 13 1.191) Our next objective is the evaluation of the elliptic integrals E(k) and K(k) by expanding these in ascending powers of k' . fc ~ 1 . 1+ u : Gy/2q.189) =2u 2u2 (18. Example 18. We have 4 ~k' 2u(l . X +2 + ' ^ (18.187) We now reexpress various quantities in terms of u. Srivastava and Varma [24] and Bhatacharya and Rau [23]. Thus k' and hence k' = v ^ ( l .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 . Saxena.186) (18.184) Hence we obtain for G close to zero: 1Gy^g l + Gy/2q~ and 1 . k! (18. Datta.^ + • (18.185) (1 + u)^2 = (1 + GVW/2 = 1 + GJ±. T h e r e f o r e it is convenient t o deal w i t h this here.18.u)1'2 = V2u{ 1 . Biswas. q ~ 2n + 1. We obtain the expansions from Ref./4 E{k) = 1 + • l n In (18.
197) .« 2 + 3u] In [ . 8 . (18.194) and (18.u) + uf\ 2 + ju2(l 16 + (u W 2 12 4u2(lu)2 +  l ) u ( l .u ( l .194) ln '^J + 2 In i/2u )+H l + 2u(l .3 .2{«(1 .3) = = ~ .. 8 + 9u 2 .195) we obtain E(k) 1 + ln uK(k) .u)} u{u(l .3 u .6 in the denominator and in the last step pick out the lowest order terms in u (i.430 and CHAPTER 18.3) .«) + 2} + V [ .« ) ^ { 2 4 .1)(1 .2)1 4 uA"(fc) In ( ^ = ) ^{«(1 .13)} + ±u3(u . 8 + 6 u 2 .u 13 \ 1 ln(± k> fc'2 +  (18.1)(1 .u) + 2} + i u ( u .3 u ( l .u)} + — u ( u . (18.{ M 2 ( 1 . 4 \ u In —= I + V W 2 1 + ln x(l u) 31/ +.u)(u .u) 2 {8 .4.3).3 y .1)(1 .e.^ L ) H[(i _ U ){4 + 3u(l .3u(6u . Anharmonic Oscillator Potentials *« = M £ H Consider first E(k): E{k) = 1 + .u ( l .6w2 + 13u} + 12u 3 (« .193) 2u(l .3) . .1)(1 .{l u ( l u) / W 2 + 2} .195) With (18.u) + 2u} + i u 2 [ 2 + (1 . ) J 4 (18.13)} + ~u3(u .u){4 + 3u(l . up to and including u2): u(u . (18.u) 2 {24 .u){24 .— ) .3 « .196) Now consider the last line here without the factor 2.3u(6« .u) + 2}] 1 H 1 o u(u .3) 1 + In ^ .18u 2 + 39M} + 12u 3 (u .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){24 .3u(6u .u 3 ( u .u ) (^){« + 5^ 1 )} + 5.u)} . 1 3 « 2 = .13)} + 12u 3 (u .3u(6u .a = (l«)(jl 2u/ 2 ) .u){24 .13)}.
.2 l_ln(1 JL\ V {2.4u .3).u H I . this becomes _2 2 u 2 / j M 2 / 3^2s 2 ~ 3G 2 2G2 H \ / 2 H / .^HfHGH Example 18.it)} . From (18. + iu^Zu2)^~u+ — +0(u3) l . we have.196) and (18. (18.uK(fe) ~ 1 + In I — = Mu2(3u.J? dz^\E V(z)\ \ 87T2 / lEVtz)]1/4 Solution: From Eqs.200) Remembering that u = Gy/2q. i.195) with (18.199) We now return t o the expression on the left of Eq. [(1 .156a) we obtain a and 7 and hence the ratio (always in the dominant order) T2~ ) w i t h h += ^ h • Since (cf.3) .18.2 (18.197) we now obtain 4 1 • E(fc) .128) and (18.e. (18 201) = Thus 3ffl« (2i/aGi/a2i/V/V"4 3ffl2 (Gii7»J4  '• .2 /o 2 431 (18. Eqs.2{w(l u) + 2}] = u(3u2 . 2 .128) and (18. (18. VB(z) = y\hTB(z) with VB(Z) ~ Bq[w±(z?  ^ [±(q  J l)}\2^V .e.«){4 + 3u(l .6: Normalization of WKB solutions Show that the normalized form of the WKB solution (18. ' 8G 2 q = 4 2 q ln f 26/2 \ o .3 The Double Well Potential Now consider the bracket [. we obtain: h ^ —^ 1 + 3G 2 V 2 8 2 /„ u 1 .196). where w±(z) — h±(z — z±). (18.. .148) (for particle mass 1/2 and in dominant order) is given by H. Inserting here from the above expansions the contributions up to and including those of order u2.] in (18.173) t o be evaluated. V^LJ 4 2 16 ^ . / to + 3GH~ \ 8 . l2 = ^ ( l + uf/2[E(k)uK(k)}.4u . i..zo) HWKB normalized _ ( h* y^eM. (18.155)) <*VA{Z) = VB{Z) a n d VWKB(Z) = 1VA{Z).198) 9 «.
93. 1 7(l. Example 18. W. pp.zo).hA/2 = h^/4 this implies . Then ((r. The constant obtained here will arise in Chapter 26 in the evaluation of the normalization constant of the WKB wave function (cf.*o)/ ^WWKBOO (18.Aw) = s/2i 2("!) V (41) and [(9 .f hi\ VWKB. 82.203) 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+. (26. by using the periodic WKB solutions below the turning points. y±(z) = IVA(Z) ± yA(z)\ = — i M O O ± 1 Ji. are given by Eqs. 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. (18.158). i. Solution: The turning points at zo and z\ on either side of the minimum at 24.7: Recalculation of tunneling deviation using W K B solutions Determine the tunneling deviation q — go of q from an odd integer go. i. and obtained as Eq.l)]![i(0 .e. Eq. (18.143).e.432 CHAPTER 18.147).35)). Anharmonic Oscillator Potentials With the standard normalization of parabolic cylinder functions. ° \Edzy/\EV(z) V(z)V4 VWKB. (18. Magnus and F./ .146) and (18. (18.g.204) . oo . 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 .3)]!23<«D [5(91)]' >A.** oo / dwD\. Oberhettinger [181]. We start from Eq. normalized = We see that the normalization constant is independent of a quantum number. 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 [ .
3. i. z{) we expect tt (18. 7T\ + ^ 4 7r\ dz . this implies sin cos sin cos Thus e.210) Choosing the point z to be z+.. i.211) 7 z l . ^(7(z) r t J z0 •I Zi I V2 dz \qh\ 1. . 6.^r + + .e. i. JV = 1. (18.206) !/2 ' + J*+dz(^qh2+±h%U(z) 1/2 7T' 4 1 =p — sin 7 f*+dz(^qh%h\U(z) ^ TV + 4 (18.g.212) tt See e.iK h\U{z) + / 4 (18.205) We now apply the boundary conditions (18.208) and cot L />G + 1/2 ^(ig4^+^))"" + J 1/2 2qh%h%U{z)\ 27 +  :± (18. Sec.i ) N c o s / •••+! provided the BohrSommerfeldWilson fZ1 dz^/E J zn quantization 4 condition holds.J^Wj +(18. from ZQ.r + ^=co S [r.e.( .207) (18. \h\U{z) 1/2 .132) and (18.2.. of course. A..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 . from z\. 1/2 .. Messiah [195].V(z) = r J zo I ZQ dz[\qh\ \ Z h%U(z) (27V + 1 ) .18. approach the minimum also from the right.Jz0 sin ^o J LJzo 4 Jzo 2 y •••4y ••• = .209) "27 In the present considerations we approach the minimum of the potential at z+ by coming from the left..2. Then at any point z € (zo.e..g.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. We could.6.
214) {«+"i} 0 for q = go = 3.217) 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.143). Banerjee and S.205). Bhattacharya [23].5. where u> is the oscillator frequency of the wells. 5. M.215). 1 1 .exp ( \/2h2T V2 7r 7 dz l / qh\ + \h\U{z) (18. One may note that the exponential factor here is not squared as in decay probabilities (squares of transmission coefficients). P. (18. in fact (with our use of qo).215) we can expand the left hand sides about these points and thus obtain . 7 . The formula thus agrees with that in the literature. We now see that Eqs. . ' /^i . We note here incidentally that this agreement demonstrates the significance of the factor of 2 in (18..27 9 .2 .216) Inserting this into Eq.213) The integral on the left can be approximated by 1. + l ) j } ^ 1_ 27' (18. (18. the level splitting is given by AE(q0.149).. and cot< (q + 1) '!}» for q = 50 = 1.e. Lifshitz [156]. (18. .215) we can insert for 7 / 7 the expression given by Eqs.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. 2IO/K.434 CHAPTER 18. 9.Zi / r.215) which results from the factor of 2 in front of the sine in the WKB formula (18. K. Landau and E. (18.h2) = = h2 h2 1 y 1/2 2{EEo)=2^{q~q0)=2^±l V2 V2h2 . and L. 7r"y in agreement with Eq. (18.90 — =F— for q0 = 1. K. S.146)). Bhatnagar [14]. they become c o t { ( ._ £/>(£<•>•) tan{(g + l Since tan ) ^ ^ . (18. (18. The latter give 7 7 1/2 exp dz dz \qh\ \qh\ + + \h%U{z) 1/2 exp I — / \h%U(z) (18.210) assume a form as in our perturbation theory. W "Cf. D. . In Eq. Anharmonic Oscillator Potentials Similarly under the same condition (l)Jvsin J Zn where it is understood that qh2/2 ~ Ey(z±).148) and (18.3. i.209) and (18. 2Sm (2 g /ft2)i/2 in agreement with the right hand side (the last step following from Eq. The prefactor 2\/2h2/n is. .
M. there is no need to have only field 'Generally a potential more singular than the centrifugal term in the (3 + l)dimensional Schrodinger equation is described as singular. However. since no probabilistic interpretation is available for the field theory matrix elements in virtue of creation and annihilation processes during the interaction. J. in an early investigation Case^ showed that potentials of the form r~~n. The centrifugal potential oc r~2 is generally considered as exceptional and is treated in detail in wellknown texts on quantum mechanics. W. Feshbach [198]. Morse and H.* However.* The physical analogy between singular field theoretic interactions and singular potential scattering of course breaks down at short distances. 'For a review from this perspective with numerous references see H. Some decades ago — before the discovery of W and Z mesons which mediate weak interactions — and before the advent of quantum chromodynamics. are not as troublesome as one might expect. W. 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.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. also in P. Giittinger and H. 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. H. Aly. f K . Case [45]. In particular Case pointed out that for a repulsive singular potential the study of scattering is mathematically welldefined and useful. 435 .Chapter 19 Singular Potentials 19. MiillerKirsten [9]. n > 2. M.
MiillerKirsten and Jianzu Zhang [226]. 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. H. In fact. N. MiillerKirsten. D referring to Dirichlet boundary conditions) in 10dimensional string theory. our presentation below is deliberately made elementary and detailed so that the reader does not shun away from it.2 19. Singular potentials arise in various other contexts. .2. H. Park. J.436 CHAPTER 19. Singular Potentials theories in mind. In the concluding section we refer to additional related literature and applications. J.e. Callan and J. 19. This chapter therefore gives the first complete solution of a Schrodinger equation with a highly singular potential. S. R. G. that with the hyperbolic cosine replacing the trigonometric cosine). § 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). for instance. this particular singular potential plays an exceptional role in view of its relation to the Mathieu equation. It is therefore natural that we study this case here in detail. K. the attractive singular potential 1/r4 was found to arise in the study of fluctuations about a "brane" (the D3 brane. J.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. Tamaryan. Hashimoto [124]. but also that the Smatrix can be calculated explicitly in both the weak and strong coupling domains. This property singles this case out from many others and attributes it the role of one of very few explicitly solvable cases. Maldacena [41]. In view of the unavoidable use of Mathieu functions expanded in series of Special Functions. some also permitting further research. (19. S. W. D. We shall see that not only can the radial Schrodinger equation be related to the modified or associated Mathieu equation (i.2) § C . Recently. roughly speaking a brane is the higher dimensional equivalent of a membrane visualized in two dimensions from which the DS brane derives its name. Manvelyan.Q. Gubser and A. Liang and Yunbo Zhang [189]. S. W.
( I + 0. (19. We begin with the derivation of various types of weakcoupling or smallh2 solutions which we construct again perturbatively. In the following we study the solutions of Eq. [l+2) ~ * A> ~ the equation converts into the periodic Mathieu equation of Chapter 17. This is rarely possible and therefore this case deserves particular attention. MiillerKirsten.^ where \f\ can be an integer and thus leads to singularities in the expansion of v (see e. The radial Schrodinger equation then assumes the form 2h2 cosh 2z . Eq. It is clear that we draw analogies to our earlier considerations of the periodic potential. R. . This has advantages compared with higher dimensional cases. h2 = ikg.g.Q. \[\ in the correspondence is nonintegral. W. It may be noticed that since I is an integer.4) first for small values of h2 and then for large values of h2. (19. r = 7 e z . 19. Manvelyan.28) below). h = ei^y/k^. J.2 The Potential 1/r4 — Case of Small h2 437 where E = k2.4) dz2 + In the literature this equation is known as the modified Mathieu equation or associated Mathieu equation. J.19. We observe that with the replacements p2 z^iz.7 that we employed also in previous chapters. Liang and Yunbo Zhang [189]. and so =4 5 and n ez = r = ~ = e^4fir. The following substitutions are advantageous: y = r1/2cf>. Our ultimate aim is the derivation of the /Smatrix for scattering off the singular 1/r4 potential. Subsequently we derive the same 5matrix from the consideration of largeh expansions and calculate the absorptivity in the case of the attractive potential. ig h \j g (19. and as we have in mind in Sec. (19. 8.3 below) g2 is negative and hence h? is real for E > 0. using the method of Dingle and Miiller of Sec.3a) In the case of the attractive potential (as in the string theory context referred to above.3b) 7 = % j. 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. (19.mo = 1/2 and h = c = 1. H.
Hv (w). Neumann or Bessel function of the second kind and Hankel functions of the first and second kinds respectively.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.Nv(w). First we make the additional substitution w = 2/icosh z. (19. VahediFaridi [10]. 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. where £>„ : = —_ + i _ + J l .6) terms amounting to 1 „ 2 d2Zu A V (19. 8.i l . w 2 .I . and H„ (w). (19. Proceeding along the lines of the perturbation method of Sec.438 19.11) We follow here largely H.9) leaves uncompensated on the right hand side of Eq.7.2 CHAPTER 19.^% dw1 1 1 —— = dw = ][Zv22Zv 4 w Zv + Zv\ = (Zv\ 2 + Zv+2]. .4) becomes dw2 w2 w2 \ dw2 w dw \ j Next we define a parameter v by the relation V2 v2 = (/ + . (19..2 A / i 2 . MullerKirsten and N. — Zv+i). where these are the Bessel function of the first kind. (19.7) We shall see that this parameter is given by the same expansion as in the case of the periodic equation. The zerothorder approximation = Zv[w) (19.6) becomes to zeroth order </>(°) = <j>v of <f> in h2 Dvcf)v = 0.1.5) so that Eq. (19. (19. H.g) This equation is wellknown as Bessel's equation or as the equation of cylindrical functions. i. W. dwA w dw { wA ) (lg. J. H. —Zu — (Z„_i + Zu+i). Eq. The solutions Zv(w) are written Jv(w).e.2. Aly.
(19. The next contribution to (^°) + t^1) therefore becomes 0(2) = /i4[(z. wz The expression (19.13).v + 2)*{u + 2._2 + {v.20) 2)*ZV_2 + {y.13) (19.u (19. (i/.u + 4fZu+A}.2 The Potential 1/r4 — Case of Small h? 439 we can rewrite the expression (19. where We observe that DUZV = Q.6). v + 2)*(v + 2.15) (19.12) (19. of course. v)Gv + (y. where the starred coefficients are defined by (19.2 ) * ( z / .13) therefore lead to the first order contribution jfU = h2[{v.2. .2)*Zi. v . it is more convenient to use these relations in order to rewrite Eq. The terms (19.2 ) * i ^ 2 + (!/. v + 2)Gv+2]. .14) (19. However.10) as a linear combination of various Zv. = §. z . Du+aZu+a but Dv+a — Dv so that DuZv+a = a(2l/ (19.17) Thus a term fiGu+a on the right hand side of Eq. (19.21) +(i/. ~a(2v + a)~ w2 * a> Zv+a = a(2u + a)Gv+a.18) Now </>(°) = Zv left uncompensated Rv .2)G„_ 2 + (y.i/4)*Z 1 / _ 4 +(u. I / ± 2 ) = 1. We assume in the following that 2u + a ^ 0 (the case of 0 has to be treated separately).19.!/+ 2)*4°J 2 ].10) in terms of functions G„ defined by Gv+a = —nZ v+a .16) ( I / . v + 2)*Z„ +2 (19. can be cancelled out by adding to </>(°) the new contribution liZu+aja{2v + a) except.i/) = 2A.. when a or 2v + a = 0. therefore 4>^ leaves uncompensated i#> = / i 2 [ ( i / ^ . and this means in (19. (19.u + 2)*Z„ +2 ]. v2)*{y2. v .10) is now particularly simple: i?i°) = h2[(v.
i/ ± 2)*. p 0 (0) = 1. v) +(i/. i/)] + • • • .4 + 58. Adding these terms and setting the coefficient of Gv equal t o zero.440 CHAPTER 19.v + 2y{v + 2.I/±2)*(I/±2. and p 0 ( 2 j ^ 0) = 0.9) 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. we obtain t h e expansion . (19. we obtain 0 = h2{v. . .v2)*{v2. subject to the boundary conditions P2i(2j) = 0 for  j  > i. (I/.455a + 1291a .23) These coefficients may also b e obtained from t h e recurrence relation P2i(2j) = p2i2(2j .1169)/i 12 .j^0 (19. P2*o(0) = 0.26) iy — 2.i/±2)*.I/±4)*. Zv + Y^h2i i=l J2 p2i{2j)Zv+2j.25)/i 8 32(al)3(a4) .l)(z/ 2 .2.g. v + 2j) (19. Evaluating t h e first few terms.v) + {v. .24) P2i2(2j + 2){v + 2j + 2. . i/ ± 2)* (i/ ± 2..v) + hA[{v.25) Finally we have to consider the terms in Gv which were left unaccounted for in Rv .4)(»/2 . Ri. ^ 64(a_1)5(a_4)4(a_9) + 0 ^ )• (1928) 1 6 . v — 2)*(v — 2. (19. v ( 4 5 a .2 + 2 9 ) ^ 64(^ 2 . v + 2)*{v + 2. etc. Singular Potentials Proceeding in this manner we obtain the solution ^ = 0(0) + 0 (1) + ^(2) + . i/ + 2j). we obtain 9 h4 = a 2(al) 3 2 (13a . . j=i. (19. + 2(y2\) — + 32(i/ 2 l) 3 (^ 2 4) —^ (9. p4(±2) = (i/.22) where p2(±2) p4(±4) = = (i/. . Reversing the expansion and setting a = (I + 1/2) 2 (not to be confused with a e. 1 1 \ 2 + 2) ~ ) s X 2 A = v 2 hA (5v2 + 7)h8 v . . i/ + 2j) + p 2 i _ 2 ( 2 j ) ( ^ + 2j. in Eq.v + 2)*(i/ + 2.2)(v + 2j .17)).v)] (19.
^ + e±(v~2)z. ±2.An(u + n). (19. = 0. £>„ = — . Thus to O(0) in h we have (jr ' = 4>v — cosh vz.19.31) It follows that Dv+2n4>v+2n = 0. (19. v)(f>v + {v. Thus v is real for small values of  h21. v . apart from a normalization factor which we have chosen (so far) such that the coefficient of Zu in <f) is 1. — We return to this question later but mention here that this problem has already been dealt with in Chapter 17.i .cosh 2z)4>v = = 2h2A(/>„ . we can rewrite the latter as j ~ .2.29) sinh vz or e±uz. Du+2n so that D„(j)v+2n = ^n(u + n)<\>v+2nSince 2 cosh 2z cosh vz 2 cosh 2z sinh vz 2 cosh 2z e ±vz = Dv .2)<t>„2 + (v. (19.30) £>„&. ±{ v+ z e = sinh(v + 2)z + sinh(z/ — 2)z.7) into Eq. so that J2 (19.4). .33) = = cosh(^ + 2)z + cosh(^ — 2)z.v2(j> = 2/i 2 (A . We are still left with the question of what will happen if 2v + a = 0 or v — ± 1 .cosh 2z)4>. Substituting (19. (19.2 The Potential 1/r4 — Case of Small h? 441 We note that this is an expansion in ascending powers of h4.32) (19. 19. 2 .34) . and is therefore real for both cases g2 > 0 and g2 < 0. (19. we may say that the first approximation <^°) leaves uncompensated terms amounting to R^ = 2ft2 (A .h2[(j)u+2 + 4>v_2] h2 [{u.3 Small h2 solutions in t e r m s of hyperbolic functions For later purposes we require yet another type of solutions. The solutions 4> of the modified Mathieu equation are now completely determined.v + 2)<t>v+2].
I/±2)*.h). Arscott [11]. 19. 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.I/) = 2 A . • (19.h).37) where p M (±2) = (I/.J/±2) = .7 — we refer to Meixner and Schafke [193].1 . .J7t0 p2i(2j)<f>u+2j. etc. In fact we could have obtained the same Ru by starting with the modified Mathieu equation for h2 replaced by — h2. Meixner and F. CHAPTER 19.2. M.h). The solutions in terms of cylindrical functions are written *J. 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.4 Notation and properties of solutions We now introduce standard notation as in established literature. > These solutions correspond in the periodic case respectively to the solutions and me^.38) (f> with (\>v = exp(vz) — Meu(z.35) The form of i ? ^ is seen to be almost identical with that of the corresponding expression for solutions in terms of cylindrical functions.h) = <(>„ +Y. 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. (19.* The solutions of the modified Mathieu equation which we are considering here are written with a first capital letter. we prefer to adhere to one equation with different solutions. > (f> with </v = sinh vz — Seu(z. 8.h2i *=1 J2 j=i.442 where (I/. The use of the symbols (v. Schafke [193]. Denning (19 36) (">" + <*) =o^7Ta)> we now have the solution  p(z. W. F. In order to avoid confusion arising from the use of different equations. (19. Singular Potentials (I/.
Also from the expansion of Jv{z) in rising powers of z. h). as Mev{z. Magnus and F.h = £ p2i(2j)(l)j exp[(^ + 2j)z] (19. see this reference p. (19.h).19. (19.45) T Below we frequently write the normalized solution as in J. W.h2) oo = Y.h)±Sey{z.44) (19. we obtain$ Ju(2hcosh(z + inir)) — Jl/(2hcosh z exp(m7r)) — exp(inuTr)Jv.43) and hence similarly M^\z + inir. (19. h) are therefore proportional to each other as a comparison of Eqs.e.(2hcosh z). h) = av{h)M^ (z. h).44) implies. Me„(z. 'See e.h).MP(z. i. h) = exp(inwn)MP(z. ' (z. h). h). .h) = Jv{2hcosh z) + ^h 2i J^ j=iJjLQ p2i(2j)Ju+2j(2hcosh z). whereas the — as yet — unnormalized solution (19.39) Writing two solutions out explicitly we have for example — apart from an overall normalization constant. W. Schafke [193]. 16. 443 (19. (19.g.41) i=l We see immediately that Meu(z + niri. 4>v = Hv (2h cosh 2) * M^\z.h) and oo i M^\z.h). Meixner and F. — oo cv2r{h2)e^+2r>.40) does not. i. 4>v = Hi.40) Cev{z. <f)v = N„(2hcosh z) MJ?\z. p.42). this normalized solution possesses contributions exp(i/z) in higher order terms.h). h) = exp{vniri)Me„(z. For the relation of Hankel functions used below.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) . Oberhettinger [181]. which. 17. (19. As emphasized in Chapter 8. Mi. (19. introduces the dominant order function into higher order contributions''' — oo i 2i Meu(z. as we remarked earlier.e. {2h cosh z) • MJ?\z.42) The solutions Meu(z.h) = exp(vz) + Y.
444 CHAPTER 19. the first relation is obtained as follows: 2M<£1 = M^l+M^l. h) .4) we obtain oo £ r=—oo [(i/ + 2r) 2 4. h) = M™(z. For later essential requirements (i.47) With this equation we obtain for nonintegral values of i/:§ ±isin WK MJ)3A\z. i.e.43).49) Inserting this expansion into the modified Mathieu equation (19. ^ . MJ>>(0.h) can be shown (cf. the explicit evaluation of the S'matrix element) we elaborate here a little on the computation of coefficients of normalized Mathieu functions. p.8.48) The series expansions of the associated Mathieu functions M„ (z.e. 178) to be convergent for  cosh z\ > 1 but uniformly convergent only when  cosh z\ > 1 for otherwise complex values of z. These points are important in our derivation of the S'matrix below. p. i. The functions Me±v{z.h) on the other hand can be shown (cf. we use the notation of Meixner and Schafke [193] and so write the expansion of the function Mev(z. As mentioned earlier.g. H™(z) = e^HfXz). 4) = exp(±^7r)i\4 3 ' 4 \ M^3'4) = M™ ± iMJ?\ (19.exp(TM"r)MW(.h2) = ] T cv2r(h2)e{y+2r)z. oiv{h) = (19.e. Meixner and Schafke [193]. 130) to converge uniformly for all finite complex values of z. Meixner and Schafke [193].50) E. Equating the coefficients to zero we obtain [{y + 2r)2 + \]cv2r = h2(cv2r+2 + c£ r _ 2 ). Singular Potentials where clearly MeJQ.46) H^z) = e^H^(z). r=—oo (19. s (19.h) Using further properties like (19. h). .\c\r + h2(c^2 + 4 r + 2 ) ] e ^ + 2 ^ = 0.h2): oo Mev(z. (19. 2M™ = eiuvMi3) + e~ivlrM^\ . we have the following for a change from v to — v. Mi 3 .h) .
+ 2 r ) 2 ] .54) 2 C2r 4 ^ h. p. \ .2 [ A . \) /. 2r (19>51) 2r For ease of reading we give the steps in rewriting this.T22r[\ A .19.52) in agreement with Meixner and Schafke [193]. Meixner and F.2 The Potential 1/r4 — Case of Small h? which we can rewrite as fc2^±2 c 445 = [A _ („ + 2r) 2 ] . see J. + o2„r 2 l2 ]  C 2r 1 2r c 2r2 or 1 c J =^[A(. eg = c~v = 1.^ + 2r + 2 ) 2 ] . 117.51) we have 2r (19. Schafke [193].[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. Alternatively taking the inverse of Eq. W.(/^. p.* For 'Actually.^ C 2r+2 (19. 122./ i 2 C ^ .53) /i2[A(^ + 2 r ) ] .^ p 2 2 ^ or c 2r+2 2r2 1 (19.(v + 2r)2} / z . Thus now c 2r . (39). . (19.^ c 2r 2r 2T~2 or 2r ^2 /l2[A_(I/ + 2r)2]£2l± 2 . h~2[\ . . Eq.
( " + 2)2] ~ h2 h. MiillerKirsten and N..5 D e r i v a t i o n of t h e S .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. + 0(/. Aly.l ) 19.12) ^ 384(^ + l)(z/ + 2)(z/ + 3)  " ^rW^hy. 128(i/+ l) 2 (i/ + 2 ) ( ! / . (19.52) and obtain c% 1 eg fc"[Afr + 2)»)h_a[A_(.1 for the evaluation of a typical case): hb + 0(hw). and also R. + i).^ ^ + 2)(i/ + o> l)(i/ 3) 0(^10)._{u+m h2 M ( " + D + 55^iy] + i ^ 5 j /i 2 4(. H.[v^. W. H.m a t r i x Our next step is the derivation of the explicit form of the ^matrix for scattering off the potential 1/r 4 since this is possibly the only singular potential permitting such a derivation in terms of known functions. i/2 + 41/ + 7 7i6 + 0 ( / i 1 0 ) . . Spector [257]. Liang and Yunbo Zhang [189]. Then di 2 1 t" + 2(^1) + • • • . Singular Potentials nonintegral values of v examples obtained in this way are (see Example 19. Solution: We put r = 1 in Eq. but follow our earlier reference H. H. M.1: Evaluation of a coefficient Evaluate explicitly the first few terms of the coefficient Cj/cfJ given in Eq. Here we insert for A the expansion (19.+4)!ll_.55 ' Example 19. MiillerKirsten. Manvelyan.446 CHAPTER 19. J . J.27) and truncate the continued fraction after the second step. W.C + l) 3 (i/ . (19. t For this purpose *We do this in the manner of the original derivation of R.2. 8 4 ( i / + 1) 128(i/ + l)2(v + 2)(vl) + 5 10 32(i/ + ^ l)(i/ + 2) + 768(^ „„ . J.Q ..55).+0^"> < 19 . VahediFaridi [10] (this paper contains several misprints which we correct here). .
(19. and then the continuation of this to infinity.57) ^ 1 .3a) and (19. The asymptotic behaviour of the Hankel functions H„ ' (w) for \w\ ^> \u\.59) which tends to zero with r. (19. In the case of the repulsive singular potential here.5) w — 2h cosh z = [ kr \ r Thus r —> 0 implies \w\ —> oo. 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.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 . I / + 2 2 l\TT +E ^ i\ E j=i.j7L0 P2i(2j)exp i[v + 2j + 2 2 . \w\ S> 1 and — IT < arg w < IT is known to be given by* (19. p. we have W r = ip + ut — 7r/4' . this wave function near the origin is the wave transmitted into this region (as distinct from the reflected and ingoing waves). Oberhettinger [181].19.2 The Potential 1/r4 — Case of Small h2 447 we require first the regular solution yreg of the radial Schrodinger equation at the origin. 2 ) 2 1/2 ex Z/7T IT H=i —J TTW P l+O 10 (19. Magnus and F.56) where by Eqs.h) 1/2 w) (19. Then y reg = = r rV2M®(z. and considering the propagation of this wavefront. We obtain the regular solution by choosing Zv(w) = H„ (w) for 5ft(z) < 0. 22.. W . .
^ 1 ) (e i 7 r z) = Hi2!>(2) e^H^iz). Oberhettinger [181].j¥=o Ki(2j)exp(Tmj) (19. 4%r g >0 4r^r ' rg See W. We now require the continuation of the regular solution y reg to solutions behaving like y± at r — oo.h) can be shown (cf.57) the condition  cosh z\ > 1 implies I cosh z\ kr + ig/r >1./l ~ 2 ~ 2z + " 4 > 0. AV . The series M^ (z.3a) and (19.62) Also with Eq. Now from Eqs.V + § 9 2 > 4hh* .h) = exp(Mr^)il4 4 ) (z. We require therefore > > the continuation of Mi. (19.448 CHAPTER 19.3b) ln • 7T Q"*4' r fg ° Vk (19. '2"4 . This means that the origin of coordinates acts • > as a sink. Meixner and Schafke [193]. ./i). Prom the relation r = (ig/h)ez we see that r — 0 > corresponds to $l(z) — —oo and r — oo to $l(z) —> oo. h) through the entire range of !&{z). but uniformly convergent only when  cosh z\ > 1 for otherwise complex values of z. '(z. In a similar manner we can define solutions y± by setting for 9fJ(z) > 0: y± = rl'2Ml?>*>(z.(2j)<gH • (19. Singular Potentials so that when t — oo : r — 0.j^Q p2.h) oo j 2 — rV flp^ + ^fc* £ i=l j=i. 2h and the square of this expression implies 2 2 ~ ' A. (19.61) In fact. 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.A14A. we can derive y_ from y+ since one can show from the circuit relations of Hankel functions^ that M^\z + m.#.60) Using the above asymptotic expressions for the Hankel functions. Magnus and F. 17. p. p.A1"—2 z '+"2 A~~ n or rA . 178) to be convergent for  cosh z\ > 1.
63) Thus there is a gap between the two regions of validity — i. A suitable set is the pair of fundamental solutions Me±v defined by (19.49). (r — r+)(r + r+)(r — r_)(r + r_) > 0.2 The Potential 1/r4 — Case of Small h2 Hence (r2 .19. This interchange in the > solution My interchanges y reg with y+. 19.ft/4 _Z z = In r/rm/4 0 r 0 =(9/k) 1/2 Fig. z = In h i— r0 4 and 7T z = In • ro { 4 for ro < r < oo.ijt/4 Rez . Thus we choose . (19. we have to assign different signs to the imaginary part of z in the two distant regions. Originally we chose the sign of m / 4 as in Eqs. the domain r_ < r < r+ — as illustrated in Fig.3b) and (19. 19.e. 449 This condition.62).1 The domains of solutions. and in order to maintain this symmetry. i.e.1. r 7r for 0 < r < ro.r2_) > 0 with r£ = rg(2 ± >/3).rl)(r2 . . (19. 130).38) or (19. These solutions converge uniformly for all finite complex values of z (cf. i Im z z = In r/r0+i7t/4 .3a). Meixner and Schafke [193]. which has to be bridged by using another set of solutions. p. Since the real part of z changes sign at r = ro. (19. is satisfied only for r < r_ < r__ or r_ < r + < r. The one further point to observe is that the variable w = 2/icosh z = (kr + ig/r) is even under the interchange z — —z.
z=—in/4. Thus we set rl/2M^\r) = = rll2[aMev(r) + (5Mev{r)}. and we determine these coefficients from this equality and the additional equality obtained from the derivatives.e.^)] dr z=in/4 dr (19. At r = ro the real part of z vanishes and then switches from negative to positive. p. Hence we match the right hand sides of Eqs.v(z)\ —m/4) + ^[r^Me.B^0. we express in the domain r_ < r < r+ the regular solution as a linear combination of solutions Me±v with coefficients a and /3.66) a = (3 .67) Next we have to continue the solution beyond the point r+ to a linear combination of solutions y+ and y_. as just explained.[ r l/2 M.65) in/A a>d[rl/2Mei/{z)]+f3>d[rl/2Me_Az)] Since Mev(z. (3) (r)] aA[rl/2Mei/(r)]+/3i:[r. i.(2)]z=w/4 + 0Me. h) (as one can check or look up in Meixner and Schafke [193]. Singular Potentials Then starting from the region r ~ 0. 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. Thus. (19.)](1964) The right hand side of the first equation now represents the regular solution in the domain r_ < r < r+. at this junction we require also the imaginary part of z to change sign. . (3 = a .v{z)\ . to r^2[AM^ + BM^\ A. + ^[r'^e^z)] dr z=—iw/4 + (31 ^[rll2Me. a4[r1/2Mev(z)] dr a'^lr^MeAz)} It follows that (19. 131).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_. i.l/2Me_i/(T. (19.450 CHAPTER 19. h) = Me^u(—z.
(19. A and B.69)) cancel out in view of the nonderivative relations. (19. ^ ] = ^ + ^ ^ .65) and (19. (19.68) From Eqs. (19. Thus we obtain from Eqs.3a) and (19. (19. (3. we use the notation W\f. with the replacements of Eq.19. (19.68) the derivatives with respect to r by this relation.70) by Me'_u (the prime meaning derivative).2Mei/{r)]+d[rl/2M] dr dr = = rxl2[AM^\r) + Ad[ri/2M(z)(r)] dr +B^[rl'2M^{r)].) d_ _ 1 d dr r dz Hence for any of the functions Mv:  [ . for instance. dr BM^(r)]. and we are left with relations expressible only in terms of z. and the second of Eqs.71) We now determine the coefficients a. = aMev(z) + 0Mev(z). (19. (19. (19.67): rl/2[(5Meu{r) + aMe_u(r)} pd[ri.65).70) by MeU and subtracting .69) Thus if we replace in the derivative relations of Eqs.3b) we infer that (z = l n r + const. For the Wronskian W of two independent solutions f(z) and g(z) which is constant and can be evaluated at any point. (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.g]:=f(z)*&g(z)WZ) dz dz Thus. (19. Then. multiplying the first of Eqs.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. the nonderivative parts (from the first term on the right of Eq.68) \j3Me„(z) + aMev(z)] P±[Me„(z)}+afz[Me„(z)] = = [AM^\z) + BM^(z)}. (19.
.58) and considering large values of \z\. (19.71) we obtain in a similar way A B = = W\MP\ Me.e.3] = [1. (19. = IT <*„.57) and (19.] W[Mev.M^] J™aV. i.4] = . Me.\ \ ( av 2iwKav ^ v .Me^} (19.75) With these relations and Eq.^ ^ a „ Q _ . h). Singular Potentials the second resulting equation from the first resulting equation. W[M^\Mlf)]W[Meu.* « * r r z H .M^] W[Mev.g. using Eqs.j] given in Meixner and Schafke [193] (p..46)) W[Mev. (19.46). we obtain the first of the following two equations. 7T W[Mev. TV W[Mev.. (19.u. 7T [1.74) Moreover. 170/171). (19. (19.Mev]' P l ' From Eqs. p.48) (or cf. (19. 7T (19.452 CHAPTER 19.Mev] = = = av.47) we obtain (av defined by Eq. .A] = W[Mi3\z). W[MJ?\ M^] W[Mev.Mevy W[M?\Mev] W[Mev.M^} W[Me„. the leading terms of the respective cylindrical functions.. or obtainable by substituting e. With these expressions the quantities A and B are found to be A= f au ^ 2isin VK \ a_u 1 a. Meixner and Schafke [193]. av{h) = Me. h)/M^\o. (19. 169) M0) = e«^Ma) _ isin U1X M(A) _ (19.76) . M^4)] + W[MJ>3). we obtain immediately W[Z.^ 47 9? MJ>4)] MeV] [3.e. we use the following circuit relation which can be derived like Eq. and Wronskians W[M^\MIJ)] = [i.77) av I 2ism VK \ ctv av ^Thus.u}W[Me.4] = . and the second is obtained similarly: W[M?\Mev\ W[Mev.73) We now use Eq. i. B = J 2 L _ e .M^]= IT e~iv*av.MiA\z)} = 2(i)(2/Trw)(dw/dz) = (4i/Tr)(2hsinhz/2hcoshz) ~ 4i/>.(0.
/ ( . (19. 2zitsm z/7T zzitsin I/TT Inserting these expressions into Eq. Z ) e ^ ] . (19.h)] ( m / 9 \ 1/2 {L\ {Aeikrei(V+l/2)*/2+Beikrei{v+l/2)*/2^ g) We set fl = ^ = MM). f c r .f c . as we recall from earlier considerations (e.h) = Me_u(z.(ite™)" 1 { 8 J ' f(k.82) The S'matrix is defined in the following way in the partial wave expansion of the scattering amplitude /(#). 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 . Z ) e ._ [ / ( f c . (19. in Chapter 16).g.l) ' We note already here that with the substitution R = ei7r7./) = (l)1/2R2^2i^)ei^)U^.46) together with the relation Mev(z. (19.h).19.2 The Potential 1/r4 — Case of Small h? The regular solution in this way continued to r ~ oo is then y reg ~ ~ 453 Tl'2[AMiz\z.(l)le^}Pi(cos 9). B=tfD'"". (19. Then f' .85a) . (19.79) where the second expression follows from Eq.80) 2^fc (1981) where \7r/ i t sin VK /(*.83) The Smatrix element is therefore given by ifc*"* .78) we obtain Ay reg « _ .h) + BM^\z. eii/7r/2 (19.
19. R 2i sin TXV. R — — = 2i sin ivy.58)) in the limit r — 0. V(r)=g2/r Fig. and recalling that yreg is proportional to a function Mc> {z. Singular Potentials r(i+k) sin 7r(7 + u) (19. (19.2. and with the left hand side following from Eqs. 19. we can write this (multiplied > by 2ismvTT.56). Fig. 9?. the reflected wave and the transmitted wave the quantities: A Ar At = = Re1' R 2zsin 7r(7 + u).454 we can rewrite the .85b) We shall obtain the 5matrix in this form in the case of large values of h2 later.2 The repulsive potential and the various waves. (19. (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. (19. We can rewrite Eq. we can define respectively as amplitudes of the incident wave. Thus in terms of the (3) variable z.Smatrix in the form s i n 7T7 Si = CHAPTER 19. (19.81) in another form from which we can deduce the amplitudes of reflected and transmitted waves. h) (see Eqs.56) to (19.86) . cf.58)) 1/2 2r e~ 2hir cosh z.
M^l)(0. (^)l4lu{h2)Jir(hez)J±u+i+r(hez). A derivation is beyond the scope of our objectives here. Liang and Yunbo Zhang [189]. (105) to (108). By inserting in Eq. Eq. Eqs. therefore we cite it from Meixner and Schafke [193].. unity minus reflection probability = transmission probability. MullerKirsten.6 E v a l u a t i o n of t h e S .2COS2VTT = (R. J. however not here. (19. Hashimoto [124] and R.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. y We observe that this relation remains valid if the real quantity R = e and the pure phase factor eW7r exchange their roles. but in the 10dimensional string theory context.88) where the factor Me„(0. if R becomes a pure phase factor and eil/7r a real exponential. (19.h) is (cf. (19. and then exploit it for the evaluation of the quantity R. i.( i J e ^ ) " 1 ! 2 = R2 + ^ . i.** 19. 178). l=—oo 1 1 lite*"* ." \R~Ji\2 1  tO _  2 i s i n VK\2 [  .88) the power expansion of the Bessel function one realizes soon that the expansion is inconvenient in view of its slow convergence.e. J. Jv+2r.e.h) = 1 Me V\ 5 / r J2 __ 0 O c»2r(h2)J„+2r(2hcoShz).h) The function Mo (0. This implies basically the evaluation of the expression R of Eq. The expansion is: 4 r ( / i 2 ) M S ( z .±)2 + (2sinj/7r) 2 . (19.28)) and R = ey real. (p. Gubser and A.84) of the ^matrix.e™*/R\2 ' ' . (19.m a t r i x Our next task is to evaluate the expression (19.39)) the associated Mathieu function expanded in terms of Bessel functions of the first kind. h) serves the use of the same coefficients as in the other expansions.79). The latter is what happens in the S'wave case of the attractive potential. Manvelyan. this expansion is given as 1 oo MJ> Hz. in which v is complex.In Meixner and Schafke [193] (p. W.89) .e.e. / i ) = Y.Q. (19. " S e e S.„„„X (1 Q Q'T'l l \Reiun e~iuw/R\2 ~ \Reiv* . H.19. unitarity is preserved. i. 180).2. 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. S.
g.90) is in some sense amazing: It permits the evaluation of one and the same quantity M±J (0.e.91) Inserting here the coefficients given in (19. i. (19. r = 2 as a check).3 9 ^ . ^ (19./i) with the help of Eq. Here in the case of the repulsive potential.49) from which we obtain for z = 0: Me„(0.49). u»^. (19. setting r = 0 in Eq. (19.h) = Jo(h)J±Ah)^^Ji(h)J±„+i(h) c0 {n ) ^ ' C^r^J2{h)J±l/2{h) cr(h2) + •••. 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. h) in many different ways. by allocating different values to r. Eq. /=—oo (19.28)). + 2 u l\2j 2 fh\2 "2+2 2 MX4 {v l)(v 4)\2 2 2(^ + 4 ^ . one obtains *. (19.6 2 ) m 6 (^_i)3(z.55).90).2_4)(l/2_9)iv2y) + <AA.90) Here the coefficients c2"{h2) are the same as those we introduced earlier for the normalized modified Mathieu functions in Eq. r = 0 and 2. (19.h) v 2 = ^2 (l)lc%ih2)Ji~r(h)JM+r{h).g.. e.456 so that in particular for z = 0 oo CHAPTER 19./>) = Co> 2 ) £ 7 " = — OO j ± > . As a matter of introduction we recall the definition of the coefficients with Eq.90) and choose r = 0 (one can choose e. Next we evaluate M^1}(0. h2 is complex but v2 is real (cf.94) . (19. The formula (19. /n ^ . Singular Potentials 4r (h )M$(0. we obtain M&(0.
2 _ ^ 4 ( ^ 2 _ 4)2 (19.9) V2 1 fh ±v 1+ 2 + • (19.!' or with the help of the inversion formula of factorials R = v\{y — 1)! smKv f h 7T 2v (^ .32i/ .l) 5 (19.97) + This expansion will be used below in the low order approximation of the absorptivity of partial waves.23) (hxQ 2 3 2 2 V ± l) (i/ T 1) (^ .] 2 \v i + {v2 . 4(z/ T H ^ 3 .1) 2u(4u + 15i/ ..12i/2 + 64u . In this procedure our attention is focussed particularly on the quantity called absorptivity in the case of the attractive potential.. From expansion (19.55) and collecting terms in ascending powers of h? (here for the case of nonintegral values of v).4)(i/ .93) and the coefficient expansions (19.96) With this explicit expansion we can evaluate the S'matrix.95) These expansions imply for the quantity R: I/! 2v ri1 i . + # "+" ( I / . _2_(h\2 .4 ) V 2 > 1 (I/)!V2/ [1+ 2_Mx2 2(i/23«/7) (h)4 (iy+l)2(i/l)(^24)' .IV 2 2 2(v =F 3i/ .2^ 2 ± 59^ .2 The Potential 1/r4 — Case of Small h2 457 Inserting the power expansion (19.7) F " (i/±l)2(^^l)(i/24)V2./I) h u . 2 (h\2 ^ l/2_1l2^ 2(i/ 2 +3f~7) /feu .96) we extract for later reference the relation sin7rz/ R IT h h 2\ 2 Av [v\(viy. We observe that this approximation involves v and h4 and hence is real.111) / / i 5 4 3 1+ 4i/ 2 + +' (z.1 ) 2 ( I / + 1 ) ( V 2 .. one obtains M£!(O.19. .
(19. + ( i + s w (E 4 . 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. 9 / « i(2yr/3)2 ss i(sin 2vr/3)2 ~ 2 sin2 TT/3. (19.458 19. Then cos 2TT/? = 1 . This is the case of real . and sin 2TT/3 = ft/.7 CHAPTER 19.101) is > zero. Eq.eiun/R2)(eiu*n ~ (19.99) 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. (19.28) that v is real.104) We now consider the case of a — 0 implying that g of Eq.100) Here we set 2^/3 = 1 + j / ? e 2™ = l + gt (19.102) (19.101) where / is complex and g is real. 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.R*2 ^ * #2 \ 11 tf2 R* (19.98) (cf. (19.^ i 4 2+^ # 1 / •. (e iun e~iv*^/R*2)' (19. _ i?* 2 2+ g # 2 i?* 2 2 3/(i + s)(4 + i?*2).2.84): Si ST 1 (11/R2)(11/R*2) .28)). where n is an integer and a and (3 are real and of 0(h4) We have with Eq. (19. In this case R — R* and so 1/i?2 ~ Ofji4). Thus we set v = n + i(ot + i(3) = (n — 0) + ia.103) Then SiSt (1+5) i .9 / .
(19.1)\} 2 h \2 iv 1 + Au {y .Q. 1 29/ ^ (1 .577. MullerKirsten.61n2 .3. (19. S. W. Liang and Yunbo Zhang [189] (there Eq. Gubser and A. J. Manvelyan.* )  14(^V = 1 T h e absorptivity A is therefore given by A = 1 . p.107a) In this result* we now have to insert the expansion of the parameter v given by Eq. 2 /l4 '•i Al=0 2[(l + 1 / 2 ) 2 . f W . Hashimoto [124].2 The Potential 1/r4 — Case of Small h2 parameters v = n — (3. At = 2i[l + 0(h2)}. Oberhettinger [181]. We can also evaluate now the amplitudes (19.3b)) h2 is real for E = k2 > 0.6C .86) and obtain in leading order for I — 0: Ai = *[l + 0(h% Ar = ±[l + 0(h2)].SiSf » 4 sin7rz/ R (19.l ) 2 2 \) +0{h«) (19. one can calculate the next term of the expansion so t h a t At=Q = 4 ^ 4h4 {7 . Here in t h e case of the attractive potential (as remarked after Eq. "This is the result effectively contained in R.107b) = Ah2[l + 0(h4)}. H.97) this can be written A 4TT2 [v\{v .107c) Here h2 is actually Vh4. i.102) we obtain Sis. 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. Using the derivative of t h e gamma function expressed as t h e psi function ip(z) = T'(z)/T(z).1 ] Oih* For S waves this implies an absorptivity given by ((—1/2)! = 1/7F) (19. = 1 . (19. With t h e help of Eq.61n/i} + ~9~ 0(h8) (19.^ ) ~ 2 2 459 1+ 9 / • & ( ! . Magnus and F.106) and evidently violates the unitarity of S.28). J. . Hence with Eq.e. (19. ( I l l ) ) in agreement with a result in S.19. and t h e values of ip(\/2) and ip{2>/2) as given in Tables^ with Euler's constant C ~ 0.105) sin7r/3 (19.
19. N. We are again concerned with the associated Mathieu equation given by Eq.3. W.109) ^We follow here mainly D. H.1. J. Aly.3 19.3.1 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. we again make use of the replacement (17. VahediFaridi [8] and G. h) for large h and upwards I = 0. completely different solutions. (19. 2 2 a = [l + 2 (19.460 CHAPTER 19. H. MiillerKirsten and Jianzu Zhang [226].4) which — for some distinction from the smallh2 treatment — we rewrite here as 2 + [2h cosh 2z .4 Fig. i. J.$ q(l. K. where the part of interest here is based on H.3 The function q(l. H. S. Wannier [278].h) (19. Tamaryan. Miiller and N.e. W.26) and hence set a1 = 2hl + 2hq + A{q.h) 10. Park.108) For the large values of h2 that we wish to consider here. 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. 2.a }tp = 0. .H. Singular Potentials 19. This is therefore a highly interesting case which found its complete solution only recently.
112) In Chapter 17 the unnormalized even and odd large/i2 solutions of the Mathieu equation were given by Eq. and a is its normalization constant. (19.113) 'We are considering nonintegral values of v.21b). i.109) in this way enables us to obtain asymptotic expansions of solutions very analogous to those of the periodic case. we defined the Floquet exponent v and observed with Eq.110) a where y+(7r) is the solution evaluated at z = TV.5) that this is given by the relation cos iris = ^ M .3 The Potential 1/r4 — Case of Large h2 461 where q is a parameter to be determined as the solution of this equation. to which we have to return in this subsection.y} where VF[y+.111) we have W[y+. however.y^] = ab\f\ is the Wronskian (in leading order) of the solutions which are even and odd about z = 0 respectively. With the boundary conditions y+(0) = l.h*y_(l. y± oc A{z)e2h sm * ± A(z)e2n sm z . y(0) = 0. We begin.3. (17. J4(0)=0. i. and A / 8 is the remainder of the laigeh2 expansion as determined perturbatively. with an essential mathematical step. Therefore the second term on the right is nonzero. 19.§ a W[y+. See also Eq.yJ\ — 1 and cos TTV . y'_(0) = 1.3. The replacement (19. We also found the boundary condition (17.h*y (19. (17. (19.l + 2y+(l. The behaviour of q as a function of h for some values of / is shown in Fig. which therefore cancels out or can be taken to be 1. which is even about z = 0. . (17. (19.21b). 19. since it is clear that large h2 considerations require a knowledge of the large/i2 behaviour of the Floquet exponent v.19.e.63).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.e.
39 2 14 /i 2 + (19. 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 . 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 .114) at z = 7r/2. . W.116) (in the first expression we have A(z) ~ Aq(z) with Aq(z) given by Eq. (17. Dingle and H.A(z)e2h sin 2 sin z sin z (19. 1 1 Or see R. B. (19. J. i. and hence with Eqs.112) from evaluation of Eqs.e. Singular Potentials and their extension to around z — TT/2 by the relations C[w{z)\ ^_2h sir Mz)}„2h. J.32). (19.111) we have^ iVo = 2 3 /2 1 + °{k and K = ¥frh l +O (19. See R. Dingle and H.115) Setting z = 0 we obtain in leading order y+(0) = 2iV0A(0) and y'_(Q) = 4hN0A{0). Miiller [73].114) a a Our first step is therefore the determination of the normalization constants No. (19. K [ (z)}2hsinz (19. B. Kin B y± oc y+(z) = N0 A(z)e2h A(z)e2h sin z + A{z)e~2h . (17. in the second expression 2h is obtained in addition from differentiation of the exponential factor). This paper contains the normalization constants of large/i 2 periodic Mathieu functions.43) we see that z = 7r/2 implies W(TT/2) = 0. Miiller [73].117) a a Referring back to Eq. Hence we require (obtained from Eq. (17. We now obtain the expressions needed in Eq. (19.462 CHAPTER 19. which is \/2 at z = 0.118a) "We emphasize again: For integral values of v the normalization constants are different.
1Q11QM cos irv + 1 ~ •==— e .h .\ { q + 3)]! 7re 4/i 24/i + (2/i 1 /2) ( 8 / l ) 9 / 2 [ _ l ( g + !)]. Thus in particular (8h) i(?i) a (3q22q + S) 27h + l\(qW " ' Inserting these various expressions now into Eq. MiillerKirsten and Zhang [226] — is cited without proof by Meixner and Schafke [193].3).119b) COS TTV + l ire This result gives the leading contribution of the lavgeh2 expansion determining the Floquet exponent v.117) are known from the matching of solutions in Eqs.! ) ] ! [ . .62a) and (17.118b) The factors a.112) we obtain COS TTU + 1 7T '2Wa[\(ql)]l[\(q+l)]\ 02h 1 ¥h + 2TT ' 2^ha [\(q . Presumably they extracted it from "between the lines" of a sophisticated paper of Langer's [159] which they cite together with others.62b) (see also the explicit calculation in Example 17.19.3 The Potential 1/r4 — Case of Large h2 and 463 as dz )z=n/2 dw (dw\ o\dzJz=n/2 1C%2 .8 g + 3) +0 ^2 (8h)^[~l(q + l)]\[l(q + 3)}\ 2eh ' ~ (19. (17.119a) This formula — derived and rediscovered in Park. (19.87 2 14 /i 2 [ i ( g . With the help of the duplication and inversion formulas of factorials the result can be reexpressed as cos( g 7r/2)[£(gl)]! .[_ l ( g 2h and hence 5 + 3)]! 'All Ah 1+2 (3g2 . 210. p. a in Eq. (19.! ) ] ! [ .2g + 3) 27/i ( 3 g 2 . (19. Tamaryan. (19.i ( g +3)]! 1 24/i {2hl>2).
h. h. —z). h.z) = A(q. z) are TP(q.121) We let Aq(z) be the solution of this equation when the right hand side is replaced by zero (i.h. z). i = 1.z) or as tp(q.120) The resulting equation for the function A(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.e.z). —h.z) := = exp[i7rg/4] j===^Aq{z) exp[2/wsinh z i(77r/4 V2ih k(q. Singular Potentials C o n s t r u c t i o n of largefa 2 solutions 19.123) We again make the important observation by looking at Eqs. we can obtain the linearly independent one either as ip(—q. h.109) remaining unchanged. (19.h) =?==.. Then straightforward integration yields > inq 4 „. h.h.4. for h — oo).s i n n z ± iq)A = ± K qj dz 2 Aih d2A dz2 (19. tp(q. .h.3. the expression (19. h(q. 1 / l + isinhz\T9/45R^oo e^ / Ag(z) = ===[r^— ~ .122) that given one solution ip(q. z) can then be written as . z) = ip(—q.3.h.h). of course.122) Correspondingly the various solutions ip(q. „ 1 A „ —A cosh z——H .124) Since this function differs from a solution ip by a factor k(q.464 CHAPTER 19. (19. .h. V— 2in (19.z)exp[±2hismh z\.109) into the equation (19. —z). With the solutions as they stand. 17.? s m h z) Vcosh z (19.h)iP(q. We define these solutions in terms of the function Ke(q.3.120) to (19.108) and set ip{q. h. Vcosh z \ 1 . In the following we require solutions H e ^ ( z ) .2 we insert the expression (19. it is still a solution but not with the symmetry property ip(q.z). . h.2.3 Following the procedure of Sec. h. dA 1. z) = ip(q. each with a specific asymptotic behaviour. —h.z) i/>(q.
as we saw earlier. One can show that the quantity < o of Wannier* (see above) is related to q by $o = & iqir/2 + OO. we h a v e ( u s i n g cosh 2rj = 2 c o s h 2 77 — 1) ry Jo ' = dr. we have to match a solution valid at Kz = — oo to a combination of solutions valid at $tz = oo. 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 .127) sin z T /'1 M TT\ — <>{ 4 . (19. * ** = > rojKe( ". i. : 7r/2. i. We observed that these satisfy the relation (19.126) For large values of the argument 2/icosh 2. in performing this cycle of replacements the function picks up a factor.h2) = OvMPfrh2). It follows t h a t S e t t i n g s i n h 77 = t a n 0 .^ s i n h 7?)]g.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 . (z. (44) of W a n n i e r .45). [ \ .fi2) can be reexpressed in terms of Hankel functions. the Bessel functions contained in the expansion of the associated Mathieu function M{. 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]. . 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 . av{h2) = ^(0J ^ Mi i . or — equivalently — by the appropriate expansion of the Bessel functions as given in Tables of Functions. / 2i<jfc + A / 8 \ ~ 2fc / drj cosh 77 1 + * _' Jo V 8fc2 c o s h 2 77 7 2fcsinh y + [ t a n . Me„(z. This is.19. i. Ju(z) TTZ VK 7T cos z T~ 4 1 + 0 ^V. (19. >• * F $ 0 * e".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]./i 2 ) M (19. . (0. 2j<.e. we h a v e 77 = 0—> 6 = 0.3 The Potential 1/r4 — Case of Large h2 465 Instead.h2).<5±1) (19 125)  (the expression on the far right in leading order for h2 large). In order to be able to obtain the Smatrix.e. achieved with the help of the Floquet solutions Me±v(z.fc 2lqk A + 1/2 + 4fc2 c o s h 2 77 A / 8 P ./h). w h e r e Iv := I drj\f a + 2k2 cosh 2*7" J0 I n s e r t i n g t h e e x p r e s s i o n for a.e.
He (z. 3te»0. exp[—ikr — in/4\ kr Re{2)(z.q.g.q.128) We now define the following set of solutions of the associated Mathieu equation in terms of the function Ke(q.vr .h) He (z.eW(z.g.130a).q.129) The solutions so defined have the following asymptotic behaviour (where e(z) = (2/icosh z)~ll2 and h2 = ikg): H. (19.h.h )~ \ ot±v cos(2/icosh z = vir/2 — 7r/4) F V2/tcosh z (19.z)./i). 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.h).3.h) (3) = = = = Ke(q.130b) r_0 rll2exp\g/r (ig) ' 19.h) Be^(z.h) = e(z)exp exp ihe^ 4 + i\ 12 i4 3^cO.466 CHAPTER 19.q.128) and consider the cosine there as composed of two exponentials whose asymptotic behaviour we identify with that of solutions of Eqs.h) = e(z) exp ihez . HeW(z. 3te»0. (19.h) = e{z) exp ihez + iexp[ikr + i7r/4] kr 7T .q.q./i) Be^(z. (2) (19. z): Re(2\z.h). (19.4 T h e c o n n e c t i o n formulas We now return to Eq. (19.q.130a) Re^(z. In this way we obtain in the domain . h.q. (19. KeW(z.130b).
(19.h) is asymptotically small (cf. — sin 777 H e ( 3 ) ( 2 .* ^ / 2 H e ( 3 ) ( Z ) g? Mev(z. (19. h. z) are proportional there.132) into Eqs.19.134) Considering now the first of relations (19. q. Eqs. (19. and He(*> exponentially decreasing. t With Eq./i 2 ) = Me_j.130a). (19.g.g.h2) a./i) — sin 7T7 Re^(z. g . q. (19.(z. we obtain in the domain tfcz <C 0 the relations: Me. h) — sin 7T7 He^(z.li 2 ) Oil.q. h) = sin 7r(7 + v) He (1) (z. (19.h2) = 2vr i a_ 2TT />) _ e ^ / 2 H e ( 4 ) ( Z j ^ Q (19.h2) Me„(z.g.125) the proportionality factor can be seen to imply the relation exp •ffa + 1) sin7r(7 + "} (19. these are exponentially increasing there.(z. h.129)) He (2) (z. and thus can be matched in this region./i). / i ) + sin7r(7^)He(4)(2.132) These relations are now valid over the entire range of z.135) sin irv 'This means.133) In a similar way one obtains the connection formulas — sin nv HeW(z. q.h) sin irv = sin7r(7 + ^)He (3) (2. e W 2 H e ( 3 ) ( ^ ? j ^ _ e M/*/2jj e (4) ( ^ q j ^ e .« * r / 2 H e ( 2 ) ( ^ g> ^ ft) _ e W 2 H e ( 2 ) ^ ^ ^ = a_ 2TT e ./i).134) in the region where the function He^(z.130b)).i ^ / 2 H e ( l ) (z> ?) (19.q. e x ^ r / 2 H e ( l ) ( 2 > g> />) _ e . h) = Ke(q.131) where the second relation was obtained by changing the sign of v in the first.3 The Potential 1/r 4 — Case of Large h2 fc>0 the relations: = 2TT 467 Msv{z. . h) = Ke(q. (19. (19. z) and He (3) (z.h). we see that the solutions (compare with Eqs. q. Substituting the Eqs. Changing the sign of z.q.131) and eliminating He") and setting again exp[m7] := we obtain the first connection formula ozwr — sin irv He (4) (z.
119b) as „4/i 1 + COS TXV ~ rrl ( g .119a) or (19. Wannier [278].e. is only the dominant term for large h2} 19. (19. i.(19. (19.135).119b).1 .m a t r i x From Eq.l ) ] ! c o s ( W 2 ) 2 2 (8h)i/ 2TT29/ 2 +0 (19. Singular Potentials The factor on the left.133)) — sin TXV He^4' (z.135) is the equation corresponding to Eq.136) ^ jrfl+l/2) (19:I35) sin 7r(7 + v) sm WKljrtla/2) S m TXV (19.( .135). The latter is defined by the following larger behaviour of the solution chosen at r = 0. H.3. (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".Sin TXV.5 D e r i v a t i o n of t h e ^ .468 CHAPTER 19.e'i7ri 1.137) We see that this expression agrees with that of Eq. cos TXV + sin TXV.119a) and (19. ' cos TXV ± v 1 + elin sin2 TXV =p v'cos 2 TXV . h) ~ — sin TXV (1) sin 7r(7 + v)e kr ikr in 4 rl/2eg/r+m/A / (ig)1/2 ( . where 8i is the phase shift. (19. Thus here the S'matrix is defined by (using Eq.138) TXVICOS TXV We obtain the behaviour of the Floquet exponent for large values of h2 from Eqs.l ) ' s i n 7 T 7 ^VK /2 ikr sin 7r(7 + v) _l)leikr i5i„—ilir/2 SxelKr . . of course. Squaring Eq. (19.133) we can now deduce the 5matrix Si — e2t51. which in our case is the solution He^4^. Wannier did not have the relation (19. exp 2^(9 + 1) and solving for sin TXJ one obtains the expression sm 7T7 sm TXV = ~iei7rq/2sin le iqK • COS 7T7. sin TXJ.139) ''Equation (19.85b) obtained earlier. (19. q. The quantity 7 is now to be determined from Eq. and therefore remarks after his Eq.l ) (l„—ikr e From this we can deduce that Si (19. (60) in the work of G. (19.
9/2 (91) with the approximation q ~ h obtainable from 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 (19.141) COS TTU irvj — i s i n WR s i n h TTUJ. irrespective of what the value of I is) by cos™ + l = c o s ( T j^J y/e1. Using Stirling's formula in the form z\ ~ e~zzz+1/2V2^.8h COS From Eq. i.h):=l\Stf ^cos rr(uji + ivi) = cos Tn/jicosh COS ^7TQ N 2 1 (19.137) together with (19. h) of the Zth partial wave. as a comparison with Eq.3 The Potential 1/r 4 — Case of Large h2 469 Since the right hand side is real for real h. i. since cos TVU is large.28) shows. (19. A{l.e.e.139) could actually be neglected) implying  cos m>\ = cosh nvi. (19.140) From this we obtain the absorptivity A(l.138) we obtain.109). (19. This is different from the behaviour for small values of \h2\.19. T h i s is r e a l for UR a n integer.e. the real part of v must be an integer. (19. . imaginary part uj (so that 1 on the left hand side of Eq. 19.4 The absorptivity of the S wave (attractive case). we can approximate the equation for q ~ h (i. for the attractive potential.§ 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.
Fubini and G. Pao and J. The diminishing fluctuations of A(l.' Rudimentary aspects of a singular potential with a general power n have also been considered previously.142) we can see the behaviour of the S'wave absorptivity as a function of h. 19. . in fact. Treiman [271]. S . Furlan [20]. H. **R. H. Challifour and R.^ A highly mathematical study of the potential ~ r~2 as an emitting or absorbing centre has been given recently. YuanBen [285]. Shabad [248]. Jones [78]. W Some of these sources can be helpful in further investigations.139) 2Tr(16h)q /32\ f e Thus with Eqs. N. (19. Cvetic. "G. Tiktopoulos and S. 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. (19.** 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. **A.4 (there are tiny fluctuations in the rapid approach to 1). Pope and T. Lii and J. N. Apart from the papers already referred to. This is sketched schematically in Fig. and. 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. Paliov and S. Dombey and R. Partial or other aspects of the weak coupling (i.470 CHAPTER 19.4 . Masson [191]. Eden [46]. Lii.P. VazquezPoritz [61].k) in the approach to unity are too small to become evident here. H. Apparently this is one of the very rare cases which permits such complete treatment." as well as other aspects. I r a n [61] and M. Limic [177]. B. D. Bertocchi. Y. Esposito [88]. small h2) case have been considered by some other authors. 11 J. Rosendorf [225]. C. A. N. A.107b) and (19. S. in further contexts beyond those already mentioned. such as phase shifts. Handelsman. Singular Potentials with near asymptotic behaviour (for h2 — oo) > Afi .4 Concluding Remarks In the above we have considered the solution of the Schrodinger equation for the potential r . 19. Cvetic. one may expect corresponding results for other singular power potentials. D. J. E. L. Lew [128].e. A. In essence. and G. see also M. F.
We concentrate in this chapter on the large order behaviour of asymptotic *C.g. M.Chapter 20 Large Order Behaviour of Perturbation Expansions 20. 471 .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. W. the subject is much older and had been explored earlier in great detail in particular by Dingle^ with the subsequent investigation of the behaviour of the late terms of asymptotic expansions of the eigenvalues of the Mathieu equation and others. B. B. and how this is related to the level splitting — in fact we shall see this in both cases of the cosine potential and that of the anharmonic oscillator. Dingle and H. J. *R.* In fact. e.* However. also convergent expansions are known and a lot of literature exists on the equation. the cosine potential is more suitable for such studies than the anharmonic oscillator since its case is simpler. T. Muller [74]. [19]. We shall also see explicitly. f R . the asymptotic solutions in different domains all have the same coefficients. Dingle [71]. This is indeed a remarkable connection. how the large order behaviour of the expansion of the eigenvalue is related to the discontinuity across the latter's cut. Wu [18]. Bender and T. 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.
1 and cases in Sec. It is a feature of a physicist's approach to a problem that he employs a method of approximation which is such that the first term of the corresponding expansion yields a rough value of the desired answer.6.2 The function ur = r\/pr vs.2. 19. one could naturally also explore the large order behaviour of these in a similar way and expect the behaviour discussed in Sec.2 and below.472 CHAPTER 20.6. r for p = 4. Large Order Behaviour of Perturbation Expansions expansions using in particular the recurrence relation of the perturbation coefficients obtainable with the perturbation method of Sec.1 The function ur = pr jr\ vs. Since the method is also applicable to convergent expansions. 8. 20. 20. Presumably this has never been done so far.2.8.7. 5. r for p = 4. Fig. This approach leads . 10 12 14 16 18 20 Fig. as evident from Example 17. 8.
4) Jo Jo its Laplace transform F(z) is called the Borel transform of E. In some cases the answer is affirmative. and the integral E(g2) is called the Borel sum. If the function E(g2) can be written as the Laplace transform E{g2)= dze'z/g F{z)=g2 dte^FigH). 20.1 and 20.1 Introductory Remarks 473 automatically to asymptotic series. i.2) or. The behaviour of such terms is illustrated in Figs.J2 2n h E (20. T V lim h2N 2 n EQ?) . the function E(g2) is given only approximately by Yln=o En92n. For an authoritative discussion of these aspects we refer to Dingle [70]. 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 traditional method of using an asymptotic expansion is to truncate the series at the least term. whereas in the case of the asymptotic expansion it is the opposite: a factorial in r divided by a power. (20.20. . even if each En is known. and the expansion is said to be Borel summable. i. The formal perturbation expansion of a quantity E(g ) . as we discussed earlier in Chapter 8. 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.3) n=0 Thus for any given value of g . E(g2) = f2 Eng2n = Eo + J2 n=0 n=0 E 2(ra+l) ^9 (20. equivalently for g = l/h. that the maximum of the absolute value of terms in a convergent expansion and the minimum of the absolute value of terms in an asymptotic expansion are reached approximately when the value of the variable is approximately equal to the number of the term. The question arises: What is the information about F(z) that can be obtained from the asymptotic expansion of E.1) is an asymptotic series if for any N lim 92^o 1 g 2N N E(g )^2En92n 71=0 2 = 0 (20. for instance.e.e. It can be seen from these. this term represents the size of the error. in the power expansion of the exponential function).2.
n _ En+\ n! En+1 ^ (!)"«!. We can therefore write down a dispersion relation representation choosing the cut from 0 to +oo: .5) then with the integral representation of the gamma function F(z) = (z — 1)!. Thus if CO F(z) = aiS(z) + J2 anzn. . . many asymptotic series do not even alternate in sign (as in the case of the series in Eq. . 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.6)).474 CHAPTER 20. Large Order Behaviour of Perturbation Expansions Eq. Q. 1 we obtain oo F(z)£(*)" = — . (and E0 = 0). J we obtain Now if />oo dteHz\ °° oo ' 0 dteH*\ TEn+1g^V . In general this is not so easy. (20. El . . 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. We observe that a function E(h) with asymptotic series ^ r=\ A has an essential singularity at h — 0. (20. (20.6) The power series of F(z) converges inside the unit circle about the origin at z = 0. that of F(z)) over a domain that is larger than its region of convergence (see Chapter 8 and Dingle [70]). In any case. In this ideal case the integral of Eq. . . n=0 (20. In fact.e. n=0 (20.4).4) can be evaluated and E(g2) thus recovered from the asymptotic expansion by Borel summation.1)? Asymptotic expansions originate from integrating a series (i. (20.n E(9 )~ 2 n=0 ^ E0 +1) = Ea^2( °° oo /»oo = J2ang2^ „n 'O n=Q dte~H\ a_i = E0.
e.g. S. (20. The potential (20.11) V.. 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.1 Introductory Remarks 475 except for possible subtractions.. i. Popov and V. e.2.10) r as shown in Fig. (20.3 and g < 0. . we have a quantum mechanical problem where probability can leak away.3 The potential V(r) gr». M../ \p\dr\.N = l. and E represents an eigenenergy.. Weinberg [228]. or can we determine Ar independently? Fig. § where 7(ff) = To exp  . (20.10) is considered in detail in this work. N = l. This is the case if the real potential has a hump of some kind... if the potential or a boundary condition is complex such that the problem is nonselfadjoint.20. if the potential is V(r) = +grN.9) The question is therefore: How can we determine 9 E(h). $s E can be calculated by the WKB procedure.2. 20... ? When ?s E ^ 0. 20. for < ^ 0. and with g = 1/h and it is found that§ En = 4 n ) ~ \lis)..
476 CHAPTER 20. (14.11) is. since this is effectively the Mathieu equation for which — for comparison purposes — extensive literat u r e was already available.73) in the W K B or semiclassical approximation. 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. V. the BreitWigner formula (10. .134)). of course. M. Equation (20. In the first such investigation it was sensible to consider the Schrodinger equation with the cosine potential. (20. Example 20.11). as we expect on the basis of Eq.2 Cosine Potential: Large Order Behaviour It is instructive to investigate various methods for obtaining the large order behaviour of an asymptotic expansion. An interesting application is provided by Example 20. 20.113)).9) for Ar. Popov and Weinberg. Straightforward RayleighSchrodinger perturbation theory usually does not enable this in view of the unwieldy form of coefficients after very few iterations. the eigenvalue expansions of the cosine potential and of anharmonic oscillators. as was (and still is) not the case for t h e equation 1 V . Large Order Behaviour of Perturbation Expansions where the frequency of particle bounces against the potential barrier is 70 ~ ^classical/27r and w c i assica i = 2ir/Tkepier oc 1/n 3 .^ The logarithmic factor in Er is a novel feature of this case.76) (see also the comment on the comparison with the Kepler problem before Eq. (11. S.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. the perturbation method of Sec. Popov and V. L.1. Thus in the following we apply these methods to our typical examples.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. Eletsky. (11. the cosine potential being t h e most completely investigated will also be seen in the following. Evaluating the integral show that 00 r=0 v ' where n is the principal quantum number of the Coulomb problem as in Eq. (20. Solution: The solution can be found in a paper of Eletsky. 8. Weinberg [83]. However. Here n is the principal quantum number of the Coulomb problem (see Eq.
60). (17. % = Aq. (17. Dingle and H. (17. or equivalently of the quantity A.l ) / 2 ] ! i![(gl)/2]! • We have seen earlier (cf. (20.13) represents effectively a partial difference equation.12) (20.M2i+1 for large i.2 Cosine Potential: Large Order Behaviour 477 with an anharmonic oscillator potential.Bq.13) In the special case j = i the boundary conditions stated after Eq.26) and (17.j) a n d M2i can be deduced.Cq. Miiller [74]. J. obey the recurrence relation jPi(q.j) of functions ipq+4j. (cf.j + 1). (17. (17.j) are known.58) and (17.l ) +2[(q + Aj)2 + l + A]Pl. i. .l ) ( g + 4 j . This has been done" and it was found that (we cite the results here since we rederive them below in detail by other methods.58) eliminate the last two contributions and one obtains the exact expression ^ ' Z j _ 4<[2» + ( g .20. M2i = 2j2 J[Pi(q.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. Eqs. With some simplifications for large values of i this equation can be solved approximately and the large order behaviour of the coefficients Pi(q. 3)? .j) +(q + 4j + l)(q + 4j + 3)i^_ x (g. are given by i .34)) that the coefficients Pi(q.2 A = Mx + ~M2 + 7^WM3 + •• • .e. W. B. Eqs.3 ) P i _ 1 ( g .J) = (g + 4 j . arising in the ith order of the perturbation expansion of the wave function oo 1 i ip — const.h) = 2h2 + 2hq+j. (20.14) Thus these coefficients can be obtained once the coefficients Pi(q. Eqs. The recurrence relation (20. 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. j .55)) X = E(q.l(q. h).
16b) With the help of Stirling's formula one can derive the following relation: i large 1 (i + m)\ \o (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. Large Order Behaviour of Perturbation Expansions where the bracketed expression at the end is the binomial coefficient. see also below) . Using this relation and the duplication formula lw('\W±^.91 l)]![i(g3)]!} 2 { [ . and from this for i even or odd it is found that Mi ~ (16)^! 1 2g2 + 3 1i .th term iW~^q 1 {[i(gl)]!} 2 2 7 r(8/ l ) i .1 22 the result becomes i t h term (i + 2n)!2 2 " 1 where n 1 (20.16a) (20.17) We see that assuming i is very large compared with m is here equivalent to assuming m is approximately zero.478 CHAPTER 20.l ( 9 + l)]![(Q + 3)]!}2 i!^" 1 1 or (with the help of the duplication formula.l (20.19) ^nOW =2(9"1) We observe that successive terms do not alternate in sign. In the following we concentrate on the cosine potential and rederive the large order behaviour obtained above by other methods. also written lCj. In the first method we do this by first obtaining the imaginary part of the eigenvalue with the help of nonselfadjoint boundary conditions.18.*. (20.•91 {[l(? + i)]![(? + 3)]!}2 (20. Hence exact Borel summation is not possible. .
. when this is g0 = 2n + l . 20.3 Cosine Potential: Complex Eigenvalues 479 20. . we follow here a formulation in the context of our earlier considerations of the Mathieu equation. this difference was found to be real.2 solutions obtained earlier. Stone and J. 2 . but now change the boundary condition at the neighbouring minimum in such a way that the difference q — qo becomes imaginary.20. . \ 1 \ \ \ \ \ 0. J.4.1 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/?. we emphasize a few points *M. Reeve [262]. We retain the previous boundary condition at one minimum. l . the difference q — q$. in the calculation of the level splitting as a result of tunneling. Previously.e. but with different boundary conditions. MiillerKirsten [200]. 20. . W.5 \ \ / \A Fig. f H .5 A i 1 1 r f \ \ M\ \ \ / ! \ \ 1 \ \ I 3 \ \\ \ /41 i \ \ 0. What changes as a result of different boundary conditions is the deviation of q from the odd integer qo.* However.3 20. At the risk of repeating what we explained earlier. n = 0 . remains unchanged. 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.4 The cosine potential shifted by TT/2. A means to achieve this was found by Stone and Reeve. . The potential we consider here then has the shape shown in Fig. i. This is a very instructive exercise which shows how the eigenvalues change with a change in boundary conditions. t To enable comparison with the work of Stone and Reeve we shift the cosine potential considered previously by 7r/2. From our previous considerations it is clear that the formal perturbation expansion of the eigenvalue X(q) in terms of the parameter g.3.
Thus one chooses not the solution with real argument at the minimum. Setting w = 2h1/2z.21) <u> + ( i « . we shift the argument by n/2. but achieves exactly what we want. Previously this situation was the same at the neighbouring minimum. Effectively we consider the tunneling from the left well in Fig. a boundary condition which at the face looks somewhat abstract.e. (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. but t h a t with complex argument there. Thus around z = 0 the potential behaves like t h a t of an oscillator well with minimum there. 20. and thus obtain ip"{z) + [A + 2h2 cos 1z\i>{z) = 0 (20.4 (around z — 0) to the one on the right (around z ~ ir).4. Large Order Behaviour of Perturbation Expansions for reasons of clarity (since some considerations here may otherwise not be sufficiently clear). and correspondingly we assume alternately even and odd oscillator eigenfunctions there.T K ^ a D_i_{q+1){±iw). as we found these for this potential. To be more precise we recall the Mathieu equation we had earlier: ip"{z) + [A . The physical wave functions there are therefore the solutions which are real and exponentially decreasing. Here we choose t h e second way.Ah2z2 • • • }^{z) = 0. and allows its imaginary part to tend to infinity. Around z ~ 0 w e can expand and obtain i)"{z) + [A + 2h2 . X + 2h2 = 2hq+—.20a) As stated.20b) with potential V(z) = — 2h2 cos 2z as shown in Fig. 8 the equation can be approximated by d2ip(w) (20. i. Now we change this situation at the minimum on the right. the degeneracy of eigenvalues of neighbouring oscillators being lifted by the finite height of the barrier separating them. 20. .480 CHAPTER 20.4. We can distort the potential there or alternatively impose a different behaviour there on the wave function. (20. which means replacing z by z±ir/2. 20.2h2 cos 2z]ip{z) = 0.
. Thus. 20. .20. we naturally constructed the even and odd solutions with these (cf. A (which we can loosely dub solutions of WKBtype).e.24) Considering t m O i n Eq. (20. we see that ips ~ cos \j2qhz for z ~ 0.23) . considering around z = 0 the ground state of the harmonic oscillator for which go = 15 and replacing go by g. whereas here we construct these from oscillatortype solutions. This somewhat abstract looking boundary condition is effectively simply one imposed on a complex solution of the equation giving the oscillator eigenfunctions. Eq.22).4. One should note the difference to our earlier case of the calculation of the level splitting: There we were seeking solutions which are even or odd about the central maximum of the potential. 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. Thus we write the boundary condition at Kz = ir: tp(±iw) — 0 for > w — 2h}/2z K z— > i. 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. (20. 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.63). 2. 3 . A derivation is given subsequently.> 0 for z > ±ioo. such that this deviation becomes imaginary. however. . that at z = 7r.3 Cosine Potential: Complex Eigenvalues 481 As explained above. We consider here explicitly only the case of the ground state around z = 0. i. we now choose an harmonicoscillator type of real solution around the left minimum of Fig. (where for the ground state q = go — 1) D_i{q+1)(±i2h1/2z) ' ± zoo (20. . . 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. by a different method. whereas here we start off with a solution even or odd about a minimum. 17.e. the even or symmetric solution about z = 0 can be taken as *• = liDfrriVhWz) +Dh{q_qo)(2h^z)}.
w) = e T Dv{w) 2TT n») AvViDv^iw).h(2h1/2 J £2=0 (20. (20. 2. We see this as follows. (20. Large Order Behaviour of Perturbation Expansions Although the two parabolic cylinder functions in Eq. Oberhettinger [181]. Magnus and F. 92. §See e. and then evaluation at u = 0.3. Thus the only nonvanishing contribution comes from differentiation of Y[—v). or logarithmic derivative of T(^) for which:§ ^(1/) r» (20.2 that one parabolic cylinder function is exponentially increasing in h and the other exponentially decreasing.27) differ by an angle IT. p.g. (20. The arguments of the functions appearing in Eq.27) The differentiation of the parabolic cylinder functions with respect to their indices is nontrivial.we obtain the following circuit formula of parabolic cylinder functions Dv(w): Du{w) T DV{.27) requires the differentiation of this expression with respect to the index v.(.24) are linearly independent around z = 0 for q ^ qo. The differentiated function Y{y) is in Tables of Functions expressed in terms of the socalled psi function. tp(v).482 CHAPTER 20. p. this can be handled in a few lines.25) qo) E2 0: (20.30) 'See e.29) Equation (20. Oberhettinger [181]. W. However.•w) + 2ir T{u) <"+!)% Dvi(iw).28) Extracting Du{—w) we have A. (20. From Tables of Special Functions'. 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. 17. Setting v = 0 in T(—v) = (—v— 1)! gives infinity. we know from the solutions of types B and C in Sec. W. Magnus and F.g. As before we set A = 2hz + 2hqA where in the present case Ei = 2hz + 2hq0 + — and A & = Ex + E2 (20.26) Thus here we can approximate ips by expanding it about E2 i)s~DQ{2hll2z) + E2 ~{DE2/4h(2h ^ z) 1 2 + DE2/. .
3 . 5 7 7 .e.20. we obtain the formulas (X) / \ 1 OO I » J ] (l+)e. with the following expression for the largez behaviour'! (here v = E2/AK) Dv{w) ~ wue~w2/4 = evlriWe~w / 4 for  arg w\ < ^TT. "See e. Oberhettinger [181]. For u — 0 one obtains the result (20. Magnus and F. (20.29) v=0 (in + In w)D0(w) + v2ire (m + lnu. Reeve [262]. . .e. pp.)e" w /4 2 l ^Dx(iw) (20.4.+ v ^ v k^x k ^ + ^ where C = Euler's constant = 0 . 92.33) = IV2~TT.//n and ij>{y) = C .27) exponentially decreasing contributions.32) which we can use here since all other problems have been resolved: d D {w) dv v (20. Thus calculations for qo ^ 1 are not given by M. This result is » not easily generalized to <jo > 1. Stone and J.34) This is the behaviour of the symmetric ground state solution away from the origin towards the limit of its domain of validity. Differentiation of this equation and subsequently setting v — 0 yields. which are valid around z ~ 7r/2. including the derivative of DE2ufl(2h1/2z)). r(i/) We now return to Eq.g. (20. . i. and "Prom W.+w2/i IW Inserting this result into ips and retaining only the dominant contributions (i. p.29).31). dropping from the derivative part of Eq. Oberhettinger [181]. i/^0 483 T(v) It follows that d dv [T(u)\ (20. we obtain 4>s D0(2h^z) D0(2h1/2. those of type A. (20. W. Magnus and F.31) dT(u) r*(v) dv 1 i/>(v) ^ o l.3 Cosine Potential: Complex Eigenvalues From formulas given in the literature we deduce the limiting behaviour^ 1. .. We now go to the WKBtype solutions. 1E2 + ± 0(e~hz2) 2 Ah Ihir ehz* E2V'< 16/i2 V2^ehz2~ 2hM2z (20.
by there replacing z by z + ir/2. i.484 CHAPTER 20.e. that in approaching z = ir. (20. Thus in the present case and with q ~ qo = 1 we obtain „2ft. y COS c\ J (20. To this end we set z' = z — ix. (20.e. cos z/2 ~ (20.40).36) V'WKB cie~2hehz' 9h  .41) 2TT .29) and (17.38) in its limiting domain towards 2 = 0. where we can then match these to the solution ips. 92) we take the formula: Dv{w) = T { y ^ \eiv^l2D^v^{iw) + e'^D^iw)] . Eqs. c2 the WKBlike linear combination e2hcos ^ W K B = ClipA + C2lpA ~ Ci z _—2/icos z — + C2 rT (20.23) demands (with constant A) ip = AD_1(±i2h1/2z') > 0 for z' + ±ioo. (20. 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. We observe that in the matching domain (i. We now return to the WKBlike solution (20. Recalling that we have shifted the potential by 7r/2.sm I We now want to match this solution to that of Eq. (20.35) We construct with constants c\.34) and (20. i.37) Comparing Eqs. x .. we obtain the required solutions from the earlier ones.34).39) Now. 2 c2e2hehz UJ2 . with expansion of cos z in the exponential) D0{2hll2z) = e~hz ~ e~2h / „2/i cos z \ — ).36) cos  .36) and consider this at the other end of its domain of validity. near z = n the boundary condition (20. smf (20.cos z — 2/icos z ipA^ Tcos  and i n A —.e. Large Order Behaviour of Perturbation Expansions extend these to their limit of validity to the left. (17. Then V'WKB becomes (with cos z ~ — 1 + z' /2.— 721 . (20.40) From Magnus and Oberhettinger [181] (p.36) we can identify the dominant large/i2 behaviour of the constant C\ as c\ = e~2h.
(20. (20. 1 2 4 1 2 8h ^^2 2 (20.3 Cosine Potential: Complex Eigenvalues Thus for v = 0 we have DQ(w) = —j=[D\(iw) V 27T 485 + Di(iw)] = real.40) can be written ^ ~ AJ^e~hz'2 + A~(i2hl^2zTlehz'2. ^'mliw <2a47) For qo = 1 we recover the result (20.43) we obtain.45) Comparing this result with tps of Eq. (20.^. i.34) we obtain E2 = i%\ = ±ih22Qe~8h. with c\ = exp(—2h) from above. for both even and odd harmonic oscillator wave functions around z — 0 with qo > 1) is nontrivial. .44) = DQ{2h / z) ± (2h7r) / e. (20.39) and (20.38) can be written TAWKB (20. (20. A = ±i4/l1/2e4h ^ t c2 = T i2(2/ivr) 1 /2 e 6/ l . 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 . The general result stated by Stone and Reeve [262] is therefore presumably guessed. J Hence Eq.43) Comparing now Eqs.46). (20.2/.20.46) Repeating these calculations for the general case (i.e.complex conjugate] (2Q42) (2032) y /  Z e _ ^ + 2(. We derive the general result in the next subsection using our own method.e.1/V)_le^_ Z With this result Eq.
At the minimum to the right of the barrier shown in Fig. i.()#0.46) we consider a half period of the cosine potential as illustrated in Fig. Fig. Wiedemann [4]. 20. the result is formula (334) there.E curre it \ j.z.5.) + 7T V> = o. 20. i. We have to impose two sets of boundary conditions. i. Large Order Behaviour of Perturbation Decaying excited states Expansions For our derivation of the generalization** of Eq. (20. W. MiillerKirsten and A.5.486 20. On the other side of the barrier at z = —TT/2 we consider different boundary conditions. Considering the original equation. Achuthan. harmonic osc 71/2 E>V / E<V z+ O \ E>V TC/2 \ \ . •_(:)*<>. (20.e.e.5 The cosine potential from — TT/2 to n/2. we impose the usual type of boundary conditions as for the minimum of a simple harmonic oscillator there. 20. and expanding cos 2z around the point z = — TT/2. V(z) . 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. Eq. J. H.e.2 CHAPTER 20.20a). .*. this equation may be approximated there as d2ip ~dz^ + E + 2h2 . at z = TT/2. (20.48) These conditions provide the required quantum number qo = 2n + 1.3.Ah2 ( z + . •+(=)* V(f) 0.
h. Proceeding as in our previous cases and substituting the expression (20.21) into the Mathieu equation we obtain. (20.3 Cosine Potential: Complex Eigenvalues 487 (using Eq. (A) Boundary conditions at the right minimum We begin with the evaluation of the boundary conditions at z = 7r/2.z)=tpA(q. B and C. (20.51a) where H Bq[w(z)} w(z) = h(«v (w) Bq[w(z)} = Bq[w(z)}.50b) which are valid as asymptotically decreasing expansions in the domain away from a minimum.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.' 4 / 1 1 / 2 COS ( Z + 7T \2 4 . 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.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.e. (20.r/2 ~ e i(2hq) V 2 .20.51b) 2i(«1)[i(9l)]'. i. as shown earlier. three pairs of solutions of types named A.z). Recapitulating these from Chapter 17. For the continuation we use again the WKB formalism. (20. we demand that ^(*)L<.21)).h. for COS I Z ± 7T 1 1 2 4 >o^\ The second pair of solutions is ipB(z) J2hsmz B. we have IPA(Z) „2hsinz Aq{z) + O il>A(q.Hz^+Ol^ Bq[w(z)}+0 h1'2 (20.
Here the upper sign refers to the even states and the lower to the odd states.53) where a (8M («!)]« 1 + 0 { \ a = 1 (8/i) K9+ ) l + O (20. These solutions are valid asymptotically with the first in the domain around z = ir/2 and the second in the domain around z = —ir/2.488 CHAPTER 20. i.54) The even and odd solutions are defined as iP±(z) = tyA(z)±iPA(z)}. ^ W H .qo . (20.e.55) Extending these solutions to the domain around the minimum at z = 7r/2. \vw). Thus we have IPB(Z) = aipA(z) and VcO*) = O ^ A O 2 ) .57) where go = 1) 3.=F2 2\1/2{28h2y/4e~Ah l + O (20. and the second around z = 7r/2. the relation q . (20. we obtain our first transcendental equation. We return to this condition after derivation of the .^ i . 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. (20. Large Order Behaviour of Perturbation Expansions the function H!/{w) being a Hermite function. We saw in Chapter 17 that some of these solutions can be matched in regions of common validity.52b) The first solution is valid asymptotically around z = —vr/2. 5.52a) = Cq[w(z)}.48) and proceeding as in Chapter 17. we obtain 1>±W = \ a a (20.56) Applying the boundary conditions (20.
20. as we shall see with the result of Eq. Inserting the approximate form of the eigenvalue into the Schrodinger equation we obtain dz2 + 2hq+ — . Thereafter we extend the latter across the turning point down to periodic WKB solutions. e x p ( . (B) Boundary condition at the left minimum The solutions of type A have been matched to the oscillatortype solutions at the right minimum. we see that the turning points at z± are given by 2hq . (20. F. (20. Jz+ ' T In principle the integral could be evaluated exactly — see P. Friedman [40]. formula 280. and we obtain the proportionality by appropriate expansion.3 Cosine Potential: Complex Eigenvalues 489 second transcendental equation from the boundary condition at the other minimum.01. These then provide complex exponentials like (20.58) Ignoring the correction A / 8 . . Some nontrivial manipulations are again required to uncover the generation of important factorials.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 . p. r ywKBlZJ ~ ~ — [4/i2cos2 2/i ]i/4 2 9 z 2 2 e x p ( £ [Ah2 cos2 z  2hq]ll2dz) ' 1 2 . 163 — but here we are interested in the WKB solution in a domain where it is proportional to a solution of type A. C O S 1/2 ^" (l) ± ' 2<2+<2 (2 °.49) and we can apply our boundary condition.Ah2 cos2 z ip = 0. 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.[ [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. I^A(Z) overlap parts of the domains of validity of these solutions.61) [Ah2 cos2 z2hq]1/2dz.2hq] / dz) We know that the solutions I/JA{Z). Byrd and M. D.e.67). Our procedure now is to match these to exponential WKB solutions.Ah2 cos2 z ~ 0. i.
) +0 z+ IT Inserting this into the integral I. we write cos2 z = l .63) We handle the logarithmic terms in Eq. 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.sin2 z = l .62) as follows.490 CHAPTER 20.» _ i y v .cos2 ( z+^) = (z+ . 2/i In the same spirit we can set h[z + l 2h 1 — cosi I z + — 7r 2/i[l + sin z\. using the relation 1 + sin z = 2sin2(2i/2 + 7r/4).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. we obtain f ~ I = = 2 1/2 Ah [z + 1 2hq + o\h \z+ 2 dz J Z+ 2h I /MH)' dz 1/2 1/2 2h 2/i /•«"(. (20.62) where we set — in accordance with our expansion *+ + 2 IT 1/2 <? + (20.59) 2h COS Z+  ~ 0. We have for the following expression contained in (20. (20. Large Order Behaviour of Perturbation Expansions Since cos(z + n/2) = — sin z.64) .
3 Cosine Potential: Complex Eigenvalues Inserting the relations (20.65) [Z[4h 2cos2z2hq]1/2d< e2he2hsin ^eg/4(g/2/t)g/4 2<?[tan(§ + f )}i/2 (20.50b). (4^2)i/4 [ i( g _3)] ! V/ l (20. <$&?(*) e2he2hsinzei/4(q/2h)i/4[tan(% + l)]'q/2 2?(4/i 2 ) 1 /4{2sin(f + f)cos(f + f ) } 1 / 2 e2heq/A(q/4y/4 /2\q/4 . .62).50a) and (20. we obtain I ~ 2h + 2hsm z Hence exp 491 V^m 2h\~'I z TT 4 — tan .20./ * [Ah2 cos2 z IZ+ 2hq\l/2dz) + f )]«/ 2 (4/i 2 COS2 z)V4 e2he2hsm zeq/4(q/2h)i/4[tan(% /2^~Q/4 (4/i 2 )V429{2sin(f + f ) c o s ( f + f)}V2 ^27*291 (4/l )V4 e ?/4( g / 4 ) 9 /4 ^ J e" ^^(2TT) /2 2 1 2 ^ ) / 4 /2\"  .60) by (4h2 cos2 z)1!4).+ qj V2 4 A. 1/2 (20.66) The WKB solution therefore becomes (approximating the denominator of (20.67) (27r) /22«+5(4/i2)V4V^y The last expression is again obtained with the help of Stirling's approximation. . (20. but mention that for q an odd integer: j(9D l + O and 1 r(93) 4(9+1) ! = ±7I . We ignore this here. 2*4(4tf)V« W ^ e[l( V D]! 1 /2y^ (20.64) into Eq. . (20.63) and (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.^ Proceeding similarly with the other WKB solution we obtain rWKB (z) ~ e x p ( . on using in the second step Eqs.
69) l(8M9/4e2^ \2 [ i ( g . We now return to the even and odd wave functions defined by Eq.49). we obtain l(8/i)g/ 4 e.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+.e 2[J(«l)]l[j(?3)]! i l[(g3)]!e (20.67) and (20.™) . (14.?/ 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. Eqs.KW4^8 4 .2 f e (27r) 1 / 2 (16/t 2 ) 1 / 4 2 (2/wz)V4[i(9_i)]! cos (2M1/2(^^)+4 7T .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. we obtain T _ ( ± ) = 0.T 7 77 r )V2 /4(2 .l^l !T ^^* T2 7 r ( ^^ /.+) + B^) HAh^/\ir..1 ) ] ! 1 (8/Qg/ e4 2fe l[(g3)]!e^ .492 CHAPTER 20. io(8/*)g/V^i6ft2)V4[±(g3)]! .2 [ ? ) Imposing the boundary condition (20.U ^K^.z+)/ VWKBW 2\2j (2^72 l(8^/4e2fe(27r)i/22V2(4fr2)V4 2 < nz+) ^WKBW [\{ql)]\ 1 (8/i). 1 2 1 ±2 . S1H (2/ KZ ) / (z f 2 (2^). i 27r(8^/ 4 2/t /4 ' r(±) .z)}].i ( 2 / i g ) 1 / 2 ( z + . Large Order Behaviour of Perturbation Expansions again using the Stirling formula. where _ 1 + [±) *) + £ (20. (14.50)).—^ (2^)^(16^2)1/4^ ^ _ _ [T+(±)exp{i(2/ig)1/2(z+_z)} 2(2/ig)V4 + r _ ( ± ) e x p { . (20. Thus with IPA(Z) and TPA(Z) taken from Eqs. (20. i.49).
(20.h) + (qq0)( BE — 90 E{qQ.47).3 ) I = }(»« !7rl/22I(. we obtain E(q. who derived it for qo = 1 and guessed the form for general values of qo. Eqs.l)j /2 e4fe(162/t2)^4_±l or ±i 7r\ 1/2 (16 2 /i 2 )9/ 4 e.h)+2h(qq0) (16h)i°+le'Sh E(qQ.h)+i%Ego. (17. (20.72) We observe that with the square root as on the right hand side both sides are complex. (20. (20. h) of Eq. 20.25)) E = A = Ei + E2. in fact. In both cases we wrote (cf.72) is our second transcendental equation.h) = = E(q0. Combining these equations by replacing (28/i2)<?/4 in Eq. (20.74) This result agrees with that of Stone and Reeve [262].57) and (20.3. (20. We now have the two equations which have to be satisfied.21) expanded about q = qo. i.e. Equation (20.3 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. for q an integer both sides are imaginary. Eqs.3 Cosine Potential: Complex Eigenvalues Using the duplication formula \{1l) this becomes 7rV 493 \ ^ .20.72).qo) El Ah' (20.72) we obtain qqo q0fiSh 2(16/i)*>e {[!(<?<)I)]!} 2 1+ 0 (20.26). (20. the result (20.73) Inserting this expression into A = E(q.57) by the left hand side of Eq.75) .h) + i l + O '4{[^ol)]!}2 E(q0. Ex = 2h2 + 2hq0 + A 1 (<? .4h (16 2 /» 2 )9/ 4 (91)]! = (16 2 /i 2 )«/ 4 = (16/i) 9 / 2 .
I g  ± l ( 2 „. H. J. 7 8 ) i=l V . W.76) he Thus SEqo is the deviation of one of these levels from the harmonic oscillator of level. For definiteness we recall the appropriate expansions together with the definition of coefficients Aj.* The existence of a formula of this type was conjectured without an explicit derivation like that given here and only for the ground state.4 R e c a l c u l a t i o n of large order behaviour Now that we have obtained QE. we recall that we obtained there for the difference 25Eqo between levels with the same oscillator quantum number qo.) {I6h)™/2e4h 1 + . f E .3. Wiedemann [4] and Example 20.W [£(901)]! < (20. by others. Eq.494 CHAPTER 20. Bogomol'nyi and V. its tunneling properties and the large order behaviour. Large Order Behaviour of Perturbation Expansions Referring back to the calculation of the level splitting with selfadjoint boundary conditions. Setting ng AEqo:=2iSfEqo. MiillerKirsten and A. cf. . we can obtain the imaginary part calculated above from the formula 4 (H) A ^ o = 27ri(<5£90)2.74b). i=l We can calculate Ai for large i from the relation Ai = * Jo 1 P°° ti^QE^dti (20.+  . The same formula can be derived in the case of the double well potential. Fateyev [26]. (20.9) the coefficients Ai of its perturbation expansion. (17. i.e. A. we can recalculate with the help of Eq.77) One can say that this simple relation summarizes the intricate connections between the discontinuity of the eigenenergy across its cut. 25Eqo=8h[ 2\1/2 . A = . (20. the discontinuity across its cut from zero to infinity. Achuthan.80) *P. * It would be interesting to see a derivation of the formula — reminiscent of an optical theorem — from first principles. We had A= or _2ft2 + 2ft. 20.2 below. B.
a = J + \<1> (2084) . We also observe that these terms do not alternate in sign.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. and the method can be applied in many other cases.13) and the coefficients M2i or M2i+i of the eigenvalue expansion can be expressed in terms of these as in Eq. *R. j .83) One now sets ftH = ( 2 % P * ( 9 .l ) +2[(q + 4j)2 + l + A}Pi_1(q. B. Dingle [74].20. (20. so that an exact Borel summation is not possible.j + 1).1). (20. The integral is recognized to be of the type of the integral representation of the gamma function.81) We observe that this result agrees with the result of Eq.j) satisfy the recurrence relation (20. J. W. J).4 Cosine Potential: Another Method of Calculation 495 with $$E(h) given by Eq. Using again n = (go . We sketch this method here slightly modified and extend it to the final formula for comparison with our previous results.t We consider again the periodic cosine potential with (unnormalized) eigenfunction expansion oo .3 ) P i _ 1 ( c ? . But as Dingle remarks.l ) ( < ? + 4 j . (20. i ^ = ^ + E?27MT E ^ifoj^W (2082) The coefficients Pi(q. (20. there is nothing special about this case. the integration is seen to give (20.. MullerKirsten [200]. 20. (20. For convenience we write Eq.74).19) for the large values of i under consideration here. f H .j) + {q + 4j + l)(q + 4j + 3)Pi1(q.13) out again: jPiiQj) = (g + 4 j .14).
1) + 2 (a — 1) a . e. a.h) 2+ 8h a p(a. (20. Pi(a) i.91) + Wa+lil/2(z) = (1 .h). 0 4j(4j + 2g) pi_i(a .85) implies p(a . Large Order Behaviour of Perturbation Expansions and considers large values of j for large values of i.1) + 2 P i _i(a) + Pi_i(a + 1).z)WaA/2(z) Setting V.86) one can convince oneself that Eq. d2WKh dz2 + 1 4 K z zM2n WKtll = 0.4 ) ^ .87) Taking terms of 0(h~l+1) of this equation.e. (20.e. the recurrence relation (20. 1  1 / 2 (4 (20. Now.e. (20. i. Abramowitz and I.4. g < 4 j .90) See M.l/2l Wa.h) +p(a + l.496 CHAPTER 20. (20. 4j(4j + 2g + 4) + / . we regain Eq. we obtain (2a .ll{z)= L . A. (20.h) = ^2pi(a) with p.88) satisfy among other relations the following recurrence relation which is of significance in the present contextt {2Kz)WK>li{z) + WK+1.^(Z).85) Considering now the generating function of the perturbation coefficients. .1/2(z) (a1)! (20.g.l. (q + 4j) 2 ~ 4j(4j + 2g).89) Replacing here K by a and \x by 1/2.83) can be written j .27h pi(a) a = 4j(4j + 2 g .a R . which are solutions of the differential equation known as Whittaker's equation. Stegun [1]. formula 13.K + ^ ] L + K^)WK. p(a.31.(a) oc . nx pi_i(a + l). (with a = j + g/2 in the coefficients) 8/? —Pi(a) = P i _x(a .85). Whittaker functions WKjfl. i. z2 (20. Taking dominant terms. (20.
another solution is obtained from this by changing the signs of K and z since this leaves the equation unchanged.„ +1)2] 7 n\z \n which with K = a and /z = 1/2 is: WAz)~e^z«Y.84)) {8h)q/2( Ah for j < 0.92) Comparing Eq..(« 1) 2 ].93) (a1)! ' where f(h). 89. The more precise form obtained from Tables of Special Functions' is w*A* z 2 e / z" n=l [f . g(h)(20. . We can infer the approximate behaviour of the Whittaker function for large values of z from the differential equation (20. p. such that p(a.86) have to be chosen such that p(a.(« .e. Oberhettinger [181].j — —j. h) is expressed in descending powers of h. i.4 Cosine Potential: Another Method of Calculation we obtain Vai.94) Hence to insure that p(a.h) = \~\Pi(&) with pi oc h l. h —> —h.g(h) are functions of h which in view of Eq.Tl ss\{a)\{al)\ (z) s=0 s — a)\(s — a — 1)! for z — oo.\)V . > > ^See W.(« .i/2(z). ^Observe that if one solution of Whittaker's equation (20. a (20. we have to remove a factor (recalling a = j + (7/2 of Eq. v . > (20.20. Magnus and F . (20. (20. Thus the approximate largez asymptotic behaviour is WKtp(z) ~ exp •FH V dz z e ^' 2 exp(/clnz). h) is expressed in descending powers of z = 8/i.88) is known. Dingle discovered the solutions^ p(a. (20.87).88) which reminds us immediately of a radial Schrodinger equation with Coulomb potential.h) ~/(fe) ' (a1)! . In the present case this means the replacements q — —q.i/2(z) + Va+lA/2(z) = 497 2+ Va.92) with Eq. (20.
J. Dingle and H. Large Order Behaviour of Perturbation Expansions and write / i \ pl<x = j+ 2<7>h) « (8h)~q/2e4h v ^+9/2..j _ g / 2  1 / 2 ( . x 4i[2» + i(« — 1)1! il! [ ^ ( 9 .95a) Inserting (20.94) here we have a series in descending powers of h. Comparing > > > this result now with Eq. This expression can be obtained from the recurrence relation (20.8 / t ) for j > 0. _ i s. this becomes / 1 \ (1)5'" ^ U + iq)<(s + iq1)\ 1 Picking out the term in hrl.498 CHAPTER 20. ^ . (20.83) and is given by 1 1 ..95b) Then for j > 0 we obtain from Eq. Muller [74].. B.+ i f or 3 < 0 (20.94) with s — s — j : > x = J + y u ( 1)J ~ f (*+&!(*+*?pi i_ Using the reflection formula (for q = 2n + 1. . we obtain the result of Dingle: (20. h\ oc (8h)q/2e~4h ( _ . n = 0.84) we conclude that the original coefficients Piiflij) a r e f° r large i given by the following expression — apart from an as yet undetermined proportionality factor cq — We determine cq by imposing as boundary condition the expression for Pi(q. i).1. 2.96) and a similar expression for j < 0 with /i — —/i. (20. j — —j. (20.1/2(8^) ( . g — —q.).l ) ] ! 11 R. W. Correspondingly we set for j > 0 / 1 A pla = j + q.
p.l ) ] ! 2 M [i + \q)\ i\\l{ql)]\(J + \q) [j + lq]\[ij]\ (20.J)} 2 i=i 2 [2i+l(ql)]!22il2 i![i(gl)]! i (20.0) = 1 as desired (although not really enforcible for large values of i).Then [2i + ±(ql)}\2* Pi{qJ) [i + \q]\ (20. Miiller [74].4 Cosine Potential: Another Method of Calculation 499 We observe that Po(q. l{Q.97a) we obtain (observe that as required this expression vanishes for j > i) Pi(q. 5.l) (i+y) 2*%(ql)}\ [2i + §(gl)]l i\[i+\ql]\ ' Inserting this expression into Eq. A special case of the formula can be found in I.e.e.157(4). is M2l = 2j2j[Pi(<l. Ryzhik [122].98) the duplication formula 22zl / y P*l>'7J<*l)'(»j»and obtain Mi% + \{q^)W + \{q.20. formula 0. note the misprint there: (r — s) ought to be (r — s)\. Dingle and H. M.100) **R.97d) Then the coefficient M<n of the eigenvalue expansion.99) In the next step we use for both factorials [2i + ^(q — 1)]! in Eq. (20.j) (i + lg)[2i + i ( g . Gradshteyn and I. Thus ° 9 (i + \q) 2«(i+Igl)! p .{j + \<l). J.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.97c) We assume that in a leading approximation we can cancel the factors (i + \q). B. (20. of Eq. .+i?_l]!}2 (20. formula (30).l ] ! (20.3)]!2 2 '+^.1 )/ 2 2 2i ' ^i\{\{ql)]\ [2i + g . i. E* J'=I i+M 2 [2i + 9 .14). S.l ] ! {[. W. (20.
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. Eq.17).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. (18. i.101) This formula allows us to cancel all factorials [i. i.. We refer back to Eq. [^ + i ( g .102) Thus the result is M m 1 2«Fi + g ..e. Large Order Behaviour of Perturbation Expansions In the following step we use the approximate formula (20. i.e. replacing 2i by i: 2 ^2 8 ^ % + *"*}' (20.e. n = 0.. the result of Bender and Wu [18].34) that £ is an expansion in ascending powers of h 6 /c 2 ..' 1 i • (20. (18. (18.2.]\.e. i\ M i large 1 . since \{q — l) + \{q — 3) = \q. i. ..86). The imaginary part we obtained is interpreted as effectively the discontinuity of the eigenvalue across its cut from h2 = 0 to infinity.i ) ] ! [ ^ +  ( g ..5 20. i. [19]. 20.86) and set (cf.o i • .Am i .e.1.l l ! with the large i behaviour in agreement with the result (20.81) for q — qo = 2 n + l . Eq.500 CHAPTER 20.l.105 ) the latter because we see from the eigenvalue expansion (18.3 ) ] ! _ 1 Hence M or.5.
in terms of K where go = 2ftT + 1. (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. 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. The result establishes the series unequivocally as an asymptotic series.e. (18. h2 _ h222K+llHr+K+ll2{l)r+1 r+ K .107) we obtain (i)r r 71" JO yirZ£{y')dy'.108) Proceeding in this way. in fact as one with Borel summability in view of the factor ( .5. We leave this to the following Example. This can then be compared with the level splitting we obtained for this potential. and hence inserting here the imaginary part from Eq.l ) r .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 (4.77) can be verified. .20.e. i. (20. one obtains the result Mr 2io+hr+1+f (l)r[r + f ] ! 7T3/2[1((?01)]! (20. the discontinuity across its cut. i.4) of Bender and Wu [19].5 Anharmonic Oscillator The Borel sum can now be formally written 501 (20. 20.106) With £= E r=0 Mr (20.(20.109) or. and then the relation (20.86).
*R. J. it can be used to obtain the other. Damburg.* A more recent status evaluation is provided by the book of LeGuillou and ZinnJustin [161]. S. W. Damburg. J. S. J. J. M. J . . J. J. V. t p . + 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. 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. Cizek. W. in t h a t when the quantity on one side is known. V. Propin. Grecchi.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. MiillerKirsten [164]. Achuthan. (357) to (371). J. M.77) was conjectured without explicit derivation and only for the ground state by Bogomol'nyi and Fateyev [26]. MiillerKirsten and A. G. An approach to arrive at the relation via instanton considerations has been examined by ZinnJustin [292]. Its usefulness has also been demonstrated there.Q .502 CHAPTER 20. Wiedemann [4]. Propin and H. Paldus. Silverstone [52]. A relation similar t o (20. Wu [18]. t The relation (20. E.^ 20. Bender and T. MiillerKirsten and A.6 General Remarks The study of the large order behaviour of perturbation theory has become an individual direction of research. T. Harrell. Grecchi. Achuthan. . J. 1982) [139].77) and its possible generality has been referred to by Brezin and ZinnJustin [36]. of International Workshop on Perturbation Theory at Large Order (Florida. E. K. Harrell. J. Silverstone [62]. M. as in later chapters. H. and is there shown to have a deep meaning (cancellation of the imaginary part of the Borel sum by explicitly imaginary terms in the perturbation expansion). J. R. H. Eqs. Gram. Wiedemann [4]. Cizek. [19] can be traced back from articles in the Proc. Graffi. Paldus and H. The relation (20. Harris. Nakai. W. J. K. " P . Large Order Behaviour of Perturbation Expansions Example 20. J. R. *Some of the literature in this field following the work of C. 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. R.77) has been obtained in the Schrodinger theory of the molecular hydrogen ion H^ . S. Solution: For the solution we refer to the literature. Liang and H.
'See in particular H. *R. We illustrate the method here by rederiving the Rutherford scattering formula with this method. ^ Our interest here focusses on quantum mechanics.e. Hibbs [95]. Chapters 2 and 5. 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.1 Introductory Remarks In this chapter we introduce the path integral formalism developed by Feynman. Kleinert [151]. Feynman and A. For the application of path integrals to the hydrogen atom (i. R. Feynman [94]. A brief introduction is also contained in R. particularly useful in cases where the behaviour of a system is to be investigated close to its classical path. The path integral is. This formalism is particularly useful for evaluating scattering or transition amplitudes. our main interest in path integrals and their uses will focus thereafter on their evaluation about solutions of classical equations. Thus we are in particular interested in rederiving the level splitting formulas obtained in earlier chapters with this method. P. however. A standard reference is the book of R. in general. Schulman [245]. 503 . Felsager [91]. unknown). A further wellknown reference is the book of L.Chapter 21 The P a t h Integral Formalism 21. P. Probably the most readable introduction.* 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. The formalism is not so useful for boundstate problems (in this case the initial wave function is. since in such cases the probability of this system choosing a path far away may be considered to be small. 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. P. S. However. Feynman [93]. However. can be found in the book of B. which explains also difficulties. Coulomb problem) we refer to existing literature.
*In Dirac's notation we write this equation later (xf.x.2 P a t h Integrals and Green's Functions The onedimensional Schrodinger equation for a particle of mass mo moving in a potential V(x) is (with mostly h = c = 1) i d2 + V(x) tff(x.ti)ip(x.3) We divide the time interval tf — ti into n + 1 equal infinitesimal elements e such that £() ti = = hi ti + e.1) Let the wave function i/> at time t = ti.504 CHAPTER 21. i. x.x).tf) = / dxK(xf.x fc _i) 2 2e2 V{xk) Thus: Given xk(tk).e—»C lim V e m0xk fc=l  V(xk) = lim n—>oo. (21.ti){x.tf.tf\ip) . the initial time. be ip{x. =J dx(xf. We are interested to know the wave function tp at a final time tf > ti and position coordinate x = Xf.t). (21. The Feynman Path Integral Formalism 21.e. */ = U + (n + l)e.4) so that the timesliced action S becomes rtf «+i S = / 'ti n+1 dtL = d£L = n—»oo.ti\tp).5) Sn = X / fc=l mo (zfe . we can compute the action S1. ti) = 5(xf . ti).2) where K is a Green's function. (21. we wish to know* ip(xf. (21.tf\x.ti). 2mo dx2 (21. Obviously K(xf.e—»0 Sn. U. 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).
1 (for convenience we write in the following in the argument of the exponential frequently S instead of S/h): oo poo po> / oo dxTl JS/h dx\ I J—oo dx2 ••• J— c X f~ x n+l Fig.28)): f dr\ exp ia a) ' (21. of which one is illustrated in Fig.6a) .tf). 21. The usefulness of the method therefore depends on the fact that some paths are more probable than others. the Fresnel integral (13. as will be shown to be possible below.21.t).ti) to its final position at (xf. Thus there is a countless number of possible paths that the system may choose in propagating from its initial position at (xi. 21.2 The Path Integral and Green's Functions 505 Above we considered time intervals of length e.1 A path in (x. Here and in the following we need one or the other of the following integrals (cf. The repetition of the quadratic factors in the argument of the exponential with indices differing by 1 suggests the construction of a recurrence relation. 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. 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.
37)) dxk exp d(xk . 2 + 2a. J—oo dxneiS/h — • 0. tf > U.506 CHAPTER 21.z f c _i) 2 } } .6c) In the evaluation of these integrals we replace a by a + ifi.0 ^7 (x f c + i .mQf Xk L ~2e^ ^ . with f™ dze"™ 2 * 2 / 2 = s/2ir/w2.7) 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.ti) '•= f g m ) (N(€\]n+i dxL. 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. dx exp z—{a.Zfci) 2 } I +{xk+i x*.xk+i) exp . n > 0.t ~Xk+l> + xk+i 2 ) +{xk .xk+x oo —oo oo ..m0 . for example. The Feynman ia Path Integral Formalism f J—c J dr) r\ exp v = 0.6b) —c ia o drjr] e x p—V € (21. ie f me\ 2a\a~) 1/2 (21.2 + (x + x f c + 1 .5)): G(xf..(x fc+ i .* W i t h the first of these integrals we obtain (compare the result with the Green's function (7.tf. (21. (21. One therefore removes the troublesome factor by defining (Sn being the sum in Eq..ir 1/2 £dxeXp[i^{2(3 dxexp\i—<2lx+ 72 iV(e) exp Z7TT.Xi.x f c _ i ) 5 \m0 J (21.s f c _i) + (xk+i .xk_i) dx exp i — { 2 a . one has j ^ dr1eifa+i^ /£ = . We then integrate and finally let /x go to zero. dxneiSn/n.8) Note t h a t in spite of the n integrations.
In the expression rx" / ?){x}eiI{^ Jx the quantity x" is only a symbol to indicate the endpoint of the path. § See e. 11 For discussions see e. B. (21.g. Felsager [91]. . However. G(xf. i. (21. p. p. (21. B. Felsager [91].5).2 The Path Integral and Green's Functions 507 Eq.Xi. 183. It is neither a variable of integration in T){x} —> Ylndxn. and will therefore not be attempted here.tf. Then one integrates over X2 from —oo to oo and obtains an exponential involving (x^ — XQ)2.2).ti) for tf > U.5).g.9) T>{x{t)}eiS/h = ^ww^IdxlIdXnelSn/h(2L10) This is the formula originally given by Feynman. (21. the Green's function K of Eq. The additional factor is needed later to combine with the number n contained in the extra factor y/n + 1 (in Eq. we arrive at a path integral without the measure factor N(e) above.21.10) can be looked at as the product of a succession of free particle Green's functions (or propagators).7) for n — 1) to give a time interval.26)) that — ignoring the potential — the multiple integral (21.tf. We shall see later (with Eq. and so on. it may be noted that with a different philosophy concerning the summation over paths. as in our discretization in Eq.ti) For tf > ti we now write* G(xf. i.24).Xi. A mathematically rigorous definition of the path integral is difficult and beyond the scope of our present aims. summation over all possible paths. 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.1. 21. (21. nor the limit of integration of some variable Xi.' These factors N(e) were introduced to cancel corresponding factors arising from the Fresnel (or Gaussian) integration. These integrations are indicated by the horizontal arrows in Fig. in fact.e.t In the following we demonstrate that the function constructed in this way is. See the square root in front of the exponential in Eq. Thus one first integrates over x\ from — oo to oo and obtains an exponential involving (x2 — xo)2. however prefixed with a new factor.U) = / JXi=x(ti) = K(xf. (21. Different constructions of these families of paths usually lead to different measures. which all vary from —oo to oo. (21.§ The factor [iV(e)]_(n+1) associated with the multiple integration is described as its measure.Xi.e. 183.tf.
16 x2 Sn+l) \ 7T2~(Xn+2 V(xn+2) Setting now xn+\ — xn+2 = n. 187 .Then we can construct a recurrence relation as follows. 21. the other with that of the action of the free particle SQ.Xi. We have G(xn+2. Felsager [91]. (21. One way to avoid this problem with the integration measure is to consider the ratio of two path integrals.72) to (23. 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.ti. Each of these linear pieces is completely characterized by its endpoints. Performing the sum now in the sense of summing and hence integrating over the vertex coordinates in Fig.3 in the evaluation of I a specific path integral (see Eqs. Expanding this in a Taylor series around x. The Feynman Path Integral Formalism The factors N(e) can be avoided by instead summing over piecewise linear paths.ti) = 1 j^p: f°° cte n +iexp } • fm0. however. Consider a time t = tf + e for position Xf = xn+2. Then we have on the right the factor G{x+rj).tf + e. so that the piecewise linear paths fill the entire space of paths. we obtain the same as before except that there is no factor N(e).ti).Xi. one with the full action S.508 CHAPTER 21. (23. We next determine the equation satisfied by G in order to verify that this is indeed the Green's function for tf > t{. . It is for these reasons that — unless required — the measure is frequently ignored. I We employ this method later in Example 23.U) imo 2 G(xn+i W)Idr.xi.exp 2e n ieV(xn+2) G(xn+2 + V. pp. the discussion in B.tf + e.xi. and one is calculating the Feynman amplitude or kernel relative to that of the free theory. Then the measure drops out.188.1 and allowing n to go to infinity. we have See.tf. we have G(xn+2.11) For reasons of transparency in the next few steps we replace xn+2 by x and suppress temporarily tf.74)).
15) We see that G(x. tf.t Thus + 0(e2 (21.ti)\ .ti) +\if(^G{x.tf.Xi. Retaining only terms up to those linear in e. U) + Q(e2 = (lieV(x) + 0(e2)) G{x. tf.6a).xi.ti) for tf > ti (21. U) = e^G + 0(e 2 ) dtf 1 d2G(x.Xi. Xi.tf.(21.ti)ip(x.U ieV(x) 2i7re\ 1 / 2 m0 ie d2 G(x.ie 1 d2 2nV0dx^G{X.6c).x.— p G ( j .Xi.tf. .tf.xi.tf + e.tf.ieV(x) + ri< —G(x.xi.2 The Path Integral and Green's Functions so that G(x.tf)= dxG(xf.xi. 2mo dx2 (21.ti) N(e) or G(x.tf.U + 0(e ). U) .tf.tf.Xi.ti) 2 —l€ — ^2 + V(x)G(x.ti) = G(x.tf.xi. ti) satisfies the timedependent Schrodinger equation.ti) = j^rr dnexp I ^rj1 . (21. we obtain: G(x.ti) +V(x)G(x.ti) . tf.U).tf.U) +••• \ 509 G(x.6b) and (21.xi.xi.Xi.Xi.tf''Xi.21.14) = G(x.tf. U) + .16) .13) We integrate with the help of the integrals (21.Xi. tf + e.tf + e. It follows that ip(xf. d and this means d i—G(x.G(x.xi. tf.tf + e.Xi.ti) = 1 d2 + V(x) G(x.Xi.
In a problem where V is small it is convenient to calculate K with a perturbation method. 21.U)6(tf . The Feynman Path Integral Formalism also satisfies the timedependent Schrodinger equation. or K(xf. The function G which satisfies the timedependent Schrodinger equation (21. as may be verified by inserting this into Eq. 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. (21.U) = G(xf. The result. Eq.17) Differentiating K with respect to tf we obtain —K(xf.*<)' ( 2 L 1 8 ) 0(tf ~ li)QfG + 5(Xf ~ X)5^f since G(xf.15) is a wave function (cf.Xi.3 The Green's Function for Potential V=0 We now calculate the Green's function K for potential V = 0.x.20) . now called So. and the Green's function KQ\ S0= Here (cf.U)G0. (21.x. We have therefore established that the Feynman integral (21. is the Fourier transform of the nonrelativistic propagator.tf. K0 = 9(tf .U) = / T){x} eiS\ [fdt \mx2.ti) = = 9{tfti)—G + G5(tfti) .tf.16).8)) Go(xf. (21. denoted by KQ.tf. 21. Hence our original function K is the same as G.tf.10) is indeed to be identified with the timedependent Green's function.510 CHAPTER 21.U).ti) = 5(xj — x). (21.1 Configuration space representation In the case of V = 0 we have from the previous section for the action.x. (21.3.U.x. Eq. as this is also called.1)). It follows therefore that {{W ~V+2m~0~^) K{Xh t/. X U) = i6{Xf ' ~ X)6{tf " U)' (2L19) Thus K is the Green's function for the problem of the Schrodinger equation.
22) • \ n/2 . f c .ti)= iX (o. (21.t) is the configuration space representation of the free particle (V — 0) nonrelativistic Green's function. TJ+1 dxn exp[i\{(xi 2 — XQ) + • • ^n) }] (21.1 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.8).3 The Green's Function for Potential and hence Go{xf.7) for here n = 1 (n+l)/2 lim ( —^~ K2meJ m /n A = 0 V 2e (21.t) = ( m0 \2mt 1/2 irriQ and writing x = Xf — Xi. (21.Xi.21.Z i ) \2KieJ \m0 J y/n + 1 1/2 m0 «mo (21. (21.25) T h e quantity Ko(x.oo /*oo / oo cfcci / J—oo cfcc2 • • • / J—oo + (x.24) ( s / .tf. n + i .Xi.tf.Xi.ti) = = G0(x.21) fe=l and t = (n + l)e = tf — U and h = 1.tf. We obtained the same expression previously — see Eq. 2t Go(xf.ti)9(t). (7./ . oo dxi 71+1 / J—c /_ dxn exp oo irriQ oo ~27 ^](a.x 0 )" n + 1 (in agreement with Eq.i ) 2 .Xi) exp \2Kie{n + 1) 2e(n + 1) Note the square root factor in front which originates as discussed after Eq.tf.37) — by explicit solution of the differential equation which the Green's function satisfies. exp where Xj = xo and Xf = xn+\ and there k = 1)..Xi. (21. In Example 21.. .Xi.tf. we have Go(xf.z / c .Xi.tf. But t = tf — ti = (n + l)e.t) K0(x.23) for Go and obtain Go(xf.23) We now insert our expression for In into Eq.ti) Ko(xf. exp 2 (21.ti) = V = 0 511 m0 lim n—>oo \ 2irie oo (n+l)/2 /oo oo /*o / dx2. Then Go(xf.
e.1: Evaluation of a path integral Verify 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. i.54). (n + l)A"_ We set y := x n +i — Xi = x n +i — XQ and (x n + 2 . Eq.t0) (21. (7.x „ + i ) 2 = (x„ + 2 .5) is carried along. ^ (xi .22) oo Indxn+leiX<x"+*x»+^ 1/2 f°° dxn+ie^:._*R±HY+l_{xo_X2f J —! r dXieiM(xlxo)2 + (x2~xi)2 ei^(x2x0) and using Eq.6a).22). (21. If the potential V(x) of Eq. 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. the latter being a Green's function with interaction like that for the harmonic oscillator.25) we can rewrite the Feynman path integral or kernel (21. For n = 1 (see below) h This result can be verified like Eq.xo)2 + (x2 . (21.2y(xn+2 .7) by direct integration with ^ .x „ + i ) 2 = ( x n + 2 Xi + Xi .x ^ 2 + y2 . Then oo 2 / (21.ti.10) in a product form. Example 21. n+l = lim / n ^{a^jexp fc=i ^°° Jxi = lim / • • • / dx\dxi • • • dxn I —1 nKx J J \2mneJ / \ 1/2 .x0. or else by induction. ••• 2 / \ 1/2 \2iriheJ dxldx2•••dxnK0(xn+l.)+(1"+2.^(Xn+1~X0^2+iX(Xn+2~Xn+1^'i.26) with n integrations and n + l factors KQ. (21. In our proof of induction we assume the result is correct for n. e 2he ^{Xn+lXn)2 = \2iriheJ lim ••• n^ooJ J x(^*) e^() . we have instead of 5*0 and KQ expressions S and K.512 CHAPTER 21.f_^y e. as / JXi f r D{x{t)}eiS°lh .tn+l•xn. (21.. Solution: The result can be obtained by cumbersome integration.)!} J — OO .tn)••• K0(xi..xi)2 = 21 xi ^2o(„.
We pass over to momentum space with the help of the following two integrals. (n+l)A n n j(a.30a) —6(t) / dp exp ipx — to Joo 1 i—t 2m. 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.a .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. i.2 y ( a . 21. ) H — (xn+2Xi) n+2 n+l n+2 y2 .Xi) + ( x n + 2 Xi)2.max exp i 2t 6(t) 1/2 \2mtJ f°° 2KimQ J dp exp J—c ipx P t 2mo (21. it . (21.2. This quantity is also described as the nonrelativistic free particle propagator.x2rriQ 1/2 toij_.2y(xn+2 . A(n + 2) 21.n+2a.25): Ko{xJ) = = (S) e{t)(^X'2 y J . (21.2 M o m e n t u m space represenation Our next task is to derive the momentum space representation of the free particle Green's function KQ. (21.3 The Green's Function for Potential V = 0 Next we set z : re+1 (x )2 — Xj).21. Using Eq. (21. n+l T h u s as r e q u i r e d 1/2 Jn+l ^ e J — oo 1/2 .29) Hence with KQ from Eq.. a dp exp [ipx + iap2] = exp — i(21.3.i) .6a). n + 2 .Q . The second integral is an integral representation of the step function.27) we have (with a = t/2m0) .28) exp (— V 2irimo J —( dp exp ipx — i p* 2mo (21.e. and + 2 2 l n + 2 ( x n + 2 . rc+2 rl 1 / s2 513 so that n dz = dy = d i n + i . e > 0 .i) H —3/ . s 2 n + 2 2 „ / ^ n + 1 / \2 (x„+2a.i j ) H n +— z 1 . The first. iir(n + 1) 1/2 jn+1 n+l L(n + 2)A'n + l I 1/2 4 (n + l ) A .
t) i(2 h?L L E iv 00 dr exnlipx T '2mo t + itr] — te ~U2^LdPJ ^E+£. ( 3 b) ° The expression E+ V — it 2m. (21.514 CHAPTER 21. i.*' where 2mo — r. The Feynman Path Integral Formalism Im t \t>0 Rex Fig. 21.51)) is fairly selfevident.Q (21. The first order correction contains one factor of the potential V. Thus in first .28) this becomes K0(x. This is the type of expansion most people describe as perturbation theory. an expansion in rising powers of the coupling constant (contained in the potential).4 Including V in First Order Perturbation We now proceed to calculate the first order correction in a perturbation expansion.31) is the nonrelativistic propagator which describes the propagation of the free particle (V — 0) wave function.e. 21.2 Contours of integration. The generalization to three space dimensions (required below in Eq. Replacing 9(t) by its integral representation (21. the second order two factors of V and so on.
+i 6XP L — {xk XkiY {iV(xhti)} 1/2 X Em fc=l — 0 (xkXki) W i t h Eq.ti.tf.tk) J_( x exp fc=l = ie(—2(xkxk 1>{x(t)}exp iSoi dtV(x.33) xG0(xi.t)] n+l .^){iV(a.i.i.. . (21. (21.U) = — lim fv{x(t)}el —— js (n+l)/2 I dxi dx2. .)} (21. dxi ( m0 V V27rie/ m0 \^He \ exp fe=.4 Including V to First Order: Perturbation Expansion h=l.U).xi.32) Since — we recall this for convenience G(xf.ti)= n+l ri_> f V{x(t)}eiSo „ iSo f ' dt[iV(x._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 . li f f 515 order (cf. ^ = k ) = lim (S^ A H*\ nKX) *—J \2irieJ r "+1 />oo X / dsi / dx2.20)) and with elS = exp i S i o / hi dtV(x... x (n+l)/2 \ lim .Xi.t) JSo dtV(x. \2meJ n+l J.t)\. dxr. Eqs.tf.t/. } e °° f l f •/ oo /"oo F ( a . (21.^ J_00 V(xk.21) this can b e written as Coo oo /•oo />oo / oo dt / J —oo ^Go(cc/.5) a n d (21.(xfc . .^.Xi. .. dxn exp i ^ —.i V e / D { a .21.t) the first order correction is (G\ ~ G — Go) Gi(xf.xfc_ OO J— OO J—OO L J.
34) We observe that we can extend the ^integration to — oo since KQ contains the step function 9{t). OO /"OO / oo dt / J—oo dxKo(xf.t) ! Jti r*f dt'V(x.i / dtV(x. i.ti). Xi.tf. t2.516 CHAPTER 21.xi. t') + • • • .xi. (21.ti) t2)K0(x2.tf.I dtV(x.t) (21. (21.ti)K0(x. The first two terms lead to identical contributions and therefore cancel the factor 1/2! in the expansion.x2. / 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'.t) =li dtV(x. in writing down the expansion exp i dtV(x.xi.36) {if •j. The Feynman Path Integral Formalism and hence correspondingly for the first order contribution Ky.t. i. Jti Consider the term of second order.t2.U) xV(xi. It can now be surmized that the next order contribution to KQ + K\ is K2 = (~i)2 / dti / dt2 I dx\ I ti. t) Ju Thus the term containing n times V contains a factor 1/n!.t2) xKo(x2.37) Jti The last two contributions do not describe the time evolution from ti to tf.t2)V(x2.tf. . dx2Ko(xf.x2.t)V(x.x.35) We also observe here that the problem of timeordering arises only in perturbation theory. The factors of "V" originate from the series expansion of the exponential exp ftf . (21.e.ti). They must therefore be deleted. This is cancelled by n! different time orderings.t)K0(x.e. = i dt2 / dx2Ki(xf.t2)V(x2.
.3 Feynman rules and representation of the Green's function.38) .ti) (21. x"^ X x"2.21. The diagrams represent mathematical quantities and hence are designed on the basis of rules.t.Xi.t) is V(p.39). U) (21.i ] P I dt J dxKn(xf.x.tf.t.t)Ko(x.3 along with the diagramatic representation of Eq.t)= f dpdEeipxiEtV(p. The Fourier transform of V(x. Here.t2 K o (x 2'l2 . x r l i 'V(x.t)V(x. This can also be written in the form of an integral equation which when iterated yields this expansion: K(xj. E) which is given by (in onetime plus onespace dimensions) V(x.tf.t)K0(x. For various reasons — such as illustration.Xi.x. called Feynman rules.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.xi. Such diagrams in the context of field theory are called Feynman diagrams.ti) i dt = K0(xf. (21.Xi.tf.40) The rules for the present case are given in Fig. (21. 21.39) dxK(xf.Xi.t) ) K K K0 'V X K o Fig. tf.ti).t)V(x. and hence the diagrams representing perturbation terms here are Feynman diagrams corresponding to those in field theory. of course.ti).E). we are concerned with quantum mechanics.ti) = K0(xf. or in momentum space as mostly in field theory. transparency or brevity — it is useful to introduce a diagramatic representation of individual perturbation terms. These may be formulated in configuration space as below.tf.Xi. 21.
41) We have to pass over to three space dimensions. dn = (L/27r) 3 dk. ^(x. (21. with p = Kk.ipf have to be properly normalized. i. in onespace plus onetime dimensions by A= dxf dxiil)*f{xf.7 = e i k ' x _ i : l i with f i/}*i/jdx. * k 1 • .Xi. here with E — k 2 /2mo. 2 The expression A 2 /T can be related to firstorder timedependent perturbation theory in quantum mechanics.e.t) = . The plane wave function normalized to one over a box of volume V = L3 is.ti)il)i(xi.= l. The Feynman Path Integral Formalism 21.tf.ti). T = duration of time for which potential is switched on. There the wave functions ipi.tj)K{xf. (21. i.k = 2wn/L. The transition amplitude A describing this process is defined by the Green's function sandwiched between the inand outstate wave functions.5 Rederivation of the Rutherford Formula As an application of the path integral method we now consider scattering of a particle off a Coulomb potential.e. and dn = dnxdnydnz. for the allowed values of k: kx = 2imx/L. . We recall for convenience from **Note that from the wave function we obtain. to one particle in all of space. (3) The transition probability per second is = \A\2 . (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.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. mo V since ip is normalized to one particle in volume V. 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).518 CHAPTER 21.
44) = u(Ek. Then H'ki can be taken out of the integrand in Eq. (10. in which T is a large but finite time.21. Then for stationary states M^t) and itm^ = f HUt'V^t'dt'.= 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. Proceeding in this manner one obtains Fermi's "golden rule". Hence (h = 1) atot = f J j2^~k dp V Vm0\A(p. H = H0 + \H'.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.T)\2 f • (2L46) We take the following expression for the timeregularized Coulomb potential.117) and Example 10. (21.4).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. «= — .k.44) and f_oo can be replaced by fQ. Eq.. The transition probability per unit time is . V(x.t) = e~^lT\ r where r = Ixl..— . (21. and with these the following perturbation ansatz: ih^* with an(t) = HV. i.e. P(*) = j r (2145) (cf.47) . # = ^an(i)VnM) n = a^(t) + \a^(t) + . u = ^\H'kfp(k). 47T 1 3 7 (21.
50) Here x — xj — Xi and t = tf — ti.tf.tf.t) = ^ J*ie*<*'*><<*'*>.xuti) = ^ J°° d p e i p x _ i ^ * .tf. (21.x. by threedimensional dxf and so on.yLi. t. Ai = dxf / dx.44) by contracting the propagators to instantaneous point interactions. is now in three space dimensions A = Ai= / dxf / d^ty*f{'x.48) where from Eq.f.tf)ipi(xf. (21.49) i.t) (21. ti) = co5(t — U)5(x — Xj) etc.>/>.44) (here in the scattering problem with initial and final energies equal).f. (21.t)K0(x. (21.tf).i x (i)V(x. £. (21. (21.t.520 CHAPTER 21.ti).. The Feynman Path Integral Formalism In calculating the Rutherford formula with the help of our previous formulas.).ti).. t)K0(x. The threedimensional generalization is (taking tf » T and ti <C —T. The lowest order approximation of A.tf)K0(xf.49) together with the following wave functions (of asymptotically free particles) 1>i(*i.e.tf)Ki(x. dt dxip*f(xf.51) Inserting this into Eq.tf)ip*(xf.x.^ / . t. (x*. we have to replace the onedimensional quantities dxf etc. also called Born approximation. This now has the form of the amplitude (21.tfx. We can relate this amplitude to the amplitude ark of Eq.U) = ~ e ^ l ^ \ ^(*/>*/) = _ L e * P .39) OO /"00 / dt / OO J —OO dxK0(xf. Thus with the ansatz Ko(x. we obtain A1 = clS(ti tf) / dxf(i)V(xf. (21. In our previous onedimensional considerations we had with Eq.52) .tf.30a) K0(xf.Xi.ti)'ipi(xi. T a large but finite time.* / . (21. so that 0(t) can be dropped) K0(xf. i.t)V(x. Xj. x.
t)exp A(p.2: piqx / Y iukawa •" dx e~Mr 47T q + M2' 2 (21.59) . (21.54) Now inserting the expression for the potential V(x.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 .56) (21. Then integrating over q and q' we obtain . (21. Then we require the following integral.k. this becomes A(p.k.i ( k 2 .iaT2/8: L With 0 0 • .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 . P2 .). which is evaluated in Example 21.~ .5 Rederivation of the Rutherford Formula we have (for ingoing momentum k and outgoing momentum p A(p. 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. £).k.p ) .p 2.T) = x exp L z/ q/2 k2 521 f dx.fdxe^rt* dxe< p) x fdte1 ' (4) 2m0 T2 e "^^J2" 16 .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.T) = ^JdXJdt%*t2/T2exp i ( k . .r) = = y . (21.T) = JdtJdx(^\V(x.21. k2 ipx + i—t + ikx — i 1 2mo 2mo (21.57) we have A(p.k. At2 dteiat~^ a2T2 f°° =e " J dre^lT2 p2k2 2mo = (~)l'2ea2T'' / 1 6 .
„.64) Inserting the result now into our expression for the total cross section. (21. we obtain .59). Using Eq. { I i _ F T l } . The Feynman Path Integral Formalism This means: The result of Fourier transforming the Yukawa potential ~ exp(—Mr)/r is the propagator (21. dE = PjV = ^ E ± .„ 2. integrating over the square of the ^dependent factor in V we have /"°° dt(e4*/"1*}2 = (^p) ' 2 instead of T.\ 2?7lo / 1 (21. (21.a ( — ) ^ .59) we can reexpress A as I /7rT2\1/2 e (P 2 k 2 )T 2 f Ajr ~) p * P . In our case this has not been done. <rtot.60) Now.T)2/T. i.< ^ £ \M VTTT2.^ . e (21.522 CHAPTER 21. k .3. This expression assumes that the integration over time t is normalized to 1." + M ] p) ' "•"• ^2 . 2 with 2mo J 2ir a2 Ct 16TT2 LU7I F [(k .P ) 2 + M2]2 4 VvrT2. otot contains the factor \A(p.T2f 8 V/ 2 .t o t= y dp ( 2 v r ) 3 "V 3 I W: 2mo y m0V k 32^a2S(^) y ( k y 22 [[ ( k _ p ))2 + M 2 ] 2 ' kP 2 (21.61) We can now Hence in our case \A\2/T has to be replaced by \A\2/(nT2/8)1/2. k2 .51 — 2 /w2 y ~ —.63) we have: Probability per unit time = 77777.65) But E=*. e 8{k) = lim 7=*—^ (21.k. write down the probability per unit time which is ( 8 \1'2 _a2 16^ .e. T ) ~ .62) We want to consider the limit T — oo. mo m (21 . For this we require the following • representation of the delta function: —k2T2 rp —k2T2/8 S(k) = lim T . 66) o . Using the result (21. y 2 [ ( k . / / P 2 _ . where M has the meaning of mass (in the case of the Yukawa potential the mass of the exchanged meson).63) We establish this result for T — 1/e in Example 21. ( 21 . .
67) T h e delta function here implies energy conservation.2: Fourier transform of the Yukawa potential Show that [ j / j eiqx dx—e .21.e .= e " f c a ^ o 2v^fa /4a = lim £^o —. esfn (21. and inserting the expression thus obtained into the final result of Example 10.p ) 2 = 2 k 2 ( l . p 2 = k 2 . 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. This result may also be obtained from lyukawa of Eq. with k = y 2moE / we obtain the differential o?m\ 4fc sin 4 (0/2)' 4 cross section (21.M £ M * = / m0kdn = m0^2m0E (k .cos 9) = 4k 2 sin 2  . —Mr 4w = .59) by replacing there q 2 by ( k — p ) 2 . giving the potential a coupling constant.67) and (21.70) .^ Jo 47T q 1 M — iq 47T i q2 + M 2 Jo G— 9 Example 21.^ . (21.e. Example 21. setting M2 = 0.. Then (x = r) foo „2jj.5 Rederivation so t h a t of the Rutherford Formula 523 M^f) .j .68) into Eq.68) Inserting (21.3: A representation of the delta function Determine the Fourier transformation of e~ax of the delta function <J(fe) = lim — / dxe~ax a*o 2K Joo for a > 0 and hence the following representation eikx = lim — .69) This is the wellknown Rutherford formula.* 9 ' d r e .65) with the cutoff M > 0. (21.e. so t h a t (21. /Yukawa = ) Solution: We choose the 2axis along the vector q.j .^ . i.4. (21. 2mo / dft. 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 .
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 ) .q).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 write the state vector of a particle of threemomentum p in the momentum space representation: p). (21.73) 5{p) = (W / dxe"P'x' 5{x) = J^rI dpePx '' (2L74) The normalizations (qp) = (27r)3<5(p . (yx) = 6(y . ' i a = lim e —0 pk 2 /e2 e^TT 21.524 The function required is CHAPTER 21. '<*> = .72) co From (21. We consider again nonrelativistic quantum mechanics. however. a. 7 r°° d 9 xdxe~ax elkx = — dx — (e~ax oo 2a Joo dx Partial integration of the right hand side implies oo /.* .oo J / )eikx. In configuration space the state of the particle is written x).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 . We have x> = I J ^ l p X p l x ) .75) .71) with respect to k yields the same as °° 2 . Va we obtain g(k) = Je~k V a = 0. 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). Differentiation of (21. In the present case.70) and (21. The Feynman Path Integral Formalism o° 2 / oo dxe~ax eikx. a simpler method suffices.71) Solution: In general one uses for the evaluation of such an integral the method of contour integration. (21.. Direct integration /4a . (21.x) (21.^0 a^C 2^/TTO. Since dxe" 1 1 1 = « / .
H. (px) = e~ipx. 159 . (21.77) (21. t\ipo)s. so thattt d — x.t) = e* H t x).t'. x' — x infinitesimal. The relation between Heisenberg picture states and Schrodinger picture states is \ipt)s = e~lHt\ip)H.t) = —iHip(x.. Thus (qPp) = p(qp) = p(27r)3<5(q . t) = (x.80) (21.The time development in the Schrodinger picture is given by \tpt)s = e~lHt\4>o)s.i'x. i.e.t).p).t'x.76) These relations imply the completeness relations 1 = J dxx><x.x.81) Comments on signs in Eq.e.^ l x ) . Then ( X  ^ . i.t).83) (21.i) = e iH4 x. (21. as had to be shown.t)} constitute a "moving reference frame" in the sense that ip{x. (21. i.0) = e""x). (21. i. ) H = (x</>o)s. at initial time ti to its final position xy at time tf.£) = ( x V ^ ' . so that \if>t=o)s = H>)H. (21.e.t\.79) Feynman's principle is expressed in terms of the propagator or transition function (x'. which describes the propagation of the particle from its initial position x. Similarly we have a position operator X with Xx) = xx).78) Pp) = pp) is to be interpreted to say: The quantity p is the eigenvalue of operator P with eigenvector p).We now introduce the Dirac bra and ket formalism.t> =iH\x.i) = K(x'. .t). The equation 1 = J ^p)(p. We consider first an infinitesimal transition.e. i.6 Path Integrals in Dirac's Notation therefore imply 525 (xp) = e ipoc . Ryder [241]. The time development is given by the Hamilton operator H. M^ip(x. (with ft = 1) d/dt\ip) = —iH\tp). Then the states {x.t\ = iH{x.82): The general Schrodinger equation has the state vector to the right.t). x. t' — t.If '/'(x.We define x. pp. we must have d/dt{x.160.21. d/dt\x.82) (21. Hence for the infinitesimal transition (x'.t) = +iH\x.e.. 4) is to satisfy the Schrodinger equation. See also L.
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). (2187) (21.91) .e.* .526 CHAPTER 21.x).89) Using this and previous relations we obtain for the Hamiltonian (21.p'). Pq) = J 7^3Pp)(pq) = J dppp>*(p .x y(x)e i p . 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.i p '. 2mo Since (21.P V<P e l )  P > e P X <2184 = /(w/(^ ' ' *" '"'' *=£Uv«.90) (P'IVIP) = y<&. (21.88) Thus P has the operator representation P= /dxx)V(x.x') + V{x)5(x' .85) the amplitude or expectation value or position space representation (x'#x) = . i.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). (21.V 2 < 5 ( x .y<&(p'ix')(x'ivix)<Xp) = f dxe.. "  » We suppose that for a single particle of mass mo moving in a potential V(x) the Hamiltonian is (21.x = V(p .
We now consider the transition of the particle from Xj.6 Path Integrals in Dirac's Notation we have in momentum space the representation (p'ff p> = ^(27r) 3 <5(p . t). EQ — Hi .p) + i{t . the signs are reversed when these operators are applied to the ket vector \x.84). (21.e.tj+i\x.fx.p') • • • j W < 5 ( p ' .21. ti to x j . tf subdivided into infinitesimal transitions (xj+i.93) + ipx P r ip(x'x) iH(t't) (2TT) 3 (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 '.p') + V(p .92).t) (cf. for the amplitude of an infinitesimal transition.tj). (21.95) when applied to the bra vector (x.94) In the above we have chosen the signs such that p = —iV and H = i— ot (21. Thus with •"•/ — ^ n > *•/ — ^ni ^ 0 — Xj.92) We now return to the expression (21.t> d f dp' P' r_dp f dV rivf*!ini\rmtf) J (27T)3y (27T p e p) e ipx +i(tt'){p'\H\p) + }eipx .j.i'x. (x'.p').t') \^(27rf5(p 2mo +V(pp' where we used (21. But e'ipx + 0[{t .82)). i. 527 (21. i. We have <x'.
d x „ _ i .ti) = x 0=Xi n^n fc=1 ? <^Pi (2TT)3_ exp / ^fa . = 1 X X<jPfc We now replace H(pfc. d x i ( x n .ti) rx/=x(i/) d x i dx2 .2.tfei) fe=i ..x2)(t2 We can rearrange the factors here in the following form c„=x/ n—1 (Xf. zmo T h e n t h e integration over all p ^ can be performed by using Eq. x exp [ .97) H hH(pn.. i.i { f l " ( p i .+ V(x f c ).xk (21.. .xn)(tn in_i)]. . x i ) ( t i .. x f c ) = ^ .. .ti. d x i . . i.dpnl dpn (2TT) 3 (2TT) 3 exp [i{pi • (xi . ..a — —e/2mo and divide b o t h > sides by 2TT.t0>(2196) We define a p a t h in configuration space by a succession of n + 1 points x o .t 0 ) + H(p2.x i ) H h pn • ( x n .e. . T h e integration over the n—1 intermediate positions..99) . We now insert for each of the infinitesimal intermediate transitions the amplitude (21. d x n .xic) by fc — Xfej H(pk. x i .94) and thus obtain a phase space expression. The resulting relation is m0 2irieJ 1/2 m0x exp ze ) " / 2TT exp Pk 2 mo 1 (21. t n  x n _ i .i /Xj=x(*i) f J [ ^ 1 dp2 3 3 J (2TT)3 (2TT) .e.528 CHAPTER 21.tixo. (Xf.tn respectively. x n which the particle passes at times tQ. . t n _ i ) t„2><xi.tf\Xi. The Feynman Path Integral Formalism we obtain (it will be seen below why we do not introduce normalization factors here) (x/..27) in which we make the replacements x —> eifc.tj) = / ••• / d x „ _ i d x n _ 2 . (21. . dx.tf\Xi.i/xj. then corresponds to the sum over all paths.X n _i)] h) (21.x 0 ) + p 2 • (x 2 .98) tk — *fcl # ( p f c .
98).103) is another form of the pathintegral for which the integration with the help of Eq.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.tfci) fe=i £(**'**){ p***^} —tki m0±2k (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.ti) = J D{x(t)}D{p(t)}expi [' dt\pi. A.8) as N(e) so t h a t the integral does not vanish for e — 0.6 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.tf\xi.101) Then Eq. i.fc=i 3n/2 r n m0 exp 2_^i{tk 2m(tk .] ex H ^I (.100) Hence with all tk — ifci = e we obtain /n dpi i _(2vr)3 exp •. (21. m& ze ) J (2vr)3 exp it p fc • xfe IP* 2 mo (21.p)]\. Truman [254]. > Thus the phase space representation of the pathintegral* (21. Smolyanov. (21. / {xf.. .21. G.e. (21.98) can be rewritten as 3n/2 2^") x / n dxfe exp * ^o** ~tfcV rnl <i 2 m ° x fc ~ ^ (21. (21.H(x. i.e. G. Tokarev and A. to / m0 \ 3 / 2 las.
t'}. £ J ) with no operator in between. Garrod [106]. We know from classical mechanics that this relation is a Legendre transform which transforms from variables x. this agrees with the corresponding discrepancy in the onedimensional case of Eq.8). (21. (21. x.104) where for the particularly simple Lagrangian L under consideration here m0±l .tf\xi. t.530 CHAPTER 21.iJ(x fc .ti) = ^lim^ N(e)3n / * JJcbc fc exp i / fc=i * dtL(x. Note that we have n factors [iV(e)]3 but only n — 1 integrations dxk. i.8). operator (21. 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'.tf\xi. Clearly for a better understanding one also wants to explore the case of sandwiched operators in the path integral representation. nl (xf.x) . We check below that this is given by x = / ( ^  p > i  ( p   <2L107) See also C.106) In order to be able to deal with this expression we first require the momentum operator representation of the operator x(t).105) D{x} n ^ o o [7V(e)]3n Udx kk=\ In this notation the measure factor is contained in ©{x}. p . Thus finally we have Feynman's principle: (Xf. . (21.ti) where nl D{x(£)}exp '.x) (21. t to x. xfc). £ /  X J . I hi dtL(x. We have now obtained the path integral representation of the matrix element ( X / . pfc) = Pfc • xfc . pk) = L(xk. The Feynman Path Integral Formalism yields automatically the correct normalization factor or metric in the configuration space path integral^ in agreement with Eq.H(xfe.V(xk) = m0±l .e.
*') = / I F / (2?FeiP"'x" / ^ 5x5 (p // K^> 5 )(. .108) (x 4 p)^^(px 5 )(x 5 e^ t 'x 6 )(x 6 p / )e.109) . x fdxG  ^ P <p"x3)<x3e^V) x (21.t"\x\x'.* * f dp = j (2^ p) ^ d .5 l^ t 'lp / > e . With the wave functions (21. .t'}.21. . (px). Thus <xV'xx'.6 Path Integrals in Dirac's Notation For the verification we use Eq.t'} = = fdx5x5(x"\eim"\x5)(x5\eim'\x'} f dx5x5{x". With contractions and integrations over delta functions the expression for the matrix element then becomes <xVxx'. (21.76) we can rewrite the factor (x4p)^(px5) as (x 4 p)x 5 (px 5 ).v ' x ' and thus {x".t"\x5){x5\x'.i p '• x .0 = J'^ j'^e^''(p"\e^''xe^'\p')e^'./ ^ I < « .107). (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.* * .77) and thus have 531 *!*') = *M = * 7 ^ I P > « . so that we obtain a cnumber position coordinate which we can shifted across other cnumbers.
*/<x"x5)(x5tfx'> + .tf subdivided into n transitions (xj+i. Inserting here the integral representation (21.• dx1 (x n . t0).i ( t " .tf — tn and x$ = XQ.£j to Xf. x / / e .0 dp x„ / (27T): 1 . (21.t'){x"\H\x') = X + •••} (21.i{t" .85) (xVxx'. ( x " .£/xxj.t n _i) x ( x n _ i .iit!' 1') 2mo + vtf) + <5(x"x').74) of the delta function.112) where x" is a cnumber. tixx 0 . (21.94) for the case with no operator in between.tj+i\x\xj.£ n xx n _i.t') (xV'xx'.x') . ] . £ n _ixx n _ 2 .113) Inserting for each of the infinitesimal intermediate transitions the amplitude .t"\x\x'. . (21. This is the expression corresponding to that of Eq. . We now consider the transition of the expectation value of operator x from Xj. The Feynman Path Integral Formalism In the second last expression here we expand the exponentials and obtain (xV'xK. (21.0 = and hence /dx 5 x 5 [(x"x 5 )(x 5 x') .£j) = / ••• / dx n _idx n _2 . Thus with x^ = xn..111) 1 . = x"6{x" .x') .£'> = x"[«5(x" .f ) i f e i p . we obtain (here and in the following the limit n —> oo at the end being understood) (x/. tn2) • • • (xi.x ' ) (2^ i^+vM + D «p(x"x') (21. we obtain with (21.110) Recalling Eq.90) we can rewrite this expression as (x".i(t" . We now proceed from there along parallel lines.ti = to as before.it / / x / (x"ffx / ) + it'x"{x"\H\x>) + • • • .532 CHAPTER 21.t') dp .i t " ( x "  #  x 5 ) ( x V ) +.tj) at equal time intervals e.
the string of factors xi X2 .tj) = / JXi 2){x(i)}x(i) exp i I I Jti dtL{.2 .2 .89) is consistent with the conjugate expression dL (2L116 * .t') .t"\x'.t') = and con 5x"Vx„{x".114). tj  xj1 x». x n _ i x n reduces to the single factor x^.. We demonstrate first that the operator representation of P given by Eq. For simplicity we consider the case of one spatial dimension.tf\x.21. (21. dyi l xi x 2 .. (21.117) "•"One may observe here that if x = x^ = x„ in the integrand of Eq. tj) = xf (x/.t"\x'.x n _ i ) ] . U)..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.( x „ ..{^( P i.105).ti) 533 rxf=x(tf) = / • • • / dXidx. Before we can consider canonical quantization in the context of path integrals we have to investigate in more detail the role played by momenta.iyxxj.\Xi.10) or (21.t"\x'.115) where (there is no integration over x n ) nl 0){x(t)}= l i m [ i V ( e ) ] .t"\x'.x2)(t2 ~ h) + • • • + H(p ••• + p „ .t') = (x" + 5x".xi) + f f dpi dp2 dp n _i dp n The momentum integrations can be carried out as before and we obtain finally (x/.114) x exp [i{pi • (xi . dxnn _i Xi X2 .x 0 ) + p 2 • (x 2 . .t') sider the variation S(x".3 " n d x f c .and hence may be put in front of the integrals to yield (xf.. x n _ i x n i JXi=x(ti) /Xj=x(tj) 7 ..« • > We reconsider therefore the transition amplitude (x".7 Canonical Quantization from Path Integrals (21. (21.112).y (27F (27F •" (2^(2^)3 exp[.{x".t') (21. . n—>oo *•»• 21..x 1 )(t 1 Z & ) +H(p2.t"\x'. .. we obtain* (xf.x . tj \ xt. Xn _i X„ xi dx. x (21.
118) JI(x)/h .t i_(dL_ x't' h\dx J t„ (21. 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.t') =  6— / 1 V°{x}t .t').(x". <9L' dt —dx + — 5 x ox ox f . V°{x}[i6I{x)/h]e* Jx' The following steps are familiar from classical mechanics: yV 51{x) S /1 dtL(x. (21. we separate the metric factors from T){x} in Eq. (21.t"\x'.118) — = (21 5x"(x". " (t). t". jt/ 37 ( £T. . " M{x)/h &(t ) = fa (L2) 2 11 and — using Eq. dx dt \ dx _ (21. For the purpose of enabling some formulations below.534 CHAPTER 21. (21. and we demand the validity of the equation of motion. Then.l&cdi. dt\Ox w(x) = r dt d^_dL(dL dx dt \ dx °L 8x + ^ T!5 X 1 ox (21.i/(x)/fi ) j ^ ) dtL(x. with N = N(t" .tt).119) The classical equation of motion follows from Hamilton's principle for which 5x = 0 at t — t'.x). dL r /•*" . Hence w « = (t).122a) = 17) 5x"Vx/.t"\x'. dL d5x / oft—5x + / dt dx dt Jt' ox Jt t' and hence <9L —oxdt + ox ox nt t.120) However: In our case here 8x ^ 0 for t — t" (although 5x — 0 at t = t').t'): 1 fx" 6(x". x) Jt Jt dL.10) and set D{x} = D°{x}/JV(i/ .
U) = T777TT / &{x}eaW\ IXi=x(ti) ) iV(A) JXi=x(U (21.t')(x'.tf\xi.x).122b) By comparison with Eq.. S. § The evolution of the system is expressed by the Feynman functional integral (xf. (21. 6 (21.115) in three space dimensions.t"\p(t")\x'.124) where with A = tf — ti. (21.7 Canonical Quantization from Path Integrals or JL(x»y\x. Here f*f D°{x} indicates that the integral is to be taken over all x with boundary conditions Xi = x(ti). (21.Xf = x(tf). 535 (21. F.t').1>) = ^{x".t'\ = l. The following properties are assumed to hold: (1) / Jti V°{x} = ^ x.126) R. I(x) = / JA dtL(x.125c) Corresponding to our relation (21.125a) (2) the completeness relation holds. / Jxi V°{x} / Jx' V°{x}. we now have in one spatial dimension {x^tflxlxuU) = J^ff'DQ{x}x{t)eiIWh. ^2\x'. Wightman [264].1 (A)iV.125b) and the measure property (3) N~\A + A') = A^. x' (21.1 (A / ).89) we see that Having explored the appearance of canonical momentum in path integrals we can proceed to extract the canonical quantization. We consider this here in quantum mechanics in one spatial dimension but parallel to field theory.21. Streater and A. .
(21.127) We can now write down an expression for a timeordered product of operators. (21.127)) (x".tf\x(ti)\xi.t') \ i ox" i ox"J or in commutator form [P.x](x. Thus if T denotes such ordering.t"\x'. (21. (21. tn lie correspondingly in consecutive order between U and tf.130) ~i5x7jX"{x".tf\T(Xl • • • xn)\xi.129) = ^ y With Eq.ti)..t').tf\T(xix2)\xi. Eq.ti) = x' 'Y^{xf.. (21.128) where tf is later than ti and t\..t') = ih(x.t'). .t"\x"\x'.ti) f ' V0{x}xlX2eu^lh.114) (xf.t') (21.133) = ih{x". Taking the functional derivative 5/iSx" of this equation we obtain 'ilx7'^' '^"^x"\x ''*') = (21.t"\X'. (2) and (3) above. Then the relation follows with the help of the properties (1).536 CHAPTER 21. (21.ti) = Xi(xf. We can rewrite this expression as the commutator relation (ITU*" x"^](x".132) The generalization to higher spatial dimensions proceeds along parallel lines. The Feynman Path Integral Formalism The time t is in general a time between ti and tf.ti) = —!— p V0{x}xx • • • xneu^h.t') + x"^J^{x".t"\x'.131) = ih(x". We have (cf. (21.t\x'. Thus (xf.t').122b) we can deduce the nontrivial canonical commutation relation arising here.t')=x"(x".tf\xi..t>). we have the relation (xf. We consider the case n = 2.t"\x'. But for t = ti we have (and equivalently for t = tf) as we can see from Eq.t"\x'.t\x'. (21.t"\x'.t'){x\t'\x2\xi.tf\xi\x'.
Chapter 22 Classical Field Configurations 22. An essential singularity is really an ambiguity rather than an infinity. i. W = e x p ( l / z ) . In physical applications such cases are rare. 102. nonzero radius of convergence in the domain of the coupling parameter. Expansions of this type arise. Whittaker and G. integrated over) beyond its radius of convergence. z = a + i/3 with a — oo or f3 —> oo or both. Such an expansion parameter is usually known to be or assumed to be small in some sense. It was therefore natural to search for alternative methods of expansion. > 537 . although the first few terms indicate a convergent behaviour.g.e. the exponentially small level splitting which occurs in "The expansion involves both positive and negative powers of the deviation. Besides. the series converges. such as the fine structure constant of quantum electrodynamics. Watson [283]. 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. N. i.g. p.e. If the expansion is mathematically well defined. In field theory it has been known for a long time that expansions in ascending powers of a large coupling constant are fairly meaningless. Consider e. the modulus of the ratio of the (r + l)th term to the rth term is larger than one. T.1 Introductory Remarks In field theory a perturbation expansion is generally understood to be an expansion in ascending powers of a coupling constant. for sufficiently large r. so that successive contributions to the first approximation do not invalidate this approximation. as we have seen in Chapter 8.e.g. Such expansions are strictly speaking.g. i. E. in the sense of rigorous convergence tests. it was known from quantum mechanics that RayleighSchrodinger perturbation theory by itself does not yield e. see e. it has some finite.e. but asymptotic. in the context of quantum mechanics are not convergent. The vast majority of perturbation expansions discussed e. divergent.
Many of these aspects are interesting by themselves .538 CHAPTER 22. and in general this change is accompanied by certain constraints.. The procedure requires a change of variables from the original ones to collective and fluctuation variables (in analogy to centre of mass and relative coordinates). the latter's topological properties if any. such > a series begins with a contribution proportional to (e.h. Classical Field Configurations the case of the symmetric double well potential or in the case of a periodic potential.. In the following we consider various typical examples. 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.. 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 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. One therefore faces the problem of quantizing a system which is subject to constraints.h2. In particular it enabled nonlinear problems to be studied and led to a consideration of topological properties. On the contrary. specific boundary conditions have to be implemented in order to yield these effects which are therefore frequently termed "nonperturbative". A method which achieves this is in particular Feynman's path integral procedure.) l/h2.h°. The study of expansions of this type has turned out to be extremely fruitful. In searching for other means of expansion. i. The fundamental extension of the method of canonical quantization (i. i. although this terminology is not precise. A further challenging task was the development of methods of quantization of theories which incorporate classical.e. such as soliton theory. the procedure discussed thus far) to systems with constraints was developed by Dirac [76] and is introduced in Chapter 27. However. cnumber.e.e. higher order corrections are of orders ±. This type of expansion is decsribed as "semiclassicar. The classical.h3. cnumber first approximations. These methods are often described as methods of collective coordinates. and the corresponding expansion is called the "loop expansion". the dominant contribution is singular (has an essential singularity) for vanishing semiclassical expansion parameter (h — 0).g. nature of the dominant approximation does not imply that this describes classical motion.. 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. There. and has led to insights which previously seemed unimaginable.
* assuming that this is acceptable to a reader with some familiarity of the basics of the more complicated field theory of electrodynamics. 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. may like to consult the article of D. The important aspect we want to draw attention to here is that of "spontaneous symmetry breaking". (22.1) ' T h e reader who wants a highly advanced overview of instantons.22. to keep this case in mind. 22. and also to see these in a somewhat broader context (by comparison with higher dimensional cases).2 The Constant Classical Field 539 and in any case deserve detailed study. Tong [272]. modified by going to imaginary time). . it is important for a better understanding of the considerations which follow. This means theories with field densities which are therefore infinitedimensional. It is evident that symmetries and their violation by the classical approximation play a significant role in the entire consideration. 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>.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.g. Thus references to field theory will be exceptional and hopefully do not irritate the reader. a static (time independent but spacedependent) solution of the classical equations of motion. V{</>) = m§0V + A0(<^)2. 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). d^] = d^d^ . t The classical first approximation in the procedure outlined above may be simply a constant (time and space independent) quantity. or even a solution to a modified field equation (e. However. vortices and kinks. 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. We shall not be concerned so much with the case of a constant first approximation. monopoles.V(4>).
£). To insure that the Hamiltonian H = f dxH with Hamiltonian density 7Y(0(x. For UIQ > 0 the potential V((f>) has a single well with minimum at </> = 0. cnumber) solutions <j)(x) = 4>c = const. B. Classical Field Configurations where d^ — (—d/d(ct).2) We ask: Are there classical (i. for m§ < 0 the potential is a double well potential with two minima at \<f)\ ^ 0. V[<t>] m 2>o o Fig. of the equation of motion? For these we must have all derivatives of <j) zero. i. .S£/<90(x. 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). m l + \04>*4> )$ = <). V ) .3) See e. 22. as illustrated in Fig.540 CHAPTER 22. £)) is bounded from below we must have Ao > 0. Felsager [91].e.1. 433. and hence here [d^ + m20}(l) = Xo(f (22.g. § (22. The EulerLagrange equation is d„ dC d(d.1 Different potentials for different signs of m§. (In models of the early universe one often considers V(0) with m\ > 0 at the early stage. p. then after cooling with a phase transition one considers expectation values 0 7^ 0 corresponding to minima at \<$>\ ^ 0.e. 22.
22. with 0 = 0. w A iP 3ml 5* "XT" (22.2 The spontaneously chosen phase.e. and so ft[<M=m6:(^* 1 + Ao(0*0)2. mg < 0. aq> For (b = const. H = —C = V.</>2)plane.5) . i. Fig. But for Ao > 0. we have 541 0. TT = —J. the equation has this invariance.2 The Constant Classical Field For Ao > 0. 22. (22. like the solution x(t) of the simple Newton equation mx(i) — —V'(x) violates the invariance of this equation under time translations t —> t + 5t (i. The Hamiltonian density H is defined by the Legendre transform H[<f>. We can convince ourselves that <fic is the field associated with a state of minimized energy.4) Here (3 is a spontaneously chosen phase like a stick held upright along H in Fig. Thus the classical solution 4>c violates the U{\) symmetry of C and of the equation of motion. whereas the field (bc becomes (a) A oi(0+<x) V~2 + Every new phase defines a different solution.e. 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.ir]=7r<fiC. the only such solution is the case we wish to consider.2 when allowed to fall chooses an unpredictable phase parallel to a radius in the (0>i. not the solution x(t)). The U(l) phase transformation <b — exp(ia)((> leaves both £ and the EulerLagrange equa> tion unaffected. TTIQ > 0.
Clearly these are the field configurations which trace out the circular bottom of the trough of the double well potential. the one for (3 = 0. For every value of the phase (3 we have a different constant cnumber field configuration 4>c. This is the condition (22.7b) In Eq.e.e. i. (22. 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). where (f>c = — T](x) = —= [Vl (x) + ilp2 (x)} • (22.g.7a) vanish on account of Eq.3ml = 2ml = ml + TA 0 A 2 = 0.e.2.2) and separating real and imaginary parts. 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. (22. one obtains the equations (d^ and (d^ + ml + ^ A 0 A 2 V = 0 ( ^ 2 .6) Inserting this into Eq. 22.7a) (22. and we wish to investigate the behaviour of the field cf> in its neighbourhood. i. . in a transverse direction.7b) the constant on the left and the coefficient of A on the right of Eq. V?) (22.3). ^ ) o + m 2 + 1A 0 A 2 ) ^1 = A (± A0A2 + m 2 ) + 0 ( ^ 2 . Suppose we choose one such configuration.e.tp2. >° for m 0 < °> . (22.3) obtained above. along the trough of V (which we can call a longitudinal direction). i. &H/d(fi = 0) if [m20 + ^>]</> = 0. (22. and partly by climbing up the parabolic wall on either side. as indicated in Fig. e. we set cf>(x) = <f)c + ri(x).Then m\ m\ = ml + A0A2 = m2. We identify the coefficients of the linear terms on the left hand sides with those of the masses of the fields ipi. Classical Field Configurations the density Ti is minimized (i.542 CHAPTER 22.
22. the existence of individual perturbation contributions).t'). (22. Thus the Green's function is similar to that in electrodynamics.22. irrespective of the question whether the perturbation series as such converges or not. (3=0 Fig.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. of course. We can see the problem here by looking at Eq. 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 £. whereas ipi(x) is y ' " " " " " ^ .e. In our later discussion of theories with nonconstant classical field configurations we shall encounter a similar. The appropriate Green's function G is the inhomogeneous solution of the the equation [8^ + m20]G(r. t. Goldstone's theorem applies to fully relativistic field theories. 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. then a massless boson exists.e.r')S(t . has to be removed in order to allow a well defined perturbation procedure (i.7b). t') = d(r . i. which. ^ = h eip / \ ^2 l . tangential to the classical path. 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. 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. There the wave function with an associated vanishing eigenvalue is called a "zero mode". though not identical phenomenon. Like the Goldstone mode in the above Higgs model it leads to a divergence in the Green's function of the theory. The elimination of the . r'.3 The spontaneously chosen phase seen from above.
To insure that the energy (see below) is positive definite we require V[$] > 0. We make one more assumption concerning V which. zero mass configurations and constraints even before the question of quantization of the fluctuation field T){x) can be considered. so that we are interested to have a parameter *For a collection of many informative papers on solitons etc. Behera and A.8) where V[<&] is a selfinteraction or potential of the field (to be specified later). The principal aspects will always be very similar so that it really pays to study simple models in considerable detail.t In particular the sextic potential can be considered along parallel lines.d^} = ~d^d^V[^) = ^2^(^j V[$].g. Rebbi and G. Khare [17]. J. . Lohe [179]. The Lagrangian density is taken to be C[^. spinless field $(x.£) in one spatial dimension. instead we shall consider static and timevarying solutions in important models of one spatial dimension. i.* 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. however. N. In the following we will not be concerned with classical cnumber solutions of the EulerLagrange equations which are simply constants. see e. MullerKirsten [291]. equates to zero the coefficient of this wouldbedivergence. Soliani [237]. *S. W. t Potentials of this type have been discussed by M.544 CHAPTER 22. 7711 = — 1 ( 1 6 Minkowski manifold). 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. Later we wish to develop a perturbation series about a classical configuration of $.* Potentials of higher polynomial order in $ than $ 4 can also be considered. Zimmerschied and H. C. A.e. we shall need only at a later stage.3 Soliton Theories in One Spatial Dimension We consider here the two basic soliton models known as <J>4 and sineGordon theories. and we choose the metric ?7oo = + 1 . (22. 22. F.
9) with quartic potential is defined by ^V\g$\ = ^V\g*.11) (22. (22. 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). 1]. In a quantum theory the field $ is an operator defined on a space of states. the socalled ^theory = ^V[g$. The first such condition to be imposed is that the energy of the solution of interest must be finite. (22.e.«{x) = V'(<p). . The field $(x. (22. ^ ^ * + V'[*]=0. the reason for this condition being that we visualize the classical solution as representative of a lump of energy.10) n*]  m4 r 25 i m2 and the sineGordon theory with cosine potential by ±V[g$} = \v[g$. The other condition which we impose (and study in detail) is that of stability. such that J > — — — = 0. In fact to begin with we restrict ourselves further and consider cnumber fields <> —• (f) which are static. but in such solutions which are subject to (ignoring the dimensional aspect) reasonable physical conditions.e. From Eq. 9 9 The EulerLagrange equation is seen to be (with c = 1) D * + V'[*] = 0. i. at least in the present case.§ Unstable classical configurations are also important and will be considered in later chapters.12) where the prime denotes differentiation with respect to $ . Before we consider quantum aspects we study classical cnumber fields which satisfy the EulerLagrange equation.l].l}. i.3 Soliton Theories in One Spatial Dimension 545 which serves as the expansion parameter of the series.t) has an explicit timedependence it is a Heisenbergpicture operator. (22. — m$ 1 — cos Im 1 I ^ $ (22. Since $(x.g] = \ v m For instance. (22.22.12) we see that they satisfy the Newtonianlike equation <j.13b) We are not interested in just any solution of this equation.13a) We write such fields <j>(x). For this purpose we assume that V[$] depends on a scaling parameter g such that Vm = V[t>.
7r) where IT is the momentum conjugate to $ defined by dC ? r = . HI tanh m(x>k) 9 Fig.4 The wellknown soliton of $4theory.16) In the static limit we obtain therefore and write this as Em = H[<t>\ + E(4>) = fdx 1 2 \dx 2 + V(</>) (22. We observe here already that the integrand of the integral in Eq.c = <£z + .14) where the overdot implies differentiation with respect to time t.17) is reminiscent of classical mechanics if we interpret x as time and V(<f>) as the potential energy of a particle at location d>. <9$ i.15) The zerozero component is equivalent to the Hamiltonian density H.17) Since we have to integrate over all space.T^ = 0. 22. .e.e.2\dx 2 ' 1 fd& + V[$].T = <I>. the classical solution will therefore have to be such that this integral is finite. i. d. (22.546 CHAPTER 22. Classical Field Configurations The energy JE?[$] of the system is defined to be the spatial integral of the Hamiltonian density 7f($. dC • Too = r?oo£ + =r$ = C + vr$ = H. n = & . (22. The energymomentum tensor T^ and the conservation law it satisfies are given by the equations dC T^ = n^C + Q^^d^. (22. (22.
In fact. — XQ).21) This sineGordon configuration is depicted in Fig. m ( x . i. in the case of the first. .5. See J. The energy is correspondingly obtained — as will be shown — as 4 m & 9 The typical shape of the configuration given by Eq.20) (22. one obtains by straightforward integration (but see also below) the soliton configuration m 4>c{x) = ± — t a n h 777(2. (22. In the sineGordon theory^ the classical or soliton configuration is (see below) found to be (x) = ±4— tan" 1 with energy 777 .x n ). "The name sineGordon is derived from the analogous KleinGordon equation or theory. Again XQ is an integration constant.18) Here XQ is an integration constant. 22. 22.13b) for the specific potentials (22.777.4. (22. the $*theory.10) and (22. it is not difficult to find classical cnumber solutions 4>c to Eq.11). 4m .3 Soliton Theories in One Spatial Dimension 547 As we shall see below. > . 22.1 .22.18) is depicted in Fig. (22. The sineGordon theory differs from the KleinGordon theory in that it possesses invariance under the shift rf> — <f> + 2n". O±m(xxo) (22. Rubinstein [240]. < = — tan [e >  ° ] 9 2 rrm/g — Fig.e.5 The soliton of sineGordon theory. around <j> = 0 the sineGordon potential behaves like the KleinGordon potential proportional to 2 4> .
Inserting the potential (22. 4m In tan tan 5 l/e±m(*xo)tan^£^)U. With we obtain g(/>(a. ± 1[tanh.e.1 "»JtKxo) 9 1 .* ( m' m 2 r)] ^2" . In the case of the sineGordon theory defined by the potential (22.18) and (22. and hence 4/71 9<t>{x.^^! m ^m <p(x0) m 4>(x) = ±—tanhm(x — XQ).20) are obtained from Eq.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. <^>(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 .13b) which we can write d \W? = !"(•».10) we obtain f<t>{x) x — XQ i 9 ! 1 = ± / J<j>(x0) dtf> 1 . or \(<P'? = V(4>)+ const.548 CHAPTER 22. 0(x) = ±4—tan" 1 j e * " 1 ^ " * 0 ) ! .9 2 4>{x) = i.o) =m7r. . Thus J<t>(xa) \/2V(6) JXOJxn U(x0) y/2V{</>) dx = x • XQ. (22. Classical Field Configurations The solutions (22. Here the constant is zero for V(<p) and <j>' zero at x = ± o o .
22) and that its second variational derivative be positive semidefinite. (22. The expressions (22. The kink solutions are also frequently described as "domain walls" in view of their analogy with the domain separating upward spins from downward spins as.17). SE((f>) 5cb(y) 0. However. Then we have to demand that the functional derivative of the energy E(<p) be zero. the curves rising monotonically from negative to positive values are known as kinks. however.21) for the energies can also be obtained from Eq. The relation (22. Jackiw [141]. (22. we obtain these expressions later (cf. the discussion of Bogomol'nyi equations) in a simpler way. 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.e. in the onedimensional Ising model. (22. the others as antikinks.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) .4 Stability of Classical Configurations 549 Prom their monotonic behaviour the solutions <f)c in these cases derive their name as "kinks". 22. for instance. A state of stability is therefore associated with minimized action and/or energy. . this does not require the action to be minimized.22.23) See for instance R. instead of doing this now.4 Stability of Classical Configurations The concept of stability can be complex. (22. i.19). Suppose now that we demand stability in the sense of minimized energy. This is called more precisely "classical stability J The EulerLagrange equation is obtained by extremizing the action or Lagrangian.
13b).e. 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). Thus we have to investigate the eigenvalue spectrum of the Schrodingerlike equation d2 + v"{<t>c ipk(x) = wkipk(x) dx2 (22.e. in fact. if 4>c{x) is a solution. Assume that S[U] is invariant under the transformations of some symmetry group G with elements g = exp(i\T) € G.e. This is.a / 0 is not the same. namely (p'c. in fact ip0(x) const. 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. it is also called a "translational mode". the function 4>c(x + a). i. and the EulerLagrange equation retains this property. (22. we obtain 4'(x) = v"(4>c ^ or d2 dx2 4>'c(x) = 0.28) . The Newtonlike equation (22. Classical Field Configurations Since y ^ ±00. The eigenfunction 4>'c{x) is for this reason called a "zero mode".13b) retains the invariance although the solution (f>c does not. The original Lagrangian was. (22. i. dx2 + V"{4>) (22. i. (22.550 CHAPTER 22. the integrated contribution vanishes. Thus the vanishing of the first variational derivative yields again Eq.26) Comparing this equation with Eq. if we apply the generator of spatial translations d/dx to the classical equation. its eigenvalues w\ > 0. (22.13b) with classical solution 4>c. written down in Lorentzinvariant form.24) For socalled "classical stability" the differential operator of this expression has to be positive semidefinite.25) Before we begin to study this equation in detail for specific potentials we can make a very important observation on very general grounds. we see that the latter has one eigenfunction. in particular the invariance under spatial translations.27) shows that the zero mode results from application of the generator of translations to the classical cnumber field configuration. In fact. of course. = 0.25). with eigenvalue w2. A theorem on zero modes Let S[U] be the action of some field U defined on Minkowski space. (22. Eq.A. dx (22. Since it arises from the violation of translational invariance by 4>c.
29) Suppose now the EulerLagrange equation possesses the classical cnumber solution Uc. We see therefore that the occurrence of these zero modes is a very general phenomenon. i. 0 = S'[UC] = S'[gUc}. (22. this integral has to be finite. in fact. (22.20) are such configurations in two particular models. we are interested in classical cnumber field configurations of finite energy. (22. As stated above.18) and (22.27). we obtain 0 = i^S'[exp(iXT)Uc]\x=o = S"[UC](TUC). i. lim [V(<f>c) + const.31) Inserting this into Eq. Invariance of S[U] under transformations of this group implies S[U} = S[gU]. (22. / oo Thus for E(4>c) to be finite. Thus TUC is a zero mode. (22.4 Stability of Classical Configurations 551 where T is a generator and A a parameter. (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. and that S[UC] is finite (or equivalently the energy).e. (22.22. it is an expression like (22.]. Then the vanishing of the first functional variation implies that 5S = 0.33) x—>±oo .32) = V'(</>). and we have seen that Eqs. We return to our considerations of stability. In the onedimensional case under discussion we can rewrite Eq.30) Differentiating this with respect to A (applying —id/dX) and setting A = 0. (22.17) we obtain oo dx[V(4>c) + const. 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. = 0.] = 0 or x—>±oo lim V{(/)c) = 0 for const.13b) as ~d~4> > ' The first integral of this expression is \(tfc? = V(<l>c) + const. We can choose the potential V such that the constant is zero.e.
c o and another for x — oo. In the case of the 3>4theory we can define a conserved current k^ by (22. n = 0 . .e. For a configuration to have finite energy. oo J—oo y The topological quantum number q is defined by 0. .34) jfc" = e^dv(l since d^k^ = 0. V{4>) = 0 at different values of </>). ± 2 . ± l .6.552 CHAPTER 22. . the classical field (bc must approach one minimum for x — . in view of the antisymmetry of eM^. 22. » In the <fr4theory defined by Eq.1).6 The minima of V in $4theory at x — ±oo. 22. This number is defined like a charge in field theory by the spatial integral of the time component of a current. the potentials have degenerate minima (i. Solutions which approach > > the same minimum for +oo and —oo are the constant solutions <f>c = ±m/g in the $ 4 theory. . Classical Field Configurations V Fig.35) and it is seen that in this case q can be ± 1 or 0. The charge Q is therefore given by dxk° = / dxe01di(bc = [ ^ ( s ) ] ^ = ±—. (22. The solutions can be characterized by an integral number called the "topological (quantum) numbcf^ "topological charge" or "winding numbed (for an illustration see Example 22. .11) we must have correspondingly for x — ±oo: » 771 9 2vrn. Thus the classical energy or vacuum configuration is not unique. In the sineGordon theory > • defined by Eq.10) we must therefore have that for x — ±oo: 4>c — ±m/g as indicated in Fig. (22. (22.
.4 Stability of Classical Configurations 553 Since we call q a topological quantum number.11) are given by ( so that for x — ±oo: > — J. For such configurations we have the energy oo 2 2 dx ^ A ( ^ ) + y ( A ^ / (22. 2g Thus even without varying E with respect to A we can see that E(X(f)c) is finite only for A = ± 1 . if it remains unchanged under continuous deformations of the (field) configuration while preserving finiteness of the energy. . zero is not a meaningful topological number. ± l ± 2 . termed topological.36) oo Consider the $ theory with 4>c — ±m/g for x — ±oo.A2)2 + 0 except for A2 = 1. one may ask how topology comes into the picture. The difference Tfl 4>c(oo) — 4>c(—oo) = 27T—An . 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.38) 4 V(\<t>c) >m ~2[l .37) for the onedimensional case here. Then for x — ±oo: > > > V(\4>c) (2 ^ 0 ) ? 4z( 1 . In the case of the sineGordon theory the minima of the potential (22. . Thus we expect the constant solutions (which are characterized by topological number zero) to be nontopological in some sense.22. Tfl \ (22. Again we see that E(\<f>c) is finite only for A = ± 1 . Such a smooth deformation (in the sense that exp(i%((2))<^c depends smoothly on 6.cos(27rnA)]. and exp(ix(0))<pc reduces to 4>C when x(#) = 0. Thus we consider the family {X(pc(x)} of (field) configurations where A is some parameter which assumes real. and we call this "topological stability. A property is. . continuous values.7T (22.7r) does not change a boundary condition lim [\<t>c{x)\ = (/>c(±oo). n = 0 . We can see this as follows. in general. x—>±oo so that under such deformations the configuration remains in the same topological sector defined by the boundary condition. First.
13b) or ^ ' ( * ) ] 2 = V(</>). where n±<XJ are the integers n corresponding to the minima of the potential at x = ±00.40) (22. J.34) is conserved independently of the equations of motion (other than Noether currents whose conservation follows from the equations of motion). pp.31) to (22. (22. (c) the time component k° is a spatial divergence. 425 . MiillerKirsten [215].43) **See e.39) E(cp) = J — 0 0 we can write / dx oo ~1 2(0'(x)) 2 + y(^) (22. its integral is nonzero only on account of nontrivial boundary conditions. if oo / dxy/2V{<l>)<f>'{x). Eqs.554 CHAPTER 22. 00 (22. Thus with one space dimension higher there.42) Thus the energy has a lower bound given by the right hand side of this inequality. (For a broader view we consider briefly in Sec. 00 (22.** (b) the time component k° depends only on <p but not on momenta (again in contrast to the case of Noether currents). the energy is minimized. W. H.41) dxy/2V{$)4>'{x). (d) the topological quantum number iV arises in a completely classical context.17)) f 00 > 0.^ = N. i. We observe (a) the topological current k^ of Eq.g. 22. (22.428. 22.9 the foregoing arguments in a higher dimensional context. From this we construct the inequality [</>'{x) = Vm^)}2 F Since (cf.e. Eq. (22.33)) that the classical configuration cf)c is the solution of the nonlinear second order differential equation (22. the factor A = exp[ix(#)] varies over a complete circle). The inequality is saturated. . (22.5 Bogomol'nyi Equations and Bounds We have seen previously (cf. Classical Field Configurations involves the difference An = rioo — n.
555 (22.44) This first order equation which saturates the inequality (22. (22. Here (V(c/)) > 0.n > 1. Eq. B.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— . Bogomol'nyi [27].^ Clearly it solves the second order equation or (22. We discuss this in the following Example. 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. n E . . 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.19). 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.40)) this occurs when 4>'(x) T V2V(4>) = 0. In the sineGordon theory we can consider.39).21).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. (22. for instance. (22.
with mass m and connected with short strings to neighbouring bobs. all being identical. rotate) only in the (y. Fig. H.7 The pendulum analogy.t) 4>{xn. Initially the pendulums are suspended freely in the gravitational field.t) 2 V dt The energy of interaction with neighbouring bobs is (recall "fcx 2 /2") k[(f>(xn+i.e. 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. z)plane and only such that their elastic strings remain taut. V dt „ 1 (d4>_ dx • i^gr(l — cos(j>) (22. Assuming that the pendulums can move (i.7. Classical Field Configurations Example 22. from the yaxis as shown in Fig.46) **See e. * / dx OO + 2^ V dt 1 K (d<t>\ 2 \7T I + ^9^(1cos 4>) (22. } n= — oo roc —> / L J . . 22. t)).t) <j>(xn.e. Construct an expression for the energy of such a classical system and its Lagrangian. rotate) only in the (y. 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.i)) The continuum limit is obtained by t a k i n g ^ (since xn = na. Chapter XI. Solution: Consider a string along the xaxis as described.556 CHAPTER 22.45) We can therefore write the Lagrangian L=TV• r dx fir P 1 2 . We can now write down an expression for the energy of such a system of (say) n pendulums.2 V dt + k[(f>(xn+i. so that oo > u. ka —> K.ij\2 Thus the The potential energy due to gravitation is (maximal for <j> = 7r) mgr(l — COS <f>{Xji. 22. total energy of the system is with the number of pendulums allowed to become infinite *= £ i ''( "' ) .cos</>(x„.t) measured e. xn+\ — xn = a) 00 a —> dx.t)\2 +mgr{\ .g. Goldstein [114]. At equidistant points xn = na along this string we attach strings with pendulum bobs. Assume that the pendulums can move (i.1: Classical analogy to sine—Gordon theory Consider a string along the xaxis.g.
t) is the fluctuation (field).50) This equation is identical with Eq.22. is therefore also called "winding numbed.49) would imply an exponential growth in the future t > 0 or in the past t < 0. We can understand what stability means in the present case. of the continuous curve in Fig. It is instructive to obtain this equation by yet another method.47) where n(x. which represents an expansion in terms of normal modes.7 linking the pendulum bobs). We are interested in studying (field) configurations $ in the neighbourhood of the classical solution cpc.t) = ° (22'48) (22. Hence we set ${x.13a) and (22. If some w\ were negative.48) becomes ( d2 \ X I [Q^2 + wl)Tpk(x) e*v(iwkt) k ^ X = k ^2v"{4>c{x))iljk(x)exp(iwkt). ' Equating coefficients of exp(—iw k t). wk would be imaginary and so the factor exp(—iwkt) in rj(x.t) = <l>c(x) + r}(x.25) obtained earlier. (22.t) is a small . (22.25) which we obtained as the condition of classical stability for eigenvalues w\ > 0. 22.49) exp(iwkt)ipk(x). 22.t): {w With the ansatz r)(x.e. Eq. we see that ipk(x) has to obey the equation 92 dx2 +V"{(t>c{x)) ^ fc (x) = w^k{x). Inserting (22. the topological quantum number. (22. each finite energy configuration beginning and ending with a classical vacuum corresponds to an integral number of rotations about the iaxis (i. t) = ^ k ~ iL2)71^ + V "^c{x))'q[x. t) of Eq. (22. (22.47) into (22.t).13b) we have for small fluctuations rj(x. and so would invalidate the procedure which assumes that rj(x.6 The Small Fluctuation Equation We now return to the Schrodingerlike equation (22.12) we obtain d2 d2 \ Since (j)c{x) obeys Eqs. The equation is called stability equation or equation of small fluctuations.6 The Small Fluctuation Equation 557 Thus if the classical vacuum corresponds to the case when all pendulums are pointing downwards. (22. This number.
and need 9 = tan" 1 j e±m^Xo) 1.20).18). cos46» = 2cos2(2(9) . 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. approaches a constant value which is nonzero in both the <54theory and the sineGordon theory. This means that the stability is not universal but only local in a certain sense.1.52) tan 6 := e±m^xo). For x — ±oo the quan> tity acting as "potential". t) have to be orthogonal to the zero mode (which is thereby circumvented) so that the Green's function of the expansion procedure exists. In fact. Inserting (j)c of (22.51) (b) SineGordon theory: Here 1 — cos m m 9 V"(<f>) = m 2 c o s f ^ .A \ Inserting <f)c of Eq. i.x0) = —2m + 6m 2 (l — sech 2 m(x — XQ)) 6m Am' cosh m{x — XQ) = (22. . In fact.50) has the general form of a timeindependent onedimensional Schrodinger equation. V"((/)c(x)).e. i. we obtain V'^c We have m cos ± 4tan" 3±m(:r—xo) (22. V"(<b) = 2m2 + 6g2(f>2. the zero mode. we obtain V"(<t>c) . (22. The equation of small fluctuations (22.e. we shall see later that the fluctuations n(x. Classical Field Configurations fluctuation. Thus for stability w\ must not be negative.558 CHAPTER 22. 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£^)*. If that is necessary.2 m 2 + 6m 2 tanh 2 m(x .
The eigenfunctions of the discrete case vanish at ± o o and are squareintegrable. we see t h a t the equation represents a Schrodinger equation with a potential which vanishes exponentially at ±oo.tan20 1 + tan 2 # Equation 559 1 — e x p { ± 2 m ( x — XQ)} I + exp{±2m(x — XQ)} sinh[=Fm(a. A  *(* + !) cosh 2 z tp(z) = 0.56) *We shall see later that the PoschlTeller equation is a limiting case of the Lame equation. and as a consequence a limiting form of the small fluctuation equation of all basic potentials. (22. ip+.e.55) The differential operator is even in z. — XQ)] cosh[±m(x — XQ)]: 1 .6 The Small Fluctuation Using the formula cos 20 we obtain cos 40 1 . We can therefore construct solutions of definite parity.ijj^ respectively. We can investigate the spectrum as follows.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 . Those of the other case are complex and periodic at infinity. even and odd solutions. ^+(0) d 0.2 (22. which satisfy the boundary conditions V>(o) = o.53) = m(x — (22.* For such a potential the spectrum can be b o t h discrete and continuous.22. (22.54) 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. . Consider the equation fd2 dz2 . i.
e.60) terminate after a finite number of terms. A * An = . A ^ An = .Z ) 2 . + l2) = 0.I) . when in the even case n ( n . i. 2n < Z.59) Setting (22.( 2 n + l .e.57) Then the equation for x{z) 1S (22.560 CHAPTER 22.( 2 n .l2 = . i.60) n=0 n=0 oo we find the recursion formulas x+(o = E ^ " > n(nl) = x(o+=l £ £&»£n. i.61) Discrete eigenvalues are obtained when the series (22. + \{\ ) 2 ~Q"m 1/2 oo «n+l (n + i ) ( n + l ) ( n + ^ ) ( n + ± .e. (22. .Z)2. (22.4 n ( 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.e.Z ) + (A + Z2) "n+1 (n + l)(n + §) &n (22. and in the odd case n+\)U+\l)+\(\ i. Since for £ — oo: * the function ^>+ is normalizable provided Z > n.58) = (1 + sinh2 z)~l/2x(z).Z ) + ^(A + Z2) = 0.
We can see the continuous spectrum of Eq. N = 0.e.2. (22.64) This is the equation of Jacobi polynomials V\a' (u/) with a = —ik of degree I in tanhz.j < I .. (22... (22.54) by setting X = k2 Then 1 .u2)—^V 2 dio and ip(z) = elkzV{to). m i. i. We can now derive the spectra of the stability equation for the potentials (22. for N = 0.( 2 .6 The Small Fluctuation Equation 561 and the function ip. (22.10) and (22.N<1.e. w2 = Am2.. (a) ^theory: the equation Inserting (22. We can summarize the results in the statement that the discrete eigenvalues are A = (ZAT) 2 . °2 i.Ny. 3m 2 . d w = tanhz.e.63) d + K(2ik  2LO)—V ' dijj' + 1(1 + l)V = 0.1.62) we see that the equation possesses discrete eigenvalues w —2 .62) where even (odd) iV are associated with even (odd) eigenfunctions.is normalizable provided 2(n + .63) we see that the continuum starts at A= 2 2^ = 0. m N<2.11).51) into (22.65a) _dz2 cosh 2 z l=2.50) we obtain with z = m(x — xo) tp(z) = 0.1: w2 = 0.\=Kl2 From (22.2Z = .22. From (22. (22. 2 . the continuum is w > Am2.
65b) dz2 cosh^J.27)) . .2 The discrete eigenvalues are now given by w ^l m^ = (lN)2. and — as expected — this eigenfunction is even. also be deduced from d(j)c/dx.50) we obtain with z = m(x — XQ) the equation (22. 22. m d .. for N = 0: to2 = 0 and the continuum starts at w2 = m2. .53) into (22. the zero modes. of course. This property can. It is now particularly interesting to look at their wave functions. Thus it has no node and so represents the eigenfunction of the lowest eigenvalue.. Thus 4>'c(x) has no node. We know that 4>c is a monotonic function. Classical Field Configurations (b) SineGordon theory: Inserting (22.66) We see that this is a nonvanishing function for any finite value of x. (22. m m ipo{x) oc —<fic{x) = ——tanhm(x — XQ) = g cosh m(x — XQ) dx g dx (22. 22. The zero mode is depicted in Fig.8 The zero mode as typical ground state. Eq.=1A=^_. . N<1. Thus either spectrum contains the expected eigenvalue zero. . %(x) Fig. In the $ 4 theory we saw that (cf. so that its derivative is nowhere zero (except at ±oo).8 and we see it has the shape of a typical ground state wave function. . i.e.562 CHAPTER 22. d .
6 The Small Fluctuation Equation 563 In the sineGordon theory the zero mode is proportional to . Finally we consider the normalization of the zero modes.68) d{mx) m(xx0) 4 m3 3 ^ g Ml J cosh / i. (22.67) J —( Mx)= and hence m w0i^ 1 2 4 m ±—tanhm(x — XQ).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. .3 .z 0 )]oo = 8 . (22.22.o) ±4 m g l __ e ±2m(xx 0 ) 1 cosh m(a. — XQ) OC Hence the conclusion is similar to that in the previous case.t a n h x . 1 o rrr tanh x — . .oc (22.e. (22.= E(<f>c). 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 . d . . 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. i.e. . without quantum corrections).
7 Existence of FiniteEnergy Solutions So far we have been considering the case of one space and one time dimensions. Classical Field Configurations 22.> 4>a{x) = 4>(ax). Affleck [5] who uses in his Eq. *See e. (2.* The general scaling argument used for this purpose was first introduced by Derrick t and is therefore referred to as Derrick's theorem.564 CHAPTER 22.T(4>). Kastrup [145]. for if such configurations exist in a realistic theory with three spatial dimensions. P. f G . They may help. J As a first example we consider the expression (22. (22.4) <j>(x) —• a<f>(ax). This question has important consequences. H. it is important to understand their physical implications.41) for the energy of the classical configuration but now with respect to a ddimensional position space.g.g. Olive [113]. i. ^For basic aspects of scale transformations see e. A. Under this transformation (the verification is given below) £(</>) ^ E(</>a) = T(cf>a) + P(</>a) = a^T^) since for instance + a" d P(0). however.2d. This difference may not always be appreciated later in our very analogous treatment of (topological) instantons and (nontopological) bounces. we consider E(<P) which we also write as (22. P(</>) = f ddxV{(f)) > 0. I. The arguments of Derrick were later rephrased by several authors whose line of reasoning we shall be using here. I. Goddard and D. Derrick [68].72) with T{$) = I' ddx]{V(j))2 > 0.71) We now consider the scale transformation (with 'a' some number € R) 4>{x) . One can have different types of scale transformations — see e. .e.73) n<(>a) = j ' d^V^)2 = a2~d / dd(ax = J' ddXl{V4>(ax)f d 4>{ax) 5ax .g. We now wish to inquire whether static finiteenergy configurations could exist in more than one spatial dimension. ddx {V4>f + v{4>) (22. to underline the fundamentally different nature of topological and nontopological finite energy configurations. *The considerations of this section extend in part beyond quantum mechanics and into field theory. H.
since d2E(<pa) da2 2(2d)T(<f>). i. a=l (22. Next we consider more complicated cases.71)..76) we see that E(<f>) can be minimized (i. we write • = lim J r . we set 0 dE(<pa) da J a = i (22.74) From (22. Can the energy also be minimized for more than one spatial dimension? To see this we stationarize E(<f>a) with respect to a.e.77) with V(4>) > 0 and x G Rd and the set of minimum configurations Mo := {4>\V{4>) = 0}. the second derivative is positive) only for d = 1. to be finite the field <P ~^ 4>oo £ Mo for x — oo.73) we see that this implies (2 .79) In order to emphasize that (f> — (p^ in any direction.x2. (22. d = 1 minimizes the energy.xd)...d)T(<f>) = dP{4>). i.75) Since both T and P are > 0 and d is integral. Thus this scaling argument excludes the existence of static finiteenergy configurations of the given theory in more than one spatial dimension. x = (x1.e.80) . It is also a virial argument in view of the virial theorem relation T{<j>) = P(<f>).e. 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.. (22.78) For the energy of a static configuration. (22.7 Existence of Finite Energy Solutions 565 We saw that the static configuration with a = 1.\ eMo(22. (22.22. In fact. Consider first the Lagrangian density £[&M = \{d^)(d^) . 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}.V(<P) (22.
e. Classical Field Configurations In the sineGordon theory (d = 1) the equivalent statement is (22. of • course.e. —00 or equivalently S = + 1 . S1. is a connected region. therefore. (ptx = <f>oo.1 manifold Sd~1\d=i <• manifold Mo <—> 0 O = 27rn^ Q <—> 0_oo = 27rn'^ kink limits (nn'/ 0) const. Table 22.and sineGordon theories we assumed $ to be real. = —00 : S = — 1 <—> ^oo = 27rn^ The topologically trivial case is given when the classical vacua associated with S1 = + 1 . and. — 1). We see.566 CHAPTER 22. i. .38). as is the case for the constant solutions. x = 00.e. that we have a set of disconnected classical vacua (</>oo><^oo.81) and (since d = 1) S = ± 1 . one set can be mapped into the other.1. If we go to two spatial dimensions (d = 2). 0oo 7^ <>< > in the case of the kink solutions) and /x associated with this a set of disconnected points in space (00. i. i. the easiest way to visualize points at infinity is to take a circle Sl of radius r and let r — 00. are the same. —1. the Lagrangian of these theories therefore does not possess the U(l) symmetry of the complex Higgs model but instead invariance under replacements $ — — $ and $ —• $ + 2TT > > respectively). ( with — = lim <pc(x) 9J *>±oo TYl \ (22. in fact. Thus in order to obtain a topologically nontrivial case we must have a different classical vacuum in association with each point at infinity.1 (A) Topologically nontrivial mapping x = +00 : S = +1 x = 00 : 5 = . Thus in this case S^1 is a disconnected set consisting of two points. Then each point at infinity is characterized by a different value of the polar angle 9. —00. (n — n = 0) (B) Topologically trivial mapping x = +00 : 5 = +1 <—> <> o = 27rn^ /o a. These observations are summarized in Table 22. 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 .
if we consider a theory in three spatial dimensions and so the unit sphere S2 = {x x{ + xl + xl = 1}. cj)(r. Then the entire set of configurations (f)^ = \4>oo\ exp(i#) would correspond to. map into. assuming invariance of C under the transformations of the group SO(3) in internal space. if cj> were real and position space twodimensional. i. Similarly. We see.e. Thus. * n reach any other <^(oo) by simple phase shifts. i. i. the manifold of classical vacua would have to possess a similar geometry if the theory is to be topologically nontrivial.83) . we have w e ca {Mx)}> and 1 3 1. phase transformations 6Q ^> 9Q + 59. 9) would depend only on r = x and hence the problem would be effectively onedimensional.3 (^real) v{4>) = / (22. Thus. $(r) : Sl . the one point S = + 1 .82) Now Mo := \ 4>(oc) ^24>2 = a2\. given any (00) = «Koo)e* = <?[<p(xMx)a2]2 567 say with 9 — 9$.22.e.e. that if the problem is to be topologically nontrivial. If position space were onedimensional. we would have only two points at infinity.1 . On the other hand. (22.7 Existence of Finite Energy Solutions Suppose.e. . 00. —00 or S = 1 . if <p is real. therefore. VMx)) with (0(oo) = linvxx) (p(r. to have topologically distinct classical configurations.• M0. by allowing <fi to have three components in some "internal" (isospin) space. 9)) Mo = {</>(oo)</>*(oo)0(oo) = a2}. the field must map position space into field space at infinity.e.g.2. for instance.e. This can be achieved e. i. i. This will be "nontrivial" in the above sense only if 0oo is independent of 9. i.
e.2: Derrick's theorem applied to gauged theories Wellknown and important theories involving a gauge field (AM if abelian and A".. Georgi and S. % . circle. The symmetry of the potential (and Lagrangian) — i. through each point at infinity). $ — $ + 27 > • T in sineGordon theory. Example 22. Classical Field Configurations The manifold A4Q is a sphere in a 3dimensional Euclidean space.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) £(</>.2. going once around S%~1 (i..e. it is effectively a surface (and so a continuous set) S2. If. the subsequent discussion requires only the general form of the Lagrangians. Olesen [220]. A) + T{A) + P{<j>). + i ( D " f t W ) a .a M J . we cover <S^_1 n times. invariance under replacements $ — —$ in $ 4 theory. 11 H. Any 4> £ MQ can be reached from any other cf> e Mo by application of an element of SO(3). .Ta} in an 50(3) invariant theory and so on — determines the transformation from one point at infinity to another (on a line.568 CHAPTER 22. the winding number is n.Av  dvAf i where 0 is a complex scalar field and the spatial dimensions are 2 or 3. depending on d) and correspondingly (with n complete windings) the transformation of one classical vacuum into another.e.V{4>a].e. a = 1. a theory being described as 'gauged' if a gauge field like the electromagnetic field is involved. i.„ where G r = 9 M £ . B. if SO (3) nonabelian) are two models defined by the following Lagrangians which we cite here solely for the purpose of being explicit. In order to preserve lim y ^ i a\ such rotations must tend to the identity at r = oo. Glashow [108]. Summarizing we see that </>(x) is a mapping from S^T1 to <S^1. sphere. . Nielsen and P. i. A) = T(4>. $ — > > $ exp{i Xlo=i Pa. so that details of these models are irrelevant. In the following Example the use of Derrick's theorem is demonstrated by application to some gauged theories. 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. $ — $exp(ia) in a U(l) symmetric theory. the appropriate rotation.</>] = ~G^Gai. (b) The GeorgiGlashow model I! with nonabelian gauge field in 3 spatial dimensions with Lagrangian density (V as under (a)) £[AM.3. L.
i.87a) C?M„(x) — \2G^{\x).84) in this way one finds that the quantities involved transform T ( ^ .A).e. l j ddxGijGai] > 0. We now consider scale transformations with A £ R.84) is positive semidefinite and so must be finite by itself. it must be stationary with respect to arbitrary field variations and so with respect to the above scale transformations also. (i)^ 2d T Considering the contributions of (22.<^[x) transforms like a vector. A) + (4 .Ax) dX Since T((f>.d /" d d x ' a</)(x') <9x' d</>(Ax) 3x M A 2 /" ^ ( A z ) d<p(Xx) A / A^ d(Ax M ).86) (22.d)T(A) .89) P{4>) are positive semidefinite and not all zero. A. T(A) = f ddxV(4>) > 0.85) Here the difference in the transformations results from their difference as scalar and vector quantities which are defined by the transformations ^V) = 0(x'>. >• A„ x ( x ) = AAM(Ax). (22. i. (2 .d T ( ^ . some contribution must be .e. A). > Since T>IJ. negative. ^ A ) = A 2 .. and so B ( ^ A . Thus we set dE(<fix.A) P{4>) 569 j / i F ( « ( P ^ ) t ( P ^ ) „ > 0. J 4 ) + A 4 . <0 >• <£A(X) = <KAx). (22. Solution: Each contribution in (22.84) and investigate the existence of finite energy classical field configurations.d T ( A ) + AdP(</>).22.87b) We consider explicitly the example of <j> — </>\(x) = <p(\x) for > T ( V(4>):= Jddx(d. (22. (22.i T(AX) = X4~dT(A). we have {^/»)A^ (22.T{A). = 0. (22. A A ) = A2dT(^.7 Existence of Finite Energy Solutions where T{<j>.dP(4>) = 0. Mx') = E with a scale transformation x — x' = Ax.d)T(4>. P(tf>A) = A " d P ( 0 ) If £ is to allow a static finiteenergy solution. 4>Y We have T«(4>A) = J ddx(d„4>x)2 = J dd .
Ginzburg and L. R.e. There are other posssibilities if we permit higher powers of derivatives as in Skyrmion models. An explicit and solvable Skyrmion model in 2 + 1 dimensions is considered in D. .8 Ginzburg—Landau Vortices We have already referred to the NielsenOlesen theory. a > 0. T^{<t>x) = \4dT(</>). Tchrakian and H.89). We first demonstrate — as a matter of interest — that the NielsenOlesen theory for static equilibrium (i. Landau [111]. if we allow a term T^ 4 ' {<f>) with fourth powers of derivatives. L. 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. H.89). tt The — now established — microscopic BCS theory of superconductivity was formulated some eight or so years after the macroscopic GinzburgLandau theory. This case is. (22.g. Equilibrium states of the superconductor are assumed to be described by the timeindependent static wave functions ip(x).t) such that  ^  2 describes the density of Cooper pairs. Thus such a contribution could counterbalance T{4>. (22.** e. A) or P{4>) for d = 3. One can thus define a scalar field or wave function ty(x. 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.570 CHAPTER 22. 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.92) The expression ^ ( x ) ! is known as the "order parameter. H. exemplified by soliton and sine—Gordon theories. of course. Skyrme [253]. (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). a = a(T) = a T —T c . (22. (e) For d > 5: In this case it is not possible to stabilize Eq. 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 (22.89) the contribution (4 — d)T(4>). ft V .90) and we would have to add to Eq. (b) For d = 2: A theory with all terms would be a candidate — in fact the NielsenOlesen theory is an example. We can therefore use our understanding of the BCS theory in interpreting the GinzburgLandau theory. (c) For d = 3: All three terms must be present as in the GeorgiGlashow model. (22. J. W. 22. MiillerKirsten [268]. timeindependent) field configurations is equivalent to the GinzburgLandau theory of superconductivity. In the following we consider a related theory.
i. is a measure of the distance from the surface of the superconductor to where . Then the static energy density becomes (cf. Sec. i.A) = iB 2 + i iXQ + 7 + ^ H 2 + /^4.^ 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. The parameter 7 serves to put the minima of the double well at V = 0.(tp) = 0.22. ^(00) = a (22. (22. 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. (22. 22.94) the solution is (in the domain 0 < x < 00) i>{x) = ^/^tanh(^x\ .93) J' (22. Returning to the three dimensional case and switching on a magnetic field (applied from outside of the superconducting material. setting S£. from x < 0) means — as is familiar from electrodynamics — replacing V by V .95) This expression can be shown to be equivalent to the integrand in the NielsenOlesen theory.2) £(^. Example 22.e. but for T < Tc it has a double well shape.8 GinzburgLandau Vortices 571 of the complex Higgs model (cf. Clearly the length (see Eq. Variation as before. which we recognize as one half of the kink solution we obtained previously as the topological instanton solution of the Schrodinger equation with double well potential (in the present case ip(x) = 0 for x < 0.e.2): For T > Tc the overall potential has a single well (minimum). assuming there is no superconducting material in this domain).94) to check the dimension) 1 = I—a.
MiillerKirsten [215]. p. Classical Field Configurations superconductor (x>0) Fig. H.g.9. 450.9 Behaviour of ip and B in the superconductor. 2mi Setting < / > \P\e vp B = ViA. yJ—a/[3. The magnetic field on the other hand. i. h [^DV'V'DV].98) where M is the magnetization resulting from atomic currents. In the Maxwell equation See e. The density of the Cooper or super current is given by the typical expression of a current.572 no superconductor CHAPTER 22.e. (22.e.M.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 . as indicated in Fig. i. 22. J. the order parameter \ip\ attains (approximately) its asymptotic value. 22. exhibits a completely different behaviour — in fact. (22. We can see this as follows. W. V x M = j s .96) ipH. (22.e. i. . 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 ).
div grad" and V • B = 0.99) Inserting (22. This is zero in the present case.. those of the socalled type I and those of type II. Hence 0 = V x H = V x ( — B . Approximating p by its equilibrium value. 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. (22. These vortices therefore carry magnetic flux which is necessarily quantized . 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.5 to 10~6 cm). There are two types of superconductors.L B = O. Also dD/dt = 0 in a static situation like the one here. Using the relation "curl curl = grad div .8 GinzburgLandau Vortices 573 the quantity j is the density of a current applied from outside. and V x (V x B) = / J 0 V x j s .e. i. (22. In the onedimensional case we have B{x) = B(0)ex/XL. K i ™*.e. V x B = [i0V x M = /lois.100) we obtain 2 m ( 0\ The ratio XL/XQ is called the GinzburgLandau parameter. 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. this becomes A B = no V x j 5 . m xi A B .22.M i.
See also P.37). (22. A. We can think of (e. 'See the remark in H. *A. 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. (22. .101) (p(r. Abrikosov [2].g) twodimensional space as being bounded by a circle at r = oo. Finiteness of E requires \<t>\ = \<fioo\ on this circle. 22. de Vega and F. In the two models of Example 22. G. in other words.9) ^e^ '>\<poo\.2 we saw that finiteness of P(V) puts no restriction on the phase of (f> so that r e (22. the space consists of two diametrically opposite points on the circle. 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).36). In all the models we considered so far we observed that the energy E contains a contribution P(V) = j ' afxVM).e.102) In the onedimensional soliton case of Eqs. De Gennes [67].9 Introduction to Homotopy Classes In this section we consider topological properties of classical finite energy configurations and introduce the concept of homotopy classes. by the element exp[i%(0)] of the internal group A(s) = e * W : SI+SI. i.9). 6')r_>00 represents a mapping from the circle at spatial infinity to the circle defined by x(#) (or.574 CHAPTER 22. Classical Field Configurations (see below). J. x—>oo Thus at spatial infinity </» has the value of a zero of the potential. of two points on a line. i. Thus the asymptotic field </>(r. Schaposnik [69] after their equation (3. (22. where exp(ix(0)) is an arbitrary phase factor. A.e.. The existence of these vortices has been confirmed experimentally.2 — identified as above with the GinzburgLandau theory — type II superconductivity occurs^ for A > 1/4 and type I superconductivity with the complete MeissnerOchsenfeld effect for A < 1/4. In the NielsenOlesen model defined in Example 22. we must have that lim V ( H ) = 0. in fact a periodic function of the polar angle with period 27r.
Thus smooth deformations *See D. 0 < 0 < 2TT.22. 2n] remains continuous if the function assumes the same value at the two endpoints. 22. i. (22. We may define this as « = ^[X(0 = 2 T T ) . The map satisfying X(x + y) = \{x)\{y) is said to be a homomorphism.1. The 1sphere is the unit circle in the space of all complex numbers. States of the strip corresponding to winding numbers 0 and 1 are illustrated in Fig.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". . Then a continuous function of (j> on [0. W. defines the orientation of this line element. Sl := {zeC\\z\ = 1}. The map A : S1 > S1 defined on S1 C R2 by (22.118)).e. 22.* We can think of this as a fixed line in space which is provided at each point with an infinitesimal direction element that is free to rotate in the plane perpendicular to the line (like the set of pendulums considered in Example 22. topological group with the usual multiplication as group operation. Misner [96].e. X is called the domain of / and y its range.X ( 0 = O)].102) is called the exponential map of S1 onto S1.1). (22.10 States of winding number 0. commutative).10. The two theories we considered above in Example 22. In mathematical language S1 is therefore a compact. abelian (i.2 imply that such a map from a circle to a circle is associated with an integer which we called the winding number n. Finkelstein and C. Naturally an integer cannot change continuously.103) Later we demonstrate that this winding number n remains unchanged under continuous deformations (those of Eq. Fig. Suppose <j>. We can illustrate the simple lowdimensional example under discussion by reference to a strip with Mobius structure.
Classical Field Configurations of the fields which preserve the fmiteness of the energy must preserve the winding number n. connectivity. no hole.) remain the same under homeomorphic transformations and are therefore called topological invariants.105) the corrections falling off faster than 1/r.576 CHAPTER 22. since otherwise the energy integral would behave like f d x f dr dr 2 J r J r lnr. A field configuration with n ^ 0 therefore cannot be deformed continuously into one with winding number zero. r^oo 1 dx(9) n r o0 (22.e. The gauge field required in the asymptotic limit (22. i. we can write 1 d h J ee rde w At h r^oo . Homeomorphism subdivides the possible spaces into disjoint equivalence classes (to be explained below). i. i.105) is described as "pure gauge" since it is determined entirely by the phase x(#). The different kinds of connectivity of these topological spaces (e. it retains this twist or hole under the continuous distortion) and is described as a homeomorphism. (22.g. In other . one hole. elements in the same class have the same topological characteristic. We observe that since time evolution is continuous.e. The continuous deformation or distortion of a set of points does not alter the topological structure of the set (e. if it has one twist as above or a hole.g. Thus two homeomorphic spaces have equivalent topological structures.i'ldx(0) 9A Thus for (22..104) to be finite we must demand that g . it cannot be deformed continuously into the socalled "classical vacuum" (the latter is a constant configuration in our soliton example). the winding number must be a constant of the motion.. 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. finite. The winding numbers n are such topological invariants.e.. For that reason n is called a "topological invariant?.104) so that d2x hJ Since A has direction e#.
we can go n times around S^. A field configuration with one loop (or twist). going once around S^.105) insure that B ^ O . is also zero at r = oo. (22. with n = 1.110) . the vector potential A. i. it is determined by a gradient (l/r)d/dd.9 Introduction to Homotopy Classes words. cannot be pure gauge everywhere. An example is the mapping H(t. Then we can construct a oneparameter family of maps or transformations which transform </>n (0) continuously into (j>h (6).106) J g j r d d g Thus the magnetic flux is quantized. into S i . (22.108) map the spatial circle S^. the number of flux quanta 2irh/g being the winding number n. if A = V x .22. i. cannot be deformed continuously into one with two loops (n = 2). However. It follows that the field Fij. 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". In terms of the phase factor \{x) = eixie\ we can rewrite the relation (22. We can calculate the magnetic flux $ through the region with $ = * rd9A0 = . we have B = V x A = curlgradx = 0. the group space of (the phase transformations of) U(l) continuously.9):=t(f>£\9) + (lt)(f)W(9) with te[0.109) A W(0) = e^ W. Therefore.e. Of course. (22. the higher order contributions in (22.107) The fields <f>(r.e. 9){^ and hence the functions (22. we can consider transformations which describe continuous transformations of a configuration with fixed n. i. the gauge field.e. Suppose two such configurations are given by O) W(9) = ei&)V\ n fixed.105) as «A(I)'AWV*<I> \(x) = J*V\ ( =• VxW)) (22. Ffiu. However.l].<b rd9—4^ = n. for B / 0 .
f T h e standard texts on homotopy theory are N. n' ^ n. Reuter [77]. Thus by varying t from 0 to 1 we deform (fin (9) continuously into (fin '(8). Classical Field Configurations which is continuous in the parameter t and is such that H(0. (fin (9) in the above example.112) is an equivalence relation — and as such it is defined by the three properties of (a) identity.3: Homotopy an equivalence relation Verify that the homotopy map H(t.111) In principle an expression like (22. the phase factors of (f)(9).9) is called a homotopy (emphasis on the second syllable). i. i. (b) symmetry and (c) transitivity. n.e.113) . We can consider (22. (22. Steenrod [261] and S. A superposition of classical configurations with different winding numbers does not minimize the energy. We verify these in the following example. Still on the level of a mathematical text is (in spite of its title) the Argonne National Laboratory Report of G. are said to be homotopic (emphasis on the third syllable) to each other and one writes ^(0)^^(9).0) = $\6).111) as a map from the space described by 9.112) are homotopic iff there exists a homotopy connecting (fin (9) and (fin (9). like <pn (0). The map H(t. S%.1].g. A brief introduction can be found in the very readable book of W. e. — S\ > (22.e. Since the passage of time is continuous. But this type of combination is excluded sind H(t. 8) = cfinl\9).T. Two curves which can be related to each other in this way.9) = <fiW(9) and H(l.578 CHAPTER 22. Example 22. (22. H. Dittrich and M. these two maps (fin from S^. The relation (22. it is clear that this is also a homotopy. to the internal group space of £7(1). and the space of t. time cannot make a field jump from one homotopy class (see below) to another. H. i.e. Solution: As stated. i.lQSl^S^ with H(0. Hu [134]. 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. 9) is an equivalence relation. H(1.e.e.110) could be written down for configurations with different winding numbers. Thomas [270]. (b) symmetry and (c) transitivity.8) itself must be a configuration with definite winding number. the unit interval I = [0.9) = ^°\0). we have to show that the map satisfies the three properties of (a) identity. i. In fact.
0). Assume <p( ' ~ (j> and <p( ' ~ <> ' /( for continuous maps <^(0'.0) for for 0<t<.t.22.e) tf2(2il. We now define the function H ( f 0 ff .9) = . we see that of the set of these maps some are equivalent and others are not. Hence <f>(0^ ~ <p 2 ). 9) := H(l . i<t<l. which therefore demonstrates that the homotopy relationship is transitive. On the other hand H(0.e. Hence the set of all such maps breaks up into equivalence classes or homotopy classes. 0) = tf2(l. Then there exist homotopies Hi and Hi connecting <j>(0> with (j>W and t^ 1 ' with </>(2) respectively.ffi(O. then </>W ~ <t>{0)• We define an H(t. In the above case of 7ri[[/(l)] we know from our earlier considerations that every equivalence class is characterized by a specific value of the winding number. 9) connects <pn (0) with itself continuously. (c) We can demonstrate the property of transitivity as follows. being the continuous map connecting </>(°) with <j>(1'>. below).0) = </>(0) is continuous there. <^(2). 0). <f>W .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. 9) = H(l. In some cases the set of homotopy classes possesses group structure (satisfying the group properties). at t = 1/2 both i?i and H 2 give c/^1'. . 0) = 4>W (0). 0) = <£{2). so that H(t.9) = <f>(0\6) and H(l. Two elements of 7ri [f7(l)] which belong to different classes are homotopically inequivalent. 9) = H(0. Moreover.9) H(0. negative or zero). 9) = <f>W (0). is such that H(0. ' . U(l)). i. so that H is a homotopy connecting <jyl>{0) with <f>^>(9).* 22.10 The Fundamental Homotopy Group 579 (a) Evidently the homotopy H(t. 0) = ^ (0) and 6(1.e. a class being a set of equivalent maps. (b) Next we check that if 0<°) ~ <p^. \ M)\ H!(2t. Then H(t. Homotopy classes are sometimes also called Chern—Pontryagin classes. Thus 7Ti[C/(l)] = vrifS'1] corresponds to (i. is isomorphic to) the set of integers Z (positive. (22 1U) At *=2!: But ^ ( ^ ) = ^ i ( M ) = #2(0. these groups are then called homotopy groups (cf. i? is a homotopy connecting </>(°) with <^>(2'. . Returning to the maps \(x) : Si ^ S1^ (i. 0). as they are called. since and tf (1.
T < 9 < 2TT. Sketch of idea demonstrating that nilS1] = Z. 22. How can we see this? We first sketch the idea. Fig. which is to be mapped into another circle Si indicated by dashed lines. but TTI[S2] = 0: S 1 is a circle. Classical Field Configurations Thus there is a denumerable infinity of such homotopic classes or sectors.1] continuously to 0 one regains xo(#) = 0(b) In the cases n = 1 and n = 2 we have Xi(0) = 9 and xi(0) = 2e for al1 e  Similar considerations are given in the wellknown book of R. This trivial map into a single point is described as a "degenerate map" and is illustrated in Fig.11 the circle drawn with the continuous line is the circle S%. p. characterized by winding numbers n are for instance (Xn(#) being a continuous function modulo 27r): (a) For n = 0: Xo(0) = 0 for all 9.115) Varying t € [0.11 for the special values of t = 1 and t = 1/2 with Xo(0) = to t(2vr 9) for for 0 < 9 < 7T. In the diagrams^ of Fig. and then introduce some of the related mathematics. 22. 22. 52. to make the result plausible. Rajaraman [235]. .e. Appropriate maps.) why 7i"i[S2] = 0.580 CHAPTER 22. n=0: t=1: t=1/2: i. T (22. S 2 the surface of a sphere in a threedimensional Euclidean space. In order to really appreciate how topology comes into the picture here one should understand (e.11 The trivial n = 0 map allowing shrinkage to a point.g.
(22.117).I** dOJn*W±e {8) X 2vr J0 — / 2Wo dO(iri) dO \ d6 l. .103)) A(x) > \'{x) where 5f(x) = \{x) + i\(x)5f(x) = A(x) + 6\(x).s. 2 the maps are wound around the dashed circle.116) In fact from this we obtain (see below) 2  n and this means r. 22. 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. 0 < X(P) < 2vr.117) 2^ d6X(x)\1(x). The maps for these cases are illustrated in Fig.117) in ±. 22. is invariant under infinitesimal (continuous) deformations.e.22.12.e.h. to the case of winding number zero. d (22. of (22. the transformations^ (the result may be looked at as 8n of expression (22.12 The n = 1 and n = 2 maps not allowing shrinkage to a point.118) is an infinitesimal (continuous) real function on the circle with (5f)o=o = (Sf)0=2n 1 See S. and hence the winding number. phase factor or element of U(l)): A(x) : Xn(x) = em*W = [Xl(x)]n.e.h. (22. i. In the cases n = 1. One can prove that the expression (22. Coleman [55].10 The Fundamental Homotopy Group 581 n=l: cutting of dashed circle n=2: Fig.s. i. For winding number n we have the mapping (i. We make another observation at this point.
Classical Field Configurations Under this transformation n changes by 5n Now.582 CHAPTER 22.e. 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. In fact. whereas 7ri(50(3)) = Z2 (i. 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. . 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. Hence i /"27r ^ i Thus we have shown explicitly that the winding number remains unchanged under continuous deformations. 1 1 See N. Thus there is only the trivial map into a point. il**( ^') r2Tr A ( 21 9 2 1) ' de\x2 *(«/W). only one homotopy class. the identity class (the group space S2 is simply connected. Steenrod [261].e. 15. an integer modulo 2) i. the cyclic group with two elements (since 50(3) is doubly connected). It is intuitively clear that any circle drawn on the surface of a sphere can be shrunk to a point.e. there is a theorem which says thatH TTg(Sn) = 0 for q < n.8. Sec. i.
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.2 Instantons and AntiInstantons A onedimensional field theory is a quantum mechanical problem.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. We consider such a theory now with action S((p) depending on the real scalar field 583 . 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. 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. Since we have not yet introduced periodic (i. different from literature. 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.e. 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!). 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.1 Introductory Remarks In Chapter 18 we considered anharmonic oscillators and calculated their eigenenergies. finite Euclidean time) classical configurations (which are nontopological). 23.
(23.1 — degenerate minima (positions of classical stability or "perturbation theory vacua") at positions 4>{t) = ±m = ± 0 O . 2mo has as the normalized ground state wave function </>o(?) = <<?0) Thus for mo = 1 we have in our notation here a. 23. n = 0.584 <)>{t). 23. Chapter 6) 1 2 . in particular E.m ^ + §^24 ' ^2 > a (23J) The potential V(</>) has — cf. Gildener and A. Fig.e. Path Integrals and Instantons S(<f>) / dt \i>2 .v{4>) <?VV m4 __ 1 . note that here the mass of the equivalent classical particle is taken to be mo = 1.2.1. and ^2(0) : m* . Patrascioiu [110] and Chapter 18.1 The double well'potential. ! " ^ j = ^ 2 . "For comparison with literature.1)) and hence E0 = mh/y/2. i. The Hamiltonian of the simple harmonic oscillator (cf. 2 2 4 ™ = ~.* /•tf CHAPTER 23.2 = 2m 2 (here m is the parameter in the potential V{4>) of Eq.2) Fig. (23.
i. as we did in Chapter 18 — but as we saw. </. 23. Fig. must have definite parity as in a physical > situation.23. These boundary conditions imply nonperturbative contributions to the eigenvalues which yield the splitting of the asymptotically degenerate oscillator . 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> ~ * ±</>.e.2 — and we can deduce from this the boundary conditions. we have the Lagrangian L(x. e. and in the above 4> corresponds to x. but there is tunneling between the two minima if the central hump is not infinitely high. The states of the system here must be even or odd under x — ±x. The corresponding wave functions ip±(x) then must be peaked at the extrema of stability — as in the examples illustrated in Fig. This tunneling affects the eigenvalues. W(0) = 0.2 Instantons and AntiInstantons 585 Classically the position <f> = 0 is a point of instability. 23. this alone does not yield the level splitting.g._(0) = 0.In mechanics of a particle of mass 1 with ~ position coordinate x(t).x) = x2 V(x). We know from the shape of the potential that the quantum mechanical energy spectrum is entirely discrete (no scattering). We can develop a perturbation theory around either of the minima.2 The wave functions of the lowest states.
i. that given by Eq.e. (23. d(f> d ' dr2 dr It follows that the equation of motion is now given by 55. classically forbidden domains then become "classically" accessible. dr We observe that Eq.7) resembles a Newton equation of motion with reversed sign of the potential as a result of our passage to Euclidean time. (23.8) V(4>) + const.2 V'(<l>) = . SE.5) and iS = dLE d dr[d(d(f)/dT)} i.0o. n) = f with the Euclidean action rTf SE(<i>) = / J TV f V{<f>}exp[SE(<P)/h} (23.£j — —oo (we assume the normalization factor or metric to > > be handled as explained in Chapter 21).4) J <bi =—(An rTf (ITLE = I JT. (23.2 m 2 ( l m z (23. The transition amplitude for this.586 CHAPTER 23. i.9) . is finite if we go to imaginary. For small h we can attempt a stationary phase method. r .3). Equation (23.<h. In the semiclassical path integral procedure that we want to use here. we consider the amplitude for transitions between the vacua or minima at 0/^ — ±</>0 = ±rn/g over a large period of time.U) = J'D{<f>}exp[iS(<l>)/h] (23. (23.3) for tf — oo. dLE = 0.£'(0) = 0.e.7) so that (23. i. We are therefore interested in the amplitude — called Feynman amplitude or kernel as we learned in Chapter 21 — (</>o.7. But quantum mechanically it can tunnel through the hump from one well to the other.  . Path Integrals and Instantons approximations into an even ground state and an odd first excited state. dr m+™ (23. A classical particle starting from rest (0 = 0) at —m/g cannot overcome the central hump and move to +m/g.7) is to be solved for 0 with the boundary conditions m 0(r) = — for Tf — oo and 0(r) = > m for n oo.2.e.tf\ . Then (0o.e. Euclidean time T = it.6) d*l dr 2 .
The solution (J>C(T) can also be looked at as describing the motion of a Newtonian particle of mass 1 in the potential —V (</>)• This is a very useful description. — (23. in the present case (T — To) = m tanh m(r — TO).23.18). .2 Instantons and AntiInstantons 587 In fact. i. is depicted in Fig.10) where TO is an integration constant.3. In the present context of Euclidean time the configuration (f)c is generally described as an "instanton". the classical path. from fa — —rn/g.e. and hence kinetic energy gained is equal to potential energy lost. Thus we know the solution from there which is the kink solution (22. 23.e. Its trajectory.3 The instanton path. 23. timeindependent) in the 1 + 1 dimensional <J>4theory which we considered previously. and having just enough energy to reach the other extremum at 4>f = m/g where it will again be at rest. Thus we can picture the particle as starting from rest at one minimum of V(</>) or maximum of —V(<p). we see that the equation is the same as the classical equation for a static soliton (i. If the constant arising in the integration of the Newtonlike equation is chosen to be zero (see below). Fig.e. i. this means the total energy of the particle in Euclidean motion is zero.
(23. the constant must be zero. thus indicating the nonperturbative nature of the expression.11) For this quantity to be finite for TJ = —oo.7) admits another configuration which is the negative of (23.13) into Eq.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. For a nonzero constant we obtain the periodic instantons discussed in Chapters 25 and 26.13).]. (23.* With this choice and inserting mo = j dr J = m we obtain* SE = m T lm4 2 g2 cosh4 m{r . Also note that we can reexpress SE in the form 2 SE f0 c (oo) T Joo \ dr j J0 C ( where ^iw = .Tj) oc exp 4 m3 m^2 (23.r 0 ) ' (23. 23. Using Eq. Tf = oo.5).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.^ . we obtain lim f+ OO. We also observe that the expression is singular for g2 —> 0. Path Integrals and Instantons The Euclidean action of the instanton solution (23. (23.10) is given by (23. We also need to know how we can calculate the difference AE of the energies of the two lowest levels from this expression. but with the same Euclidean action (23.4).13).' .12) dimr) cosh m(T — TO) m tanh x tanh x 00 _ 4m*_ (23.7"/! ).TJ T fro. The configuration (23.588 CHAPTER 23. Inserting the result (23. This antiinstanton configuration rises monotonically from right to left as shown in Fig.10) and is called the antikink or antiinstanton configuration and obviously (following the arguments of Chapter 22) carries topological charge — 1.4.14) But this is not yet enough.8) we obtain T f SE dr[2V((j)c) + const. 'Note that this integral may be evaluated for finite limits r = ± T and presents no problem in the limit T oo. The classical equation (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. Questions which arise at this point are: Is the nonmonotonic configuration (23. from the sketch or from the variation of SE.5.15b) has the shape shown in Fig. #(r) = </>C(T . (23.TQ are very far apart their sum differs from an exact solution only by an exponentially small quantity. The sum (23.TO) + ^ c ( r . Clearly we are here interested in configurations of type (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.g. Both configurations comunicate between the two wells of the potential at < = —m/g and (ft = m/g.g. b u t if TQ.15b) a mending together of an instanton and an antiinstanton at r = T\ with —oo = To < T\ < T2 — 00 for TQ.15b) is not exactly a solution. e. as we can see e. 0c(r) = &(T).15a) we have a simple superposition.10) we can ask ourselves: Does it make sense to consider the sum of an instanton localized at TQ as indicated in Fig. and an antiinstanton localized at (say) TQ.5. 23.Writing ••• + L ••• and performing the variation in the usual way with 5<j) = 0 at r = ± 0 0 .15b) since we require configurations from vacuum to vacuum. 23. 23. in the case of (23.23.15b) a solution of the classical equation? Is it a topological configuration — if so. (23.2 Instantons and AntiInstantons •<l> 589 ^ ^ ^ 0 ^ ^ ^ _ _ ^ Fig.r0).TO far apart and T'Q 3> TQ.15a) or # ( r ) = 0(rro)^c(rro)0(rir)e(TTi)^c(r^)0(T2r). so correspondingly one defines the instanton or kink with topological charge + 1 and the antiinstanton or antikink with topological charge — 1.15b) In the case of (23. / > Looking at the instanton (23.4 The antiinstanton.
since (23. at oo the variation is <<> = 0). g cosh m(r — TQ) (23. m/g / o / o ^ 0 1 \ c'\ o\ T2 ' ^^__ m/g Fig. 23.15a) is a configuration which starts from 4> = —i^/g and ends there. Ignoring exponentially small contributions the Euclidean action of the configuration (23. In fact.15a) is not exactly a classical solution.e x p { .5 Superposition of an instanton and a widely separated antiinstanton.16) .T 0 ) } ] . it also does not exactly stationarize the Euclidean action. we see that in the domain [—oo.r 0 ) At r = Ti the variation Sep is finite but nonzero so that we are left with an exponentially small contribution which.2 m ( T o . of course.15a) can be argued to be zero. vanishes as T\ —> ±oo (of course. 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 . T\] with LECIT = 0.CHAPTER 23. 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. Path Integrals and Instantons '. #(r) dr 1 1 m 9 cosh m(r — ro) cosh m{r — T'Q)_ cosh2 m(r — TQ m cosh2 m(r — TQ) g cosh m{r — ro) m 1 [1 . 5^ In order to answer the second question we observe that (23. Finally.
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).
T')dT' (23. We observe that if we discretize Eq. the kernels to the left and to the right would be associated with step functions 8(T — r') and 9(T' — T ) . 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.lnJV(T)}] N(T) N(T) (23. = 1 52 K(n. In the limit S —* 0 .64) Inserting the explicit expression for the kernel given in Eq.64) as above and consider the determinant of the expression. r i ) + .66) •Dr) .Ti) i=l : exp \ £ K(T.n) o K(TJ. so that (with T *T) 1/2 •Dz N(T) (23. ^This is the method explained by B. T.63b).23. r ' ) is to be evaluated at r = T ' . (23. are summed over all values from 1 to n.a { l n i V ( r ) . . 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.64) D (23.TI) K(T2. .63a).n) K(Tj. 192. Martin [43]. 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. (23.TI) 0 K(T2. Thus the factor a inserted above has to be taken as 1/2..T2) I exp I K(n. the value of the determinant is half (or arithmetic mean value) along the diagonal.64). where a f 1 See R.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. the kernel K(r. (23.TJ) The factorials arise since every determinant of I rows and I columns appears l\ times when i.67) i=l v ) = lim exp V In ( 1 + ' L K(T. Cameron and W. 2! . whereas the determinant arises only once in the original determinant Dn. n — oo the expansion becomes > D lim Dn = 1 + n—*oo J K(r1.T1)dT1 + J dn J dr2. X(TI.n) i=l N lim exp £*(T.65) The quantity D is really the Fredholm determinant which is obtained as above by solving the Fredholm integral equation with constant integration limits.62) — and with insertion of a discretization dependent factor a still to be explained — we have in the present case D: .n) z=l x (23. H. as in Eq..e.7 Y. Our final step is to relate the VolterraFredholm determinant to the required functional determinant.. (23. n )d? (23..j. we obtain exactly the nonvanishing terms on the right of Eq.N{T) Finally we add the following remarks concerning the discretization (23. Felsager [91]. Thus D„ = l + 5 ^ K ( r i . This determinant can be expanded in powers of K (more precisely in powers of a parameter A which we attach to K).68) D above. [44]. p.exp • /_:T JV(T')" dr' = e x p [ . Thus in our case one of the kernels is zero and we have to decide what we do when.
59) the endpoint conditions »)(±T) = 0 and evaluate the integral.73) . With partial integration (in the second step) we obtain (since (cf. and hence may be ignored. In = dzi dz2 • • • dzn e x p . Briefly. In every integration / dzi the contribution €7. (23. so that (with V^N(T)/N(n)) 7t = J V{z}^[JT_Tdr[\[z(r)+ia^)' — lim n—*oo. rn+i = 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).72) The endpoint integration dz(T) = dz(rn+i) can be combined with this.Z i .J dz(T) J T>{z} J da•Dz hJ dz{T) v{z} I f_A\^+W)i{T)} J H § I"exp [ .2o) 2 rc^O.(T) + /_ dr — .71) Hence (with h = 1) l = j .59)) + f(T)] = — f dz(T) f°° dae^W+f™. T + N(T) / T' = T JT dr' N(T') .l + £7t) «i (23. .0.( .69) Incorporating this condition into the functional integral Jo with a delta function 1 = / dz(T)S[z(T) we have (cf. I0 = — f dz(T) f 1>{z} f da T'Z x exp T>r\ «/>5^> + ix(*<T> + '"< T >/• T *>m™ 7V2(T') (23. represents a constant translation.N(T) . (23. Eq.Z ( T ).e—*0 In.57)) z(—T) —> 0) ^)/>'£^') = iv(r)/_T/T'A(__l_)2(T0 T JV(T') T . n J=.22)) In = (7T6)"/2 Vn + 1 exp e(n+l) ( Z n + l . N(T') . (23. Path Integrals and Instantons Example 23.— . f(T) = N{T) J^ dr'^^z(r') (23.3: Implementation of the endpoint constraints Using a Lagrange multiplier a insert into the functional integral (23. t h e range of T is subdivided into n + 1 equal elements of length e with (n + l)e = 2T and TO = —T. I2 and using induction one obtains (cf. (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. Thus 0 = 2 (rf) + f(rf) or 0 = z{T) + f(T). Solution: From Eq. Evaluating Ii. (23.608 CHAPTER 23.70) and this now requires also integration over the endpoint coordinate z(T). (21. Eq.
/ d£o <9r0 5[F(TQ. J co § dr Recall that / f ^ doce^ = yfVfp. . 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.12). Ignoring the remaining metric factor with the reasoning explained > between Eqs. N.^ 23. a = const. 11 L.3 T h e Faddeev—Popov constraint insertion The FaddeevPopov constraint insertion implies the complete removal of the zero mode.10) and (21. ^E. D.4. Faddeev and V.74) where we reinserted h. Eq. i. dxv 0(r) = (f>c(r) + V(r) and set 1 TOO a — const.. (3. £n = ~2 dTT](T)7]n(l (23.75) 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. This result verifies Eq. 1/2 T W 2 (r)J (23.11). Patrascioiu [110]." The method consists essentially in replacing the single integration J d£o by a double integration involving a delta function. j / dTT]n(T)r]m(l r .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. / — / / We started off with V(T) = ^CnVniT). Popov [89].£Q) — aTo]dTod£o. (23. Gildener and A.76) Since the zero mode T]O(T) is proportional to d(f>c{r)/dT with — SE. (23.e. (21.59) and agrees with that in the literature.23.
.76) we set 4>c{r) = ^CnVnir).78) <t>ij) = Y.J drm(r)Mr + T0) (23. We now impose (i. e. in the direction of the zero mode. demand validity of) the constraint £o = 0.75) requires.^n{r). this constraint implies that we go to the subspace which is orthogonal to the zero mode. In analogy to (23. We establish the function ^(To>£o) with the following reasoning. / dTm{r)n{T).83) **The collective coordinate in the present context is also called "instanton time".77) The set {r/ n (r)} constitutes a basis of the fluctuation space about the classical configuration. i /*oo (23. The number £o is the component of the fluctuation rj(r). i.81) 2 / dr»7o(r)0c(r) = .79) (23. In the integral of Eq.^O) with dependence on ro which the replacement (23. i. if the set {Vn(T~)} spans the Hilbert space of fluctuation states about 0C (at collective coordinate position r ) . Then cn = ~2 / dr(pc(T)rjn(T).610 this means that CHAPTER 23. (23. (23. The delta function 8[F(TQ.£o)/a.g. Path Integrals and Instantons _^ ^ w . (23. in the case of the soliton of the static 1 + 1 dimensional theory this is the kink coordinate.80) i r or •i /*oo dr7?o(r)0(r) = 0. where L = in + cn. in the direction T]O(T) at the point TQ of the parameter r which parametrizes the trajectory of the classical configuration </>c(r).e.82) This condition says that (j)(r) is not to possess any component in the direction of T)O(T). (23. Thus. A position ro of this classical configuration is called the collective coordinate** of (the field) 4>{T). We consider the following integral / ( r 0 ) := .81) the rdependence is integrated out and hence the integral does not provide the function F(TO. at r = ro. (23. (23.£Q) — aro] contained in Eq.75) enforces the collective coordinate ro to be given by F(ro.e.e.
(23. Felsager [91].23..82) we can replace here 7?O(T)0C(T) by 77O(T)T7(T).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 ) .£O) F(r0^0) we have defined as + ro) + T?(T)] = / ( T 0 ) + p£o.88) ro0 Then in Eq.£O) OTQ = = £ " .4 Field Fluctuations 611 for an infinitesimal translational shift TO of the position of 4>c(r). . We have AFP = /F(T0. 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) . jdrrj0(T)4>c(r) + JdTm(r)(p'c(T) X/SET0.75) 1 = AFP f dT06[{F{rQ. so t h a t with F(TQ.Zo)y/SETQ We now define A FP '• =0 • (23. (23.87) l/M) (23. (23.91) " F o r more discussion of this point and related aspects see also B.<t>c{r)} .y/SET0 = 0.90) Now tt 0^C(T + TQ) ro=0 <9T0 dr (23. p. 158.y/S^ro] (23. We evaluate this quantity as follows in the leading approximation.86) = Jdrrjoir^Mr F(TO./ drri0(T)<pc(T) + PJ (23.(23.85) PJ W i t h Eq.89) Here the factor App is called the FaddeevPopov determinant. .84) (23.^) .
0AFP8 4o H P .e. i.91) into Eq. iytQ 0C(TO + dr0)  4>C{TQ) = dT0^'c(T0) = dr0 7 0 (r 0 ). 0 d r (23. 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.93).e.e. if ^ + d£o = 0. 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. Inserting (23. we obtain (remembering that % = pd(j>c/dT/y/S^) AFP = 4 = f dr ( ^ V \/!5E JT \ dr * = yfe.77) between the zero mode d(^ c (r)/dr and the fluctuation vector in its direction. Path Integrals and Instantons Thus a shift TQ in the direction of the zero mode (i.93) where AT = TT — 00. ~ There is a quick way to obtain the result (23. i.90). 7 Thus there is no fluctuation along the direction of the zero mode in the sum. P J AT (23. i. =~ (23. tangential to (J)C(T)) leaves this and hence the square of this quantity. in d(j)c(To) + dr/fa). 770(7")./ dT0AFP. the "static energy" invariant. (23.92) 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. Consider.94) ' ' T h i s is an important result which will be required later in the explicit evaluation of path integrals. P the relative sign being of no significance.75) now becomes d£0AFP8[f(T0) = = / dr0 / J&T J + p£0 .612 CHAPTER 23. using the relation (23.e. fs d(/)c{ro) = We have also drj(T0) = d£0r]o(To) + ^2d£ir)i{To). /(TO) ^/SE~T0} V~SET0 c d£. if the coefficient of 770(70) vanishes. .
we can set I 2TT ^o = l'o\ wlp2'' which with Eq.98) . We determine I'Q by observing t h a t this differs from IQ of Eq.4 Field 23.96) lim .60a) for I0 and Eq.96) does not yet allow a comparison with (23. (23. (23.35a) since the dependence on Tf — Ti = 2T of both expressions differs significantly.55b). We have to take into account also multiinstanton configurations which almost stationarize the action integral and satisfy the same boundary conditions as the instanton.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.4 Fluctuations T h e single instanton contribution 613 We now return to the multiple integral (23. The source of lack of similarity is attributed to the fact t h a t the contribution of a single instanton to the p a t h integral is not enough. We can now perform the Gaussian integrations and write the result lim T^oo p 4 m3 exp 3^J (23.4. (23.23.52a) and replace in this J d£o by the expression (23. We observe t h a t the result (23.93).43)).97) is a constant still to be evaluated.55b) implies I'o w0p m \ ' LOQP '2K irh) m \ ' irh) 4:\/6mp mAT (23. so t h a t h = det r)/(n\ppATexp 4 m3' ATAFPdet(^ \ a£ jexp (23. 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. Since w\ is for finite T given by WQ of Eq.60a) in not involving the zero mode contributions. (23.A T A F P i " o e x p T—>oo p where I'o n^O. (23.
i.* Finally we rewrite the expression (23. Patrascioiu [110]. 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 .101) 2T > oo (cf.5 I n s t a n t o n . 0c M. (3.100) in such a way that it exhibits the crucial factors of which it is composed.e.2 m T e x p 2 irgh ' T»oo (23. we drag the factor p2 along for ease of comparison with literature. Gildener and A.55b)) w0 = v0e VQ 23. (23.4. Patrascioiu [110] make the replacements \QP <> 2g 2 and p?GP <» 2m 2 . Although we choose p2 = 1.TiMiy+'Mr «>.100) We observe that the dependence on the normalization constant p cancels out.614 CHAPTER 23. Gildener and A. .32).20). 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. the approximation is then known as the "dilute gas approximation". the result becomes h lim 2 m r ^ ^ 1 e . in the present case we also have to include multiinstanton configurations in the form of one instanton plus any number of instantonantiinstanton pairs which describe back and forth tunneling between the two wells. ^See E.4>C exp h (23. Thus — the subscript 1 indicating that it is the Feynman amplitude for the oneinstanton contribution — lim where with A T ATAFPI0 W0 SE{. (23. their Eq.102) *For comparison with the parameters of E.a n t i .99) limlATAFF/0f^^eA^exp 4 m3 3 / ^ Inserting the explicit expressions for IQ and App.r). In view of their normalization of the normal modes. ~ rKo')e{Tx . *For the validity of the calculation the instantons or kinks have to be far apart. This means we are concerned with configurations like* 2n+l d(r . they have to introduce the inverse of that factor on the far right of their expression (3.
17b) and there perform first the integrations over the intermediate configurations whose endpoints are not fixed. T2 = T/3. T3 = T) d(f>l((pf.102) distinguishes between instanton (rising from left to right) and antiinstanton (rising from right to left).4>i.Ti) = X^/' T /I&> r i> ( 2 n+1) n=0 (23.T2}((t)2. 7 i = . the total Feynman amplitude is given by (^/. this means we have to consider the amplitude (23.<t>f) + 2 d</>{ SE{4>fi4>i) + ld2SE 2 d(/){ ( h .23.l ) i + 1 in the sum of Eq.<t>ff + h .<j>i) .<t>2)eSE{<t>2.Tl\(j)i.(r) = ^ 2 n + 1 ) ( r ) + 77(r) with r ^ ) « 0. . (23. Since these endpoint configurations.2 T / 3 .r / 3 .<j>i) + 2 d<& 1 r)2 *? k? + (<f>2 .r/l&. The factor ( . (23. 23.T2\(f)l. Taking these configurations into account. 1 0 6 a ) and SE{(J>I. .103) For instance for n = 1 one has the configuration <jyc (T) which has the form shown in Fig.fo) = Id2 St SE(<Pf.105) where we write 'proportional' since only the classical instanton action depends on the endpoint configurations.4>l)eSE{4>l. (23. (n = 0 implies the single instanton solution). r0(3) = 2T/3.Ti} OO J — OO /OO OO OC OO d(f>2 J — OO LeSE{<l>f. are varied. Consider (with T 0 = T. 2 .(l>i) (23.Tf\(p2.6. (3) (23. Proceeding as before we now consider first in analogy with Eq.r 0 ( 1 ) = .36) (/. r 0 ( 2 ) = 0. we can expand the Euclidean actions here involved as follows (the first derivative vanishing at a solution): SE{<j>f. here (f>2.H)2 + .104) Considering the case n — 1 of a kink with one additional kinkantikink pair.4 Field Fluctuations 615 for n = 1.Tl}{4>l. .106b) SE{<t>l.4>I) ld2SE SE(c/>i.<Pi)2 + • • • . ( 2 3 .
e.105) can in other respects be treated in analogy with the single instanton amplitude. The allowed positions of the individual kinks have to be such that their order is maintained. On the assumption that the kinks of the instantons are widely separated.e.e.4>i) / d(f>2 e x p {<h . It follows that — always within our approximations — c (^TflfaTjW SE{<t>f. of course. With our remark after Eq. It has to be remembered.4>i) { 2 " 2d SE . ±m/g (23.(01 .106a).4>f) d4>\ 4>s72 (23. i. i. (23.e.